CN118713290A - A method and device for load regulation management and diversified interactive services - Google Patents

Abstract

The present invention discloses a method and device for load regulation management and diversified interactive services. First, core A collects data in real time through a communication interface, and sends the collected data to core B through an isolation unit; core B preprocesses the received data to obtain preprocessed data; core B stores the preprocessed data to form a preprocessed data set; core B calls the preprocessed data set and performs intelligent analysis on it to generate analysis results; core B generates a refined load control strategy based on the analysis results; core B displays the refined load control strategy through a user interface, and generates a personalized energy-saving optimization plan in combination with the energy consumption profile of each household. The present invention realizes efficient energy utilization and cost optimization through intelligent and refined load management. At the same time, it provides diversified user interactive services, improves system efficiency and service quality, ensures stable operation of the power grid, and helps energy conservation, emission reduction and the development of smart grids.

Description

Method and device for load regulation management and diversified interactive service

Technical Field

The invention relates to the technical field of air conditioner monitoring systems, in particular to a method and a device for load regulation management and diversified interactive service.

Background

The traditional energy management system mode can not meet the requirements of a novel energy system when monitoring energy sources such as electric power around a single main body, and a user side needs to adopt a big data information integration mode to carry out scientific and systematic comprehensive energy management, so that not only is the new energy consumption optimized and the energy utilization efficiency improved, but also the energy conservation of single equipment is converted to the intelligent and systematic directions. The user side needs to collect, monitor, predict and manage the data of each unit of the energy system, thereby laying a good foundation for high-quality energy utilization and service provision, and the purpose of building the power load center in each place is also achieved. The comprehensive user side needs to automatically adjust according to the power requirements of different stages by taking a power grid, a distributed power generation system, an energy storage system, an adjustable load and an electric vehicle charging pile as physical bases according to a dynamic energy price mechanism, so that the stable operation of the power system is ensured. The user needs to monitor and analyze the energy consumption data, and can timely detect the phenomena of energy waste, leakage and the like.

Existing energy management systems, when faced with the need for new energy systems, expose a series of problems:

Firstly, the traditional single-body management mode is limited to simple monitoring and management of electric power and other energy sources, and is difficult to cope with complex application scenes such as distributed energy sources, micro-grids, multi-energy complementation and the like, so that the integration and optimization capability of multiple energy sources is deficient.

Secondly, big data information on the user side is limited in integration and analysis capability, and multiple data sources such as intelligent ammeter, internet of things sensor, distributed energy equipment and the like cannot be effectively integrated, so that deep scientific analysis and systematic management of an energy system are hindered.

Moreover, under the background of large-scale access of new energy, the traditional system often has obvious short plates in the aspects of new energy consumption flexibility, overall energy utilization efficiency improvement, especially in the aspects of demand response, peak-valley balance and the like. Meanwhile, the energy management at the user side is still in a low development stage, and the management of all components of the energy system such as distributed power generation, energy storage, load management, electric vehicle charging facilities and the like cannot realize comprehensive, fine and collaborative optimization, so that the market demand change caused by a dynamic energy price mechanism is difficult to effectively realize.

In addition, the existing energy management system has low automation and intelligent level, and the user side is difficult to accurately predict, intelligently schedule and automatically adjust the energy consumption by utilizing advanced information technology and artificial intelligence technology, so that challenges are faced in ensuring the stable operation and efficient utilization of the power system. The energy consumption monitoring means and technology of the user side are lagged, so that real-time and comprehensive energy consumption monitoring cannot be realized, and the phenomena of energy waste, leakage and the like are rapidly identified and prevented, so that the realization of the fine management of energy use and the energy-saving and environment-friendly targets is seriously influenced.

Disclosure of Invention

The invention aims to solve the problems of low load regulation and control management efficiency, insufficient safety and user service individuation loss of a power system.

According to an aspect of the present application, there is provided a method for load regulation management and diversified interactive services, implemented based on a load regulation management system, the load regulation management system including an a core, a B core, and an isolation unit disposed between the a core and the B core, the method comprising the steps of:

s1, carrying out data acquisition on the core A in real time through a communication interface, and sending acquired data to the core B through an isolation unit;

S2, preprocessing the received data by the B core to obtain preprocessed data;

s3, storing the preprocessed data by the core B to form a preprocessed data set;

S4, the core B calls a preprocessing data set and carries out intelligent analysis on the preprocessing data set to generate an analysis result;

s5, generating a refined load control strategy by the core B according to the analysis result;

And S6, displaying the refined load control strategy through a user interface by the core B, and generating a personalized energy-saving optimization scheme by combining the energy consumption profile of each user.

According to one aspect of the present application, a device for load regulation management and diversified interaction services is provided, the device comprising an a core, an isolation unit and a B core;

The core A performs data acquisition on the intelligent ammeter, the Internet of things sensor and the intelligent equipment in real time through a communication interface, performs preprocessing on the acquired data, and transmits the preprocessed data to the core B through an isolation unit;

the isolation unit performs safety isolation and data exchange between the core A and the core B;

and the core B receives the data transmitted by the core A, performs data processing, data storage, data analysis and fine load control strategy generation, displays the data through a user interface, and combines the energy consumption profile of each household to generate a personalized energy-saving optimization scheme.

Advantageous effects

According to the intelligent mobile phone intelligent control system, working parameters of electrical equipment such as an air conditioner are flexibly controlled according to different time periods and power demand conditions, safety isolation between national network power consumption information acquisition and user side data processing and control strategy execution is ensured through the dual-function structure of the A core and the B core, power consumption data are prevented from being tampered by users, a user-friendly interactive interface is provided, and the users can conveniently and rapidly acquire power consumption information and set the control strategy like operating the intelligent mobile phone. In addition, the method integrates the internet of things, edge computing and machine learning technologies, realizes integration of real-time and fine energy data acquisition, analysis and intelligent regulation, and dynamically optimizes the control strategy according to history and real-time data by utilizing reinforcement learning and evolution optimization algorithm so as to be more suitable for actual demands.

Drawings

FIG. 1 is a flow chart of the present invention.

FIG. 2 is a diagram of a hardware device architecture of the present invention.

Fig. 3 is a schematic view of an a-core frame of the present invention.

Fig. 4 is a schematic view of a B core frame of the present invention.

FIG. 5 is a schematic diagram of an isolated functional unit framework of the present invention.

FIG. 6 is a diagram of a user interface according to an embodiment of the present invention.

Detailed Description

As shown in fig. 1, according to an aspect of the present application, there is provided a method of load regulation management and a diversified interactive service, implemented based on a load regulation management system including an a core, a B core, and an isolation unit disposed between the a core and the B core, the method comprising the steps of:

s1, carrying out data acquisition on the core A in real time through a communication interface, and sending acquired data to the core B through an isolation unit;

S2, preprocessing the received data by the B core to obtain preprocessed data;

s3, storing the preprocessed data by the core B to form a preprocessed data set;

S4, the core B calls a preprocessing data set and carries out intelligent analysis on the preprocessing data set to generate an analysis result;

s5, generating a refined load control strategy by the core B according to the analysis result;

And S6, displaying the refined load control strategy through a user interface by the core B, and generating a personalized energy-saving optimization scheme by combining the energy consumption profile of each user.

According to one aspect of the application, step S1 is further:

S11, the A core performs data acquisition on the intelligent ammeter, the Internet of things sensor and the intelligent equipment in real time through a communication interface; the data includes power data, environmental data, and device status data;

S12, carrying out preliminary processing on the acquired data by the core A, wherein the preliminary processing comprises outlier screening and time stamp adding;

and S13, the core A sends the data after preliminary processing to the core B through the isolation unit.

According to one aspect of the application, step S2 is further:

S21, checking the received data by the B core, and eliminating incomplete or illegal data;

s22, detecting and correcting abnormal values of the checked data by the core B according to the configured threshold rule;

S23, carrying out mapping coding on the corrected data by the B core, and converting the time data into a uniform time stamp format;

And S24, the B core performs feature extraction on the data which is subjected to mapping coding and conversion into a uniform timestamp format according to a load control strategy.

According to one aspect of the application, step S3 is further:

S31, the core B stores the preprocessed data into an InfluxDB time sequence database through an ETL tool at a second-level granularity to form a time sequence data set;

S32, storing the aggregated data into a MySQL relational database by the core B to form an aggregated data set;

S33, storing unstructured data into a MongoDB document database by the core B to form an unstructured data set;

And S34, integrating the time sequence data set, the aggregation data set and the unstructured data set by the core B to form a preprocessing data set.

According to one aspect of the application, step S4 is further:

S41, invoking time sequence data in a preprocessing data set by a core B, constructing and training a short-term load prediction model by using a time sequence prediction algorithm, and performing prediction analysis on the electricity consumption of a specified time period in the future by using the short-term load prediction model to obtain electricity consumption prediction data of the specified time period in the future;

S42, invoking time series data in the preprocessing data set by the core B, constructing an isolated forest model by using a local anomaly factor algorithm, and detecting anomaly values in power consumption data in real time by the isolated forest model to obtain anomaly detection data;

S43, invoking the aggregation data and the unstructured data in the preprocessing data set by the core B, carrying out electricity consumption behavior portraits on users through a K-Means clustering algorithm, and analyzing aiming at different types of users to obtain electricity consumption behavior portraits of the users.

According to one aspect of the present application, step S5 is further:

S51, the core B predicts the electricity consumption of each device in a future appointed time period by using a random forest or XGBoost algorithm according to electricity consumption prediction data, abnormality detection data and user electricity consumption behavior portrait data in the appointed time period, and combines electricity price steps and demand response rewards to construct a cost model;

s52, taking the minimum cost as an optimization target, adopting a particle swarm optimization PSO algorithm, and solving an optimal equipment start-stop combination scheme;

S53, using a model predictive control technology by the core B, and rolling and optimizing a refined load control strategy of N time periods in the future in real time according to the power consumption predictive data updated in real time and an optimal equipment start-stop combination scheme; wherein N is a natural number.

According to one aspect of the application, step 6 is further:

s61, calling aggregate data in the preprocessing data set by the core B, generating a telegram table, and displaying the telegram table and a refined load control strategy through a user interface;

S62, outputting a control instruction by the core B through a data terminal, and remotely controlling a controllable load device or setting an automatic optimization mode through an automatic control instruction;

and S63, the core B carries out comprehensive analysis and calculation according to the electricity consumption report and the control instruction by combining the energy consumption profile of each household to generate a personalized energy-saving optimization scheme, wherein the energy consumption profile of each household comprises an electric appliance list, the number of residents and electricity consumption preference of each household.

In this embodiment, first, the a core collects the electric energy data of the total meter and the branch meter from the smart electric meter through the interfaces such as RS485 and WIFI, collects the environmental data such as the temperature and humidity from the sensor of the internet of things, and collects the device state data from the smart device such as the air conditioner and the lighting lamp. And screening out obvious abnormal values of the acquired data, marking a time stamp, and transmitting the data to the core B through the isolation unit. And B core checks the received data, eliminates incomplete or illegal data, adopts median or average value replacement according to the configured threshold rule, and detects and corrects abnormal values of the digital data. The category type data such as the device status is map-encoded. The time data is converted into a unified time stamp format, and relevant characteristics such as electricity consumption peak time and periodicity rule are extracted aiming at a machine learning model.

And storing the preprocessed data into an InfluxDB time sequence database by an ETL tool at a second-level granularity so as to perform time sequence analysis. And storing the aggregated data, such as daily and monthly electricity, into a MySQL relational database for generating a report. Unstructured data, such as alarm logs, are stored in a MongoDB document database.

A short-term load prediction model was trained using Facebook propset or LSTM algorithms to predict the power usage for the next 1 hour, 1 day.

And (3) establishing an isolated forest model by using a local anomaly factor (LOF) algorithm, detecting an abnormal value in electricity utilization data in real time, and finding out abnormal conditions such as electric power leakage, equipment faults and the like.

The K-Means clustering algorithm is used for carrying out electricity behavior portraits on users, and the electricity behavior portraits are divided into early-onset electricity users, night cats and the like, so that personalized services can be conveniently provided.

Then, the power consumption of each device for a period of time in the future is predicted by using algorithms such as random forests, XGBoost and the like, and a cost model is built by combining electricity price steps, demand response rewards and the like. And taking the minimum cost as an optimization target, solving the optimal equipment start-stop combination by using a particle swarm optimization PSO algorithm, and carrying out fine scheduling on controllable loads such as an air conditioner, a charging pile and the like. And (3) using a Model Predictive Control (MPC) technology, and performing real-time rolling optimization on control strategies of N moments in the future according to load prediction results.

And finally, the user checks the electricity consumption analysis reports such as the electricity consumption of the present day and the month, the electricity consumption ratio of the equipment and the like through the touch screen interface of the B core. The user can also carry out remote centralized control on equipment such as an air conditioner, a charging pile and the like through the APP, or set an automatic optimization mode. The system provides a personalized energy-saving optimization scheme according to the electric appliance list, the number of residents, the electricity preference and the like of the user.

The method integrates the internet of things, edge calculation and machine learning technologies, and realizes the integration of load fine acquisition, real-time analysis and intelligent regulation. And applying front-edge algorithms such as reinforcement learning, evolution optimization and the like, continuously learning and optimizing on the basis of massive historical data and real-time data, and generating a self-adaptive optimal control strategy.

The scheme is expanded into comprehensive energy service from single power Demand Side Management (DSM), and covers energy varieties such as electricity, heat, cold, water and the like, and has huge market prospect. Through the large-scale popularization and application to industrial parks, commercial buildings and residential communities, the efficiency of an energy system can be remarkably improved, and the carbon emission is reduced.

According to one aspect of the present application, step S13 is further:

s131, establishing encryption connection between the isolation unit and the core A, wherein a transmission scheduler is configured in the isolation unit;

s131a, the isolation unit receives a data transmission request initiated by the A core and sends a data packet to the A core;

s131b, after receiving a data transmission request, the isolation unit establishes an encryption channel and a buffer pool, and allocates a session ID for the transmission;

s131c, the isolation unit dynamically generates a disposable session key for the session, and returns the session ID and the session key to the A core;

s132, after encryption connection is established, the A core encrypts data and caches the data to a cache pool of the isolation unit;

s132a, the core A encrypts the subsequent data packet by using the session key and sends ciphertext data and the session ID to the isolation unit;

s132b, the isolation unit performs session ID verification, and decrypts the ciphertext data by using the session key to obtain plaintext data;

S132c, the isolation unit allocates a transmission channel for the decrypted plaintext data based on the priority of the data and the load balancing strategy;

S132d, the isolation unit caches the clear text data to a cache pool, and waits for transmission scheduling;

s133, the isolation unit transmits the plaintext data in the cache pool;

s133a, a transmission scheduler of the isolation unit inquires the transmission state of each channel in real time;

s133b, selecting an idle channel by a transmission scheduler, and calling plaintext data with highest priority from a cache pool;

s133c, using an encryption channel to send the plaintext data to the B core;

S133d, core B receives the plaintext data sent by the isolation unit and replies acknowledgement character ACK to the isolation unit;

s134, after the data transmission is finished, the isolation unit confirms and destroys the data;

s134a, the isolation unit receives an ACK reply of the acknowledgement character of the B core, and confirms that the data transmission is successful;

S134b, immediately destroying the cached plaintext data copy and a session key corresponding to the plaintext data copy by the isolation unit; the isolation unit releases the buffer space and channel resources occupied by the plaintext data copy;

S135, connection termination and resource recovery;

S135a, after all data transmission is completed, the A core sends a connection termination request to the isolation unit;

S135b, the isolation unit recovers the session ID and closes the encryption channel;

s135c, the isolation unit performs destructive cleaning on the encryption channel to ensure that no data residue exists in the encryption channel.

In a further embodiment, the information interaction process of the isolation functional unit specifically includes:

S131, establishing connection:

S131a, the isolation unit receives a data transmission request initiated by the A side and sends a first data packet.

S131b, after the isolation unit receives the request, an encryption channel is established, and a session ID is allocated for the transmission.

S131c, the isolation unit dynamically generates a one-time session key for the session and returns the session ID and the key to the A side.

S132, data encryption:

S132a, A side encrypts the subsequent packet by using the session key, and sends the ciphertext data together with the session ID to the quarantine unit.

S132b, the quarantine unit verifies the session ID and decrypts the data using the session key.

And S132c, the isolation unit allocates a transmission channel for the decrypted data based on strategies such as data priority, load balancing and the like.

S132d, the isolation unit caches the clear text data and waits for transmission scheduling.

S133, data transmission:

s133a, a transmission scheduler of the isolation unit inquires the transmission state of each channel in real time.

S133b, the scheduler picks the idle channel and takes out the data with the highest priority from the cache.

And S133c, the dispatcher sends the data to the B test by using the encrypted channel.

And S133d, receiving data by the B side, demodulating the plaintext data, and replying acknowledgement character ACK to the isolation unit.

S134, data destruction:

s134a, the isolation unit receives the ACK reply of the acknowledgement character of the B side, and confirms that the data transmission is successful.

S134b, the isolation unit immediately destroys the plaintext data copy in the cache.

And S134c, the isolation unit immediately destroys the session key corresponding to the data.

S134d, the isolation unit releases the buffer space and channel resources occupied by the data.

S135, connection termination:

and S135a, after all data transmission is completed, the A side sends a connection termination request to the isolation unit.

S135b, the isolation unit recovers the session ID and closes the encryption channel.

And S135c, the isolation unit performs destructive cleaning on the encrypted channel to ensure that no data residue is left.

According to the embodiment, through the one-time pad scheme and dynamic key management, data security is guaranteed, and the data are destroyed immediately after being used, so that both security and efficiency are achieved. The transmission bottleneck of the isolation unit is broken through the cooperative optimization of the softness and the hardness, so that the safety and the performance are not contradicted. In a word, the embodiment not only solves the potential safety hazard of the data isolation unit, but also breaks through the efficiency bottleneck of the traditional isolation scheme, and realizes important leaps in the aspects of confidentiality, integrity, availability and the like of data interaction.

According to one aspect of the application, step S41 further comprises constructing a short-term load prediction model based on the attention mechanism and the countermeasure generation network, comprising the following specific steps:

S41a, analyzing time series data in a preprocessing data set by using an attention mechanism, and highlighting key influence factors by dynamically adjusting the weight of each time step;

S41b, adopting a generator network in a short-term load prediction model to carry out data enhancement, and synthesizing a predetermined amount of vivid load curve samples;

S41c, performing countermeasure training on the load curve sample and the preliminary prediction result through a discriminator network in the short-term load prediction model, and transmitting a feedback signal to the short-term load prediction model for iterative updating;

S41d, using a trained short-term load prediction model to conduct prediction analysis on the electricity consumption of a specified time period in the future.

In some embodiments, steps S4 and S5 may also be:

S4a, constructing a short-term load prediction model:

The weight of each time step in the time sequence data is dynamically adjusted by using an attention mechanism, key influence factors are highlighted, and the prediction precision is improved;

Adopting a generator network in a short-term load prediction model to carry out data enhancement, synthesizing a predetermined amount of vivid load curve samples, and relieving the data sparseness problem;

And (3) performing countermeasure training on the prediction result through a discriminator network in the short-term load prediction model, forcing the short-term load prediction model to be continuously self-perfected, and generating a more accurate and more robust prediction value.

S4b, an abnormality detection model:

and constructing an anomaly detection algorithm for LSTM-CNN-WT multichannel fusion, and extracting data features from the time dimension, the space dimension and the frequency dimension.

And carrying out self-adaptive weighted fusion on the characteristics of each dimension, training an abnormal scoring model, and giving an abnormal probability value to each piece of data.

Cross-verifying on the basis of anomaly scores by using an isolated forest and a single classification SVM algorithm to realize anomaly judgment with high confidence;

In the time dimension, the dynamic change trend of the load is modeled by using a long and short term memory network (LSTM).

In the spatial dimension, a Convolutional Neural Network (CNN) is utilized to capture the spatial correlation between each region and each device.

In the frequency dimension, the periodic characteristics of the load curve are described from a time-frequency domain perspective using Wavelet Transform (WT).

S4c, a user portrait clustering model:

a multi-mode self-learning user portrait clustering algorithm is constructed, structured data and unstructured data are fused, and user portraits are automatically learned.

The feature combination and importance ranking is performed on structured data such as electricity consumption, electricity fees and the like by using a XGBoost or LightGBM model GBDT.

Feature embedding is performed on unstructured data, such as power utilization time series, using an unsupervised representation learning algorithm, such as Word2Vec, seq2Seq, etc.

The structured features and unstructured embedments are stitched into a unified high-dimensional user representation vector.

And (3) using a dimension reduction algorithm such as t-SNE and the like to realize visualization on the user portrait vector so as to reveal the similarity structure among users.

And combining the user vectors subjected to dimension reduction by using a plurality of clustering algorithms such as spectral clustering, DBSCAN and the like, and adaptively finding out the optimal division of the user group.

In other embodiments, the specific steps of step S5 policy optimization are as follows:

s5a, a device-level optimization control model:

and constructing a hierarchical cascading multi-agent reinforcement learning optimization model to realize the cooperation of global optimization and local optimization.

The whole system is divided into a plurality of layers, such as a power transmission and distribution control layer, a building micro-grid control layer, a device end control layer and the like.

An agent is arranged at each level and is responsible for optimizing the cost benefits of the level, such as peak load reduction, operation efficiency, fault risk and the like.

And the intelligent agents of all levels are coupled through objective functions and constraint conditions to form a layered game problem.

And using MADDPG reinforcement learning algorithm to solve Nash equilibrium points of the layered game, and realizing dynamic equilibrium of benefits of each level.

And decoding the game balancing strategy into control instructions of each level, such as transformer tap adjustment, energy storage charge and discharge control, air conditioner temperature setting and the like, so as to realize hierarchical cascading optimization control of the system.

S5b, a demand side response optimization model:

And organizing massive small-micro-demand side resources into flexibly-scheduled virtual power plants through a hybrid agent group game demand response mechanism.

Modeling each demand side resource as an autonomous agent, and endowing the autonomous agent with the properties such as utility functions, strategy functions, learning algorithms and the like.

Various demand response tasks such as peak clipping and valley filling, frequency adjustment, standby capacity and the like are designed, and a task pool is formed for real-time release.

And each demand side agent dynamically forms a alliance by adopting game rules similar to an auction mechanism to compete and claim tasks according to the self utility preference.

The winning alliance completes the claimed demand response task by using respective adjustable load, distributed power supply, energy storage and other resources through the cooperation of internal parts.

The alliance internally adopts cooperative game equilibrium theory such as Shapley value and the like to fairly distribute the obtained task benefits, and adjusts respective strategy functions accordingly, so as to learn and seek a new round of optimal alliance combination.

S5c, energy management optimization model:

and constructing an ecological niche segmentation energy flow diagram optimization model, and realizing optimal scheduling of energy conversion and storage from the ecological perspective.

Referring to the food chain structure of the ecosystem, the energy utilization devices inside the building are abstracted into multi-nutrient level energy flow diagrams. Each nutritional level corresponds to a type of energy modality, such as electrical energy, thermal energy (high temperature), thermal energy (low temperature), mechanical energy, chemical energy, and the like. Various energy devices such as photovoltaics, gas turbines, heat pumps, storage batteries and the like are regarded as 'species' in the drawings and occupy specific ecological niches.

The energy exchange of the species in the ecological niche is characterized by a glancing food coefficient matrix, and corresponds to the energy conversion efficiency in reality.

And using Ecopath model, adopting Lotka-Volterra population competition equation to solve steady-state equilibrium solution of the complex ecological system, and obtaining optimal operation combination of various energy devices.

Based on Ecopath static description, a Ecosim dynamic simulation mechanism can be introduced to simulate the dynamic evolution track of the ecological system under the disturbance of the time-varying environment, so that the real-time rolling optimization of energy management is realized.

And solving a dynamic balance equation of the food network by taking the total carbon emission and the fossil energy consumption of the whole park as control targets to obtain an optimal operation combination strategy of various energy devices under different environmental conditions such as illumination, temperature, humidity, people flow and the like.

The optimal strategy is converted into a control instruction which can be executed by the equipment, and the control instruction is issued in real time through an automation platform such as an Energy Management System (EMS), an energy storage management system (BEMS) and the like. Meanwhile, the running state of the equipment is fed back to the energy ecological model in real time, model parameters are corrected online, and the environment fluctuation change is continuously adapted.

In the embodiment, key influence factors can be more accurately mined by introducing a attention mechanism and a load prediction model of the GAN, and vivid data are generated by utilizing small samples, so that the prediction precision and robustness are greatly improved. By adopting the multi-view fusion anomaly detection algorithm, the running mode of the energy system can be described from multiple dimensions such as time, space, frequency and the like, various anomaly conditions can be found in time, and the anomaly conditions can be accurately positioned. The multi-mode self-learning user portrait clustering algorithm is adopted, so that the user can fully go deep into the utilization preferences of the park tenants, and powerful support is provided for personalized services. The multi-agent reinforcement learning optimization model based on layered cascade realizes the maximization of the whole benefit by cross-layer collaborative optimization while reducing the running cost and considering the benefits of all parties. The enthusiasm of the tail users with equal length of middle and small merchants can be fully mobilized through the demand response mechanism of the hybrid agent group game, fragmented energy resources are excavated, and a flexible self-adaptive virtual power plant is constructed. The energy management optimization model of the ecological niche segmentation theory is used for consulting the energy chain from an ecological perspective, so that the energy configuration is dynamically optimized, and meanwhile, the environmental influence is minimized.

As shown in fig. 2, according to an aspect of the present application, there is further provided an apparatus for load regulation management and diversified interaction services, the apparatus including an a core, an isolation unit, and a B core;

The core A performs data acquisition on the intelligent ammeter, the Internet of things sensor and the intelligent equipment in real time through a communication interface, performs preprocessing on the acquired data, and transmits the preprocessed data to the core B through an isolation unit;

the isolation unit performs safety isolation and data exchange between the core A and the core B;

and the core B receives the data transmitted by the core A, performs data processing, data storage, data analysis and fine load control strategy generation, displays the data through a user interface, and combines the energy consumption profile of each household to generate a personalized energy-saving optimization scheme.

In some embodiments, the a-core is primarily responsible for data acquisition and edge computation. And various sensors and intelligent equipment are connected through interfaces such as RS485 and WIFI, and data such as electric energy, environment and equipment state are collected in real time. The a core may also perform preliminary processing, such as verification, time stamping, etc., on the acquired data.

The isolation unit is used for carrying out safe isolation and data exchange between the A core and the B core. The isolation unit ensures the isolation of the two security domains of electricity collection and edge control, prevents B core faults from affecting A core normal collection, and also avoids users or external factors from interfering with the collection process.

Regarding the manner of isolation: the A core and the B core belong to two independent security domains, and are communicated through a special isolation unit. The isolation unit adopts technologies such as photoelectric isolation and electromagnetic shielding to prevent electric interference and signal leakage between the two chips. The communication interfaces between the A core and the isolation unit, and between the isolation unit and the B core, such as SPI, I2C, and the like, are all provided with independent physical layer protection circuits, such as TVS tubes, ESD protection devices, and the like, so as to prevent static electricity and surge from damaging the interface circuits. External memories such as NAND FLASH of the A core and the B core use a special encryption chip (such as SM4 encryption chip) to encrypt the stored data, preventing the memory from being directly read. The isolation unit integrates a TPM trusted platform module, and utilizes a built-in cryptographic algorithm and safe storage to carry out multi-factor authentication on the A core and the B core, and only the authenticated chip allows data to be transmitted through the isolation unit.

B core: carrying the main data processing and application service logic. After data is transmitted from the A core to the B core through the isolation unit, the data is subjected to pretreatment, storage, analysis, optimization and other processes in the B core. The software architecture of the B core is divided into two parts, namely a data layer and a service layer.

Data layer: the method comprises modules of data preprocessing, storage, analysis and the like, and corresponds to the steps S2-S4 in the scheme before me. The layer refines the data value by utilizing machine learning and big data technology.

Service layer: the method comprises modules such as strategy optimization and interactive service, and corresponds to the steps S5-S6 in the scheme. The layer converts the analysis result of the data layer into a business strategy and service which can be landed, and the business strategy and service are delivered to an end user through channels such as a Web front end or a mobile APP.

Communication link: the A core communicates with the data layer of the B core through the isolation unit, and the acquired data is transmitted. The service layer of the B core interacts with external user terminals through standard communication links such as 4G/5G, ethernet and the like.

Thus, the software architecture and hardware of the whole system form an organic whole body with compact fusion and clear layers. Data flows, is derived, aggregated between the various tiers, driving various business applications and intelligent services. The A core and the isolation unit are responsible for safe and reliable data input, and the B core is responsible for maximum mining and conversion output of data value.

On the basis of the A core and the B core of hardware, a layered software architecture is constructed:

Data acquisition layer: and the system is responsible for collecting original data from various internet of things sensors, intelligent electric meters, power equipment and the like.

Data preprocessing layer: and (3) preprocessing such as cleaning, normalizing, extracting features and the like is carried out on the collected original data, so that preparation is carried out for subsequent analysis.

Data storage layer: and storing the preprocessed data into a time sequence database, a relational database, a document database and the like, so as to realize the persistent storage and the efficient query of the data.

Data analysis layer: and carrying out intelligent analysis on the electricity consumption data such as load prediction, anomaly detection, user image and the like by using a machine learning algorithm such as a random forest, a support vector machine and the like.

Policy optimization layer: based on the analysis result, the intelligent optimization technologies such as reinforcement learning, evolutionary algorithm and the like are applied in combination with external signals such as electricity price policies, demand response and the like to generate a refined load control strategy.

Service application layer: and diversified interactive services such as electricity analysis reports, intelligent household appliance control, energy efficiency optimization suggestions and the like are provided for users.

According to another aspect of the application, a device for load management refinement and interactive service diversification, a hardware structure comprises an A core functional area, a B core functional area, an isolation functional unit and a power supply unit. The power supply unit supplies power to the whole system and comprises an AC-DC power supply and a backup power supply, wherein the backup power supply consists of a super capacitor and a battery; the A core functional area consists of an A core minimum system, a network port, bluetooth, USB, a radio station unit and a plurality of functional modules, and the functional modules can be used in a plurality of combinations according to the actual requirements of users; the B core functional area consists of a B core minimum system, a network port, WIFI, a touch screen, a plurality of LED indicator lamps, a camera, a microphone, a loudspeaker and the like; the isolation functional unit can realize the functions of safety isolation and information exchange of the safety domains of the two functional areas of the core A and the core B.

The power supply unit consists of an AC-DC power supply and a backup power supply, and the backup power supply consists of a super capacitor and a battery. After the AC-DC main power supply is insufficient or disappears, the backup power supply can be automatically switched on, the continuous normal working time of the device is kept to be not less than three minutes, and the outage information can be reported to the main station platform. The AC-DC power supply supports single-phase-crossing direct-current self-adaptive power supply input, and the power supply voltage is 80-264V. The charging time of the backup power supply is not more than 2 hours, and the maintenance-free time of the super capacitor is not less than 10 years.

The A core functional area consists of an A core minimum system, a network port, bluetooth, USB, a radio station unit and a plurality of functional modules, and the functional modules can be used in a plurality of combinations according to the actual requirements of users. The A-core minimum system consists of a processor, a memory, a power supply management, a watchdog, a hardware encryption chip and the like. The A core can realize the electricity consumption information acquisition requirement of the national network electric power center.

The B core functional area is composed of a B core minimum system, a network port, WIFI, a touch screen, a plurality of LED indicator lamps, a camera, a microphone, a loudspeaker and the like. The minimum system of the B core consists of a processor, a memory, a power supply management, a watchdog, a recovery trigger button and the like. And the core B adopts an android operating system, corresponding demand service APP is installed according to actual demands, the use of common users is facilitated, and the diversification of interaction with the users can be realized based on touch screen interaction.

The isolation functional unit can realize the functions of safety isolation and information exchange of the safety domains of the two functional areas of the core A and the core B.

As shown in fig. 3, in this embodiment, the hardware block diagram of the a-core minimum system mainly includes a processor, a memory, a power management, a watchdog, a hardware encryption chip, and the like. The A core processor selects 4 cores and 1.0GHz main frequency, the memory capacity is 1GB, and the data storage area capacity is 8GB. Considering system reliability, the a-core configures 64kb FRAM memory and 512MB SLC NAND memory. The FRAM memory is used for backup storage of important and key parameters; the SLC NAND memory is used for storing a starting and guiding program, an operating system and application software. In addition, this storage is also a backup storage area for critical parameters and important data. When in use, the boot area should be protected, and unauthorized read-write operations are prohibited. Considering the read-write lifetime, the EMMC memory should be set to the enhancement mode pSLC first when in use, increasing the read-write lifetime. The A-core processing system should be configured with watchdog circuitry and should be able to restart the system when the system is in an abnormal state.

As shown in FIG. 4, the B-core minimum system hardware block diagram mainly comprises a processor, a memory, power management, a watchdog and a recovery trigger button. The A core processor selects 4 cores, 1.5GHz main frequency and 8GB of memory capacity. To increase the stability and abnormal self-recovery capability of the system, the B-core system should be configured with dual EMMC memory while additionally configured with SLC NAND, with EMMC1 memory capacity 32gb, EMMC2 memory capacity 8gb, and SLC NAND memory capacity 512MB. The EMMC1 is used for storing the operating system, the application program and the data when the system normally operates, the EMMC2 can only be used for storing the backup operating system and the application program, and the operating system and the application program of the EMMC1 can be restored through the backup when the system operates abnormally. SLC NAND is used to store system boot programs and UBOOT, and is the boot storage medium for the system. In addition, this storage is also a backup storage area for critical parameters and important data. When in use, the boot area and the parameter data backup area should be isolated, the boot area should be strictly protected, and unauthorized read-write operations should be prohibited. The recovery trigger key is used for triggering recovery operation at the terminal side, and when the key is pressed, the terminal can start a recovery function on the premise of having recovery conditions. The restoration triggering key should be arranged in the terminal seal area, and the key is protected by the terminal seal. Considering the read-write lifetime, the EMMC memory should be set to the enhancement mode pSLC first when in use, increasing the read-write lifetime. The B-core processing system configures a watchdog circuit that should be able to restart the system when the system is in an abnormal state.

As shown in fig. 5, each communication channel between the a core and the B core has two isolation modes of physical isolation and protocol isolation. The A core and the B core can be communicated through a network port connection, the network port connection is provided with a hardware cut-off switch, and any one of the A core and the B core can cut off the physical connection through a control signal. The A core and the B core can communicate through two independent SPI interfaces, when data flow flows from the A core to the B core, the two cores communicate through SPI-1 communication ports, at the moment, the A core SPI-1 belongs to a master end, the B core SPI-1 belongs to a slave end, and the A core and the B core can unilaterally cut off physical connection of an SPI-1 communication channel through control signals. When data flow flows from the core B to the core A, the two cores communicate through the SPI-2 communication port, at the moment, the core B SPI-2 belongs to the master end, the core A SPI-2 belongs to the slave end, and the core A and the core B can unilaterally cut off the physical connection of the SPI-2 communication channel through control signals.

As shown in fig. 6, in the example of the user operation interface, the operation of obtaining the electricity consumption of each electric device such as an air conditioner, illumination, charging pile, photovoltaic and the like is simple and clear, and the operation of issuing the control strategy instruction is simple.

The invention has the advantages that: low cost and high cost performance, and is a novel power load management system. Fine automatic adjustment, which is performed according to the power requirements of different stages, for example: the response stage can realize automatic adjustment like temperature, operation frequency and short wind speed of air-conditioning equipment, and the influence on the body feeling of a user is minimized while the effective voltage drop and the electric load are achieved. Data security: the system has the advantages that the absolute isolation between the safe environment for electricity collection and the unsafe environment at the user side is realized between the electricity collection of the A-core functional national network and the data arrangement and control strategy function at the user side of the B-core functional national network, and the data of the electricity collection end cannot be tampered by the user, so that the double requirements of users such as government, hotel and enterprises on data safety and information acquisition are met. User-friendly interactions: based on the android operating system, a 10.1-inch capacitive touch screen is adopted, so that the threshold used by a user is reduced, and the graphical interface is operated to acquire electricity information and issue control strategy instructions as simple as operating a personal mobile phone.

The preferred embodiments of the present invention have been described in detail above, but the present invention is not limited to the specific details of the above embodiments, and various equivalent changes can be made to the technical solution of the present invention within the scope of the technical concept of the present invention, and all the equivalent changes belong to the protection scope of the present invention.

Claims (10)

1. A method for load regulation management and diversified interaction service, based on the implementation of a load regulation management system, the load regulation management system comprises an A core, a B core and an isolation unit arranged between the A core and the B core, and the method is characterized by comprising the following steps:

s1, carrying out data acquisition on the core A in real time through a communication interface, and sending acquired data to the core B through an isolation unit;

S2, preprocessing the received data by the B core to obtain preprocessed data;

s3, storing the preprocessed data by the core B to form a preprocessed data set;

S4, the core B calls a preprocessing data set and carries out intelligent analysis on the preprocessing data set to generate an analysis result;

s5, generating a refined load control strategy by the core B according to the analysis result;

And S6, displaying the refined load control strategy through a user interface by the core B, and generating a personalized energy-saving optimization scheme by combining the energy consumption profile of each user.

2. The method of load control management and diversification interaction service according to claim 1, wherein step S1 further comprises:

S11, the A core performs data acquisition on the intelligent ammeter, the Internet of things sensor and the intelligent equipment in real time through a communication interface; the data includes power data, environmental data, and device status data;

S12, carrying out preliminary processing on the acquired data by the core A, wherein the preliminary processing comprises outlier screening and time stamp adding;

and S13, the core A sends the data after preliminary processing to the core B through the isolation unit.

3. The method of load control management and diversification interaction service according to claim 2, wherein step S2 further comprises:

S21, checking the received data by the B core, and eliminating incomplete or illegal data;

s22, detecting and correcting abnormal values of the checked data by the core B according to the configured threshold rule;

S23, carrying out mapping coding on the corrected data by the B core, and converting the time data into a uniform time stamp format;

And S24, the B core performs feature extraction on the data which is subjected to mapping coding and conversion into a uniform timestamp format according to a load control strategy.

4. The method for load control management and diversification of interactive services according to claim 3, wherein step S3 further comprises:

S31, the core B stores the preprocessed data into an InfluxDB time sequence database through an ETL tool at a second-level granularity to form a time sequence data set;

S32, storing the aggregated data into a MySQL relational database by the core B to form an aggregated data set;

S33, storing unstructured data into a MongoDB document database by the core B to form an unstructured data set;

And S34, integrating the time sequence data set, the aggregation data set and the unstructured data set by the core B to form a preprocessing data set.

5. The method of load control management and diversification interaction service according to claim 4, wherein step S4 further comprises:

S41, invoking time sequence data in a preprocessing data set by a core B, constructing and training a short-term load prediction model by using a time sequence prediction algorithm, and performing prediction analysis on the electricity consumption of a specified time period in the future by using the short-term load prediction model to obtain electricity consumption prediction data of the specified time period in the future;

S42, invoking time series data in the preprocessing data set by the core B, constructing an isolated forest model by using a local anomaly factor algorithm, and detecting anomaly values in power consumption data in real time by the isolated forest model to obtain anomaly detection data;

S43, invoking the aggregation data and the unstructured data in the preprocessing data set by the core B, carrying out electricity consumption behavior portraits on users through a K-Means clustering algorithm, and analyzing aiming at different types of users to obtain electricity consumption behavior portraits of the users.

6. The method of load control management and diversification interaction service according to claim 5, wherein step S5 further comprises:

S51, the core B predicts the electricity consumption of each device in a future appointed time period by using a random forest or XGBoost algorithm according to electricity consumption prediction data, abnormality detection data and user electricity consumption behavior portrait data in the appointed time period, and combines electricity price steps and demand response rewards to construct a cost model;

s52, taking the minimum cost as an optimization target, adopting a particle swarm optimization PSO algorithm, and solving an optimal equipment start-stop combination scheme;

S53, using a model predictive control technology by the core B, and rolling and optimizing a refined load control strategy of N time periods in the future in real time according to the power consumption predictive data updated in real time and an optimal equipment start-stop combination scheme; wherein N is a natural number.

7. The method of load control management and diversification of interactive services according to claim 6, wherein step 6 is further:

s61, calling aggregate data in the preprocessing data set by the core B, generating a telegram table, and displaying the telegram table and a refined load control strategy through a user interface;

S62, outputting a control instruction by the core B through a data terminal, and remotely controlling a controllable load device or setting an automatic optimization mode through an automatic control instruction;

and S63, the core B carries out comprehensive analysis and calculation according to the electricity consumption report and the control instruction by combining the energy consumption profile of each household to generate a personalized energy-saving optimization scheme, wherein the energy consumption profile of each household comprises an electric appliance list, the number of residents and electricity consumption preference of each household.

8. The method of load control management and diversification interaction service according to claim 7, wherein step S13 further comprises:

s131, establishing encryption connection between the isolation unit and the core A, wherein a transmission scheduler is configured in the isolation unit;

s131a, the isolation unit receives a data transmission request initiated by the A core and sends a data packet to the A core;

s131b, after receiving a data transmission request, the isolation unit establishes an encryption channel and a buffer pool, and allocates a session ID for the transmission;

s131c, the isolation unit dynamically generates a disposable session key for the session, and returns the session ID and the session key to the core A;

s132, after encryption connection is established, the A core encrypts data and caches the data to a cache pool of the isolation unit;

s132a, the core A encrypts the subsequent data packet by using the session key and sends ciphertext data and the session ID to the isolation unit;

s132b, the isolation unit performs session ID verification, and decrypts the ciphertext data by using the session key to obtain plaintext data;

S132c, the isolation unit allocates a transmission channel for the decrypted plaintext data based on the priority of the data and the load balancing strategy;

S132d, the isolation unit caches the clear text data to a cache pool, and waits for transmission scheduling;

s133, the isolation unit transmits the plaintext data in the cache pool;

s133a, a transmission scheduler of the isolation unit inquires the transmission state of each channel in real time;

s133b, selecting an idle channel by a transmission scheduler, and calling plaintext data with highest priority from a cache pool;

s133c, using an encryption channel to send the plaintext data to the B core;

S133d, core B receives the plaintext data sent by the isolation unit and replies acknowledgement character ACK to the isolation unit;

s134, after the data transmission is finished, the isolation unit confirms and destroys the data;

s134a, the isolation unit receives an ACK reply of the acknowledgement character of the B core, and confirms that the data transmission is successful;

S134b, immediately destroying the cached plaintext data copy and a session key corresponding to the plaintext data copy by the isolation unit; the isolation unit releases the buffer space and channel resources occupied by the plaintext data copy;

S135, connection termination and resource recovery;

S135a, after all data transmission is completed, the A core sends a connection termination request to the isolation unit;

S135b, the isolation unit recovers the session ID and closes the encryption channel;

s135c, the isolation unit performs destructive cleaning on the encryption channel to ensure that no data residue exists in the encryption channel.

9. The method of load regulation management and diversification of interactive services according to claim 7, wherein step S41 further comprises constructing a short-term load prediction model based on the attention mechanism and the countermeasure generation network, and comprises the following steps:

S41a, analyzing time series data in a preprocessing data set by using an attention mechanism, and highlighting key influence factors by dynamically adjusting the weight of each time step;

S41b, adopting a generator network in a short-term load prediction model to carry out data enhancement, and synthesizing a predetermined amount of vivid load curve samples;

S41c, performing countermeasure training on the load curve sample and the preliminary prediction result through a discriminator network in the short-term load prediction model, and transmitting a feedback signal to the short-term load prediction model for iterative updating;

s41d, using a trained short-term load prediction model to conduct prediction analysis on the electricity consumption of the appointed time period in the future, and obtaining electricity consumption prediction data of the appointed time period in the future.

10. An apparatus for load regulation management and diversification interactive service based on any one of claims 1-9, wherein the apparatus comprises an a core, an isolation unit, and a B core;

The core A performs data acquisition on the intelligent ammeter, the Internet of things sensor and the intelligent equipment in real time through a communication interface, performs preprocessing on the acquired data, and transmits the preprocessed data to the core B through an isolation unit;

the isolation unit performs safety isolation and data exchange between the core A and the core B;

and the core B receives the data transmitted by the core A, performs data processing, data storage, data analysis and fine load control strategy generation, displays the data through a user interface, and combines the energy consumption profile of each household to generate a personalized energy-saving optimization scheme.