CN118798641A

CN118798641A - Neural network-based operating status evaluation method and device, and electronic equipment

Info

Publication number: CN118798641A
Application number: CN202410894753.5A
Authority: CN
Inventors: 王钊; 曾爽; 张宝群; 董毅; 丁屹峰; 康琦; 王立永; 赵永良; 王雅群; 王超; 赵宇彤; 马麟; 刘畅; 张恒; 梁安琪
Original assignee: State Grid Beijing Electric Power Co Ltd; State Grid Corp of China SGCC
Current assignee: State Grid Beijing Electric Power Co Ltd; State Grid Corp of China SGCC
Priority date: 2024-07-04
Filing date: 2024-07-04
Publication date: 2024-10-18

Abstract

The invention discloses an operation state evaluation method and device based on a neural network and electronic equipment, and relates to the field of smart grids, wherein the evaluation method comprises the following steps: acquiring a state index data set from a distributed photovoltaic grid-connected system based on a preset grid-connected safe operation index system; preprocessing state index data in the state index data set to obtain a preprocessed state index data set; constructing a state index vector based on the preprocessed state index data set, and acquiring a weight vector of the state index; inputting the state index vector and the weight vector into a neural network model, and outputting the risk level of the distributed photovoltaic grid-connected system; and generating an operation state evaluation result of the distributed photovoltaic grid-connected system based on the risk level. The invention solves the technical problems of lower accuracy of the evaluation result in the related art based on the mode of evaluating the operation state of the photovoltaic system by a mathematical method.

Description

Operation state evaluation method and device based on neural network and electronic equipment

Technical Field

The invention relates to the field of smart grids, in particular to a neural network-based running state evaluation method and device and electronic equipment.

Background

With the rapid development and application of photovoltaic power generation technology, distributed photovoltaic grid-connected power generation has become an important power generation mode, and the distributed photovoltaic grid-connected system has the advantages of being high in flexibility, high in reliability, environment-friendly and the like, and a plurality of photovoltaic power generation units are distributed at different positions and run in a grid-connected mode. However, the operation characteristics of the distributed photovoltaic grid-connected system are affected by various factors, such as weather changes, photovoltaic module quality, system configuration, and the like, so that evaluation and research on the operation characteristics are required. After the distributed photovoltaic grid connection, the power flow distribution of the distribution system can be changed by injecting electric energy into the distribution system, so that the voltage of a line load node is increased, and voltage deviation is caused. Because distributed photovoltaic can also be influenced by external environment and scheduling requirements, the power output of the distributed photovoltaic is unstable, and the problems of node voltage fluctuation, voltage instantaneous drop, voltage flicker, harmonic distortion, three-phase imbalance and the like can be caused. The operation characteristic evaluation research of the distributed photovoltaic grid-connected system can help optimize the system design, improve the system efficiency and reduce the system cost, thereby promoting the development and application of the distributed photovoltaic grid-connected technology.

In the related art, a mathematical method is often adopted for the operation state evaluation based on the neural network, the mathematical method comprises a fuzzy evaluation method, an approximate ideal solution ordering method, a gray correlation analysis method, a rank sum ratio method and the like, wherein the fuzzy state evaluation method is an evaluation method based on a membership theory, the operation state is quantized by using a membership function, the uncertainty of the membership is considered, the accuracy is processed, and the influence of a randomness factor is ignored. The approach to ideal solution ordering method shows the overall similarity between the state to be evaluated and the ideal state in a function curve mode, however, the method cannot well reflect the correlation between the change trend of each index and the overall state. The gray correlation analysis method judges the running state of the system by comparing the correlation degree of each index of the system and the development trend of the ideal system, can improve the error generated by the asymmetry of information, well illustrates the relation between each index variation trend and the whole state, however, because the subjectivity of the optimal value is stronger, certain defects exist on the whole judgment of the system scheme; the rank sum ratio method combines the parameter and non-parameter evaluation methods together, the calculated rank sum ratio is in the range of 0 to 1 no matter aiming at relative indexes or absolute indexes, the method is simple and easy to operate, the result is clear, the method is not limited by the distribution characteristics of index data, and the comprehensive performance of the photovoltaic power station is evaluated by comparing the sizes of the rank sum ratio. The mathematical method uses clear language description reasoning process, does not need a large amount of historical data, but is excessively dependent on subjective experience of an operation and maintenance expert to evaluate, so that the evaluation result has lower accuracy.

In view of the above problems, no effective solution has been proposed at present.

Disclosure of Invention

The embodiment of the invention provides a neural network-based operation state evaluation method and device and electronic equipment, which are used for at least solving the technical problem of lower accuracy of an evaluation result in a mode of evaluating the operation state of a photovoltaic system based on a mathematical method in the related technology.

According to an aspect of the embodiment of the present invention, there is provided an operation state evaluation method based on a neural network, including: acquiring a state index data set from a distributed photovoltaic grid-connected system based on a preset grid-connected safe operation index system; preprocessing state index data in the state index data set to obtain a preprocessed state index data set; constructing a state index vector based on the preprocessed state index data set, and acquiring a weight vector of the state index; inputting the state index vector and the weight vector into a neural network model, and outputting the risk level of the distributed photovoltaic grid-connected system, wherein the neural network model is a model which is obtained through pre-training and is used for evaluating the running risk of the distributed photovoltaic grid-connected system, and the neural network model at least comprises a self-attention network, a time domain convolution network and a long-term and short-term memory network; and generating an operation state evaluation result of the distributed photovoltaic grid-connected system based on the risk level.

Optionally, inputting the state index vector and the weight vector into a neural network model, and outputting the risk level of the distributed photovoltaic grid-connected system includes: inputting the state index vector and the weight vector to the neural network model based on an input layer of the neural network model; outputting a weight feature vector based on a self-attention network of the neural network model; taking the weight feature vector as input data of a time domain convolution network of the neural network model, and outputting a time sequence feature vector based on the time domain convolution network; and taking the time sequence feature vector as input data of a long-short-period memory network of the neural network model, and outputting the risk level of the distributed photovoltaic grid-connected system based on the long-short-period memory network.

Optionally, before collecting the status index data set from the distributed photovoltaic grid-connected system based on the preset grid-connected safe operation index system, the method further comprises: carrying out importance assignment on the state indexes in the grid-connected safe operation index system to obtain a relative importance value between every two state indexes; constructing a judgment matrix based on the relative importance values; calculating the eigenvalue and eigenvector of the judgment matrix; and carrying out consistency verification based on the characteristic values to obtain a verification result, and taking the characteristic vector as a weight vector of the state index under the condition that the verification result indicates that the judgment matrix passes verification.

Optionally, before collecting the status index data set from the distributed photovoltaic grid-connected system based on the preset grid-connected safe operation index system, the method further comprises: selecting an electric energy quality index for evaluating the running state of the distributed photovoltaic grid-connected system, wherein the electric energy quality index comprises at least one of the following components: voltage deviation, voltage flicker, voltage fluctuation, three-phase imbalance, frequency deviation, and harmonic content; selecting an electric energy index for evaluating the running state of the distributed photovoltaic grid-connected system, wherein the electric energy index comprises at least one of the following: power factor, power generation efficiency, active power change rate; selecting an environmental index for evaluating the running state of the distributed photovoltaic grid-connected system, wherein the environmental index comprises at least one of the following: ambient temperature, illumination intensity; and constructing the grid-connected safe operation index system based on the electric energy quality index, the electric energy index and the environment index.

Optionally, the step of preprocessing the state index data in the state index data set includes: traversing the state index data, and deleting abnormal values in the state index data; and establishing an interpolation function, and complementing the missing value in the state index data.

Optionally, the neural network model is pre-trained, and the step of training the neural network model includes: collecting N simulation state index data sets based on a simulation model of the distributed photovoltaic grid-connected system, wherein N is a positive integer; configuring an operation state label for each simulation state index data set; constructing a sample set based on the simulation state index data set and the running state label, and dividing the sample set based on a preset dividing proportion to obtain a training set and a testing set; setting up an initial neural network model; performing iterative training on the initial neural network model based on the training set to obtain a trained neural network model; and testing the trained neural network model based on the test set to obtain a test result, and obtaining the trained neural network model under the condition that the test result indicates that the neural network model passes the test.

Optionally, the step of collecting N simulation state index data sets based on the simulation model of the distributed photovoltaic grid-connected system includes: building a simulation model of the distributed photovoltaic grid-connected system; simulating M operation states in the simulation model of the distributed photovoltaic grid-connected system based on state indexes in the grid-connected safe operation index system, wherein M is a positive integer; and collecting a plurality of simulation state index data sets of the simulation model of the distributed photovoltaic grid-connected system under various running states to obtain N simulation state index data sets.

According to another aspect of the embodiment of the present invention, there is also provided an operation state evaluation device based on a neural network, including: the system comprises an acquisition unit, a control unit and a control unit, wherein the acquisition unit is used for acquiring a state index data set from a distributed photovoltaic grid-connected system based on a preset grid-connected safe operation index system; the processing unit is used for preprocessing the state index data in the state index data set to obtain a preprocessed state index data set; the construction unit is used for constructing a state index vector based on the preprocessed state index data set and acquiring a weight vector of the state index; the output unit is used for inputting the state index vector and the weight vector into a neural network model and outputting the risk level of the distributed photovoltaic grid-connected system, wherein the neural network model is a model which is obtained through training in advance and used for evaluating the running risk of the distributed photovoltaic grid-connected system, and the neural network model at least comprises a self-attention network, a time domain convolution network and a long-period memory network; and the generation unit is used for generating an operation state evaluation result of the distributed photovoltaic grid-connected system based on the risk level.

Optionally, the output unit includes: a first input module for inputting the state index vector and the weight vector to the neural network model based on an input layer of the neural network model; the first output module is used for outputting a weight characteristic vector based on the self-attention network of the neural network model; the second output module is used for taking the weight feature vector as input data of a time domain convolution network of the neural network model and outputting a time sequence feature vector based on the time domain convolution network; and the third output module is used for taking the time sequence feature vector as input data of a long-period memory network of the neural network model and outputting the risk level of the distributed photovoltaic grid-connected system based on the long-period memory network.

Optionally, the operation state evaluation device based on the neural network further includes: the first assignment module is used for carrying out importance assignment on the state indexes in the grid-connected safe operation index system to obtain relative importance values between every two state indexes; the first construction module is used for constructing a judgment matrix based on the relative importance values; the first calculation module is used for calculating the eigenvalue and eigenvector of the judgment matrix; and the first verification module is used for carrying out consistency verification based on the characteristic value to obtain a verification result, and taking the characteristic vector as a weight vector of the state index under the condition that the verification result indicates that the judgment matrix passes verification.

Optionally, the operation state evaluation device based on the neural network further includes: the first selecting module is used for selecting an electric energy quality index for evaluating the running state of the distributed photovoltaic grid-connected system, wherein the electric energy quality index comprises at least one of the following: voltage deviation, voltage flicker, voltage fluctuation, three-phase imbalance, frequency deviation, and harmonic content; the second selecting module is used for selecting an electric energy index for evaluating the running state of the distributed photovoltaic grid-connected system, wherein the electric energy index comprises at least one of the following components: power factor, power generation efficiency, active power change rate; the third selecting module is used for selecting an environmental index for evaluating the running state of the distributed photovoltaic grid-connected system, wherein the environmental index comprises at least one of the following: ambient temperature, illumination intensity; the second construction module is used for constructing the grid-connected safe operation index system based on the electric energy quality index, the electric energy index and the environment index.

Optionally, the processing unit includes: the first deleting module is used for traversing the state index data and deleting abnormal values in the state index data; and the first complementing module is used for establishing an interpolation function and complementing the missing value in the state index data.

Optionally, the operation state evaluation device based on the neural network further includes: the first acquisition module is used for acquiring N simulation state index data sets based on a simulation model of the distributed photovoltaic grid-connected system, wherein N is a positive integer; the first configuration module is used for configuring an operation state label for each simulation state index data set; the third construction module is used for constructing a sample set based on the simulation state index data set and the running state label, and dividing the sample set based on a preset dividing proportion to obtain a training set and a testing set; the first building module is used for building an initial neural network model; the first training module is used for carrying out iterative training on the initial neural network model based on the training set to obtain a trained neural network model; the first test module is used for testing the trained neural network model based on the test set to obtain a test result, and obtaining the trained neural network model under the condition that the test result indicates that the neural network model passes the test.

Optionally, the first acquisition module includes: the first building sub-module is used for building the simulation model of the distributed photovoltaic grid-connected system; the first simulation sub-module is used for simulating M running states in the distributed photovoltaic grid-connected system simulation model based on the state indexes in the grid-connected safe running index system, wherein M is a positive integer; the first acquisition sub-module is used for acquiring a plurality of simulation state index data sets of the simulation model of the distributed photovoltaic grid-connected system under various running states to obtain N simulation state index data sets.

According to another aspect of the embodiment of the present invention, there is further provided an electronic device, where the computer readable storage medium includes a stored computer program, and when the computer program runs, the device where the computer readable storage medium is controlled to execute any one of the operation state evaluation methods based on the neural network.

According to another aspect of the embodiments of the present invention, there is also provided a computer program product including one or more processors and a memory for storing one or more programs, wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to implement any of the above-described neural network based operation state evaluation methods.

In the present application, the method comprises the following steps: the method comprises the steps of firstly collecting a state index data set from a distributed photovoltaic grid-connected system based on a preset grid-connected safe operation index system, preprocessing state index data in the state index data set to obtain a preprocessed state index data set, then constructing a state index vector based on the preprocessed state index data set, acquiring a weight vector of the state index, inputting the state index vector and the weight vector into a neural network model, and outputting a risk grade of the distributed photovoltaic grid-connected system, wherein the neural network model is a model which is obtained in advance and used for evaluating the operation risk of the distributed photovoltaic grid-connected system, and is composed of a self-attention network, a time domain convolution network and a long-short-term memory network, and finally generating an operation state evaluation result of the distributed photovoltaic grid-connected system based on the risk grade.

According to the application, multi-dimensional index data are acquired from the distributed photovoltaic grid-connected system to be evaluated based on a grid-connected safe operation index system, a multi-dimensional evaluation standard is provided for state analysis, then the state index data are analyzed based on index vectors and a pre-trained neural network model, the risk level of the distributed photovoltaic grid-connected system is acquired, the running state of the distributed photovoltaic grid-connected system is further obtained, the characteristics in the data can be deeply mined based on the analysis of the multi-layer neural network in the neural network model, and the importance degree of each characteristic is marked based on the weight value configured for each characteristic, so that the accuracy of state evaluation is improved, and the technical problems that in the related art, the running state of the photovoltaic system is evaluated based on a mathematical method, and the accuracy of an evaluation result is low are solved.

Drawings

The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the application and do not constitute a limitation on the application. In the drawings:

FIG. 1 is a flow chart of an alternative neural network based operational state assessment method, according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of an alternative grid-tie safe operation index body in accordance with an embodiment of the present invention;

FIG. 3 is a schematic diagram of an alternative neural network according to an embodiment of the present invention;

FIG. 4 is a schematic diagram of an alternative neural network-based operational state assessment device, according to an embodiment of the present invention;

Fig. 5 is a block diagram of a hardware structure of an electronic device (or mobile device) based on a neural network operation state evaluation method according to an embodiment of the present invention.

Detailed Description

In order that those skilled in the art will better understand the present invention, a technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in which it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, shall fall within the scope of the present invention.

It should be noted that the terms "first," "second," and the like in the description and the claims of the present invention and the above figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the invention described herein may be implemented in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

To facilitate an understanding of the invention by those skilled in the art, some terms or nouns involved in the various embodiments of the invention are explained below:

SENet, squeze-and-Excitation Networks, is a convolutional neural network model in the field of deep learning, and improves the perceptibility of the network to the features by introducing a squeze-and-Excitation (SE) module, wherein squeze represents compression operation and expression represents Excitation operation.

The time domain convolutional network Temporal Convolutional Network, abbreviated as TCN, is a deep learning model and is mainly used for processing sequence data.

The Long Short-Term Memory network, LSTM for Short, is one kind of cyclic neural network for Long-Term dependence in processing Long sequence data.

Batch normalization Batch Normalization, BN for short.

The ReLU activation function, namely the full-scale RECTIFIED LINEAR Unit, is one of the activation functions commonly used in deep learning;

A BOOST circuit, a direct current-to-direct current (DC-DC) converter circuit, for boosting an input voltage to an output voltage;

A PI regulator is a proportional-integral (PI) controller used to control a system, typically to regulate the performance of the system;

DQ conversion is a data processing technique for converting raw data into a more meaningful or more convenient form for processing;

LCL filters are a common type of filter used to filter harmonics in power systems, consisting of a series circuit of one inductance (L) and two capacitances (C);

logistic functions, i.e., sigmoid activation functions, are Sigmoid functions commonly used in mathematics and statistics.

Softmax, a mathematical treatment method, is commonly used to address multi-classification problems.

It should be noted that, the operation state evaluation method and the device based on the neural network in the application can be used in the smart grid field under the condition that the operation state of the distributed photovoltaic grid-connected system is evaluated based on the neural network, and can also be used in any field except the smart grid field under the condition that the operation state of the distributed photovoltaic grid-connected system is evaluated based on the neural network.

It should be noted that, related information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) related to the present application are information and data authorized by a user or sufficiently authorized by each party, and the related data are collected, stored, used, processed, transmitted, provided, disclosed, applied, etc. processed, all in compliance with related laws and regulations and standards of related areas, necessary security measures are taken, no prejudice to public order and custom are provided with corresponding operation entries for the user to select authorization or rejection. For example, an interface is provided between the system and the relevant user or institution, before acquiring the relevant information, the system needs to send an acquisition request to the user or institution through the interface, and acquire the relevant information after receiving the consent information fed back by the user or institution.

In the application, when the client information is collected and analyzed, a corresponding operation entrance is provided for the user, so that the user can choose to agree or reject the automatic decision result; if the user selects refusal, the expert decision flow is entered.

The following embodiments of the present invention are applicable to various neural network-based operation state evaluation systems/applications/devices. The invention provides a neural network model based on an attention mechanism, a time domain convolution network and a long-term and short-term memory network, and the neural network model is used for evaluating the running state of a distributed photovoltaic grid-connected system, so that the problem that the running state evaluation effect is poor due to low big data processing efficiency and poor capability of mining data characteristics in the traditional mathematical method for evaluating the running state of the distributed photovoltaic grid-connected system at present is solved.

The present invention will be described in detail with reference to the following examples.

Example 1

According to an embodiment of the present invention, there is provided an embodiment of a neural network-based operation state evaluation method, it should be noted that the steps illustrated in the flowchart of the drawings may be performed in a computer system such as a set of computer-executable instructions, and that although a logical order is illustrated in the flowchart, in some cases, the steps illustrated or described may be performed in an order different from that herein.

FIG. 1 is a flowchart of an alternative neural network-based operational state assessment method, as shown in FIG. 1, according to an embodiment of the present invention, comprising the steps of:

step S101, collecting a state index data set from a distributed photovoltaic grid-connected system based on a preset grid-connected safe operation index system;

step S102, preprocessing state index data in the state index data set to obtain a preprocessed state index data set;

step S103, constructing a state index vector based on the preprocessed state index data set, and acquiring a weight vector of the state index;

Step S104, inputting a state index vector and a weight vector into a neural network model, and outputting a risk level of the distributed photovoltaic grid-connected system, wherein the neural network model is a model which is obtained by training in advance and is used for evaluating the running risk of the distributed photovoltaic grid-connected system, and at least comprises a self-attention network, a time domain convolution network and a long-period memory network;

Step S105, generating an operation state evaluation result of the distributed photovoltaic grid-connected system based on the risk level.

Through the steps, a state index data set is acquired from the distributed photovoltaic grid-connected system based on a preset grid-connected safe operation index system, the state index data in the state index data set is preprocessed to obtain a preprocessed state index data set, a state index vector is built based on the preprocessed state index data set, a weight vector of the state index is obtained, the state index vector and the weight vector are input into a neural network model, the risk level of the distributed photovoltaic grid-connected system is output, the neural network model is a model which is obtained by training in advance and used for evaluating the operation risk of the distributed photovoltaic grid-connected system, the neural network model at least comprises a self-attention network, a time domain convolution network and a long-short-term memory network, and finally an operation state evaluation result of the distributed photovoltaic grid-connected system is generated based on the risk level.

In the embodiment, multi-dimensional index data are collected from a distributed photovoltaic grid-connected system to be evaluated based on a grid-connected safe operation index system, a multi-dimensional evaluation standard is provided for state analysis, then the state index data are analyzed based on index vectors and a pre-trained neural network model, the risk level of the distributed photovoltaic grid-connected system is obtained, the running state of the distributed photovoltaic grid-connected system is further obtained, the characteristics in the data can be deeply mined based on the analysis of the multi-layer neural network in the neural network model, the importance degree of each characteristic is marked based on the weight value configured for each characteristic, and therefore the accuracy of state evaluation is improved, and the technical problems that in the related art, the running state of the photovoltaic system is evaluated based on a mathematical method, and the accuracy of an evaluation result is low are solved.

Embodiments of the present invention will be described in detail with reference to the following steps.

It should be noted that, the distributed photovoltaic power generation is used as a current hot power generation mode, solar energy can be converted into electric energy, the distributed photovoltaic power generation system has the characteristics and advantages of distributed layout, environmental friendliness, flexibility and high expandability, and as a part of the intelligent power grid, the distributed photovoltaic grid-connected system can realize more efficient energy management through energy storage equipment and intelligent control technology, promote the development of the intelligent power grid, and the distributed photovoltaic grid-connected system is influenced by external environment, lines and other factors, and may have unstable operation states, so that the working efficiency is low, and therefore, the evaluation of the operation states has important significance on the distributed photovoltaic grid-connected operation system.

The state evaluation method based on the neural network provided by the embodiment of the invention aims at the state evaluation problem of the distributed photovoltaic operation system, takes index data as input of a neural network model and takes evaluation results as output. Compared with the traditional mathematical method, the state evaluation method based on the neural network does not need to establish a complex mathematical physical model, can continuously optimize the evaluation model according to actual operation data samples, and has extremely strong processing capacity for big data. Meanwhile, a time domain convolution network is added in the neural network model, so that the neural network model has strong capture capability of long-term dependence information, and the evaluation accuracy can be remarkably improved.

Meanwhile, the embodiment of the invention introduces an attention mechanism in the neural network model, wherein the attention mechanism is a resource allocation strategy, and limited computing resources are used for processing more important information, so that the information overload problem is solved. When the neural network processes a large amount of input information, key information can be focused and processed through a focusing mechanism, so that the efficiency of the network is improved. The self-attention network adopted by the embodiment of the invention is a network structure applying an attention mechanism, and different weights are distributed to different parts of the input index data, so that the neural network can pay attention to important parts in the input data, namely the parts with the most obvious characteristics, thereby improving the accuracy of state evaluation with larger data scale and not increasing the complexity of the network.

Optionally, before collecting the status index data set from the distributed photovoltaic grid-connected system based on the preset grid-connected safe operation index system, the method further comprises: selecting an electric energy quality index for evaluating the running state of the distributed photovoltaic grid-connected system, wherein the electric energy quality index comprises at least one of the following components: voltage deviation, voltage flicker, voltage fluctuation, three-phase imbalance, frequency deviation, and harmonic content; selecting an electric energy index for evaluating the running state of the distributed photovoltaic grid-connected system, wherein the electric energy index comprises at least one of the following: power factor, power generation efficiency, active power change rate; selecting an environmental index for evaluating the running state of the distributed photovoltaic grid-connected system, wherein the environmental index comprises at least one of the following: ambient temperature, illumination intensity; and constructing a grid-connected safe operation index system based on the electric energy quality index, the electric energy index and the environment index.

It should be noted that, according to the embodiment of the present invention, based on the analysis of the state operation data in the historical time period, the cause that may cause the unstable operation state or the poor operation state of the distributed photovoltaic grid-connected system is determined, and the power quality index, the power index and the environmental index may be selected as the indexes to be evaluated, so as to construct a grid-connected safe operation index system of the distributed photovoltaic grid-connected system, and fig. 2 is a schematic diagram of an optional grid-connected safe operation index body according to the embodiment of the present invention, as shown in fig. 2, the grid-connected safe operation index system includes three parts, namely, the power quality index, the power index and the environmental index, where the power quality index includes: voltage deviation, voltage flicker, voltage fluctuation, three-phase imbalance, frequency deviation, and harmonic content; the power metrics may include: power factor, power generation efficiency, active power change rate; the external environmental indicators may include: the range of the data to be acquired can be defined based on the grid-connected safe operation index system, so that the multidimensional index data is analyzed, and the operation state of the distributed photovoltaic grid-connected system can be estimated more accurately.

Optionally, before collecting the status index data set from the distributed photovoltaic grid-connected system based on the preset grid-connected safe operation index system, the method further comprises: carrying out importance assignment on state indexes in a grid-connected safe operation index system to obtain relative importance values between every two state indexes; constructing a judgment matrix based on the relative importance values; calculating the eigenvalue and eigenvector of the judgment matrix; and carrying out consistency verification based on the characteristic values to obtain a verification result, and taking the characteristic vector as a weight vector of the state index under the condition that the verification result indicates that the judgment matrix passes verification.

It should be noted that, after the multidimensional index to be evaluated is obtained, a weight value needs to be configured for each index to characterize the importance of each index, and the configuration of the weight value of the index can be realized through a hierarchical analysis method, which specifically includes: obtaining the mutual importance degree between every two indexes through expert scoring, wherein the mutual importance degree is determined according to a 1-9 reciprocal scale method, namely when two elements are compared with equal importance, the scale is 1; when the former is a little more important than the latter, the scale is 3; when the former is significantly more important than the latter, the scale is 5; when the former is extremely important than the latter, the scale is 7; when the former is more important than the latter, the scales of 9,2, 4, 6 and 8 are the median values of the adjacent judgment conditions, a judgment matrix can be constructed for the state indexes in the grid-connected safe operation index system through expert scoring, and then the maximum eigenvalue of the judgment matrix and the corresponding eigenvector are calculated.

In order to prevent the judgment error, consistency test is needed to be carried out on the judgment matrix, if the test fails, the judgment is proved to be inaccurate, and the construction of the judgment matrix is carried out again. If the consistency test is passed, the judgment matrix eigenvector is output as a weight vector. The consistency check process comprises the following steps: the consistency index c.i= (λmax-n)/(n-1) is calculated, where λmax is the maximum eigenvalue and n is the number of state indexes, finding the average random consistency index r.i. and then calculating the consistency index ratio c.r= (c.i.) v/r.i.. When the C.R. is smaller than the preset threshold, the consistency test of the determined judgment matrix is considered to be passed, otherwise, the judgment matrix is modified.

For example, if there are three indices, index a, index b, and index c, the relative importance among the three indices needs to be determined first. Assuming scoring by an expert, we get the following results:

the index a has an important value of 3 relative to the index b;

The index a has an important value of 5 relative to the index c;

the index b has an important value of 2 with respect to the index c.

The decision matrix is constructed as follows:

And calculating the maximum eigenvalue and eigenvector of the judgment matrix to obtain an eigenvector w= [0.45,0.35,0.20], wherein the weight of the index a is configured to be 0.45, the weight of the index b is configured to be 0.35, and the weight of the index c is configured to be 0.20.

Step S101, collecting a state index data set from a distributed photovoltaic grid-connected system based on a preset grid-connected safe operation index system.

When the operation state of the distributed photovoltaic grid-connected system is evaluated in real time, firstly, relevant state index data are collected again from the distributed photovoltaic grid-connected system according to the state index to be evaluated recorded in the grid-connected safe operation index system, for example: and obtaining data such as voltage, current, harmonic wave, illumination intensity parameter, temperature parameter, active power and the like of the grid-connected point to obtain a state index data set.

Step S102, preprocessing the state index data in the state index data set to obtain a preprocessed state index data set.

After the state index data set is obtained, the state index data needs to be preprocessed, where the preprocessing operation includes: deleting the abnormal value of the data directly; and establishing a proper interpolation function f (x) for the missing value of the data, wherein the missing value is correspondingly interpolated by the function value obtained by the corresponding point.

Step S103, constructing a state index vector based on the preprocessed state index data set, and acquiring a weight vector of the state index.

After preprocessing the state index data, a one-dimensional state index vector is constructed based on the preprocessed state index data set, and a weight vector of the state index is obtained based on a weight value configured for the state index in advance.

And step S104, inputting the state index vector and the weight vector into the neural network model, and outputting the risk level of the distributed photovoltaic grid-connected system.

According to the embodiment of the invention, the state index vector is analyzed by adopting a pre-trained neural network model, so that the risk level of the distributed photovoltaic grid-connected system is obtained, the neural network model at least comprises an input layer, a convolution layer, a self-attention network, a time domain convolution network, a long-period memory network, a short-period memory network, a discarding layer, a full-connection layer, a softmax layer and an output layer, wherein the self-attention network is a network layer constructed based on an attention mechanism and can pay attention to and process key information, namely operation state characteristics, so that the identification accuracy and the data processing efficiency of the neural network are improved, the time domain convolution network can improve the accuracy of extracting the characteristics of state index data in terms of time sequences, and the long-period memory network and the short-period memory network can strengthen the memory capacity of the neural network and solve the problem of instability of gradients, thereby improving the accuracy of the neural network on the identification of the operation state of the distributed photovoltaic grid-connected system.

It should be noted that, the neural network model in the embodiment of the present invention is obtained by pre-training, and the training process includes: step one, data generation and preprocessing, wherein training data in the embodiment of the invention does not depend on historical data, but simulates the running states of a distributed photovoltaic grid-connected system under various conditions through simulation software, so as to obtain a training sample; step two, configuring an operation state label for the generated simulated electric energy signal; step three, a sample set is constructed, and the sample set is divided into a training set and a testing set; step four, an initial neural network model is built; step five, iterative training is carried out on the neural network model based on the training set; and step six, testing the trained neural network model based on the test set.

Optionally, the step of collecting N simulation state index data sets based on the simulation model of the distributed photovoltaic grid-connected system includes: building a simulation model of the distributed photovoltaic grid-connected system; simulating M operation states in a distributed photovoltaic grid-connected system simulation model based on state indexes in a grid-connected safe operation index system, wherein M is a positive integer; and collecting a plurality of simulation state index data sets of the simulation model of the distributed photovoltaic grid-connected system under various running states to obtain N simulation state index data sets.

For the first step, the data generation and preprocessing specifically includes: the distributed photovoltaic grid-connected operation simulation model is built, the photovoltaic modules can adopt simulation software self-carrying models, the number of the photovoltaic modules is set to be 9 multiplied by 17 (the number of the photovoltaic modules can be set based on actual demands), the rated maximum power is 30kW (the rated maximum power can be set based on actual demands), and the illumination intensity and the environmental temperature are set to be changed values. The photovoltaic module is boosted by a BOOST circuit, meanwhile, the tracking control of the maximum power point is completed, and the algorithm is realized by programming simulation software by adopting a disturbance observation method. The boosted voltage is stabilized by a direct current bus and then is converted from direct current to alternating current by a voltage type three-phase bridge inverter, and the inverter control strategy adopts voltage-current double closed-loop control, namely a voltage outer ring and a current inner ring. The dc bus voltage needs to be kept stable, so that the dc bus voltage is used as a current reference value of the current inner loop after passing through a PI regulator. In addition, because the PI controller cannot realize no-dead-difference control of the ac quantity, DQ conversion is performed on the ac quantity by factor, and the dc quantity is decomposed and then controlled. The controlled current is filtered by an LCL filter to obtain grid-connected current meeting the grid-connected requirement, and then the environment where the distributed photovoltaic grid-connected system model is located is regulated on the basis of the grid-connected current meeting the requirement, so that different running states are simulated, for example: the illumination intensity is adjusted to simulate the partial shadow shielding of the photovoltaic modules, the number of the series-parallel photovoltaic modules is adjusted to simulate the conditions of disconnection and short-circuit faults of the photovoltaic modules, and the greater the number of the adjustment, the heavier the fault degree and the higher the risk level to the running state. And the capacity of increasing the output power of the photovoltaic module is used for simulating the condition of grid-connected point voltage out-of-limit. The illumination intensity and the ambient temperature fluctuation are regulated to simulate the condition of photovoltaic output power fluctuation, so as to simulate the problems of voltage fluctuation, flicker and the like; a nonlinear load is connected in the circuit to simulate the situation of harmonic injection, etc.

For each simulated situation, respectively measuring state index data corresponding to the state index indicated by the grid-connected safe operation index system, including: and taking simulation state index data such as voltage deviation, voltage flicker, voltage fluctuation, three-phase unbalance degree, frequency deviation, harmonic content, power factor, power generation efficiency, active power change rate, ambient temperature, illumination intensity and the like as sample data, and finally preprocessing the measured sample data.

The step two, the configuration operation state label for each simulation state index data set specifically includes: a risk level is configured for each simulated state index data set as an operational state label.

It should be noted that fig. 3 is an architecture diagram of an alternative neural network according to an embodiment of the present invention, and as shown in fig. 3, the neural network model mainly includes: input layer (Input), convolution layer (Convolution), self-attention network (SENet), time domain convolution network (TCN), long-short-term memory network (LSTM), full connection layer (FullyConnected), drop layer (Dropout), full connection layer (regrantion), and the following details of constructing an initial neural network model by combining the above-mentioned fig. 3:

Firstly, an input layer is built, data is input as a one-dimensional sequence, the dimension of the input layer is the same as the dimension of a simulated electric energy signal vector, the input layer is connected to a one-dimensional convolution layer, as shown in fig. 3, the number of convolution kernels of the one-dimensional convolution layer is set to be f, the dimension of each convolution kernel is k, and the sliding step length is s.

Then, a self-attention network is built, as shown in fig. 3, the self-attention network is composed of a global average pooling layer (GlobleAverage Polling) and two full-connection layers (Fullyconnected), the outputs of the one-dimensional convolution layers are sequentially connected with the global average pooling layer and the two full-connection layers, the dimensions of the two full-connection layers are respectively 1×1×c/r and 1×1×c (C is the channel number and r is the scaling ratio), the two full-connection layers respectively adopt ReLU and Sigmoid as activation functions, and finally, the value of each channel of the output vector of the second full-connection layer is multiplied with the corresponding channel of the output feature diagram of the convolution layer, so that the output (Scale) of the self-attention network is obtained.

Then, a time domain convolution network of a neural network model is built, and the self-attention network is connected with the time domain convolution network, wherein the time domain convolution network mainly comprises three parts: causal convolution, dilation convolution, and residual connection.

The causal convolution calculates the neurons yt, yt-1, yt-2 … and y0 at the time t in the next layer (hidden layer) by using the neurons xt, xt-1 and xt-2 … x0 at the time t and the time before the time t of the input layer, so that the causal relation exists between network layers, the feature extraction of the electric energy signal in the time sequence is realized, zero filling is performed on the left side of the input sequence, and the same dimension of input and output data is ensured.

The expansion convolution is that a cavity is injected into a receptive field of the common convolution, namely, the receptive field range is enlarged by skipping a certain length of input mode through a filter, and the set expansion factor d represents that one neuron in each d neurons xt is selected to participate in calculation of next-layer neuron yt. As the depth of the network increases, d generally increases exponentially with 2, and the receptive field expands exponentially. And combining the dilation convolution and the causal convolution to construct a dilation causal convolution layer, so that a larger field of view is provided with fewer layers, and the problem of long-distance data dependence in a time sequence model is solved.

Residual connection, and the introduction of the residual connection avoids the gradient disappearance phenomenon possibly occurring due to network deepening.

As shown in fig. 3, when the time domain convolution networks are built, m time domain convolution modules are set in each time domain convolution network, 1 time domain convolution module is composed of two expansion causal convolution layers (Convolution), the output of the self-attention network is connected to the first time domain convolution module (the expansion factor (DilitionRate Casual) is 1), the module is provided with two identical expansion causal convolution layers, each convolution layer is connected with the next expansion causal convolution layer after Batch Normalization (BN) and ReLU activation functions, the output of the second expansion causal convolution layer is added with the corresponding element of the feature vector output by the self-attention network, and the output of the second expansion causal convolution layer is output to the next time domain convolution module after ReLU activation functions. The m time domain convolution modules are connected to form a time domain convolution network of the neural network model.

And (3) building a long-period memory network, connecting the time domain convolution network with the long-period memory network, and setting the number of hidden layer units of the long-period memory network to be H. The long-term and short-term memory network is introduced into a gating mechanism to control the information transmission path, and the gating value range is (0, 1) which indicates that the information is allowed to pass through according to the proportion.

The long-period memory network comprises three gates, namely an input gate it, a forget gate ft and an output gate ot, wherein the input gate it controls how much information in the candidate state at the current moment needs to be stored; the forgetting gate ft controls how much information is to be discarded in the internal state ct-1 input from the last moment to the current moment; the output gate ot controls how much information of the internal state ct at the current time is output as the external state ht. The gate activation function may be a Logistic function and the candidate state is a state activation function using a hyperbolic tangent function (tanh).

The long-term and short-term memory network calculation process comprises the following steps: calculating the values of three gates and candidate states by using the external state ht-1 at the previous moment and the input xt at the current moment; updating the memory cell ct by combining the forget gate ft and the input gate it; and in combination with the output gate ot, the information of the internal state is transferred to the external state ht.

The long-period memory network and the short-period memory network are connected with a plurality of circulating units, so that new information and information accumulated before forgetting parts are selectively added, long-period memory is built for the neural network, and the problem of gradient explosion or disappearance of the neural network in the training process is solved.

As shown in fig. 3, the feature vector output by the long-short term memory network is processed by links such as a discarding layer, a full-connection layer, a regression layer and the like, and finally connected to the output layer to output the evaluation result of the neural network algorithm.

For the fifth step, performing iterative training on the neural network model based on the training set and the sixth step, performing testing on the trained neural network model based on the testing set specifically includes:

Setting training parameters of the neural network model, selecting a proper gradient descent algorithm and setting the training parameters, wherein the training parameters can comprise: maximum number of rounds, minimum lot size, initial learning rate, etc.

The neural network model is trained, and the training set data is used as the training of the neural network model, and the iterative process of the neural network model is to adjust the value (the value of the convolution kernel, the LSTM layer weight, the full connection layer weight and the like) of each neural network layer. The parameters are dynamically adjusted through a back propagation algorithm, namely, the network parameters are adjusted layer by layer in the back direction after each small batch of training, and the gradient is calculated through a gradient descent algorithm, so that the neural network parameters are changed towards the direction that the error is minimized. Sample data in the training set is used for network training, so that errors of risk levels output by the neural network and real risk levels meet preset requirements when the maximum training round is reached.

And predicting the risk level of the simulation state index data in the test set by using the trained neural network model, and comparing the risk level calculated by the neural network model with the real risk level to evaluate the training effect.

Optionally, the step of inputting the state index vector and the weight vector into the neural network model and outputting the risk level of the distributed photovoltaic grid-connected system includes: inputting the state index vector and the weight vector into the neural network model based on an input layer of the neural network model; outputting a weight feature vector by a self-attention network based on a neural network model; the weight feature vector is used as input data of a time domain convolution network of the neural network model, and the time sequence feature vector is output based on the time domain convolution network; and taking the time sequence feature vector as input data of a long-short-period memory network of the neural network model, and outputting the risk level of the distributed photovoltaic grid-connected system based on the long-short-period memory network.

It should be noted that, when the self-attention network works, the channel weight vector is output through the global average pooling layer and two fully connected processes for the input signal vector, then the channel weight vector and the signal feature vector output by the convolution layer are calculated to obtain the weight feature vector as the output of the self-attention layer, and the self-attention network can assign different weights to different parts of the data, so that the features of the disturbance occurrence part are more prominent, and the neural network is beneficial to feature extraction, identification and classification.

It should be noted that, the time domain convolution network includes M time domain convolution modules, and the output of the time domain convolution network is obtained through layer-by-layer calculation of the time domain convolution modules, specifically: the method comprises the steps of inputting a weight feature vector output by a self-attention network as input data of a time domain convolution network to a first time domain convolution module belonging to the convolution network, inputting the input data to a first expansion causal convolution layer of the time domain convolution module after batch normalization processing (BN) and a ReLU activation function, outputting the input data to a next expansion causal convolution layer after convolution processing, calculating to obtain a time sequence feature vector, namely, outputting the first time domain convolution network, adding the time sequence feature vector output by each time domain convolution network and the weight feature vector input to the time domain convolution network to obtain the input of the next time domain convolution module, and so on until the final time sequence feature vector is obtained through calculation of M time domain convolution modules and is used as output data of the time domain convolution network.

The time sequence feature vector output by the time domain convolution network is input into the long-short-term memory network, and the risk level of the distributed photovoltaic grid-connected system is output through calculation of the long-short-term memory network.

It should be noted that, the risk level based on the output of the neural network model can be used for evaluating the running state of the distributed photovoltaic grid-connected system, and the countermeasures adopted by different risk levels are different, so that a necessary basis is provided for making a treatment scheme for a management end, the running state of the distributed photovoltaic grid-connected system is evaluated in real time, and the efficiency and the safety of the distributed photovoltaic grid-connected system can be guaranteed to the greatest extent.

The following describes in detail another embodiment.

Example two

The operation state evaluation device based on the neural network provided in this embodiment includes a plurality of implementation units, where each implementation unit corresponds to each implementation step in the first embodiment, and a specific implementation and a beneficial effect of each implementation unit may refer to the foregoing method embodiment and are not described herein again.

Fig. 4 is a schematic diagram of an alternative neural network-based operation state evaluation device according to an embodiment of the present invention, and as shown in fig. 4, the neural network-based operation state evaluation device may include: an acquisition unit 41, a processing unit 42, a construction unit 43, an output unit 44, a generation unit 45, wherein,

The collecting unit 41 is used for collecting a state index data set from the distributed photovoltaic grid-connected system based on a preset grid-connected safe operation index system;

a processing unit 42, configured to preprocess state index data in the state index data set to obtain a preprocessed state index data set;

a construction unit 43, configured to construct a state index vector based on the preprocessed state index data set, and acquire a weight vector of the state index;

The output unit 44 is configured to input the state index vector and the weight vector into a neural network model, and output a risk level of the distributed photovoltaic grid-connected system, where the neural network model is a model that is obtained by training in advance and is used to evaluate an operation risk of the distributed photovoltaic grid-connected system, and the neural network model is at least composed of a self-attention network, a time domain convolution network and a long-short-term memory network;

the generating unit 45 is configured to generate an operation state evaluation result of the distributed photovoltaic grid-connected system based on the risk level.

The operation state evaluation device based on the neural network acquires a state index data set from the distributed photovoltaic grid-connected system based on a preset grid-connected safe operation index system through the acquisition unit 41; preprocessing state index data in the state index data set by the processing unit 42 to obtain a preprocessed state index data set; constructing a state index vector based on the preprocessed state index data set by a construction unit 43, and acquiring a weight vector of the state index; the state index vector and the weight vector are input into a neural network model through an output unit 44, and the risk level of the distributed photovoltaic grid-connected system is output, wherein the neural network model is a model which is obtained through pre-training and is used for evaluating the running risk of the distributed photovoltaic grid-connected system, and the neural network model at least comprises a self-attention network, a time domain convolution network and a long-period memory network; the operation state evaluation result of the distributed photovoltaic grid-connected system is generated based on the risk level by the generation unit 45.

Optionally, the output unit 44 includes: the first input module is used for inputting the state index vector and the weight vector into the neural network model based on an input layer of the neural network model; the first output module is used for outputting weight feature vectors based on the self-attention network of the neural network model; the second output module is used for taking the weight feature vector as input data of a time domain convolution network of the neural network model and outputting a time sequence feature vector based on the time domain convolution network; and the third output module is used for taking the time sequence feature vector as input data of the long-short-period memory network of the neural network model and outputting the risk level of the distributed photovoltaic grid-connected system based on the long-short-period memory network.

Optionally, the operation state evaluation device based on the neural network further includes: the first assignment module is used for carrying out importance assignment on the state indexes in the grid-connected safe operation index system to obtain relative importance values between every two state indexes; the first construction module is used for constructing a judgment matrix based on the relative importance values; the first calculation module is used for calculating the eigenvalue and eigenvector of the judgment matrix; the first verification module is used for carrying out consistency verification based on the characteristic values to obtain a verification result, and taking the characteristic vector as a weight vector of the state index under the condition that the verification result indicates that the judgment matrix passes the verification.

Optionally, the operation state evaluation device based on the neural network further includes: the first selecting module is used for selecting an electric energy quality index for evaluating the running state of the distributed photovoltaic grid-connected system, wherein the electric energy quality index comprises at least one of the following: voltage deviation, voltage flicker, voltage fluctuation, three-phase imbalance, frequency deviation, and harmonic content; the second selecting module is used for selecting an electric energy index for evaluating the running state of the distributed photovoltaic grid-connected system, wherein the electric energy index comprises at least one of the following: power factor, power generation efficiency, active power change rate; the third selecting module is used for selecting an environment index for evaluating the running state of the distributed photovoltaic grid-connected system, wherein the environment index comprises at least one of the following: ambient temperature, illumination intensity; the second construction module is used for constructing a grid-connected safe operation index system based on the electric energy quality index, the electric energy index and the environment index.

Optionally, the processing unit 42 includes: the first deleting module is used for traversing the state index data and deleting abnormal values in the state index data; the first complementing module is used for establishing an interpolation function and complementing the missing value in the state index data.

Optionally, the first acquisition module includes: the first building sub-module is used for building a simulation model of the distributed photovoltaic grid-connected system; the first simulation sub-module is used for simulating M running states in a simulation model of the distributed photovoltaic grid-connected system based on state indexes in a grid-connected safe running index system, wherein M is a positive integer; the first acquisition sub-module is used for acquiring a plurality of simulation state index data sets of the simulation model of the distributed photovoltaic grid-connected system under various running states to obtain N simulation state index data sets.

The neural network-based operation state evaluation device may further include a processor and a memory, wherein the acquisition unit 41, the processing unit 42, the construction unit 43, the output unit 44, the generation unit 45, and the like are stored in the memory as program units, and the processor executes the program units stored in the memory to implement the corresponding functions.

The processor includes a kernel, and the kernel fetches a corresponding program unit from the memory. The kernel can be set to one or more, and the running state of the distributed photovoltaic grid-connected system is evaluated by adjusting the kernel parameters.

The memory may include volatile memory in a computer-readable medium, random Access Memory (RAM) and/or nonvolatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM), which includes at least one memory chip.

According to another aspect of the embodiment of the present invention, there is also provided a computer readable storage medium, where the computer readable storage medium includes a stored computer program, and when the computer program is executed, the device in which the computer readable storage medium is located is controlled to execute any one of the above methods for evaluating an operation state based on a neural network.

According to another aspect of the embodiments of the present invention, there is also provided an electronic device including one or more processors and a memory for storing one or more programs, wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to implement any of the above-described neural network based operation state evaluation methods.

According to another aspect of the embodiments of the present invention, there is also provided a computer program product, including a computer program, wherein the computer program when executed by a processor implements any one of the above methods for estimating an operation state based on a neural network.

The application also provides a computer program product adapted to perform, when executed on a data processing device, a program initialized with the method steps of: acquiring a state index data set from a distributed photovoltaic grid-connected system based on a preset grid-connected safe operation index system; preprocessing state index data in the state index data set to obtain a preprocessed state index data set; constructing a state index vector based on the preprocessed state index data set, and acquiring a weight vector of the state index; the method comprises the steps of inputting a state index vector and a weight vector into a neural network model, and outputting a risk level of a distributed photovoltaic grid-connected system, wherein the neural network model is a model which is obtained through pre-training and used for evaluating the running risk of the distributed photovoltaic grid-connected system, and at least comprises a self-attention network, a time domain convolution network and a long-term and short-term memory network; and generating an operation state evaluation result of the distributed photovoltaic grid-connected system based on the risk level.

The application also provides a computer program product, which, when being executed on a data processing device, is further adapted to carry out a program initialized with the method steps of: the method comprises the steps of inputting a state index vector and a weight vector into a neural network model, and outputting a risk level of a distributed photovoltaic grid-connected system, wherein the steps comprise: inputting the state index vector and the weight vector into the neural network model based on an input layer of the neural network model; outputting a weight feature vector by a self-attention network based on a neural network model; the weight feature vector is used as input data of a time domain convolution network of the neural network model, and the time sequence feature vector is output based on the time domain convolution network; and taking the time sequence feature vector as input data of a long-short-period memory network of the neural network model, and outputting the risk level of the distributed photovoltaic grid-connected system based on the long-short-period memory network.

The application also provides a computer program product, which, when being executed on a data processing device, is further adapted to carry out a program initialized with the method steps of: before the state index data set is collected from the distributed photovoltaic grid-connected system based on the preset grid-connected safe operation index system, the method further comprises the following steps: carrying out importance assignment on state indexes in a grid-connected safe operation index system to obtain relative importance values between every two state indexes; constructing a judgment matrix based on the relative importance values; calculating the eigenvalue and eigenvector of the judgment matrix; and carrying out consistency verification based on the characteristic values to obtain a verification result, and taking the characteristic vector as a weight vector of the state index under the condition that the verification result indicates that the judgment matrix passes verification.

The application also provides a computer program product, which, when being executed on a data processing device, is further adapted to carry out a program initialized with the method steps of: the step of preprocessing the state index data in the state index data set comprises the following steps: traversing the state index data, and deleting abnormal values in the state index data; and establishing an interpolation function, and complementing the missing value in the state index data.

The application also provides a computer program product, which, when being executed on a data processing device, is further adapted to carry out a program initialized with the method steps of: the neural network model is obtained through pre-training, and the step of training the neural network model comprises the following steps: collecting N simulation state index data sets based on a simulation model of the distributed photovoltaic grid-connected system, wherein N is a positive integer; configuring an operation state label for each simulation state index data set; constructing a sample set based on the simulation state index data set and the running state label, and dividing the sample set based on a preset dividing proportion to obtain a training set and a testing set; setting up an initial neural network model; performing iterative training on the initial neural network model based on the training set to obtain a trained neural network model; and testing the trained neural network model based on the test set to obtain a test result, and obtaining the trained neural network model under the condition that the test result indicates that the neural network model passes the test.

The application also provides a computer program product, which, when being executed on a data processing device, is further adapted to carry out a program initialized with the method steps of: the step of collecting N simulation state index data sets based on the simulation model of the distributed photovoltaic grid-connected system comprises the following steps: building a simulation model of the distributed photovoltaic grid-connected system; simulating M operation states in a distributed photovoltaic grid-connected system simulation model based on state indexes in a grid-connected safe operation index system, wherein M is a positive integer; and collecting a plurality of simulation state index data sets of the simulation model of the distributed photovoltaic grid-connected system under various running states to obtain N simulation state index data sets.

Fig. 5 is a block diagram of a hardware structure of an electronic device (or mobile device) based on a neural network operation state evaluation method according to an embodiment of the present invention. As shown in fig. 5, the electronic device may include one or more processors 502 (shown in fig. 5 as 502a, 502b, … …,502 n) (the processor 502 may include, but is not limited to, a microprocessor MCU, a programmable logic device FPGA, etc.) a memory 504 for storing data. In addition, the method may further include: a display, an input/output interface (I/O interface), a Universal Serial Bus (USB) port (which may be included as one of the ports of the I/O interface), a network interface, a keyboard, a power supply, and/or a camera. It will be appreciated by those of ordinary skill in the art that the configuration shown in fig. 5 is merely illustrative and is not intended to limit the configuration of the electronic device described above. For example, the electronic device may also include more or fewer components than shown in FIG. 5, or have a different configuration than shown in FIG. 5.

The foregoing embodiment numbers of the present invention are merely for the purpose of description, and do not represent the advantages or disadvantages of the embodiments.

In the foregoing embodiments of the present invention, the descriptions of the embodiments are emphasized, and for a portion of this disclosure that is not described in detail in this embodiment, reference is made to the related descriptions of other embodiments.

In the several embodiments provided in the present application, it should be understood that the disclosed technology may be implemented in other manners. The above-described embodiments of the apparatus are merely exemplary, and the division of the units, for example, may be a logic function division, and may be implemented in another manner, for example, a plurality of units or components may be combined or may be integrated into another system, or some features may be omitted, or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be through some interfaces, units or modules, or may be in electrical or other forms.

The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in the embodiments of the present invention may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in software functional units.

The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied essentially or in part or all of the technical solution or in part in the form of a software product stored in a storage medium, including instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a usb disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a removable hard disk, a magnetic disk, or an optical disk, or other various media capable of storing program codes.

The foregoing is merely a preferred embodiment of the present invention and it should be noted that modifications and adaptations to those skilled in the art may be made without departing from the principles of the present invention, which are intended to be comprehended within the scope of the present invention.

Claims

1. A method for evaluating operating status based on a neural network, comprising:

Collecting a set of status indicator data from the distributed photovoltaic grid-connected system based on a preset grid-connected safety operation indicator system;

Preprocessing the status indicator data in the status indicator data set to obtain a preprocessed status indicator data set;

Constructing a state indicator vector based on the preprocessed state indicator data set, and obtaining a weight vector of the state indicator;

Inputting the state indicator vector and the weight vector into a neural network model, and outputting the risk level of the distributed photovoltaic grid-connected system, wherein the neural network model is a pre-trained model for evaluating the operation risk of the distributed photovoltaic grid-connected system, and the neural network model is composed of at least a self-attention network, a time domain convolutional network, and a long short-term memory network;

An operating status assessment result of the distributed photovoltaic grid-connected system is generated based on the risk level.

2. The evaluation method according to claim 1, characterized in that the step of inputting the state indicator vector and the weight vector into a neural network model and outputting the risk level of the distributed photovoltaic grid-connected system comprises:

Inputting the state indicator vector and the weight vector into the neural network model based on the input layer of the neural network model;

A self-attention network output weight feature vector based on the neural network model;

Using the weight feature vector as input data of a time-domain convolutional network of the neural network model, and outputting a time series feature vector based on the time-domain convolutional network;

The time series feature vector is used as input data of a long short-term memory network of the neural network model, and the risk level of the distributed photovoltaic grid-connected system is output based on the long short-term memory network.

3. The evaluation method according to claim 1 is characterized in that before collecting the state indicator data set from the distributed photovoltaic grid-connected system based on the preset grid-connected safety operation indicator system, it also includes:

Assigning importance to the status indicators in the grid-connected safe operation indicator system to obtain a relative importance value between every two status indicators;

Constructing a judgment matrix based on the relative importance values;

Calculating the eigenvalues and eigenvectors of the judgment matrix;

A consistency check is performed based on the eigenvalue to obtain a check result. When the check result indicates that the judgment matrix passes the check, the eigenvector is used as a weight vector of the state indicator.

4. The evaluation method according to claim 1 is characterized in that before collecting the state indicator data set from the distributed photovoltaic grid-connected system based on the preset grid-connected safety operation indicator system, it also includes:

Selecting power quality indicators for evaluating the operating status of the distributed photovoltaic grid-connected system, wherein the power quality indicators include at least one of the following: voltage deviation, voltage flicker, voltage fluctuation, three-phase unbalance, frequency deviation, and harmonic content;

Selecting an electric energy index for evaluating the operating status of the distributed photovoltaic grid-connected system, wherein the electric energy index includes at least one of the following: power factor, power generation efficiency, and active power change rate;

Selecting environmental indicators for evaluating the operating status of the distributed photovoltaic grid-connected system, wherein the environmental indicators include at least one of the following: ambient temperature, light intensity;

The grid-connected safe operation index system is constructed based on the power quality index, the power index and the environmental index.

5. The evaluation method according to claim 1, characterized in that the step of preprocessing the status indicator data in the status indicator data set comprises:

Traversing the state indicator data, and deleting abnormal values in the state indicator data;

An interpolation function is established to fill in the missing values in the state indicator data.

6. The evaluation method according to claim 1, characterized in that the neural network model is pre-trained, and the step of training the neural network model comprises:

Based on the distributed photovoltaic grid-connected system simulation model, N simulation status indicator data sets are collected, where N is a positive integer;

Configuring an operation status label for each simulation status indicator data set;

Constructing a sample set based on the simulation state indicator data set and the operation state label, and dividing the sample set based on a preset division ratio to obtain a training set and a test set;

Build the initial neural network model;

Iteratively training the initial neural network model based on the training set to obtain a trained neural network model;

The trained neural network model is tested based on the test set to obtain a test result. When the test result indicates that the neural network model passes the test, the trained neural network model is obtained.

7. The evaluation method according to claim 6, characterized in that the step of collecting N simulation status indicator data sets based on the distributed photovoltaic grid-connected system simulation model comprises:

Building a simulation model of the distributed photovoltaic grid-connected system;

Based on the state indicators in the grid-connected safety operation indicator system, M operating states are simulated in the distributed photovoltaic grid-connected system simulation model, where M is a positive integer;

A plurality of simulation state indicator data sets of the distributed photovoltaic grid-connected system simulation model under various operating states are collected to obtain N simulation state indicator data sets.

8. A distributed photovoltaic grid-connected system operation status assessment device, characterized by comprising:

A collection unit, used to collect a state indicator data set from a distributed photovoltaic grid-connected system based on a preset grid-connected safety operation indicator system;

A processing unit, used for preprocessing the state indicator data in the state indicator data set to obtain a preprocessed state indicator data set;

A construction unit, configured to construct a state indicator vector based on the preprocessed state indicator data set, and obtain a weight vector of the state indicator;

An output unit, used to input the state indicator vector and the weight vector into a neural network model, and output the risk level of the distributed photovoltaic grid-connected system, wherein the neural network model is a pre-trained model for evaluating the operation risk of the distributed photovoltaic grid-connected system, and the neural network model is composed of at least a self-attention network, a time domain convolutional network, and a long short-term memory network;

A generating unit is used to generate an operating status assessment result of the distributed photovoltaic grid-connected system based on the risk level.

9. A computer-readable storage medium, characterized in that the computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the neural network-based operating status assessment method according to any one of claims 1 to 7.

10. An electronic device, characterized in that it comprises one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the neural network-based operating status assessment method described in any one of claims 1 to 7.