Virtual power plant resource scheduling method and system

Abstract

The invention relates to the technical field of energy internet and intelligent power grids, in particular to a virtual power plant resource scheduling method and system. According to the virtual power plant resource scheduling method, a cloud-edge architecture is adopted, the architecture comprises a cloud platform and an edge layer which communicate with each other, and the cloud platform carries a photovoltaic prediction model, an energy storage charging and discharging model, a load response model and a cooperative regulation and control model. Based on an optimization decision-making method of four types of core models and a closed-loop regulation and control mechanism, time sequence data and structured data of a base-layer equipment end are collected through an edge computing node, subjected to primary processing and transmitted to a cloud platform, and the cloud platform judges an original strategy of the edge computing node and then corrects the original strategy to generate a precise regulation and control instruction. Therefore, the overall data processing speed and the regulation and control precision are improved, and the technical problem that the resource scheduling precision is low when an existing scheduling method is used for power generation capacity prediction and user load coping is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy internet and smart grid, in particular to a virtual power plant resource scheduling method and system. BACKGROUND

[0002] In the current power system, the traditional scheduling method has significant defects in dealing with the volatility of renewable energy and the randomness of user load.

[0003] Renewable energy such as solar energy and wind energy is greatly affected by natural conditions and has strong intermittency and volatility. This makes the power generation prediction error large and it is difficult to accurately match power supply and demand. For example, the sudden appearance of clouds on a sunny day will cause the power generation of a photovoltaic power station to drop sharply; the instability of wind speed will cause the output of a wind farm to fluctuate greatly. User load presents randomness due to production and living habits, seasonal changes, and unexpected events. For example, during the summer heat, the use of air conditioning and other refrigeration equipment will cause the load to rise sharply; during the holiday period, the regular industrial load will decrease while the specific industrial load will increase. These random factors make it difficult to accurately predict load demand in the power system, increasing the difficulty of maintaining power supply and demand balance, and thus causing insufficient frequency modulation and peak shaving accuracy, affecting the stability and economy of the power system. The existing virtual power plant heterogeneous resource scheduling architecture uses edge computing nodes to reduce cloud computing pressure and calculates the prediction results at the edge nodes to achieve computing optimization, but this architecture cannot effectively improve the accuracy of power generation capacity prediction and user load response resource scheduling.

[0004] Therefore, there is an urgent need for a new technical solution to solve the above problems to improve the ability of virtual power plants to respond to the volatility of renewable energy and the randomness of user load and to improve the frequency modulation and peak shaving accuracy of the power system. SUMMARY

[0005] In order to solve the technical problem of low resource scheduling accuracy in power generation capacity prediction and user load response of the existing scheduling method, the present application provides a virtual power plant resource scheduling method and system.

[0006] A virtual power plant resource scheduling method realizes a closed-loop regulation and control of "edge feedback + cloud correction + constraint verification + user incentive" through a "cloud-edge" architecture of a cloud platform and an edge layer: Edge feedback: a plurality of edge computing nodes of the edge layer extract key information from the operation data of photovoltaic power stations, energy storage devices, and user loads, and analyze whether the key information has abnormal conditions and potential problems to generate edge analysis results; Cloud correction: the cloud platform substitutes the running data into the collaborative regulation model to verify the edge analysis result, if correct, generate the preliminary instruction; otherwise, based on the running data, through the photovoltaic prediction model, the energy storage charging and discharging model, the load response model, re-generate the secondary instruction; the preliminary instruction / secondary instruction combined with real-time market information and system constraint conditions to generate the preliminary regulation instruction after constraint; Constraint verification: the cloud platform substitutes the preliminary regulation instruction into the collaborative regulation model, and the collaborative regulation model based on the safety constraints of the power system, the operation constraints of the equipment, the market rule constraints to obtain the target regulation instruction; User incentive: through a number of edge computing nodes, the target regulation instruction is issued to the grassroots user end and the monitoring equipment end, and the user feedback samples about the target regulation instruction are counted, which are used for deep learning of the photovoltaic prediction model, the energy storage charging and discharging model, the load response model, and the collaborative regulation model. If the user feedback sample, the user is given a score incentive.

[0007] Further, the key information includes the trend of change of power generation and the load fluctuation.

[0008] Further, the edge analysis result includes: there are abnormal situations and potential problems / no abnormal situations and potential problems; And / or, the abnormal situation and potential problem include: the trend of change of power generation is unstable, the load fluctuation is irregular.

[0009] Further, the generation method of the preliminary instruction includes: The collaborative regulation model judges whether the result of the edge computing node predicting the power generation of the photovoltaic power station is correct based on the running data, if correct, the preliminary instruction is to arrange the power generation plan in advance according to the prediction result; The collaborative regulation model judges whether the result of the edge computing node judging whether the power generation of the photovoltaic power station is excessive is correct based on the running data, if correct, the preliminary instruction is: when the photovoltaic power station generates excessive power, control the energy storage device to store the excess power, when the power supply is insufficient, the preliminary instruction is to release the power to participate in power supply; The collaborative regulation model judges whether the analysis of the user's electricity behavior mode, load characteristics and government incentive response mechanism by the edge computing node is correct based on the running data, if correct, the preliminary instruction is: use the means of price incentive and subsidy policy to guide users to reduce electricity load when power supply is tight, and increase electricity load when power is excessive.

[0010] Further, the photovoltaic prediction model training method includes: collecting historical meteorological data, geographic location information and photovoltaic power station running data from InfluxDB, using long short-term memory network LSTM algorithm for model training, obtaining a photovoltaic prediction model suitable for the virtual power plant, which is used to predict the photovoltaic output of different time scales.

[0011] Further, the energy storage charging and discharging model training method includes: obtaining parameters of energy storage devices from MySQL, extracting historical operation data from InfluxDB, combining power market price data in Redis to build an energy storage charging and discharging optimization model; And / or, the load response model training method includes: analyzing user electricity data including electricity time and electricity load size from InfluxDB, combining price incentives and subsidy policies in MySQL to build a load response model for generating strategies to guide users to adjust load; And / or, the coordinated regulation model training method includes: based on the data including power system structure and power grid constraint conditions stored in MySQL, using genetic algorithm to build a coordinated regulation model.

[0012] Further, the secondary instruction generation method includes: According to the photovoltaic prediction model, collect historical meteorological data, geographic location information and photovoltaic power station operation data, and predict the power generation of the photovoltaic power station. The result of the secondary instruction is to arrange the power generation plan in advance according to the prediction result; When the energy storage charging and discharging model detects that the photovoltaic power station generates excess power, the secondary instruction is to control the energy storage device to charge and store excess power. When the energy storage charging and discharging model detects that the power supply is insufficient, the secondary instruction is to release the power to participate in power supply; Through the load response model, analyze the user's electricity behavior mode, load characteristics and government incentive response mechanism. The secondary instruction is to use price incentives and subsidy policies to guide users to reduce electricity load when power supply is tight and increase electricity load when power is surplus.

[0013] The application also provides a virtual power plant resource scheduling system adopting the virtual power plant resource scheduling method, the virtual power plant resource scheduling system comprising: a cloud platform and a plurality of edge computing nodes. The "cloud-edge" architecture of the cloud platform and the plurality of edge computing nodes realizes the closed-loop regulation and control of "edge feedback + cloud correction + constraint verification + user incentive". The plurality of edge computing nodes are used to extract key information in the operation data of photovoltaic power stations, energy storage devices and user loads, and analyze whether the key information has abnormal conditions and potential problems to generate edge analysis results; the cloud platform is used to verify the edge analysis results by substituting the operation data into the collaborative regulation and control model, and if correct, a preliminary instruction is generated; otherwise, based on the operation data, a secondary instruction is regenerated through a photovoltaic prediction model, an energy storage charging and discharging model and a load response model; the preliminary instruction / secondary instruction is combined with real-time market information and system constraint conditions to generate a preliminary regulation and control instruction after constraint; wherein the plurality of edge computing nodes are also used to issue the target regulation and control instruction to the grassroots user end and the monitoring device end, and to count the feedback samples of the target regulation and control instruction of the user, which are used for deep learning of the photovoltaic prediction model, the energy storage charging and discharging model, the load response model and the collaborative regulation and control model, and the user is given a score incentive if the user feedback sample.

[0014] Further, the operation data of photovoltaic power stations, energy storage devices and user loads are collected by deploying grassroots monitoring device ends in the edge layer.

[0015] Further, the grassroots monitoring device end comprises sensors of photovoltaic panels, energy storage devices and user power consumption ends, and the sensors comprise power sensors and temperature sensors.

[0016] Compared with the prior art, the application has the following beneficial effects: 1. Based on the optimization decision method of the four types of core models and the closed-loop regulation and control mechanism of "edge feedback + cloud correction + constraint verification + user incentive", the edge computing nodes collect and preliminarily process the time series data and structured data (operation data) of the grassroots device end, then arrange the key information such as the power generation power change trend and the load fluctuation situation to generate a preliminary strategy and transmit it to the cloud platform, the cloud platform verifies and processes the preliminary strategy according to the four types of core models (with prediction ability), and if necessary, modifies the original strategy of the edge computing nodes to generate accurate regulation and control instructions, thereby ensuring the processing speed while improving the processing accuracy. The rationality of the regulation and control instructions is verified by using device rated parameters, market rules and other data, and finally the target regulation and control instruction is output, and the feedback of the user and the grassroots device end is monitored in the subsequent stage to improve the number of real data samples, so as to improve the accuracy of the overall data and the accuracy of the regulation and control according to a large number of real sample data, thereby solving the technical problem of low resource scheduling accuracy of the existing virtual power plant architecture in power generation capacity prediction and user load response.

[0017] 2. Give full play to its advantages based on global data integration of multi-source heterogeneous resources, form an efficient energy complementary mechanism, improve the efficiency of adjusting power supply and demand balance, and ensure reasonable energy scheduling and stable operation of power grid.

[0018] 3. The data collection adopts a "cloud-edge" collaborative mode, which solves the problems of traditional cloud centralized collection, which can store full data but has high transmission delay and cannot meet the high-precision response demand of frequency modulation, and pure edge collection, which has fast response but limited storage capacity and is difficult to support long-term big data analysis. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 It is an aggregation resource collection and control schematic diagram. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0021] The embodiments of the present application provide a virtual power plant resource scheduling method, which is committed to improving the dynamic aggregation, collaborative scheduling and optimization of multi-source heterogeneous resources (distributed photovoltaic, distributed wind power, energy storage device, controllable load, electric vehicle, charging pile, battery swap station, etc.) of virtual power plant. Aiming at solving the problem that the traditional scheduling method is difficult to deal with the volatility of renewable energy and the randomness of user load, resulting in insufficient frequency modulation and peak regulation accuracy. The method adopts a "cloud-edge" architecture, which includes a cloud platform and an edge layer (several edge computing nodes + basic monitoring equipment end) that communicate with each other. The cloud platform is equipped with a photovoltaic prediction model, an energy storage charging and discharging model, a load response model, and a collaborative control model (i.e. four core models). And on the basis of the "cloud-edge" architecture, a closed-loop control mechanism of "edge feedback + cloud correction + constraint verification + user incentive" is added. And to improve the frequency modulation and peak regulation accuracy, and to improve the user feedback enthusiasm to obtain a large number of real data samples, so as to improve the overall data control accuracy of the virtual power plant, and solve the technical problems of low resource scheduling accuracy in power generation capacity prediction and user load response of the existing virtual power plant architecture.

[0022] Among them, the decision-making method of photovoltaic prediction model, energy storage charging and discharging model, load response model, and collaborative control model is as follows: (1) Photovoltaic prediction model: Using machine learning algorithms, combined with historical meteorological data (such as light intensity, temperature, humidity, etc.), geographic location information, and photovoltaic power station operation data, the power generation of photovoltaic power stations is accurately predicted. For example, through the long short-term memory network (LSTM) model, the long-term dependence relationship in time series data is effectively captured, and the photovoltaic output at different time scales (such as short-term hourly, ultra-short-term minute) is predicted, providing reliable power generation prediction information for power dispatching, helping to arrange power generation plans in advance, and reducing the impact of photovoltaic power generation fluctuations.

[0023] (2) Energy storage charging and discharging model: Considering factors such as energy storage device charging and discharging efficiency, remaining power (SOC), life consumption, and electricity market price signals, an energy storage charging and discharging optimization model is established. When renewable energy generation is excessive, control the energy storage device to charge at the optimal charging power to store excess electricity; when power supply is insufficient or electricity price is high, release electricity according to the optimized discharging strategy to participate in power supply. In this way, the regulating effect of energy storage devices is utilized to smooth the volatility of renewable energy generation, improving the stability and reliability of the power system.

[0024] (3) Load response model: Analyze user electricity consumption behavior patterns, load characteristics, and government incentive response mechanisms to build a load response model. Through price incentives (such as time-of-use electricity prices, real-time electricity prices), subsidies, and other means, guide users to reduce electricity consumption when power supply is tight, and increase electricity consumption when power is surplus. For example, for industrial users, according to their production processes and equipment characteristics, develop individualized load regulation schemes to flexibly adjust the load without affecting production, enhancing the adaptability of the power system to user load randomness.

[0025] (4) Coordinated regulation model: Based on the aggregated resource operating state of the virtual power plant, considering factors such as renewable energy generation, energy storage devices, load response, and grid constraints, the aggregated resources are virtually managed, and the optimal operation strategy of the system is solved through optimization algorithms (genetic algorithm) according to resource types, resource regions, and resource operating states.

[0026] Through the above four core models, the generation situation can be effectively predicted to cope with the precision reduction problem caused by power generation fluctuations due to environmental mutations at different times.

[0027] The virtual power plant resource scheduling method (using a closed-loop control mechanism, where each step is a small mechanism) includes steps S100, S200, S300, and S400.

[0028] S100, edge feedback: several edge computing nodes of the edge layer extract key information from the operation data of the photovoltaic power station, energy storage device and user load, and analyze whether the key information has abnormal conditions and potential problems to generate edge analysis results.

[0029] In this step, the base layer monitoring device end, i.e. various types of intelligent monitoring gateway devices. The operation data includes time series data and structured data. The time series data includes power generation power change trend, load fluctuation, etc., and the structured data includes user information, device parameters, etc. The edge layer includes edge computing nodes and various types of intelligent monitoring gateway devices (monitoring device end). Various types of intelligent monitoring gateway devices collect operation data of renewable energy generation devices, energy storage devices, user loads, etc. in real time, and quickly feed back these data to the local edge computing nodes through the MQTT protocol. The edge computing nodes preliminarily process and analyze the data, extract key information such as power generation power change trend, load fluctuation, etc., and temporarily store these time series data in InfluxDB, discover abnormal conditions and potential problems in system operation in a timely manner, and generate edge analysis results after comprehensively analyzing these conditions, so as to facilitate the cloud platform to process.

[0030] S200, cloud correction: the edge computing nodes upload the operation data and edge analysis results to the cloud platform; the cloud platform substitutes the operation data into the collaborative regulation and control model to verify the edge analysis results, and if correct, generates a preliminary instruction; otherwise, based on the operation data, the photovoltaic prediction model, the energy storage charging and discharging model, and the load response model are used to generate a secondary instruction; the preliminary instruction / secondary instruction is combined with real-time market information and system constraint conditions to generate a preliminary regulation and control instruction after being constrained.

[0031] In this step, edge computing nodes upload processed information to the cloud platform via the MQTT protocol. The cloud platform utilizes powerful computing capabilities and big data analytics to comprehensively evaluate and deeply analyze the operational status of the entire virtual power plant system. When generating secondary instructions, the system uses historical meteorological data, geographic location information, and photovoltaic power plant operation data collected by the photovoltaic prediction model to predict the power generation of the photovoltaic power plant. The secondary instructions are then used to schedule power generation plans in advance based on these predictions. When the energy storage charging and discharging model detects excess power generation from the photovoltaic power plant, the secondary instructions control the energy storage devices to charge and store the excess energy. When the energy storage charging and discharging model detects insufficient power supply, the secondary instructions release energy to participate in power supply. Finally, by analyzing user electricity consumption patterns, load characteristics, and government incentive response mechanisms through the load response model, the secondary instructions use price incentives and subsidy policies to guide users to reduce their electricity load when power supply is tight and increase their electricity load when power is surplus. The cloud platform stores the received time-series data (i.e., global time-series data) in InfluxDB, and stores structured data such as user information and device parameters in MySQL. Simultaneously, it uses Redis to cache frequently accessed data such as real-time photovoltaic forecast results and current electricity prices. Based on the calculation results of the optimized decision-making model (four core models), combined with real-time market information and system constraints, the edge analysis results of the edge computing nodes in the edge layer are corrected and optimized to generate more accurate preliminary control instructions.

[0032] S300, Constraint Verification: The cloud platform inputs the preliminary control instructions into the collaborative control model, which performs constraint verification based on the power system's security constraints, equipment operation constraints, and market rule constraints to obtain the target control instructions.

[0033] In this step, after generating the initial control instructions, the instructions undergo rigorous constraint verification to ensure that their execution does not violate power system safety constraints, equipment operation constraints, or market rules. During verification, equipment rated parameters and market rules are read from MySQL, and real-time grid status data is retrieved from Redis. For example, the verification checks whether the instructions would cause grid line overload, equipment exceeding its rated operating range, or violation of market trading rules. Only control instructions that pass the constraint verification are issued to the edge layer for execution.

[0034] S400, User Incentive: The target control command is sent to the grassroots user end and monitoring equipment end through several edge computing nodes, and the feedback samples of users on the target control command are collected for deep learning of photovoltaic prediction model, energy storage charging and discharging model, load response model and coordinated control model. If the user feedback sample is positive, the user will be given a bonus incentive.

[0035] In this step, after the edge layer receives and executes the control command, it will issue demand response incentives. Feedback points are allocated based on the actual effects of each user's participation, continuously optimizing the incentive mechanism to enhance users' enthusiasm and initiative in load response, forming a virtuous cycle of closed-loop control. For example, if user A actively participates in feedback, user A's points will increase to improve user credibility and other indicators. This incentive mechanism effectively increases user enthusiasm, thereby increasing the number of real data samples. The increase in real data samples helps improve the authenticity of average data and reduce average error, thus improving the overall control accuracy. Simultaneously, these real data samples can be used as training data to optimize the four core models, including the collaborative control model.

[0036] In summary, this invention utilizes an optimization decision-making method based on four core models and a closed-loop control mechanism of "edge feedback + cloud correction + constraint verification + user incentives." Edge computing nodes collect and pre-process time-series and structured (operational) data from grassroots equipment. The resulting edge analysis results, along with key information such as power generation trends and load fluctuations, are then transmitted to the cloud platform. The cloud platform processes the time-series, structured, and labeled data based on the four core models and verifies the edge analysis results from the edge computing nodes. If the edge analysis results are correct, preliminary control commands are directly generated, or precise preliminary control commands are generated after correction. The rationality of the control commands is verified using data constraints such as equipment rated parameters and market rules. Finally, the target control command is output. Subsequent monitoring of user and grassroots equipment feedback increases the number of real data samples, thereby improving the overall data accuracy and control precision. Existing control methods typically employ edge computing nodes for localized data processing, directly transmitting the results (i.e., preliminary strategies) to the cloud platform. The cloud platform receives these different processing results from various edge computing nodes and integrates and summarizes them. However, because edge computing nodes omit a significant amount of crucial data (operational data), providing only hypothetical "critical data," the cloud platform cannot recognize the actual data received by each node, making it unable to distinguish between true and false data. This results in the generation of control commands with even greater errors after integrating multiple sets of error data. Furthermore, existing control methods fail to recognize the importance of user-feedback data samples, relying solely on data from automated equipment. If this data contains errors, the final cloud-based decision will have even greater errors. Therefore, compared to existing technologies, this invention effectively improves the ability to collect and identify real data, enabling global analysis based on real data to obtain high-precision control commands based on global real data. This solves the technical problem of low resource scheduling accuracy in existing scheduling methods for power generation capacity prediction and user load response.

[0037] The present invention also provides a virtual power plant resource scheduling system, which adopts the above-mentioned virtual power plant resource scheduling method. The virtual power plant resource scheduling system includes: a cloud platform, several edge computing nodes, and a basic monitoring equipment terminal.

[0038] Several edge computing nodes and basic monitoring equipment belong to the edge layer. The basic monitoring equipment includes sensors installed on photovoltaic panels, energy storage devices, and user power terminals to collect real-time data such as power generation, remaining power, and power load. These sensors include power sensors and temperature sensors, used to collect data from photovoltaic, energy storage, and user power terminals: Photovoltaics: three-phase current, three-phase voltage, active power, reactive power, power generation, power factor, and internal air temperature, etc. Energy Storage: three-phase current, three-phase voltage, active power, reactive power, grid connection / off-grid status, charging / discharging status, ambient temperature, and maximum operating capacity, etc. User power terminals: three-phase current, three-phase voltage, active power, reactive power, power generation, and power factor, etc. Edge computing nodes are used to extract key information from the operational data of photovoltaic power plants, energy storage devices, and user loads, and analyze whether there are any anomalies or potential problems in this key information to generate edge analysis results. Edge computing nodes utilize embedded devices with data processing and analysis capabilities, connecting to intelligent monitoring equipment and interacting with it via the MQTT protocol. InfluxDB is configured to store local time-series data. The cloud platform employs high-performance servers and big data processing software, deploying InfluxDB for global time-series data storage, MySQL for structured data such as user information and device parameters, and Redis as a high-frequency data cache, enabling data interaction and information processing with the edge computing nodes. The cloud platform incorporates a closed-loop control mechanism of "edge feedback - cloud correction - constraint verification - user incentives"; several edge computing nodes communicate with the cloud platform; and the grassroots monitoring equipment communicates with several edge computing nodes. The cloud platform includes four core models (photovoltaic prediction model, energy storage charging and discharging model, load response model, and coordinated control model). The training and deployment methods for each model are as follows: The photovoltaic prediction model is developed by the cloud platform, which collects historical meteorological data, geographical location information and photovoltaic power plant operation data from InfluxDB. The model is trained using the Long Short-Term Memory (LSTM) algorithm to obtain a photovoltaic prediction model suitable for the virtual power plant. The photovoltaic prediction model is then deployed on the cloud platform, which stores the prediction results in Redis for use in predicting photovoltaic output at different time scales.

[0039] The energy storage charging and discharging model involves the cloud platform obtaining parameters of the energy storage device, including charging and discharging efficiency and rated capacity, from MySQL, extracting historical operating data from InfluxDB, and combining them with electricity market price data from Redis to construct the energy storage charging and discharging model, which is then deployed on the cloud platform.

[0040] The load response model is constructed by analyzing user electricity consumption data, including electricity usage time and load size, from InfluxDB on the cloud platform. Combined with price incentives and subsidy policies in MySQL, the model is deployed on the cloud platform to generate strategies that guide users to adjust their load.

[0041] The collaborative control model is based on the characteristics of the power system structure and grid constraints stored in MySQL. It uses a distributed collaborative optimization algorithm (such as Genetic Algorithm, GA) to build the collaborative control model and deploy it on the cloud platform. During operation, it obtains real-time data from InfluxDB and Redis.

[0042] like Figure 1 As shown, Figure 1 This is a schematic diagram of the aggregated resource acquisition and control mechanism (i.e., closed-loop control mechanism) of the present invention. The starting point is the resource device (i.e., the basic monitoring device / intelligent monitoring device) shown in the diagram: (1) Edge feedback: The intelligent monitoring device collects data in real time and transmits it to the edge computing node via the MQTT protocol. The edge computing node processes the data, such as filtering and transforming it, and extracts key information such as the trend of power generation change and load fluctuation, and stores it in the local InfluxDB. If an abnormal situation is found (such as a sudden drop in power generation or a sudden increase in load), it is marked in time and uploaded to the cloud platform via the MQTT protocol.

[0043] (2) Cloud correction: The cloud platform receives the information uploaded by the edge computing nodes, stores it in the cloud InfluxDB, and combines the calculation results of the photovoltaic prediction model, energy storage charging and discharging model, load response model and coordinated control model cached in Redis, as well as the real-time market information and system constraints in MySQL, to correct and optimize the control strategy initially generated by the edge layer and generate accurate control instructions.

[0044] (3) Constraint Verification: The cloud platform performs constraint verification on the generated control commands, reads equipment rated parameters and market rules from MySQL, and obtains real-time power grid status data from Redis to check for problems such as power grid line overload, equipment exceeding rated operating range, and violation of market transaction rules. If problems are found, the command is returned for correction; if the verification passes, the command is sent to the edge computing node via the MQTT protocol.

[0045] (4) User incentives (not mentioned in the figure): Edge computing nodes send control instructions to user terminals and device control terminals to execute corresponding control operations. After the operation is completed, the user response is statistically analyzed, such as the response load and response time, and user feedback information (i.e., real data samples) is collected to optimize the incentive allocation mechanism and related models (such as the four types of core models).

[0046] In summary, the virtual power plant system provided by this invention, based on an optimization decision-making method using four core models and a closed-loop control mechanism of "edge feedback + cloud correction + constraint verification + user incentives," effectively addresses the shortcomings of traditional dispatching methods in dealing with the volatility of renewable energy and the randomness of user loads. Specifically, the photovoltaic prediction model improves the accuracy of renewable energy power generation prediction, providing a reliable basis for power dispatch; the energy storage charging and discharging model utilizes the regulating role of energy storage devices to smooth the volatility of renewable energy power generation; the load response model enhances the power system's adaptability to the randomness of user loads; and the collaborative control model achieves the optimal operating strategy globally. In the closed-loop control mechanism, the MQTT protocol ensures the real-time and reliable transmission of data between the edge and cloud, InfluxDB efficiently stores massive amounts of time-series data, MySQL standardizes the management of structured data, and Redis improves the access speed of high-frequency data, collectively ensuring the real-time, accurate, and effective nature of the control strategy. Edge feedback can promptly detect system problems; cloud correction optimizes control commands; constraint verification ensures the security and compliance of commands; and user incentives increase user participation, forming a virtuous cycle. This invention significantly improves frequency regulation and peak shaving accuracy, ensuring the safe and stable operation of the power system and enhancing its overall performance and economy. It aims to address the problem of insufficient frequency regulation and peak shaving accuracy caused by the inability of traditional virtual power plant dispatching methods to cope with the volatility of renewable energy and the randomness of user loads: 1. It fully leverages the advantages of integrating multi-source heterogeneous resources to form an efficient energy synergy and complementarity mechanism, improving the efficiency of power supply and demand balance regulation and ensuring rational energy dispatch and stable grid operation. 2. Data acquisition adopts a "cloud-edge" collaborative approach, solving the drawbacks of traditional centralized cloud acquisition, which, while capable of storing full data, suffers from high transmission latency and cannot meet the high-precision response requirements of frequency regulation, and pure edge acquisition, while offering fast response, has limited storage capacity and cannot support long-term big data analysis. 3. It develops new multi-source heterogeneous resource aggregation technologies and advanced intelligent model algorithms to improve the overall performance and operating efficiency of the virtual power plant.

[0047] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0048] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A virtual power plant resource scheduling method, characterized in that, The "edge feedback + cloud correction + constraint verification + user incentive" closed-loop regulation is realized through the "cloud-edge" architecture of the cloud platform and the edge layer: Edge feedback: Several edge computing nodes of the edge layer extract key information from the operation data of photovoltaic power stations, energy storage devices, and user loads, and analyze whether the key information has abnormal conditions and potential problems to generate edge analysis results; Cloud correction: The cloud platform substitutes the operation data into the collaborative regulation model to verify the edge analysis results. If correct, a preliminary instruction is generated. Otherwise, based on the operation data, a secondary instruction is generated through the photovoltaic prediction model, the energy storage charging and discharging model, and the load response model. The preliminary instruction / secondary instruction is combined with real-time market information and system constraint conditions to generate a preliminary regulation instruction after constraint; Constraint verification: The cloud platform substitutes the preliminary regulation instruction into the collaborative regulation model, and the collaborative regulation model performs constraint verification based on the safety constraints, device operation constraints, and market rule constraints of the power system to obtain a target regulation instruction; User incentive: Through several edge computing nodes, the target regulation instruction is issued to the grassroots user end and the monitoring device end, and the feedback samples of the users about the target regulation instruction are counted for deep learning of the photovoltaic prediction model, the energy storage charging and discharging model, the load response model, and the collaborative regulation model. If the user feedback sample is correct, the user is given a score for incentive.

2. The virtual power plant resource scheduling method of claim 1, wherein, The key information includes the trend of power generation and the load fluctuation.

3. The virtual power plant resource scheduling method of claim 1, wherein, The edge analysis results include: abnormal conditions and potential problems / no abnormal conditions and potential problems; And / or, the abnormal conditions and potential problems include: unstable trend of power generation, irregular load fluctuation.

4. The virtual power plant resource scheduling method of claim 1, wherein, The generation method of the preliminary instruction includes: The collaborative regulation model judges whether the result of the edge computing node predicting the power generation of the photovoltaic power station based on the operation data is correct. If correct, the preliminary instruction is to arrange the power generation plan in advance according to the prediction result; The collaborative regulation model judges whether the result of the edge computing node judging whether the photovoltaic power station is overproducing based on the operation data is correct. If correct, the preliminary instruction is to control the energy storage device to store excess power when the photovoltaic power station is overproducing, and to release the power to participate in power supply when the power supply is insufficient; The collaborative regulation model judges whether the analysis of the user's electricity consumption behavior pattern, load characteristics, and government incentive response mechanism by the edge computing node based on the operation data is correct. If correct, the preliminary instruction is to guide the user to reduce the electricity load when the power supply is tight and to increase the electricity load when the power is surplus through price incentive and subsidy policy.

5. The virtual power plant resource scheduling method of claim 1, wherein, The photovoltaic prediction model training method includes: collecting historical meteorological data, geographic location information, and photovoltaic power station operation data from InfluxDB, using the long short-term memory network LSTM algorithm for model training, and obtaining a photovoltaic prediction model suitable for the virtual power plant for predicting photovoltaic output at different time scales.

6. The virtual power plant resource scheduling method of claim 1, wherein, The energy storage charging and discharging model training method includes: obtaining the parameters of the energy storage device from MySQL, extracting historical operation data from InfluxDB, and combining the electricity market price data in Redis to build an energy storage charging and discharging optimization model; And / or, the load response model training method comprises: analyzing the user's electricity data including electricity time and electricity load size from InfluxDB, combining the price incentive and subsidy policy in MySQL to build a load response model for generating a strategy to guide the user to adjust the load; And / or, the coordinated regulation model training method comprises: based on the data including power system structure and grid constraint conditions stored in MySQL, using genetic algorithm to build a coordinated regulation model.

7. The virtual power plant resource scheduling method of claim 1, wherein, The generation method of the secondary instruction comprises: According to the historical meteorological data, geographic location information and photovoltaic power station operation data collected by the photovoltaic prediction model, the power generation of the photovoltaic power station is predicted, and the secondary instruction is arranged in advance according to the prediction result. When the energy storage charging and discharging model detects that the photovoltaic power station generates excess electricity, the secondary instruction is to control the energy storage equipment to charge and store excess electricity; when the energy storage charging and discharging model detects that the power supply is insufficient, the secondary instruction is to release the electricity to participate in power supply. Through the load response model, the user's electricity behavior mode, load characteristics and government incentive response mechanism are analyzed, and the secondary instruction is used to guide the user to reduce the electricity load when the power supply is tight and to increase the electricity load when the power is surplus by means of price incentive and subsidy policy.

8. A virtual power plant resource scheduling system employing the virtual power plant resource scheduling method according to any one of claims 1 to 7, characterized by The virtual power plant resource scheduling system comprises: a cloud platform and a plurality of edge computing nodes; the "cloud-edge" architecture of the cloud platform and the plurality of edge computing nodes realizes the closed-loop regulation of "edge feedback + cloud correction + constraint verification + user incentive"; The plurality of edge computing nodes are used to extract key information from the operation data of photovoltaic power stations, energy storage devices and user loads, and analyze whether the key information has abnormal conditions and potential problems to generate edge analysis results; The cloud platform is used to verify the edge analysis results by substituting the operation data into the coordinated regulation model, and if correct, a preliminary instruction is generated; otherwise, based on the operation data, a secondary instruction is regenerated by the photovoltaic prediction model, the energy storage charging and discharging model and the load response model; the preliminary instruction / secondary instruction is combined with real-time market information and system constraint conditions to generate a preliminary regulation instruction after constraint; The plurality of edge computing nodes are also used to issue the target regulation instruction to the grassroots user end and the monitoring equipment end, and to count the feedback samples of the target regulation instruction of the user, which are used for deep learning of the photovoltaic prediction model, the energy storage charging and discharging model, the load response model and the coordinated regulation model, and the user is given a score incentive if the user feedback sample. 9.The virtual power plant resource scheduling system of claim 8, wherein, The operation data of photovoltaic power stations, energy storage devices and user loads are collected by deploying grassroots monitoring equipment ends in the edge layer. 10.The virtual power plant resource scheduling system of claim 8, wherein, The grassroots monitoring equipment end comprises sensors of photovoltaic panels, energy storage devices and user electricity ends, and the sensors comprise power sensors and temperature sensors.

Images