CN111382789B - Power load identification method and system based on machine learning - Google Patents

Abstract

The power load identification method and the system based on machine learning provided by the application are characterized in that the basis of actually measured electrical parameter data including current, voltage, power and the like is in a unified format, and on the basis of long-time extraction, collection, analysis, induction and training of power load characteristics, the type of an electrical appliance in use can be accurately identified under the condition that a plurality of pieces of power load overall basis electrical parameter data for a period of time including voltage, current, active power, reactive power and the like are known. Therefore, the machine learning model training method and system for power load identification provided by the application do not need to manually adjust parameters, and compared with the traditional method, such as time domain waveform matching, feature point matching, spectrum analysis and the like, the method and system have high matching accuracy, can automatically learn and automatically acquire the feature parameters required by power load identification, thereby improving the application range of the model and improving the accuracy of power load identification.

Description

Power load identification method and system based on machine learning

Technical field

The present application relates to the technical field of electric load detection, and in particular to a method and system for electric load identification based on machine learning.

Background technique

The power load characteristics are the rules in which the active power and reactive power absorbed by the power load from the power supply of the power system change with the voltage at the load end point and the system frequency. The power load characteristics are an important part of the power system. The power load characteristics are used to identify Electrical equipment plays an important role in the development of smart grid technology.

The most commonly used methods for power load identification are intrusive and non-intrusive identification methods. Among them, the intrusive identification method requires the establishment of a monitoring system and the installation of sensors at each load. Although this method can directly obtain the measurement data of the load, the installation cost is high, the installation process is complicated, and the maintenance is relatively difficult; the non-intrusive identification method Only by installing monitoring equipment at the main entrance of the power supply can the loads in the entire system be decomposed, monitored and identified. Specifically, the non-intrusive identification method is based on the extraction and identification of electrical load imprint features; among them, the electrical load imprint features can reflect the unique information of the electrical equipment in operation, such as voltage, active power waveform, starting current, etc.; in These load imprint characteristics will appear repeatedly during equipment operation. Based on this, we can identify the electrical equipment.

Among them, the design and extraction of load imprint features is the main difficulty of the entire method; feature design generally uses relatively simple steady/transient characteristics of current, voltage, active power and reactive power and their combinations. However, artificially designed signal features require manual adjustment of parameters, which has the problem of low complexity and low dimensionality. At the same time, traditional matching algorithms such as time domain waveform matching, feature point matching, and spectral analysis have low matching accuracy. Furthermore, the accuracy of electrical load identification is not high, and the practical application effect is not ideal; and data modeling is a very necessary and important task. For electrical equipment with relatively stable operating conditions, such as televisions, electric kettles, computers, etc., load identification is relatively difficult. If there are many working conditions, such as fully automatic washing machines, the power consumption changes a lot during work, and the load It is also very difficult to identify. In response to the above problems, extraction technology based on steady-state features cannot effectively deal with some scenes with high recognition difficulty.

Contents of the invention

This application provides an electric load identification method and system based on machine learning to solve the technical problem in the existing method that artificially designed signal characteristics require manual adjustment of parameters, resulting in low accuracy of electric load identification.

In order to solve the above technical problems, the embodiments of this application disclose the following technical solutions:

In a first aspect, embodiments of the present application disclose a method for identifying electric power loads based on machine learning. The method includes:

Obtain the historical electrical parameter data set of each electrical appliance;

Cleaning the historical electrical parameter data set of each electrical appliance;

Divide the cleaned historical electrical parameter data set of a single electrical appliance into the original training set, verification set and test set in proportion;

Intercept data fragments from the original training set, verification set and test set along the timeline to generate a training set, verification set and test set composed of data fragments;

Establish a convolutional neural network model based on denoising autoencoder for each target electrical appliance;

For each target electrical appliance, train the convolutional neural network model based on the denoising autoencoder based on the training set, verification set and test set composed of data segments to obtain an optimized model for each target electrical appliance;

Collect the current data of the user's electric load, input the current data into the optimization model of each target electrical appliance, isolate the working status of the electrical appliance, and output the category results of the electric load.

Optionally, obtaining the historical electrical parameter data set of each electrical appliance includes:

Integrate and summarize today’s public data sets to obtain the first data set;

Install an electric energy meter at the user's general incoming line end to obtain the electrical parameters of the overall and individual electrical loads in one or more spaces to obtain a second data set;

A historical electrical parameter data set of each electrical appliance is obtained based on the first data set and the second data set.

Optionally, the original training set, verification set and test set are intercepted along the timeline to generate a training set, verification set and test set composed of data fragments, including:

Use a sliding window with a length of n and a moving step of 1 to intercept data fragments from the original training set, verification set and test set along the time axis direction to generate a training set, verification set and test set composed of data fragments.

Optionally, the cleaning of the historical electrical parameter data set of each electrical appliance includes: unification of data format, down-sampling to a specified frequency, and voltage normalization.

Optional, the unification of the data format includes:

according to Convert active power into a value between [0, 1], where:

S[i] represents the sampling value, which is the instantaneous active power, C is the electric load type, sa is the sample data, and s _c is the active power of the electric load c.

Optionally, the downsampling to a specified frequency includes:

If the sampling rate is lower than 1Hz, it will be recorded at the original sampling rate;

If the sampling rate is higher than 1Hz, downsample the sampling rate to 1Hz;

Among them, downsampling the sampling rate to 1Hz includes:

Use the value of the sampling point every 1 second and discard all other sampling values within 1 second;

Calculate the average of the original sampling points within adjacent 1 second as the 1 second boundary data value;

Calculate the median value of the original sampling points within adjacent 1 second as the 1 second boundary data value.

Optionally, the voltage normalization includes:

according to Normalize the voltage to the same fluctuation range, where:

Power _normalized represents the normalized power value, Power represents the measured power value, and Voltagenominal represents the nominal voltage value. Voltageobserved represents the measured voltage value.

Optionally, the cleaning of the historical electrical parameter data set of each electrical appliance also includes detecting gaps, normal operating times, and identifying the top K power loads with energy consumption, where k is an adjustable parameter.

Optionally, the convolutional neural network model based on the denoising autoencoder is trained for each target electrical appliance based on the training set, verification set and test set composed of data segments to obtain an optimized model for each target electrical appliance, include:

Use a training set composed of target electrical appliance data segments to train the parameters of the convolutional neural network model based on the denoising autoencoder;

Verify and test the different models obtained in different training stages on the verification set composed of data fragments until the best effect is used as the corresponding model of the target appliance;

Use the test set composed of data fragments to perform performance testing on the corresponding model of the target electrical appliance until the performance is optimal to obtain the optimized model of the target electrical appliance.

In the second aspect, this application is based on the above-mentioned machine learning-based power load identification method. This application also provides a machine learning-based power load identification system. The system includes:

The data set acquisition module is used to obtain the historical electrical parameter data set of each electrical appliance;

A data set cleaning module, used to clean the historical electrical parameter data set of each electrical appliance;

The data set division module is used to divide the cleaned historical electrical parameter data set of a single electrical appliance into the original training set, verification set and test set in proportion;

A data set fragment interception module is used to intercept data fragments from the original training set, verification set and test set along the timeline, and generate a training set, verification set and test set composed of data fragments;

The model building module is used to establish a convolutional neural network model based on the denoising autoencoder for each target electrical appliance;

A model optimization module, used to train the convolutional neural network model based on the denoising autoencoder based on the training set, verification set and test set composed of data fragments for each target appliance to obtain an optimization model for each target appliance;

The electric load category output module is used to collect the current data of the user's electric load, input the current data into the optimization model of each target electrical appliance, isolate the working status of the electrical appliance, and output the category result of the electric load.

Compared with the existing technology, the beneficial effects of this application are:

It can be seen from the above technical solution that the machine learning-based power load identification method and system provided by this embodiment are specifically based on the measured electrical parameter data including current, voltage, active power and reactive power, etc., and unify the basic electrical parameter data Format, based on our long-term extraction, collection, analysis, induction and training of electric load characteristics, we can know the overall basic electrical parameter data of several electric loads for a period of time, including voltage, current, active power, and reactive power. etc., correctly identify the type of electrical appliance being used. Therefore, the machine learning model training method and system for power load identification provided by this application does not require manual adjustment of parameters. Compared with traditional methods such as time domain waveform matching, feature point matching and spectrum analysis, the matching accuracy is higher. The application can learn independently and automatically obtain the characteristic parameters needed to identify electric loads, thereby improving the applicable scope of the model and improving the accuracy of electric load identification.

It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and do not limit the present application.

Description of the drawings

In order to explain the technical solutions of the present application more clearly, the drawings required to be used in the embodiments will be briefly introduced below. Obviously, for those of ordinary skill in the art, without exerting creative efforts, Additional drawings can be obtained from these drawings.

Figure 1 is a schematic flow chart of a machine learning-based power load identification method provided by an embodiment of the present application;

Figure 2 is a schematic structural diagram of a power load identification system based on machine learning provided by an embodiment of the present application;

Figure 3 is a schematic diagram of overall basic electrical parameter data monitoring by a DSP single-phase multi-function meter provided by an embodiment of the present application.

Detailed ways

In order to enable those in the technical field to better understand the technical solutions in the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described The embodiments are only some of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts should fall within the scope of protection of this application.

Electric load imprint characteristics can reflect the unique information reflecting the power consumption status of an electrical equipment during operation, such as voltage, active power waveform, current, etc. During the operation of the equipment, these load imprints will appear repeatedly. Based on In this way, we can identify each electrical equipment.

In the first aspect, an embodiment of the present application discloses a method for identifying electric power loads based on machine learning, as shown in Figure 1. The method includes:

S110: Obtain the historical electrical parameter data set of each electrical appliance.

The acquisition of historical electrical parameter data sets of each electrical appliance includes:

Integrate and summarize today’s public data sets to obtain the first data set;

Install an electric energy meter at the user's general incoming line end to obtain the electrical parameters of the overall and individual electrical loads in one or more spaces to obtain a second data set;

A historical electrical parameter data set of each electrical appliance is obtained based on the first data set and the second data set.

In the embodiment of this application, in addition to summarizing today's public data sets, an electric energy meter is also installed at the user's total incoming line end to obtain electrical parameters.

Because in the data measurement to obtain the transient and steady-state signals of the total load, there will be the following measurement errors: First, the inconsistency of the measuring devices, that is, for the same electrical equipment, different measuring devices have different measurements. Second, the sensor will cause data loss during the process of compressing and transmitting raw data. Precisely because data collection and transmission will cause data deviation or loss, it is necessary to process the data to improve the noise immunity of the load identification method; third, it is necessary to study the impact of the data sampling cycle on load identification, and explore the relationship between data sampling overhead and A balance of system modeling complexity.

The data set we collect is a record of energy consumption. We use multiple sets of monitoring instruments and equipment to monitor various electrical appliances in a room or space within a certain time range, using two data collection modes: low frequency and high frequency. The acquisition frequency of low-frequency acquisition mode is 1Hz, while the high-frequency acquisition can reach 10kHz to 100kHz. Low-frequency signals are mainly extracted for load steady-state characteristics, while high-frequency signals can obtain load transient characteristics and high-frequency harmonic characteristics. Generally speaking, high-frequency signals can include more load power characteristics, which is beneficial to model training and accuracy improvement, but it also puts forward higher requirements for data collection, transmission, compression and processing capabilities, improving the system complexity. Throughout the research process, we need to make trade-offs between the complexity and accuracy of the system based on the actual situation.

In addition, we use DSP single-phase multi-function electric energy meters to monitor the overall basic electrical parameter data (including voltage, current, active power, reactive power, etc.), as shown in Figure 3. Figure 3 is based on Figure 3. Schematic diagram of the overall basic electrical parameter data monitoring provided by the DSP single-phase multi-function meter provided in the application embodiment. The meter uses RS485 remote link to the monitoring control panel. The data collection is queried once every minute and is linked to the real-time data collection server in time. The advantages of this equipment Yes: DIN35MM guide rail installation, easy to load and unload; communication rate can reach 9600bps, fast transmission rate; adopts six-way switch input and output to meet the requirements for measurement data input and output; DSP chip can be reconstructed and processed according to actual needs Developed to meet the requirements of the experimental environment.

For example: we use a DSP single-phase multi-function electric energy meter to collect overall basic electrical parameter data (including voltage, current, active power, reactive power, etc.), and the sampling data within the same parameter are arranged in chronological order. Taking active power as an example, p _i represents the voltage degree detected by the electric energy meter at the i-th sampling time point. The original collected data is arranged in time order as the active power sequence P _active = {p ₁ , p ₂ ,…p _i ,…}, forms the output of the power load characteristic data sampling module and serves as the input of the power load data cleaning module.

S120: Clean the historical electrical parameter data set of each electrical appliance.

After obtaining the information of electric load data, electric load data cleaning will be the core technology of this system and method, which mainly includes automatic screening and abnormal data cleaning, realizing noise identification and separation, down-sampling, discard rate, and data normalization. , missing data compensation and processing, Top-k, erroneous data elimination, etc.

This system needs to create data CSV files and complete the work of cleaning the data set by deleting incomplete data (some instrument and equipment data have incomplete or lost data due to different timestamps). Once the dataset is created and the CSV file is generated for import, the data resides in our in-memory data structure, which can be used throughout the training process (saving or loading data to disk). In order to solve the problem of inconsistent data formats in different data sets, we need to perform several preprocessing tasks.

We use a DSP single-phase multi-function electric energy meter to collect overall basic electrical parameter data (including voltage, current, active power, reactive power, etc.). The sampling data within the same parameter are arranged in chronological order. Taking active power as an example, p _i represents the voltage degree detected by the electric energy meter at the i-th sampling time point. The original collected data is arranged in time order as the active power sequence P _active = {p ₁ , p ₂ ,…p _i ,…}, forms the output of the power load characteristic data sampling module and serves as the input of the power load data cleaning module.

In the data cleaning module, we down-sample the electrical timing data to the specified frequency f according to the actual application requirements. Assuming that the frequency f corresponds to the interval period of 5 sampling points in the original sampling data, the down-sampled active power data sequence is P ' _active = {p ₁ , p ₆ , p ₁₁ ,...p _5i+1 ,...}={q ₁ , q ₂ ,...q _i ,...}. The obtained relatively low-frequency time series data is then normalized and other preprocessing operations are performed to generate continuous time series data that can be directly used by the model for training. Since the power load characteristic identification model accepts multiple electrical parameter inputs at the same time, but the input dimension of a specific electrical parameter is limited and has a fixed length n, it is impossible to input all the continuous time series data of this electrical parameter at one time, so the length is The sliding window method of n continuously slides with a step size l. After each sliding, a data subsequence of fixed length n in the electrical parameter sequence is intercepted, for example, the active power subsequence P _i ={q _in+1 ,q _in+2 ,…q _in+n }, which is used as the input of the power load characteristic identification module.

Different electrical appliances have separate corresponding deep neural network models in the load characteristic identification module. The data subsequences are simultaneously input into the deep neural network models corresponding to different electrical appliances. The n sampled values of each electrical parameter in the same time period correspond to the deep neural network models. The corresponding n nodes of the network input layer, through the forward propagation operation of the network, output the mth target electrical appliance operating status sequence S _m,i in this data subsequence = {s _m,in+1 ,s _m,in+2 , …s _m,in+n }. The length of the state sequence is n but is not limited to n and is determined by the model architecture. Based on the working status of each electrical appliance, the types of all electrical appliances in use in this sequence can be determined.

Specifically include:

(1)Unified format

Since the format of the original data set is not uniform, this method needs to extract the characteristics of each data set for evaluation. In order to avoid great interference in the judgment due to the large difference in power consumption of different electrical appliances, the data needs to be cleaned, and it is also The normalization operation is to convert it into a value between [0, 1], S[i] represents the sampled value, which is the instantaneous active power, C is the electric load type, sa is the sample data, s _c is the active power of the electric load c Power, the formula is as follows:

(2) Downsampling

The sampling rate of the device monitor in the data set is between 0.008Hz and 16kHz, so this system will use aggregation functions such as mean, mode, and median to downsample the data set to the specified frequency.

If the sampling rate is lower than 1Hz, it will be recorded at the original sampling rate.

If the sampling rate is higher than 1Hz, the sampling rate is downsampled to 1Hz. Specific downsampling methods include:

1. Use the value of the sampling point every 1 second and discard all other sampling values within 1 second;

2. Calculate the average of the original sampling points within adjacent 1 second as the 1 second boundary data value;

3. Calculate the median value of the original sampling points within adjacent 1 second as the 1 second boundary data value;

(3)Voltage normalization

Because the collected voltage fluctuates, for example, the same data set shows the voltage changes from 180-250V, while the voltage in another data set changes in the range of 118-123V. The system must take into account the effects of these voltage fluctuations as they can significantly affect power consumption.

according to Normalize the voltage to the same fluctuation range, where:

Power _normalized represents the normalized power value, Power represents the measured power value, Voltagenominal represents the nominal voltage value, and Voltageobserved represents the measured voltage value.

(4)Top-k

Generally speaking, our identification system targets the large energy-consuming devices ranked in the top k (where k is an adjustable parameter) rather than all devices. There are three reasons. First, the top k power-consuming devices can already detect the overall Power consumption provides most of the reference information; secondly, these devices have the most significant characteristics, and the rest of the devices can be considered to only generate noise; thirdly, modeling and identifying a larger proportion of power-consuming devices will greatly improve the accuracy of the data. reliability.

During the data cleaning process, this system will also address other common issues with the data set, such as: equipment sensors not reporting readings, small data missing, removal of abnormal values such as observed voltages exceeding twice the rated voltage, and loss of main power data Data and more.

(5)Detection gap

Many algorithms today assume that the communication of each data acquisition device is continuous. However, the actual situation is that sometimes the data acquisition device is disconnected or malfunctions. If we set a parameter value, when a disconnection or failure occurs, If the fault time is greater than the set parameter value, it can be considered that there will be a gap in a continuous power data sample. For example, we calculate the difference between the timestamps of adjacent sampling points. If it is greater than a certain parameter such as 10 seconds, it is considered that there is a gap in the data set. Data sequences with gaps cannot be directly used in the system training set and test set. All training data and test data sequences must be data sequences without gaps in the middle.

(6)Uptime

Uptime is the total time recorded by the sensor. It is the duration of the last timestamp, minus the first timestamp, minus any existing gaps.

S130: Divide the cleaned historical electrical parameter data set of a single electrical appliance into an original training set, a verification set and a test set in proportion.

S140: Intercept data fragments from the original training set, verification set and test set along the timeline, and generate a training set, verification set and test set composed of data fragments.

For example, in the data cleaning module, we down-sample the electrical timing data to the specified frequency f according to the actual application requirements. Assuming that the frequency f corresponds to the interval period of 5 sampling points in the original sampling data, then the down-sampled active power data sequence It is P' _active = {p ₁ , p ₆ , p ₁₁ ,…p _5i+1 ,…}={q ₁ , q ₂ ,…q _i ,…}. The obtained relatively low-frequency time series data is then normalized and other preprocessing operations are performed to generate continuous time series data that can be directly used by the model for training. Since the power load characteristic identification model accepts multiple electrical parameter inputs at the same time, but the input dimension of a specific electrical parameter is limited and has a fixed length n, it is impossible to input all the continuous time series data of this electrical parameter at one time, so the length is The sliding window method of n continuously slides with a step size l. After each sliding, a data subsequence of fixed length n in the electrical parameter sequence is intercepted, for example, the active power subsequence P _i ={q _in+1 ,q _in+2 ,…q _in+n }, which is used as the input of the power load characteristic identification module.

S150: Establish a convolutional neural network model based on the denoising autoencoder for each target electrical appliance.

In the embodiment of the present invention, through the power load characteristic identification module designed by us, the electrical appliances in use can be identified by monitoring the power load data of the bus when the power load characteristics of the branch lines are unknown.

The design of the power load feature identification module is based on a deep neural network. It does not require complex artificially designed feature parameters to automatically obtain the feature parameters required to identify the power load. In this patent, data acquisition and data cleaning are intelligent distribution and consumption information. Preparatory work for load feature identification, the power load feature identification module is the key technology of this patent.

Think of the power load decomposition as a "noise reduction" task, in which the power of the target appliance is equivalent to the "clean" main signal, while the power generated by other appliances at the same time is the background "noise". The purpose of the task is to reduce the power of the target appliance to Separate from background "noise" power.

Build a deep neural network based on the concept of Denoising AutoEncoder (DAE), use convolutional layers for the first and last layer of the network, and use its position, scaling and distortion invariance to extract useful low-level features, such as 1000 The characteristic of a step change in watts is likely to be a useful feature regardless of where it occurs in the power segment.

The architecture and detailed parameters of the convolutional neural network model based on denoising autoencoders are shown in the following table:

Table 1 Architecture and parameters of the convolutional neural network model based on denoising autoencoder

In the power load feature identification module, such as learning rate, batch size, back propagation optimization model, dropout function, activation function selection, etc., different parameter combinations will bring different machine learning effects, and our system supports different By optimizing the parameter combination, there is no need to manually design the characteristic parameters of the electric load, and the type of electrical equipment can be accurately identified in the electric load characteristic identification module.

S160: For each target electrical appliance, train the convolutional neural network model based on the denoising autoencoder based on the training set, verification set and test set composed of data segments to obtain an optimized model for each target electrical appliance.

S170: Collect the current data of the user's electric load, input the current data into the optimization model of each target electrical appliance, isolate the working status of the electrical appliance, and output the category result of the electric load.

Different electrical appliances have separate corresponding deep neural network models in the load characteristic identification module. The data subsequences are simultaneously input into the deep neural network models corresponding to different electrical appliances. The n sampled values of each electrical parameter in the same time period correspond to the deep neural network models. The corresponding n nodes of the network input layer, through the forward propagation operation of the network, output the mth target electrical appliance operating status sequence S _m,i in this data subsequence = {s _m,in+1 ,s _m,in+2 , …s _m,in+n }. The length of the state sequence is n but is not limited to n and is determined by the model architecture. Based on the working status of each electrical appliance, the types of all electrical appliances in use in this sequence can be determined.

We input the basic electrical parameter data (including current, voltage, active power, reactive power) processed by the aforementioned power load acquisition and cleaning module into the power load identification module as input data. The goal is to infer each user through the electrical data. Equipment category for electrical equipment.

The data and identification methods selected are as follows:

Select the basic electrical parameter data of the total circuit (including current, voltage, active power, and reactive power);

Select the above data from 1 minute to 5 minutes, taking the sampling rate of 1Hz as an example, there will be 60 to 300 sampling points;

Process the above data into several sequences of 16 data points as one sequence, and eliminate sequences with gaps;

Load the pre-trained machine learning model program for power load identification, which can be implemented using any existing machine learning or neural network programming framework, such as tensor flow, keras, etc.);

Input the data sequence into the machine learning model, and the identification module outputs the current equipment category data corresponding to the sampling point at each moment (expressed by integers, such as 1 = air conditioner, 2 = kettle, 3 = washing machine, 4 = light, etc.).

We input the basic electrical parameter data processed by the aforementioned power load collection and cleaning module into the power load identification module, including current, voltage, active power, and reactive power as input data. The goal is to infer each electrical equipment through the electrical data. device category.

The data and identification methods selected are as follows:

Select the basic electrical parameter data of the total circuit, such as current, voltage, active power, and reactive power;

Select the above data from 1 minute to 5 minutes, taking the sampling rate of 1Hz as an example, there will be 60 to 300 sampling points;

Process the above data into several sequences of 16 data points as one sequence, and eliminate sequences with gaps;

Load the pre-trained machine learning model program for power load identification, which can be implemented using any existing machine learning or neural network programming framework, such as tensor flow, keras, etc.;

Input the data sequence into the machine learning model, and the identification module outputs the current equipment category data corresponding to the sampling point at each moment (expressed by integers, such as 1 = air conditioner, 2 = kettle, 3 = washing machine, 4 = light, etc.).

The power load identification method and system based on machine learning provided in this embodiment is specifically based on the measured electrical parameter data including current, voltage, active power and reactive power, etc., and unifies the basic electrical parameter data into a unified format. In our long-term Based on the extraction, collection, analysis, induction and training of electric load characteristics, it is possible to accurately determine the overall basic electrical parameter data of several electric loads for a period of time, including voltage, current, active power, reactive power, etc. Identify the type of appliance being used. Therefore, the machine learning model training method and system for power load identification provided by this application does not require manual adjustment of parameters. Compared with traditional methods such as time domain waveform matching, feature point matching and spectrum analysis, the matching accuracy is higher. The application can learn independently and automatically obtain the characteristic parameters needed to identify electric loads, thereby improving the applicable scope of the model and improving the accuracy of electric load identification.

In the second aspect, this application is based on the above-mentioned machine learning-based power load identification method. This application also provides a machine learning-based power load identification system, as shown in Figure 2. The system includes:

The data set acquisition module is used to obtain the historical electrical parameter data set of each electrical appliance;

A data set cleaning module, used to clean the historical electrical parameter data set of each electrical appliance;

The data set division module is used to divide the cleaned historical electrical parameter data set of a single electrical appliance into the original training set, verification set and test set in proportion;

A data set fragment interception module is used to intercept data fragments from the original training set, verification set and test set along the timeline, and generate a training set, verification set and test set composed of data fragments;

The model building module is used to establish a convolutional neural network model based on the denoising autoencoder for each target electrical appliance;

A model optimization module, used to train the convolutional neural network model based on the denoising autoencoder based on the training set, verification set and test set composed of data fragments for each target appliance to obtain an optimization model for each target appliance;

The electric load category output module is used to collect the current data of the user's electric load, input the current data into the optimization model of each target electrical appliance, isolate the working status of the electrical appliance, and output the category result of the electric load.

Since the above embodiments are described by reference and combination with other methods, different embodiments all have the same parts, and the same and similar parts between the various embodiments in this specification can be referred to each other. No further details will be given here.

Other embodiments of the present application will be readily apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of this application and include common knowledge or customary technical means in the technical field that are not disclosed in this application. . It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the application being indicated by the following claims.

The above-described embodiments of the present application do not limit the scope of protection of the present application.

Claims (8)

1. A power load identification method based on machine learning, characterized in that the method includes: Obtain the historical electrical parameter data set of each electrical appliance; Cleaning the historical electrical parameter data set of each electrical appliance; Divide the cleaned historical electrical parameter data set of a single electrical appliance into the original training set, verification set and test set in proportion; Intercept data fragments from the original training set, verification set and test set along the timeline to generate a training set, verification set and test set composed of data fragments; Establish a convolutional neural network model based on denoising autoencoder for each target electrical appliance; For each target electrical appliance, train the convolutional neural network model based on the denoising autoencoder based on the training set, verification set and test set composed of data segments to obtain an optimized model for each target electrical appliance; Collect the current data of the user's electric load, input the current data into the optimization model of each target electrical appliance, isolate the working status of the electrical appliance, and output the category results of the electric load; The cleaning of the historical electrical parameter data set of each electrical appliance includes: automatic screening and abnormal data cleaning, noise identification and separation, down-sampling, discard rate, data normalization, missing data compensation and processing, Top-k and Error data removal; The acquisition of historical electrical parameter data sets of each electrical appliance includes: Integrate and summarize today’s public data sets to obtain the first data set; Install an electric energy meter at the user's general incoming line end to obtain the electrical parameters of the overall and individual electrical loads in one or more spaces to obtain a second data set; A historical electrical parameter data set of each electrical appliance is obtained based on the first data set and the second data set. 2. The power load identification method based on machine learning according to claim 1, characterized in that, the original training set, the verification set and the test set are intercepted along the time axis to generate data fragments. The training set, validation set and test set include: Use a sliding window with a length of n and a moving step of 1 to intercept data fragments from the original training set, verification set and test set along the time axis direction to generate a training set, verification set and test set composed of data fragments. 3. The power load identification method based on machine learning according to claim 1, characterized in that it also includes the unification of data formats: according to Convert active power into a value between [0, 1], where: S[i] represents the sampled value, which is the instantaneous active power, C is the electric load type, is the sample data, s _c is the active power of the electric load c. 4. The power load identification method based on machine learning according to claim 1, further comprising down-sampling to a specified frequency: If the sampling rate is lower than 1Hz, it will be recorded at the original sampling rate; If the sampling rate is higher than 1Hz, downsample the sampling rate to 1Hz; Among them, downsampling the sampling rate to 1Hz includes: Use the value of the sampling point every 1 second and discard all other sampling values within 1 second; Calculate the average of the original sampling points within adjacent 1 second as the 1 second boundary data value; Calculate the median value of the original sampling points within adjacent 1 second as the 1 second boundary data value. 5. The power load identification method based on machine learning according to claim 1, characterized in that it also includes voltage normalization: According to Power =/> Power normalizes the voltage to the same fluctuation range, where: Power Represents the normalized power value, Power represents the measured power value, Voltage _nominal represents the nominal voltage value, and Voltage _observed represents the measured voltage value. 6. The power load identification method based on machine learning according to claim 1, characterized in that the cleaning of the historical electrical parameter data set of each electrical appliance further includes detecting gaps, normal operating time and identifying energy consumption rankings. The electric load of the first K positions, where k is an adjustable parameter. 7. The power load identification method based on machine learning according to claim 1, characterized in that, for each target electrical appliance, the training set, verification set and test set composed of data fragments are used to identify the power load based on noise reduction auto-encoding. The convolutional neural network model of the device is trained to obtain the optimization model of each target electrical appliance, including: Parameters of the convolutional neural network model based on the denoising autoencoder for training using a training set composed of target electrical appliance data segments; Verify and test the different models obtained in different training stages on the verification set composed of data fragments until the best effect is used as the corresponding model of the target appliance; Use the test set composed of data fragments to perform performance testing on the corresponding model of the target electrical appliance until the performance is optimal to obtain the optimized model of the target electrical appliance. 8. A power load identification system based on machine learning, characterized in that the system includes: The data set acquisition module is used to obtain the historical electrical parameter data set of each electrical appliance; the acquisition of the historical electrical parameter data set of each electrical appliance includes: integrating and summarizing current public data sets to obtain the first data set; in the user's total incoming line The end-installed electric energy meter obtains the electrical parameters of the overall and individual electrical loads in one or more spaces to obtain a second data set; and obtains historical electrical parameter data of each electrical appliance based on the first data set and the second data set. set; The data set cleaning module is used to clean the historical electrical parameter data set of each electrical appliance; the cleaning of the historical electrical parameter data set of each electrical appliance includes: automatic screening and abnormal data cleaning, realizing the identification and separation of noise, and down-sampling , discard rate, data normalization, missing data compensation and processing, Top-k and erroneous data elimination; The data set division module is used to divide the cleaned historical electrical parameter data set of a single electrical appliance into the original training set, verification set and test set in proportion; A data set fragment interception module is used to intercept data fragments from the original training set, verification set and test set along the timeline, and generate a training set, verification set and test set composed of data fragments; The model building module is used to establish a convolutional neural network model based on the denoising autoencoder for each target electrical appliance; A model optimization module, used to train the convolutional neural network model based on the denoising autoencoder based on the training set, verification set and test set composed of data fragments for each target appliance to obtain an optimization model for each target appliance; The electric load category output module is used to collect the current data of the user's electric load, input the current data into the optimization model of each target electrical appliance, isolate the working status of the electrical appliance, and output the category result of the electric load.