A resource scheduling method, apparatus, equipment, medium, and product for a virtual power plant.

Abstract

This invention discloses a resource scheduling method, apparatus, equipment, medium, and product for a virtual power plant. Based on collected user behavior data, a user scheduling strategy is calculated; wherein the user scheduling strategy characterizes the resource allocation preference for target users; based on the user behavior data and the user scheduling strategy, a load forecast value for the target user is calculated; the load forecast value is smoothed using a preset interpolation algorithm to obtain a processed load forecast value; wherein the interpolation algorithm is obtained by combining a linear interpolation algorithm and a locally weighted regression interpolation algorithm; based on the processed load forecast value and the load demand of the target user, a resource scheduling strategy for the user in the virtual power plant is generated; wherein the resource scheduling strategy includes a resource scheduling value. Using this invention, dynamic changes in user behavior can be effectively captured, improving the flexibility and accuracy of resource scheduling in the virtual power plant.

Description

Technical Field

[0001] This invention relates to the field of power system dispatching technology, and in particular to a resource dispatching method, apparatus, equipment, medium and product for a virtual power plant. Background Technology

[0002] Virtual power plants optimize energy utilization by aggregating distributed energy resources, but existing scheduling methods have significant shortcomings in handling user behavior: fixed-rule-based methods, such as time-of-use pricing, simplify user behavior to a static model and cannot adapt to dynamic changes; prediction-based methods, such as LSTM (Long Short-Term Memory Network for Load Forecasting), rely on historical data, and prediction errors accumulate when user behavior fluctuates, lacking real-time feedback; centralized control methods, such as a central EMS (Energy Management System), while enabling unified scheduling, suffer from high system complexity, large communication delays, and difficulty in scaling. In real-world cases, sudden increases in user air conditioning load or charging delays often lead to scheduling failures. Therefore, a novel collaborative scheduling method is urgently needed that can capture dynamic changes in user behavior in real time and achieve two-way interaction between users and the system. Summary of the Invention

[0003] The purpose of this invention is to provide a resource scheduling method, apparatus, equipment, medium, and product for virtual power plants, which can effectively capture dynamic changes in user behavior and improve the flexibility and accuracy of resource scheduling in virtual power plants.

[0004] To achieve the above objectives, embodiments of the present invention provide a resource scheduling method for a virtual power plant, comprising:

[0005] Based on the collected user behavior data, a user scheduling strategy is calculated; wherein, the user scheduling strategy is used to characterize the resource allocation tendency for target users;

[0006] Calculate the load prediction value for the target user based on the user behavior data and the user scheduling strategy;

[0007] The load forecast value is smoothed using a preset interpolation algorithm to obtain the processed load forecast value; wherein, the interpolation algorithm is obtained by combining a linear interpolation algorithm and a locally weighted regression interpolation algorithm;

[0008] Based on the processed load forecast and the load demand of the target user, a resource scheduling strategy for the virtual power plant is generated for the user; wherein, the resource scheduling strategy includes resource scheduling values.

[0009] As an improvement to the above solution, the step of generating a user scheduling strategy based on the collected user behavior data includes:

[0010] Collect user behavior data; wherein, the user behavior data includes the target user's load demand, response latency, and preference window;

[0011] The load demand, the response delay, and the preference window are standardized and converted into feature vectors.

[0012] The feature vector, the feature vectors of the target user's neighboring users, and the adjacency matrix are input into a preset user behavior chain feedback model for processing, and the user scheduling strategy of the target user is output. The adjacency matrix is ​​used to characterize the degree of mutual influence between the target user and the neighboring users, and the neighboring users are other users who have behavioral associations with the target user. The user behavior chain feedback model is constructed and trained based on deep reinforcement learning.

[0013] As an improvement to the above scheme, the step of using a preset interpolation algorithm to smooth the load forecast value to obtain the processed load forecast value includes:

[0014] Based on the actual load value of the target user at time (t-1) and the predicted load value at time (t+1), a preset linear interpolation algorithm is used to calculate the predicted load value at time t, which is denoted as the first predicted load value.

[0015] Based on the actual load values ​​of neighboring users at time t, the local weighted regression interpolation algorithm is used to calculate the predicted load value of the target user at time t, which is then used as the second predicted load value; wherein, the neighboring users are other users who have behavioral associations with the target user;

[0016] The load forecast value is calculated based on the first load forecast value, the second load forecast value, and the preset adaptive weighting coefficient.

[0017] As an improvement to the above solution, the method further includes:

[0018] The roulette wheel algorithm is used to calculate the scheduling selection probability of the target user;

[0019] The policy scheduling priority of the target user is determined based on the scheduling selection probability; wherein the scheduling selection probability is positively correlated with the policy scheduling priority.

[0020] The resource scheduling policy is executed on the target user according to the policy scheduling priority.

[0021] As an improvement to the above scheme, the step of using the roulette wheel algorithm to calculate the scheduling selection probability of the target user includes:

[0022] Calculate the utility value of the target user; wherein the utility value is used to evaluate the value of the user's behavior;

[0023] Calculate the scheduling selection probability of the target user based on the utility value;

[0024] A time-series scheduling control method is introduced to adjust the scheduling selection probability based on the historical behavior weights of the target user, thereby obtaining the final scheduling selection probability; wherein, the historical behavior weights are used to characterize the target user's responsiveness and reliability to resource scheduling strategies within a historical period.

[0025] As an improvement to the above scheme, after generating the resource scheduling strategy of the virtual power plant for the user based on the processed load forecast value and the load demand of the target user, the method further includes:

[0026] Calculate the scheduling error; wherein the scheduling error is used to represent the difference between the predicted load and the actual load for all users;

[0027] Using a global-local two-level game strategy, the utility function values ​​of the global game and the local game are calculated, as well as the global adjustment factor and the local adjustment factor. The goal of the global game layer is to optimize the resource allocation of the entire virtual power plant to balance the grid load and user demand, while the goal of the local game layer is to optimize the resource scheduling strategy of each user.

[0028] Calculate the game theory adjustment term based on the utility function value of the global game, the utility function value of the local game, the global adjustment factor, and the local adjustment factor;

[0029] The resource scheduling value is modified according to the game theory adjustment term to form the final resource scheduling strategy.

[0030] This invention also provides a resource scheduling device for a virtual power plant, comprising:

[0031] The user scheduling strategy generation module is used to generate user scheduling strategies based on collected user behavior data; wherein, the user scheduling strategy is used to characterize the resource allocation preference for target users;

[0032] The load forecast calculation module is used to calculate the load forecast value of the target user based on the user behavior data and the user scheduling strategy.

[0033] The load forecast processing module is used to smooth the load forecast value using a preset interpolation algorithm to obtain the processed load forecast value; wherein, the interpolation algorithm is obtained by combining a linear interpolation algorithm and a local weighted regression interpolation algorithm;

[0034] The resource scheduling strategy generation module is used to generate a resource scheduling strategy for the virtual power plant for the user based on the processed load forecast value and the load demand of the target user; wherein, the resource scheduling strategy includes resource scheduling values.

[0035] This invention also provides a resource scheduling device for a virtual power plant, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the resource scheduling method for the virtual power plant as described in any of the preceding embodiments.

[0036] This invention also provides a computer-readable storage medium, which includes a stored computer program, wherein the computer program, when running, controls the device where the computer-readable storage medium is located to execute the resource scheduling method of the virtual power plant as described in any of the preceding embodiments.

[0037] This invention also provides a computer program product, which includes a computer program or computer instructions. When the computer program or computer instructions are executed by a processor, they implement the resource scheduling method for a virtual power plant as described in any of the above embodiments.

[0038] Compared with existing technologies, the resource scheduling method, apparatus, equipment, medium, and products of the virtual power plant disclosed in this invention, by collecting rich user behavior data and combining it with improved calculation models and algorithms, can deeply explore user electricity consumption patterns and demands, making the calculated user scheduling strategies and load forecasts more accurate. Load forecast curve smoothing further improves the stability and reliability of load forecasting, thereby generating more accurate resource scheduling strategies, avoiding over-allocation or under-allocation of resources. The virtual power plant can better balance the power supply and demand relationship, improving the scheduling efficiency and accuracy of the virtual power plant. Simultaneously, by guiding users to adjust their electricity consumption behavior, it achieves optimized allocation of power resources in time and space, improving the stability and reliability of the power system. By formulating personalized user scheduling strategies, considering the characteristics and needs of user electricity consumption behavior, and guiding users to participate in power resource scheduling while meeting their normal electricity needs, it can reduce electricity costs. Attached Figure Description

[0039] Figure 1 This is a flowchart illustrating a resource scheduling method for a virtual power plant provided in an embodiment of the present invention;

[0040] Figure 2 This is a flowchart illustrating a preferred embodiment of the resource scheduling method for a virtual power plant according to the present invention.

[0041] Figure 3 This is a schematic diagram of the structure of a resource scheduling device for a virtual power plant provided in an embodiment of the present invention;

[0042] Figure 4 This is a schematic diagram of the structure of a resource scheduling device for a virtual power plant provided in an embodiment of the present invention. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] In the description of this application, it should be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.

[0045] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0046] See Figure 1 This is a flowchart illustrating a resource scheduling method for a virtual power plant according to an embodiment of the present invention. The embodiment of the present invention provides a resource scheduling method for a virtual power plant, including steps S11 to S14:

[0047] S11. Calculate a user scheduling strategy based on the collected user behavior data; wherein, the user scheduling strategy is used to characterize the resource allocation tendency towards the target user;

[0048] S12. Calculate the load prediction value of the target user based on the user behavior data and the user scheduling strategy;

[0049] S13. The load forecast value is smoothed using a preset interpolation algorithm to obtain the processed load forecast value; wherein, the interpolation algorithm is obtained by combining a linear interpolation algorithm and a local weighted regression interpolation algorithm;

[0050] S14. Based on the processed load forecast value and the load demand of the target user, generate the resource scheduling strategy of the virtual power plant for the user; wherein, the resource scheduling strategy includes resource scheduling values.

[0051] In this embodiment of the invention, user behavior data is collected in real time through a smart terminal device. This user behavior data includes the behavior data of the target user and neighboring users. A user scheduling strategy calculation model is constructed, using preprocessed user behavior data as input. By training the model, the relationship between user electricity consumption patterns and resource allocation preferences is explored. Based on the trained model and the real-time collected user behavior data, a user scheduling strategy for each target user is calculated. This strategy quantitatively represents the resource allocation preference for the target user, such as allocation priority and the proportion of schedulable resources.

[0052] Features related to load forecasting are extracted from user behavior data. Combined with the user scheduling strategy, the load forecast value for the target user is predicted. This load forecast value is then smoothed using an interpolation algorithm. Finally, the difference between the processed load forecast value and the actual load demand of the target user is analyzed to determine the direction and extent of resource scheduling. Based on the analysis results and the overall resource status of the virtual power plant, a resource scheduling strategy for the target user is generated. The resource scheduling strategy includes specific resource scheduling values, such as adjusting the user's power consumption and allocating power resources. Simultaneously, the resource scheduling strategy is distributed to the target user's smart terminal device via a communication network to guide the user in adjusting their electricity consumption behavior.

[0053] By employing the technical means of this invention, and through the collection of abundant user behavior data, combined with improved calculation models and algorithms, it is possible to deeply explore user electricity consumption patterns and demands, making the calculated user dispatch strategies and load forecasts more accurate. Load forecast curve smoothing further improves the stability and reliability of load forecasting, thereby generating more precise resource dispatch strategies, avoiding over-allocation or under-allocation of resources. Virtual power plants can better balance power supply and demand, improving their dispatch efficiency and accuracy. Simultaneously, by guiding users to adjust their electricity consumption behavior, optimal allocation of power resources in time and space is achieved, improving the stability and reliability of the power system. By formulating personalized user dispatch strategies that consider user electricity consumption characteristics and demands, and guiding users to participate in power resource dispatch while meeting their normal electricity needs, electricity costs can be reduced.

[0054] As a preferred embodiment, the present invention further implements the above embodiments. Step S11, namely, generating a user scheduling strategy based on the collected user behavior data, includes steps S111 to S113:

[0055] S111. Collect user behavior data; wherein, the user behavior data includes the target user's load demand, response latency, and preference window;

[0056] S112. Standardize the load demand, the response delay, and the preference window to convert them into feature vectors;

[0057] S113. Input the feature vector, the feature vectors of the target user's neighboring users, and the adjacency matrix into a preset user behavior chain feedback model for processing, and output the user scheduling strategy of the target user; wherein, the adjacency matrix is ​​used to characterize the degree of mutual influence between the target user and neighboring users, and the neighboring users are other users who have behavioral associations with the target user; the user behavior chain feedback model is constructed and trained based on deep reinforcement learning.

[0058] In this embodiment of the invention, a user behavior data acquisition module collects user electricity consumption data in real time, including information such as user load demand, usage preferences, and response latency. By constructing a user behavior chain feedback model, user electricity consumption behavior can be effectively analyzed and predicted, and a preliminary scheduling strategy can be generated based on this.

[0059] Specifically, the system collects users' load demand in real time through smart terminal devices (such as smart meters). Response delay Preference Window Information such as user behavior data is collected and standardized to convert user behavior into feature vectors, which are then used for model training and policy optimization. The calculation formula is as follows:

[0060]

[0061] in, , , , The mean and standard deviation of the load data, and These are the weighting coefficients.

[0062] The user behavior chain feedback model is built on deep reinforcement learning (DRL), which learns the dynamic characteristics of user behavior through neural networks. The feedback function... It is obtained through training a neural network and is used to determine the user's current state. Dynamically adjust scheduling strategies.

[0063] Scheduling policy for each user Based on its current state Influence from neighbors (adjacency matrix) The selection is made using (representation). The dynamic update equation for user behavior is:

[0064]

[0065] in: User In time User scheduling strategy. It is a feedback function, learned through a neural network, representing the state of the user. The scheduling strategy is dynamically adjusted based on the status of neighboring users. User The set of neighbors, representing the user Other sets of users with behavioral associations. It is an adjacency matrix, representing users Other users The strength of the behavioral association between users reflects the degree of mutual influence between them. User In time The state vector contains information such as user load demand, response latency, and preferences. User In time The state vector represents the mutual influence between users, and the feedback function... The scheduling strategy will be adjusted based on the impact of these user behaviors.

[0066] In this model, the user scheduling strategy is implemented through a deep reinforcement learning model. This model, which we're learning about, makes scheduling decisions based on a user's current state and the behavior of their neighbors (calculated through a weighted adjacency matrix). The model's goal is to dynamically adjust and optimize user behavior strategies, enabling the virtual power plant to adapt to constantly changing electricity demand and user behavior.

[0067] Through the recursive feedback mechanism of neural networks, the system can optimize the scheduling strategy at each time step and dynamically adjust the relationship between user behavior and load scheduling, thereby improving the overall scheduling efficiency and stability of power resources.

[0068] In a preferred embodiment, step S12, namely calculating the load prediction value of the target user based on the user behavior data and the user scheduling strategy, includes:

[0069] Based on the user behavior data, a baseline load forecast value is generated using a traditional time series model or a machine learning model.

[0070] The user scheduling strategy is used as a dynamic adjustment factor to correct the baseline load forecast value, thereby obtaining the load forecast value for the target user.

[0071] For example, if If the user accepts delayed charging, then the predicted charging load for period t+1 will be reduced; if If the user responds to the real-time electricity price, then the weight of adjustable load is increased.

[0072] Preferably, the modified formula is as follows:

[0073]

[0074] in, The load forecast value at time t+1 This is the baseline load forecast value. This represents the strategy impact coefficient. , is a dynamic adjustment factor transformed from the user scheduling strategy, used to represent the user's responsiveness.

[0075] As a preferred embodiment, the present invention further implements the above embodiments. Step S13, namely, smoothing the load forecast value using a preset interpolation algorithm to obtain the processed load forecast value, includes steps S131 to S133:

[0076] S131. Based on the actual load value of the target user at time (t-1) and the predicted load value at time (t+1), a preset linear interpolation algorithm is used to calculate the predicted load value at time t, which is denoted as the first predicted load value.

[0077] S132. Based on the actual load values ​​of neighboring users at time t, the local weighted regression interpolation algorithm is used to calculate the load prediction value of the target user at time t, which is used as the second load prediction value; wherein, the neighboring users are other users who have behavioral associations with the target user;

[0078] S133. Calculate the load forecast value based on the first load forecast value, the second load forecast value, and the preset adaptive weighting coefficient.

[0079] In this embodiment of the invention, after completing user behavior modeling, the system enters the prediction optimization stage. The linear interpolation optimization module receives user feature vectors and improves the accuracy and stability of virtual power plant resource scheduling through a series of refined prediction optimization steps. To effectively cope with load fluctuations and optimize resource scheduling, the system combines traditional linear interpolation methods with innovative locally weighted multinomial regression interpolation technology. This innovative solution not only smooths load fluctuations but also has an adaptive adjustment mechanism, capable of adjusting load forecasts in real time based on feedback. Thus, the system can maintain high load forecast accuracy during normal operation and react quickly to sudden fluctuations, ensuring the stability and efficiency of scheduling.

[0080] Traditional linear interpolation methods can smooth load fluctuations, but their accuracy is insufficient when load demand changes drastically or unexpected events occur. To address this issue, this invention combines locally weighted multinomial regression interpolation technology to perform locally weighted regression fitting on historical load data, generating more accurate load forecasts.

[0081] First, the load forecast value at time t is calculated using the traditional linear interpolation algorithm, denoted as the first load forecast value. The traditional linear interpolation formula is as follows:

[0082]

[0083] in, For users At time step The first load forecast value, For time step ( The actual load value, For time steps The load forecast value. It is a time interval.

[0084] To further improve accuracy, this invention introduces a locally weighted regression interpolation method to calculate the load forecast value of the target user at time t, which is used as the second load forecast value. The locally weighted regression equation is as follows:

[0085]

[0086] in, For users At time step The second load forecast value; For users The set of neighbors, representing the user Other sets of users with behavioral associations; The weighting coefficients are dynamically adjusted based on time distance and data importance, and are typically expressed using a Gaussian function or an exponential decay function. For users At time step The load value.

[0087] Based on the first load forecast value, the second load forecast value, and the preset adaptive weighting coefficient, the load forecast value is calculated using the following dynamic fusion formula:

[0088]

[0089] in, As an adaptive weighting coefficient, optionally, when the load is stable, =0.7; when the load fluctuates greatly, =0.3.

[0090] By combining linear interpolation and local weighted regression interpolation, the system can fully utilize the simplicity of linear interpolation and the accuracy of local regression when facing complex load fluctuations, thereby improving the stability and accuracy of predictions.

[0091] Preferably, considering the unpredictability of load fluctuations, traditional interpolation methods have limited ability to cope with extreme fluctuations. This invention employs an adaptive adjustment mechanism, dynamically adjusting the weighting coefficients in the regression model based on real-time feedback and scheduling errors. Especially when load fluctuations are large, the system automatically strengthens the weighting of historical data to ensure the accuracy of load forecasting. The adaptive adjustment formula is as follows:

[0092]

[0093] in, The standard deviation of the Gaussian function determines the weighting range. The scheduling error for the current period measures the difference between the forecast and the actual load. This represents the maximum scheduling error for the current time period, used for normalization.

[0094] This mechanism adjusts the weighting coefficients based on scheduling errors and historical load changes, enabling the model to flexibly respond to different load changes and ensuring scheduling accuracy and stability.

[0095] In a preferred embodiment, step S14, namely generating the resource scheduling strategy of the virtual power plant for the user based on the processed load forecast value and the load demand of the target user, includes:

[0096] Based on the processed load forecast and the target user's load demand, the virtual power plant generates a resource scheduling strategy for the user according to the following scheduling formula:

[0097]

[0098] in, The resource scheduling value for user i at time step t; This is the load forecast value after smoothing. This represents the actual load demand. These are adjustment coefficients, used to control the correction of prediction errors and the impact of historical scheduling, respectively.

[0099] In this embodiment of the invention, the scheduling system performs optimized scheduling based on the smoothed load forecast. During the scheduling process, a preliminary load forecast is first obtained through linear interpolation, then further optimized through local weighted regression interpolation, and finally the scheduling strategy is adjusted based on real-time feedback from the system.

[0100] Through multi-stage scheduling optimization, the system can flexibly adjust its scheduling strategy based on real-time load demand and resource status, thereby avoiding scheduling errors caused by sudden changes in load demand. This scheduling optimization allows the system to make flexible adjustments based on predicted load and real-time resource scheduling needs, avoiding scheduling errors caused by sudden changes in load demand.

[0101] As a preferred embodiment, the present invention is further implemented based on any of the above embodiments, see [link to previous embodiments]. Figure 2 This is a flowchart illustrating a preferred embodiment of the resource scheduling method for a virtual power plant according to the present invention. The method further includes steps S21 to S23:

[0102] S21. Calculate the scheduling selection probability of the target user using the roulette wheel algorithm;

[0103] S22. Determine the policy scheduling priority of the target user based on the scheduling selection probability; wherein, the scheduling selection probability is positively correlated with the policy scheduling priority;

[0104] S23. Execute the resource scheduling policy on the target user according to the policy scheduling priority.

[0105] To ensure the efficient operation of the virtual power plant and its ability to respond to constantly changing load demands and unforeseen events, dispatch optimization relies not only on accurate load forecasting but also on real-time adaptability. Building upon the reliable basis for dispatch decisions achieved by optimizing the accuracy and stability of load forecasting, the system further optimizes the dispatch strategy using a roulette wheel algorithm, intelligently and dynamically allocating resources based on real-time feedback and optimized load forecast results. This method, through fitness weighting and time-series dispatch control, ensures that the virtual power plant can efficiently and evenly allocate resources, while avoiding excessive concentration or dispersion of resources.

[0106] In this embodiment of the invention, a roulette wheel algorithm is used to calculate the scheduling selection probability of the target user. The scheduling selection probability is used to characterize the policy scheduling priority for the target user. Generally, the higher the scheduling selection probability, the higher the policy scheduling priority. Then, based on the policy scheduling priority, the resource scheduling policy is executed for the target user.

[0107] Preferably, step S21, namely, calculating the scheduling selection probability of the target user using the roulette wheel algorithm, includes steps S211 to S213:

[0108] S211. Calculate the utility value of the target user; wherein the utility value is used to evaluate the value of the user's behavior;

[0109] S212. Calculate the scheduling selection probability of the target user based on the utility value;

[0110] S213. Introduce a time-series scheduling control method to adjust the scheduling selection probability based on the historical behavior weights of the target user, thereby obtaining the final scheduling selection probability; wherein, the historical behavior weights are used to characterize the target user's responsiveness and reliability to resource scheduling strategies within a historical period.

[0111] In this embodiment of the invention, to better optimize the scheduling strategy of the virtual power plant, a spatiotemporally coupled utility function is first designed, which considers the spatiotemporal distribution characteristics of the electric vehicle (EV) cluster. The design of this utility function helps the system consider the distance between EVs and charging stations when scheduling resources, thereby achieving geographical optimization of resources. This utility function can be expressed as:

[0112]

[0113] in, For users and charging piles At time step The utility value. Indicates user and charging piles At time step Optimize geographical distribution by considering distance. Indicates time step The overall electricity demand reflects market demand. and This is an adjustment coefficient used to balance the effects of different utility factors.

[0114] In the roulette wheel algorithm, each user's scheduling strategy dynamically allocates selection probabilities based on utility values. A user's scheduling selection probability can be calculated using the following formula:

[0115]

[0116] in, For users At time step The scheduling selection probability. For users At time step The utility value.

[0117] To avoid imbalances caused by centralized scheduling, this invention innovatively introduces time-series scheduling control, adjusting the user's scheduling selection probability based on historical load performance and market demand fluctuations. The adjusted selection probability is:

[0118]

[0119] in, For users At time step The historical behavior weights are dynamically adjusted based on the user's historical load performance and response capabilities. For users At time step The current utility value.

[0120] By adjusting the dynamic weighted selection probability, the system can respond more flexibly to load changes, ensure a more balanced allocation of resources among different users, effectively prevent users with low loads from consuming too many resources, and improve the fairness of overall scheduling.

[0121] Preferably, the weight of historical behavior is dynamically adjusted based on the user's historical performance and current load demand. Prioritize scheduling for users with high load demand and good response times. Historical behavior weights are dynamically adjusted within each scheduling cycle according to the following formula:

[0122]

[0123] in, For users Weights based on historical behavior in the previous cycle. This is the maximum value of all user utility values ​​within the current period, used to normalize user utility. The scheduling error for the current period measures the difference between the actual scheduling and the predicted scheduling. This represents the maximum scheduling error within the current period, used to normalize the scheduling error. and This is an adjustment coefficient used to balance the impact of utility and scheduling error on the adjustment of user behavior weights.

[0124] Through this dynamic timing adjustment mechanism, the system can flexibly adjust the user's scheduling priority based on historical behavior and real-time feedback, ensuring that the system responds more accurately and promptly to load fluctuations.

[0125] As a preferred embodiment, the present invention further implements the above embodiments. In step S14, that is, after generating the resource scheduling strategy of the virtual power plant for the user based on the processed load forecast value and the load demand of the target user, the method further includes steps S24 to S27:

[0126] S24. Calculate the scheduling error; wherein the scheduling error is used to represent the difference between the predicted load and the actual load for all users;

[0127] S25. Calculate the utility function values ​​of the global game and the local game through a global-local two-level game strategy, and calculate the global adjustment factor and the local adjustment factor; wherein, the goal of the global game layer is to optimize the resource allocation of the entire virtual power plant to balance the grid load and user demand, and the goal of the local game layer is to optimize the resource scheduling strategy of each user.

[0128] S26. Calculate the game theory adjustment term based on the utility function value of the global game, the utility function value of the local game, the global adjustment factor, and the local adjustment factor;

[0129] S27. The resource scheduling value is modified according to the game theory adjustment term to form the final resource scheduling strategy.

[0130] In this embodiment of the invention, scheduling error The following calculation process is used to determine how the system will reallocate resources and optimize energy use based on the scheduling error in each time period. Assume the scheduling error is... This represents the difference between the predicted load and the actual load for all users, reflecting the accuracy of the entire virtual power plant system's scheduling. The calculation formula is:

[0131]

[0132] in, and users respectively At time step Forecasted load and actual load; This represents the total number of users within the virtual power plant.

[0133] The system dynamically adjusts the parameters of the regression model and resource scheduling based on the scheduling error in each time period to ensure the accuracy and stability of scheduling and avoid resource waste caused by load fluctuations.

[0134] Furthermore, the scheduling strategy of the virtual power plant is optimized using a roulette wheel algorithm. Each user's scheduling strategy is dynamically adjusted based on the calculation of a utility function, which is the core of the optimized scheduling strategy. Within a game theory framework, global and local game layers further utilize these utility functions to coordinate resource allocation and optimize scheduling strategies. Through dynamic adjustment of utility functions and optimization of game strategies, the system ensures more balanced and efficient scheduling among users and resources, thereby achieving adaptive scheduling and maximized resource utilization of the virtual power plant in complex power environments.

[0135] Specifically, the goal of the global game theory layer is to optimize the resource allocation of the entire virtual power plant to balance grid load and user demand. In this process, the utility function... Used to measure each user at time step The utility function optimization formula for global game theory can be written in the following form: (The formula is used to determine the efficiency and load demand, thus providing a basis for resource scheduling.)

[0136]

[0137] At the local game level, each user selects a scheduling strategy based on their own load demand and response capacity. The utility function plays a crucial role in local games, reflecting user preferences and response latency to help the system allocate appropriate resources and scheduling strategies to users. Each user's utility function can be expressed as:

[0138]

[0139] in, To predict load, This represents the actual load. This is for scheduling error; and The adjustment coefficients control the impact of prediction error and scheduling error on utility, respectively.

[0140] This utility function allows users to adjust their dispatching strategies based on the difference between their load forecasting errors and actual load demand, ensuring that the virtual power plant can be dispatched efficiently in a variable load environment.

[0141] To further enhance the system's adaptability and scheduling accuracy, this invention introduces a dynamic penalty factor and a dual adjustment factor to optimize the strategy update process in the game. The penalty factor and adjustment factor help the system cope with fluctuations in user behavior, load changes, and market volatility.

[0142] Each participant (i.e., a user or a scheduling resource) chooses a strategy based on their own utility function. During the game, the system maximizes overall utility by optimizing the participants' strategy choices. The introduction of penalty and adjustment factors can optimize game strategies and prevent excessive deviations from the intended strategy.

[0143] Dynamic penalty factor This penalty factor is used to penalize when the strategy deviates from the system objective, thereby guiding the system towards the optimal solution. The penalty factor can be dynamically adjusted based on user behavior feedback and changes in global scheduling, avoiding excessive or unreasonable scheduling strategies.

[0144]

[0145] in, It is a dynamic penalty factor; These are the system's current global utility and optimal utility, respectively; This is an adjustment coefficient used to control the strength of the penalty factor. When the system utility deviates from the optimal value, the penalty factor increases, thereby prompting the system to adjust its strategy to reduce the deviation.

[0146] To accurately balance global and local scheduling strategies, the system introduces dual adjustment factors: a global adjustment factor and a local adjustment factor. The global and local adjustment factors dynamically adjust the update rate and accuracy of the scheduling strategy, ensuring that the scheduling process is both responsive to changes and maintains decision stability. The update formula for the global adjustment factor is as follows:

[0147]

[0148] The formula for the local adjustment factor is:

[0149]

[0150] in: and For global and local adjustment factors; and The sensitivity coefficient controls the relationship between the adjustment factor and the strategy bias. and These represent the scheduling error for the current time period and the maximum scheduling error, respectively.

[0151] By dynamically adjusting global and local adjustment factors, the system can flexibly adjust its dispatching strategy according to changes in load demand, actual resource utilization, and market fluctuations, ensuring efficient dispatching of virtual power plant resources.

[0152] Finally, through the above multi-level game adjustment process, the system calculates the game theory adjustment term based on the optimization results of global and local strategies, utility function, dynamic penalty factor and adjustment factor, and then dynamically adjusts the resource scheduling value of the target user.

[0153] The final resource scheduling value is the initial calculated resource scheduling value plus the game theory adjustment term. The sum, that is:

[0154]

[0155] In this process, the final scheduling output can be expressed by the following formula:

[0156]

[0157] in, For users At time step The resource scheduling value. For users At time step The load forecast value. For users At time step The actual load value. The scheduling error represents the difference between the predicted load and the actual load. The desired target utility of the system. For the global utility function at time step The current value. This represents the scheduling error for the current time period. This represents the maximum scheduling error for the current time period, used to normalize the scheduling error. , , , , These are adjustment coefficients, used to control the impact of prediction error, historical scheduling, scheduling error, global adjustment factor, and local adjustment factor on scheduling optimization, respectively.

[0158] Using the technical means of this invention, a dynamic adaptive collaborative scheduling system architecture for virtual power plants based on user behavior chain feedback was constructed, achieving intelligent evolution of scheduling strategies through multi-module collaborative innovation. For example... Figure 2As shown, this solution adopts a four-layer progressive architecture of "data-driven modeling - prediction optimization - strategy selection - game theory adjustment". First, it constructs a user behavior chain feedback model through deep reinforcement learning to capture the implicit features of user electricity consumption behavior in real time. Second, it innovatively integrates local weighted regression interpolation and adaptive adjustment mechanisms to improve load prediction accuracy. Then, it uses a time-series optimized roulette wheel algorithm to achieve dynamic optimization of resource allocation strategies. Finally, it achieves intelligent coordination between supply and demand through a global-local two-level game theory framework. The modules are organically linked through a distributed control architecture, with user behavior feedback data running through the entire process, forming a dynamic closed-loop optimization system of "perception-prediction-decision-execution", which significantly improves the system's adaptability to complex user behaviors and its ability to balance multi-objective optimization.

[0159] See Figure 3 This is a schematic diagram of the structure of a resource scheduling device for a virtual power plant provided in an embodiment of the present invention. The embodiment of the present invention provides a resource scheduling device 10 for a virtual power plant, comprising:

[0160] The user scheduling strategy generation module 11 is used to generate a user scheduling strategy based on the collected user behavior data; wherein, the user scheduling strategy is used to characterize the resource allocation tendency for the target user;

[0161] The load forecast calculation module 12 is used to calculate the load forecast value of the target user based on the user behavior data and the user scheduling strategy.

[0162] The load forecast processing module 13 is used to perform curve smoothing on the load forecast value using a preset interpolation algorithm to obtain the processed load forecast value; wherein, the interpolation algorithm is obtained by combining a linear interpolation algorithm and a local weighted regression interpolation algorithm;

[0163] The resource scheduling strategy generation module 14 is used to generate a resource scheduling strategy for the virtual power plant to the user based on the processed load forecast value and the load demand of the target user; wherein, the resource scheduling strategy includes resource scheduling values.

[0164] It should be noted that the resource scheduling device for a virtual power plant provided in this embodiment of the invention is used to execute all the process steps of the resource scheduling method for a virtual power plant in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0165] See Figure 4This is a schematic diagram of the structure of a resource scheduling device for a virtual power plant provided in an embodiment of the present invention. The present invention also provides a resource scheduling device 20 for a virtual power plant, including a processor 21, a memory 22, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the resource scheduling method for a virtual power plant as described in any of the above embodiments.

[0166] This invention also provides a computer-readable storage medium, which includes a stored computer program, wherein the computer program, when running, controls the device where the computer-readable storage medium is located to execute the resource scheduling method of the virtual power plant as described in any of the above embodiments.

[0167] This invention also provides a computer program product, which includes a computer program or computer instructions. When the computer program or computer instructions are executed by a processor, they implement the resource scheduling method for a virtual power plant as described in any of the above embodiments.

[0168] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0169] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A resource scheduling method for a virtual power plant, characterized in that, include: Based on the collected user behavior data, a user scheduling strategy is calculated; wherein, the user scheduling strategy is used to characterize the resource allocation tendency for target users; Calculate the load prediction value for the target user based on the user behavior data and the user scheduling strategy; The load forecast value is smoothed using a preset interpolation algorithm to obtain the processed load forecast value; wherein, the interpolation algorithm is obtained by combining a linear interpolation algorithm and a locally weighted regression interpolation algorithm; Based on the processed load forecast and the load demand of the target user, a resource scheduling strategy for the virtual power plant is generated for the user; wherein, the resource scheduling strategy includes resource scheduling values; The step of calculating the user scheduling strategy based on the collected user behavior data includes: Collect user behavior data; wherein, the user behavior data includes the target user's load demand, response latency, and preference window; The load demand, the response delay, and the preference window are standardized and converted into feature vectors. The feature vector, the feature vectors of the target user's neighboring users, and the adjacency matrix are input into a preset user behavior chain feedback model for processing, and the user scheduling strategy for the target user is output. The adjacency matrix is ​​used to characterize the degree of mutual influence between the target user and its neighboring users, where the neighboring users are other users with behavioral associations with the target user. The user behavior chain feedback model is constructed and trained based on deep reinforcement learning. The step of calculating the load prediction value for the target user based on the user behavior data and the user scheduling strategy includes: Based on the user behavior data, a baseline load forecast value is generated using a traditional time series model or a machine learning model. The user scheduling strategy is used as a dynamic adjustment factor to correct the baseline load forecast value, thereby obtaining the load forecast value for the target user.

2. The resource scheduling method for a virtual power plant as described in claim 1, characterized in that, The process of smoothing the load forecast value using a preset interpolation algorithm to obtain the processed load forecast value includes: Based on the actual load value of the target user at time (t-1) and the predicted load value at time (t+1), a preset linear interpolation algorithm is used to calculate the predicted load value at time t, which is denoted as the first predicted load value. Based on the actual load values ​​of neighboring users at time t, a local weighted regression interpolation algorithm is used to calculate the predicted load value of the target user at time t, which is then used as the second predicted load value; wherein, the neighboring users are other users who have behavioral associations with the target user; The load forecast value is calculated based on the first load forecast value, the second load forecast value, and the preset adaptive weighting coefficient.

3. The resource scheduling method for a virtual power plant as described in claim 1, characterized in that, The method further includes: The roulette wheel algorithm is used to calculate the scheduling selection probability of the target user; The policy scheduling priority of the target user is determined based on the scheduling selection probability; wherein the scheduling selection probability is positively correlated with the policy scheduling priority. The resource scheduling policy is executed on the target user according to the policy scheduling priority.

4. The resource scheduling method for a virtual power plant as described in claim 3, characterized in that, The step of using the roulette wheel algorithm to calculate the scheduling selection probability of the target user includes: Calculate the utility value of the target user; wherein the utility value is used to evaluate the value of the user's behavior; Calculate the scheduling selection probability of the target user based on the utility value; A time-series scheduling control method is introduced to adjust the scheduling selection probability based on the historical behavior weights of the target user, thereby obtaining the final scheduling selection probability; wherein, the historical behavior weights are used to characterize the target user's responsiveness and reliability to resource scheduling strategies within a historical period.

5. The resource scheduling method for a virtual power plant as described in claim 1, characterized in that, After generating the resource scheduling strategy of the virtual power plant for the user based on the processed load forecast value and the load demand of the target user, the method further includes: Calculate the scheduling error; wherein the scheduling error is used to represent the difference between the predicted load and the actual load for all users; Using a global-local two-level game strategy, the utility function values ​​of the global game and the local game are calculated, as well as the global adjustment factor and the local adjustment factor. The goal of the global game layer is to optimize the resource allocation of the entire virtual power plant to balance the grid load and user demand, while the goal of the local game layer is to optimize the resource scheduling strategy of each user. Calculate the game theory adjustment term based on the utility function value of the global game, the utility function value of the local game, the global adjustment factor, and the local adjustment factor; The resource scheduling value is modified according to the game theory adjustment term to form the final resource scheduling strategy.

6. A resource scheduling device for a virtual power plant, characterized in that, include: The user scheduling strategy generation module is used to calculate the user scheduling strategy based on the collected user behavior data; wherein, the user scheduling strategy is used to characterize the resource allocation tendency for the target user; The load forecast calculation module is used to calculate the load forecast value of the target user based on the user behavior data and the user scheduling strategy. The load forecast processing module is used to smooth the load forecast value using a preset interpolation algorithm to obtain the processed load forecast value; wherein, the interpolation algorithm is obtained by combining a linear interpolation algorithm and a local weighted regression interpolation algorithm; A resource scheduling strategy generation module is used to generate a resource scheduling strategy for the virtual power plant to the user based on the processed load forecast value and the load demand of the target user; wherein, the resource scheduling strategy includes resource scheduling values; The user scheduling policy generation module is specifically used for: Collect user behavior data; wherein, the user behavior data includes the target user's load demand, response latency, and preference window; The load demand, the response delay, and the preference window are standardized and converted into feature vectors. The feature vector, the feature vectors of the target user's neighboring users, and the adjacency matrix are input into a preset user behavior chain feedback model for processing, and the user scheduling strategy for the target user is output. The adjacency matrix is ​​used to characterize the degree of mutual influence between the target user and its neighboring users, where the neighboring users are other users with behavioral associations with the target user. The user behavior chain feedback model is constructed and trained based on deep reinforcement learning. The load forecast calculation module is specifically used for: Based on the user behavior data, a baseline load forecast value is generated using a traditional time series model or a machine learning model. The user scheduling strategy is used as a dynamic adjustment factor to correct the baseline load forecast value, thereby obtaining the load forecast value for the target user.

7. A resource scheduling device for a virtual power plant, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the resource scheduling method for a virtual power plant as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the resource scheduling method of the virtual power plant as described in any one of claims 1 to 5.

9. A computer program product, characterized in that, The computer program product includes a computer program or computer instructions, which, when executed by a processor, implement the resource scheduling method for a virtual power plant as described in any one of claims 1 to 5.

Images