An intelligent load scheduling method and system for a virtual power plant user terminal

Abstract

The present invention discloses an intelligent load scheduling method and system for a virtual power plant user terminal. The method comprises: obtaining original load data of the virtual power plant user terminal; performing feature extraction based on the original load data to obtain a feature vector; performing load demand forecasting based on the feature vector to obtain a predicted load demand; and adopting a load scheduling model based on game theory based on the predicted load demand to obtain a load scheduling scheme. The load scheduling model includes a local game layer and a global game layer. During each round of the game, the load scheduling model determines the load decision weight of the user terminal based on a fault tolerance factor, and adjusts the load decision of the user terminal based on the fault tolerance factor. The present invention can effectively improve the flexibility and accuracy of load scheduling at the user terminal of the virtual power plant.

Description

Technical Field

[0001] The present invention relates to the field of smart grid control technology, and in particular to an intelligent load scheduling method and system for a virtual power plant user end. Background Art

[0002] In the operation of virtual power plants, user-side load management and control are key factors in improving power system flexibility and stability. Traditional load scheduling methods generally rely on centralized control, which can lead to long response times, insufficient accuracy, and network latency when handling large-scale distributed loads. The development of edge computing technology enables data processing and decision-making close to the load source, enabling more real-time and accurate load scheduling.

[0003] Most existing edge computing methods fail to fully consider the dynamic characteristics of power loads and demand fluctuations, lack intelligent adaptive adjustment mechanisms, and their load scheduling flexibility and accuracy are insufficient. Summary of the Invention

[0004] In order to solve the above technical problems, the present invention provides an intelligent load scheduling method and system for the user end of a virtual power plant, which can effectively improve the flexibility and accuracy of load scheduling.

[0005] An embodiment of the present invention provides an intelligent load scheduling method for a virtual power plant user terminal, comprising:

[0006] Obtain the original load data of the virtual power plant user end;

[0007] Perform feature extraction based on the original load data to obtain a feature vector;

[0008] Performing load demand forecasting based on the characteristic vector to obtain a forecasted load demand;

[0009] According to the predicted load demand, a load dispatching model based on game theory is used to obtain a load dispatching plan;

[0010] The load scheduling model includes a local game layer and a global game layer. During each round of the game, the load scheduling model determines the load decision weight of the user end according to the fault tolerance factor, and adjusts the load decision of the user end based on the fault tolerance factor.

[0011] As an improvement to the above solution, the fault tolerance factor is determined according to the load variation amplitude, equipment health status and external disturbance factors.

[0012] As an improvement to the above solution, the feature extraction is performed based on the original load data to obtain a feature vector, including:

[0013] Performing noise filtering and outlier removal on the original load data to obtain first load data;

[0014] performing data synchronization and standardization processing on the first load data to obtain second load data;

[0015] performing data processing on the second load data using an adaptive fuzzy rule to obtain third load data;

[0016] According to the third load data, a variational mode decomposition method is used to extract features to obtain a feature vector; the feature vector includes a long-term trend component and its dynamic variation, a periodic component and its dynamic variation, and a random fluctuation component.

[0017] As an improvement to the above solution, the method of processing the second load data using an adaptive fuzzy rule to obtain third load data includes:

[0018] Based on a preset fuzzy rule, according to the fluctuation range of the second load data, the parameters of the fuzzy rule are adjusted to obtain an adjusted rule width and a fuzzy set center;

[0019] calculating the membership degree of each fuzzy rule according to the second load data, the adjusted rule width and the fuzzy set center;

[0020] The membership degrees of the fuzzy rules are weighted averaged to calculate and obtain the third load data.

[0021] As an improvement to the above solution, the variational mode decomposition method is used to extract features based on the third load data to obtain a feature vector, including:

[0022] determining the number of decomposition layers of variational modal decomposition according to the fluctuation amplitude of the third load data; the greater the fluctuation amplitude, the more the number of decomposition layers;

[0023] performing variational mode decomposition on the third load data according to the number of decomposition levels to obtain a long-term trend component, a periodic component, and a random fluctuation component corresponding to the third load data;

[0024] A characteristic vector is determined based on the long-term trend component, the periodic component, the random fluctuation component and the corresponding characteristic component weights; wherein the characteristic component weights are dynamically adjusted according to the fluctuation amplitude of each component.

[0025] As an improvement to the above solution, the load demand forecasting according to the characteristic vector to obtain the predicted load demand includes:

[0026] According to the characteristic vector, a long short-term memory network model is used to predict the load demand to obtain the predicted load demand;

[0027] Before load demand forecasting is performed, a weighting coefficient corresponding to the current eigenvector is calculated based on the attenuation coefficient, fluctuation amplitude and time difference of the current eigenvector, and the parameters of the long short-term memory network model are updated based on the weighting coefficient.

[0028] As an improvement to the above solution, the calculation formula of the feature component weight is as follows:

[0029]

[0030] in, For the The weight of the feature components, is the weighted attenuation coefficient, is the fluctuation amplitude of the characteristic component.

[0031] As an improvement to the above solution, the calculation formula of the load decision weight of the user terminal is:

[0032]

[0033] in, is the load decision weight of the i-th user terminal, and is the utility function of the i-th user terminal and the j-th user terminal, and is the fault tolerance factor of the i-th user terminal and the j-th user terminal, and N is the total number of user terminals in the system.

[0034] As an improvement to the above solution, the adjusting the load decision of the user terminal based on the fault tolerance factor includes:

[0035] The load decision of the i-th user terminal in the t+1 round of the game is calculated according to the following formula: :

[0036]

[0037] in, The load decision of the i-th user terminal in the t-th round of the game, To adjust the step size, is the utility function Load decision making The gradient, To adjust the parameters, is the fault tolerance factor of the i-th user terminal.

[0038] The embodiment of the present invention further provides an intelligent load dispatching system for a virtual power plant user terminal, comprising:

[0039] Data acquisition module, used to obtain the original load data of the virtual power plant user end;

[0040] A feature extraction module, configured to extract features based on the original load data to obtain a feature vector;

[0041] A load forecasting module, configured to forecast load demand based on the characteristic vector to obtain a predicted load demand;

[0042] A load scheduling module, configured to obtain a load scheduling plan based on the predicted load demand and a load scheduling model based on game theory;

[0043] The load scheduling model includes a local game layer and a global game layer. During each round of the game, the load scheduling model determines the load decision weight of the user end according to the fault tolerance factor, and adjusts the load decision of the user end based on the fault tolerance factor.

[0044] Compared with the prior art, the beneficial effects of the intelligent load scheduling method and system for the user end of a virtual power plant provided by an embodiment of the present invention are: by extracting features from the original load data, and predicting the load demand based on the extracted features, and then using a load scheduling model based on game theory to obtain a load scheduling plan based on the predicted load demand, wherein the load scheduling model determines the load decision weight of the user end according to the fault tolerance factor in each round of the game, and adjusts the load decision of the user end based on the fault tolerance factor, thereby realizing global and local load coordinated scheduling, ensuring the efficient utilization of power resources, ensuring the robustness of the system, and improving the accuracy and flexibility of the load scheduling of the virtual power plant; by optimizing the data processing process by using an adaptive fuzzy rule adjustment mechanism during the data processing process, the adaptability and accuracy of the load data are further improved; by using variational mode decomposition and adaptive decomposition mechanism to analyze the load data, and dynamically adjusting the number of decomposition layers according to the data fluctuation amplitude, the modeling accuracy can be improved and the computational complexity can be reduced; by using an adaptively updated long short-term memory network to predict the load demand, high-precision load demand prediction is achieved, further improving the accuracy of the load scheduling of the virtual power plant. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is a flow chart of an intelligent load scheduling method for a virtual power plant user end provided by an embodiment of the present invention;

[0046] Figure 2 This is a structural diagram of an intelligent load dispatching system for a virtual power plant user end provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0048] See also Figure 1 , Figure 1 The figure is a flow chart of an intelligent load dispatching method for a virtual power plant user end provided by an embodiment of the present invention. The intelligent load dispatching method for a virtual power plant user end includes:

[0049] S1: Obtain the original load data of the virtual power plant user end;

[0050] Specifically, by deploying edge computing nodes at the user end of the virtual power plant, various power load data, including instantaneous power, are collected in real time through smart meters, load monitoring equipment, sensors and other equipment. , current ,Voltage etc., forming a multi-dimensional load data set , that is, the original load data, its expression is:

[0051]

[0052] in, Edge computing nodes are deployed close to the user end, which can effectively reduce data transmission delays and ensure the real-time and integrity of collected data.

[0053] S2: Extracting features based on the original load data to obtain a feature vector;

[0054] As one of the optional embodiments, extracting features based on the original load data to obtain a feature vector includes:

[0055] Performing noise filtering and outlier removal on the original load data to obtain first load data;

[0056] performing data synchronization and standardization processing on the first load data to obtain second load data;

[0057] performing data processing on the second load data using an adaptive fuzzy rule to obtain third load data;

[0058] According to the third load data, a variational mode decomposition method is used to extract features to obtain a feature vector; the feature vector includes a long-term trend component and its dynamic variation, a periodic component and its dynamic variation, and a random fluctuation component.

[0059] Specifically, after collecting the original load data, in order to eliminate external interference and abnormal data, the original load data needs to be cleaned, including noise filtering and outlier removal;

[0060] Noise filtering: Use fuzzy membership function to evaluate data quality and identify noise data:

[0061]

[0062] in, For data The membership degree of is the mean of the data, is the standard deviation. When the value is less than the set threshold, it is regarded as noise data and is eliminated.

[0063] Outlier elimination: Determine the degree of data deviation through statistical analysis, and eliminate and replace data that deviates from the normal range:

[0064]

[0065] in, To control the parameters, determine the threshold range of data deviation, First load data is obtained for data cleaning.

[0066] Furthermore, to ensure the consistency of data in time and space, the cleaned first load data is synchronized and standardized;

[0067] Data synchronization: Align data from different users based on timestamps to solve the problem of asynchronous transmission of multi-source data:

[0068]

[0069] in, It is the time offset, and synchronization calibration is achieved by adjusting the timestamp of the data.

[0070] Data standardization: Normalize data of different dimensions to ensure that the data is analyzed under a unified standard:

[0071]

[0072] in, and are the minimum and maximum values ​​of the data, respectively. is the second load data after normalization.

[0073] Furthermore, to further improve the adaptability and accuracy of data processing, an embodiment of the present invention designs an adaptive fuzzy rule adjustment mechanism to optimize the data processing process, which can dynamically adjust the parameters of the fuzzy rules based on real-time data fluctuations.

[0074] As one of the optional embodiments, the step of processing the second load data using an adaptive fuzzy rule to obtain third load data includes:

[0075] Based on a preset fuzzy rule, according to the fluctuation range of the second load data, the parameters of the fuzzy rule are adjusted to obtain an adjusted rule width and a fuzzy set center;

[0076] calculating the membership degree of each fuzzy rule according to the second load data, the adjusted rule width and the fuzzy set center;

[0077] The membership degrees of the fuzzy rules are weighted averaged to calculate and obtain the third load data.

[0078] Specifically, the present invention designs a rule base containing multiple rules, covering key factors such as load fluctuations, demand forecast errors, and power supply changes:

[0079] Rule 1: When load fluctuations are large, increase backup power reserves. Specifically, when load fluctuations exceed 10%, increase the charging of energy storage devices or activate backup generators to ensure system stability.

[0080] Rule 2: When the load is stable and low, optimize load distribution and reduce backup power requirements. Specifically, when the load fluctuation rate is less than 5%, optimize load distribution and reduce the use of backup power sources to improve system efficiency.

[0081] Rule 3: Dispatch distributed energy resources in advance when the load increases rapidly. Specifically, when the load growth rate is greater than 5% / minute, dispatch distributed energy resources or energy storage systems to alleviate the load pressure.

[0082] Rule 4: Adjust load distribution when grid frequency fluctuates. When the grid frequency is abnormal (e.g., frequency <49 Hz or >51 Hz), adjust load distribution, activate backup generators, or adjust load response to ensure grid stability.

[0083] Based on the above fuzzy rules, fuzzy reasoning is performed on the second load data, and the membership degree of each fuzzy rule is calculated. The membership degree reflects the matching degree between the input data (i.e., the second load data) and the fuzzy set.

[0084] Specifically, the fuzzy set of the second load data is set as a combination of multiple rules, and each rule has a corresponding membership function for calculating the matching degree of the input data to each rule.

[0085] Calculate the membership degree for the i-th rule:

[0086]

[0087] in, For the The fuzzy set center of the rule represents the central value of the load range described by the rule; is the width of the fuzzy rule. By calculating the membership degree of each rule, we can determine the matching degree of the input data under each rule, and then get the strength of each rule.

[0088] Furthermore, in order to cope with different load environments, the embodiment of the present invention adopts an adaptive adjustment mechanism to update the parameters of the fuzzy rule in real time. Whenever new second load data is obtained, the rule parameters, especially the rule width, are adjusted according to the fluctuation of the second load data. and fuzzy set center , ensuring that the fuzzy rules can accurately reflect the current load status.

[0089] Specifically, the rule parameters are updated according to the following formula:

[0090]

[0091]

[0092] in, and is the change of fuzzy set center and fuzzy rule width, and is the adjustment coefficient, is the fluctuation amplitude of the second load data:

[0093]

[0094] The embodiment of the present invention automatically optimizes fuzzy rules according to load variation trends, thereby enhancing the system's adaptability to abnormal fluctuations and dynamic data.

[0095] Furthermore, the data processed by fuzzy rules is defuzzified and the data processing results are output through weighted average method, that is, the third load data :

[0096]

[0097] in, is the membership degree of the i-th rule, is the weight of the fuzzy rule.

[0098] After data collection, cleaning, synchronization standardization and adaptive fuzzy rule processing, the edge computing node outputs high-quality data results: the third load data and transmit it to the virtual power plant control center to provide accurate data support for subsequent load analysis and scheduling.

[0099] Furthermore, in view of the nonlinear, time-varying and periodic characteristics of the user-side load of the virtual power plant, the embodiment of the present invention adopts variational mode decomposition (VMD) and adaptive decomposition mechanism to decompose the original load data into different components, so as to accurately extract the multi-level characteristics of the user-side load data and improve the accuracy of data modeling.

[0100] As one of the optional embodiments, the method of extracting features using a variational mode decomposition method based on the third load data to obtain a feature vector includes:

[0101] determining the number of decomposition layers of variational modal decomposition according to the fluctuation amplitude of the third load data; the greater the fluctuation amplitude, the more the number of decomposition layers;

[0102] performing variational mode decomposition on the third load data according to the number of decomposition levels to obtain a long-term trend component, a periodic component, and a random fluctuation component corresponding to the third load data;

[0103] A characteristic vector is determined based on the long-term trend component, the periodic component, the random fluctuation component and the corresponding characteristic component weights; wherein the characteristic component weights are dynamically adjusted according to the fluctuation amplitude of each component.

[0104] Specifically, the third load data is decomposed into long-term trend components , periodic component and random fluctuation components :

[0105]

[0106] Among them, the long-term trend component Reflects the overall growth or decline trend of the load; periodic component Capture daily, weekly and seasonal load variations; random fluctuation components Represents short-term fluctuations and noise caused by external disturbances.

[0107] Based on traditional time series decomposition, this paper designs an adaptive decomposition mechanism that dynamically adjusts decomposition parameters to automatically adapt to the changing characteristics of different load data. In particular, it dynamically adjusts the number of decomposition levels based on the fluctuation amplitude of the load data, effectively improving modeling accuracy and reducing computational complexity. First, the fluctuation amplitude of the load data is monitored in real time, and the volatility of the data is evaluated based on statistical characteristics (such as standard deviation, mean, fluctuation amplitude, etc.). The fluctuation amplitude evaluation formula is as follows:

[0108]

[0109] in, is the fluctuation amplitude of the third load data at time t; is the value of the i-th third load data at time t; is the window size used to calculate local fluctuations. The above formula can be used to determine the fluctuation amplitude of the third load data. When the fluctuation amplitude A(t) is greater than the preset threshold, it indicates that the current load data fluctuates significantly. The number of decomposition levels is automatically increased, and vice versa.

[0110] In the case of large fluctuations in load data, the number of VMD decomposition layers is increased to refine the decomposition process. A higher number of layers can better capture fluctuations and trends in different frequency bands, but it also increases computational complexity. To avoid overly complex calculations, the embodiment of the present invention dynamically adjusts the most appropriate number of decomposition layers according to the fluctuation amplitude. The specific adjustment formula is as follows:

[0111]

[0112] in, The minimum number of decomposition layers is preferably set to 2 layers; is the adjustment coefficient. When data fluctuations are large, the number of decomposition layers is increased to improve modeling accuracy; when fluctuations are small, the number of decomposition layers is reduced to reduce computational complexity.

[0113] Furthermore, based on the time series decomposition, the features of each component are extracted to construct a feature vector containing multi-dimensional information: First, the decomposed components are integrated with the historical changes to construct a feature vector :

[0114]

[0115] in, is the long-term trend component, is the periodic component is the random fluctuation component, is the dynamic change of the long-term trend component, It is the dynamic change of the periodic component, reflecting the changing trend of the load.

[0116] Furthermore, the eigenvector is determined by the long-term trend component, the periodic component, the random fluctuation component, and the corresponding characteristic component weights, where the characteristic component weights are dynamically adjusted according to the fluctuation amplitude of each component. Specifically, by calculating the importance score of each characteristic component, the characteristic component weights are dynamically adjusted to ensure that the model can strengthen the influence of key features when the load changes drastically, thereby improving the response speed of the prediction. Specifically, the calculation formula of the characteristic component weights is as follows:

[0117]

[0118] in, For the The weight of the feature components, is the weighted attenuation coefficient, is the fluctuation amplitude of the characteristic component.

[0119] S3: performing load demand prediction based on the characteristic vector to obtain predicted load demand;

[0120] As one of the optional embodiments, performing load demand forecasting according to the characteristic vector to obtain the predicted load demand includes:

[0121] According to the characteristic vector, a long short-term memory network model is used to predict the load demand to obtain the predicted load demand;

[0122] Before load demand forecasting is performed, a weighting coefficient corresponding to the current eigenvector is calculated based on the attenuation coefficient, fluctuation amplitude and time difference of the current eigenvector, and the parameters of the long short-term memory network model are updated based on the weighting coefficient.

[0123] Specifically, to achieve high-precision load forecasting, the embodiment of the present invention adopts a long short-term memory network (LSTM) to perform load demand forecasting, and achieves high-precision load demand forecasting through real-time data updating and adaptive optimization of model parameters.

[0124] The LSTM network is used to model the time series characteristics of the load, and its input is the characteristic vector of the load , the output is the predicted load demand. LSTM internally transmits information through memory cells, and the state of the memory cell determines the output at the current moment. The output update formula is as follows:

[0125]

[0126] in, It is a hidden state, indicating the feature information at the current moment; is the input feature vector; is the weight matrix, is the bias, and f is the activation function. Each time step of LSTM processes the input feature vector and captures the long-term and short-term dependency characteristics of the load by passing the hidden state and cell state at the current moment in the network.

[0127] Furthermore, in the process of load demand forecasting, the long short-term memory network model is adaptively updated through a preset weighted sliding window. Specifically, assuming there is a window containing n input data points, each data point in the window will have a corresponding weight coefficient , this weighted coefficient determines the influence of the data point on the model update.

[0128] Specifically, the weighting coefficient According to the attenuation coefficient , fluctuation range , time difference The calculation formula is:

[0129]

[0130] in, Indicates the current time Hedi Data point moment The time difference between is the attenuation coefficient, which is used to control the intensity of time decay and can be set to 1; It is The fluctuation range of each data point indicates the degree of change of the load data. Controls the attenuation effect of load data over time, that is, the closer the data point is to the current moment, the greater its weight should be. It will affect its weight, and data points with larger fluctuations will have a greater impact on the update of the model. The time difference between the data point and the current moment It will also affect the weights. As time goes by, the influence of data points farther away from the current moment on the model should gradually decrease.

[0131] When new data arrives, according to the calculated weight coefficient Update the LSTM model. The update process is as follows:

[0132] On the one hand, by calculating the weight coefficient of each data point , adjust the learning rate of the model. A larger weighting coefficient corresponds to a larger learning rate, which means that the new load data has a greater impact on the model and the update will be faster. On the other hand, for the current data, the system adjusts the learning rate according to the weighting coefficient. Dynamically update the parameters of LSTM to improve the prediction accuracy. By adjusting the weights of the model and bias , so that the model can better fit the newly arrived data points. The update formula is expressed as:

[0133]

[0134]

[0135] in, and represents the model parameters at the current moment, and is the parameter update calculated by backpropagation.

[0136] S4: According to the predicted load demand, a load scheduling model based on game theory is used to obtain a load scheduling plan; wherein, the load scheduling model includes a local game layer and a global game layer. During each round of the game, the load scheduling model determines the load decision weight of the user end according to the fault tolerance factor, and adjusts the load decision of the user end based on the fault tolerance factor.

[0137] Specifically, the embodiment of the present invention combines game theory with fault tolerance mechanism, and realizes dynamic distribution and optimization of load through multi-level game and real-time fault tolerance adjustment. In the load dispatch process, the load demand forecast value obtained by LSTM network is It will serve as the input for load scheduling at each user end, thereby achieving reasonable load distribution.

[0138] The load dispatch model uses a strategy that combines local and global game strategies. The local game layer optimizes load dispatch by considering the load demand and device status of individual users, while the global game layer coordinates the load decisions of different users to ensure overall system stability.

[0139] Each user terminal in the local game layer calculates its own load demand, equipment status and predicted load change trend (i.e. ) to formulate a dispatching strategy. The goal of the local game is to optimize load dispatch and maximize its utility function, thereby reducing load fluctuations, avoiding overloads, and improving equipment operation stability. The utility function of the local game is as follows:

[0140]

[0141] in, For the The current load of each user terminal; Load demand predicted for LSTM; is the load variation, indicating the change of load; and is the adjustment coefficient, which controls the stability of the target load and the smoothness of load fluctuation respectively; For the Through this utility function, the user side will optimize its own load scheduling strategy to reduce equipment damage and load imbalance, while ensuring that load fluctuations are within a reasonable range to avoid overload or power waste.

[0142] The global game layer needs to coordinate the load decisions of all users to ensure system stability and security. The goal of the global game is to reasonably distribute the load of all users in the virtual power plant to prevent the load fluctuation of any single user from affecting the global stability. The specific formula is as follows:

[0143]

[0144] in, is the total number of clients in the system; and is the adjustment coefficient, which is used to control the balance between load target and load change; is the load decision value of the i-th user terminal at time t; is the target load value of the i-th user terminal at time t; is the load change of the i-th user terminal at time t, The key to the global game is to coordinate the load demands of different user terminals and ensure load balance among all levels of the system, thereby reducing the impact of power fluctuations on equipment and avoiding system instability or equipment damage.

[0145] The local and global games interact through a feedback mechanism. Load decisions generated by local games feed back into the global game, optimizing the system's load scheduling, ensuring global stability while meeting the needs of all users. Conversely, the scheduling results of the global game are fed back to each user to adjust their load decisions. Through repeated game cycles, the system gradually optimizes load scheduling and improves the rationality and stability of overall load distribution.

[0146] Furthermore, during actual load scheduling, the system may experience uncertainty due to factors such as equipment failures, power fluctuations, or information delays. To address these emergencies, the present invention adopts a dynamic fault-tolerant game mechanism that can flexibly adjust the scheduling strategy in the face of failures or load fluctuations to maintain system stability.

[0147] As one of the optional embodiments, the fault tolerance factor is determined according to the load variation amplitude, equipment health status and external disturbance factors.

[0148] Specifically, in the load scheduling process, by introducing the fault tolerance factor , in order to dynamically adjust the game strategy. The fault tolerance factor is determined based on factors such as the magnitude of load changes, equipment health status, and external disturbances. The formula for the fault tolerance factor is:

[0149]

[0150] in, Indicates the equipment health factor, which reflects the impact of the equipment's health status on load scheduling. If the equipment is in a faulty state, A smaller value indicates that the carrying capacity of the equipment is reduced; The load fluctuation factor reflects the impact of the load change on the scheduling strategy. The greater the load fluctuation, the greater the fault tolerance factor, and the system tends to adjust the load distribution strategy. Represents the external environmental factor, considering the impact of external disturbances on load scheduling. If the external conditions are unstable, the fault tolerance factor increases to improve the robustness of the system; 、 、 is an adjustment coefficient used to control the contribution of each factor to the fault tolerance factor. The embodiments of the present invention dynamically adjust the fault tolerance factor to ensure that the system can flexibly adjust the load distribution strategy in the face of sudden load fluctuations, equipment failures, or external interference, thereby ensuring the robustness of scheduling.

[0151] Furthermore, the fault tolerance factor will directly affect the weight of the user end in the load decision-making process. When the system is in an unstable state, the user end with a larger fault tolerance factor will bear more load distribution responsibilities to ensure the stability of the overall load scheduling of the system. The load decision weight at the user end is calculated by dynamically adjusting the fault tolerance factor as follows:

[0152]

[0153] in, is the load decision weight of the i-th user terminal, and is the utility function of the i-th user terminal and the j-th user terminal, and is the fault tolerance factor for the i-th and j-th client, and N is the total number of clients in the system. Clients with larger fault tolerance factors receive more load, ensuring they can cope with sudden equipment failures or load fluctuations, and preventing excessive or uneven system load distribution.

[0154] Furthermore, during each round of the game, the system dynamically adjusts the load distribution of each user terminal according to the value of the fault tolerance factor. For each user terminal, the adjustment of its load decision depends not only on the result of the current game strategy, but also on the fault tolerance factor. The specific adjustment formula is:

[0155]

[0156] in, is the load decision of the i-th user terminal in the t+1-th round of the game, The load decision of the i-th user terminal in the t-th round of the game, To adjust the step size, is the utility function Load decision making The gradient, To adjust the parameters, is the fault tolerance factor of the i-th user terminal. Through this formula, the system can adjust the load distribution strategy according to the change of the fault tolerance factor, so that the system can maintain good stability and robustness in the face of equipment failures, load fluctuations or external interference.

[0157] After the load dispatch model generates a scheduling plan, the user-side devices are controlled to adjust load distribution based on the plan, for example, by adjusting the output power of power generation equipment or changing the charging and discharging strategy of energy storage devices. The user-side devices execute the instructions, implement load control, and provide feedback on their operating status to the virtual power plant control center. The control center then adjusts the load forecast for future time periods based on the deviation between the feedback data and the forecast results. The control strategy is fine-tuned based on actual conditions and environmental changes to optimize load distribution across the entire system.

[0158] The embodiments of this invention aim to enhance the load management and dispatching capabilities of virtual power plants, ensuring accurate dispatching and efficient control amidst dynamic power demand and supply fluctuations. By leveraging advanced edge computing technology and intelligent dispatching algorithms, the operational efficiency and system stability of virtual power plants are optimized, meeting the demands for flexible dispatching and precise control in modern power systems.

[0159] By deploying edge computing nodes on the user side, real-time load data is collected, and combined with noise filtering and data synchronization technologies, data accuracy and real-time performance are guaranteed. To improve data reliability and processing efficiency, fuzzy rules are used to intervene in loads that meet regulation conditions. The rule base covers multiple aspects such as load regulation, load forecasting, and load optimization. By formulating precise regulation thresholds and optimization strategies, the system can ensure efficient response to load fluctuations, effectively avoiding the data delays and errors that may exist in traditional centralized computing, and providing a reliable foundation for subsequent load analysis and scheduling.

[0160] By adaptively analyzing and predicting historical load data and real-time load change information, we can accurately capture changing trends in load demand. By dynamically adjusting the prediction results, we can flexibly respond to load demands in different time periods and automatically adjust the control strategy to cope with large load fluctuations. This adaptive control strategy not only significantly improves the accuracy of load scheduling, but also enhances the virtual power plant's regulatory capabilities and system stability in complex power environments, ensuring that power supply and demand can be balanced during peak periods or periods of large demand fluctuations.

[0161] Game theory and fault-tolerance mechanisms further optimize the load coordination and scheduling process at the user end of the virtual power plant. The load scheduling model establishes a real-time information feedback mechanism between edge computing nodes and the virtual power plant control center, enabling global and local coordinated load scheduling. Game theory models are used to optimize the scheduling strategies of each load unit, maximizing the interests of all parties involved. Fault-tolerance mechanisms ensure system robustness, enabling the system to automatically adjust and maintain stable operation even if some nodes fail. In complex power environments such as power supply fluctuations and sudden changes in demand, the system can automatically adjust load distribution strategies to ensure efficient utilization of power resources while maintaining stable system operation.

[0162] Through this intelligent distributed control mechanism, virtual power plants can achieve precise load scheduling and efficient control in dynamic and complex power environments, thereby significantly improving the overall operating efficiency, system stability and reliability of virtual power plants, providing technical support for intelligent scheduling and load management of virtual power plants, and laying the foundation for flexible scheduling and optimized operation of future power systems.

[0163] Correspondingly, the present invention also provides an intelligent load dispatching system for a virtual power plant user end, which can implement all processes of the intelligent load dispatching method for a virtual power plant user end in the above embodiment.

[0164] See also Figure 2 , Figure 2 This is a schematic diagram of the structure of an intelligent load dispatching system for a virtual power plant user end provided by an embodiment of the present invention. The intelligent load dispatching system for a virtual power plant user end includes:

[0165] The data acquisition module 201 is used to obtain the original load data of the virtual power plant user end;

[0166] A feature extraction module 202 is configured to extract features based on the original load data to obtain a feature vector;

[0167] A load forecasting module 203 is configured to forecast load demand based on the characteristic vector to obtain a predicted load demand;

[0168] A load scheduling module 204 is configured to obtain a load scheduling solution based on the predicted load demand using a load scheduling model based on game theory;

[0169] The load scheduling model includes a local game layer and a global game layer. During each round of the game, the load scheduling model determines the load decision weight of the user end according to the fault tolerance factor, and adjusts the load decision of the user end based on the fault tolerance factor.

[0170] Preferably, the fault tolerance factor is determined according to load variation amplitude, equipment health status and external disturbance factors.

[0171] Preferably, the feature extraction module 202 is specifically used to:

[0172] Performing noise filtering and outlier removal on the original load data to obtain first load data;

[0173] performing data synchronization and standardization processing on the first load data to obtain second load data;

[0174] performing data processing on the second load data using an adaptive fuzzy rule to obtain third load data;

[0175] According to the third load data, a variational mode decomposition method is used to extract features to obtain a feature vector; the feature vector includes a long-term trend component and its dynamic variation, a periodic component and its dynamic variation, and a random fluctuation component.

[0176] Preferably, the adopting of an adaptive fuzzy rule to process the second load data to obtain the third load data comprises:

[0177] Based on a preset fuzzy rule, according to the fluctuation range of the second load data, the parameters of the fuzzy rule are adjusted to obtain an adjusted rule width and a fuzzy set center;

[0178] calculating the membership degree of each fuzzy rule according to the second load data, the adjusted rule width and the fuzzy set center;

[0179] The membership degrees of the fuzzy rules are weighted averaged to calculate and obtain the third load data.

[0180] Preferably, the method of extracting features using a variational mode decomposition method based on the third load data to obtain a feature vector includes:

[0181] determining the number of decomposition layers of variational modal decomposition according to the fluctuation amplitude of the third load data; the greater the fluctuation amplitude, the more the number of decomposition layers;

[0182] performing variational mode decomposition on the third load data according to the number of decomposition levels to obtain a long-term trend component, a periodic component, and a random fluctuation component corresponding to the third load data;

[0183] A characteristic vector is determined based on the long-term trend component, the periodic component, the random fluctuation component and the corresponding characteristic component weights; wherein the characteristic component weights are dynamically adjusted according to the fluctuation amplitude of each component.

[0184] Preferably, the load forecasting module 203 is specifically configured to:

[0185] According to the characteristic vector, a long short-term memory network model is used to predict the load demand to obtain the predicted load demand;

[0186] Before load demand forecasting is performed, a weighting coefficient corresponding to the current eigenvector is calculated based on the attenuation coefficient, fluctuation amplitude and time difference of the current eigenvector, and the parameters of the long short-term memory network model are updated based on the weighting coefficient.

[0187] Preferably, the calculation formula of the feature component weight is as follows:

[0188]

[0189] in, For the The weight of the feature components, is the weighted attenuation coefficient, is the fluctuation amplitude of the characteristic component.

[0190] Preferably, the calculation formula of the load decision weight of the user terminal is:

[0191]

[0192] in, is the load decision weight of the i-th user terminal, and is the utility function of the i-th user terminal and the j-th user terminal, and is the fault tolerance factor of the i-th user terminal and the j-th user terminal, and N is the total number of user terminals in the system.

[0193] Preferably, the adjusting the load decision of the user terminal based on the fault tolerance factor includes:

[0194] The load decision of the i-th user terminal in the t+1 round of the game is calculated according to the following formula: :

[0195]

[0196] in, The load decision of the i-th user terminal in the t-th round of the game, To adjust the step size, is the utility function Load decision making The gradient, To adjust the parameters, is the fault tolerance factor of the i-th user terminal.

[0197] In the specific implementation, the working principle, control process and technical effect of the intelligent load scheduling system at the user end of the virtual power plant provided by the embodiment of the present invention are the same as those of the intelligent load scheduling method at the user end of the virtual power plant in the above embodiment, and will not be repeated here.

[0198] The embodiment of the present invention provides an intelligent load scheduling method and system for the user end of a virtual power plant, which has the following beneficial effects: by extracting features from original load data, and predicting load demand based on the extracted features, and then using a load scheduling model based on game theory to obtain a load scheduling plan based on the predicted load demand, wherein the load scheduling model determines the load decision weight of the user end according to the fault tolerance factor in each round of the game, and adjusts the load decision of the user end based on the fault tolerance factor, thereby realizing global and local load coordinated scheduling, ensuring the efficient utilization of power resources, guaranteeing the robustness of the system, and improving the accuracy and flexibility of the load scheduling of the virtual power plant; by optimizing the data processing process by using an adaptive fuzzy rule adjustment mechanism during the data processing process, the adaptability and accuracy of the load data are further improved; by using variational mode decomposition and an adaptive decomposition mechanism to analyze load data, and dynamically adjusting the number of decomposition layers according to the data fluctuation amplitude, the modeling accuracy can be improved and the computational complexity can be reduced; by using an adaptively updated long short-term memory network to predict load demand, high-precision load demand prediction is achieved, further improving the accuracy of the load scheduling of the virtual power plant.

[0199] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. An intelligent load dispatching method for a virtual power plant user terminal, characterized in that: include: Obtain the original load data of the virtual power plant user end; Perform feature extraction based on the original load data to obtain a feature vector; Performing load demand forecasting based on the characteristic vector to obtain a forecasted load demand; According to the predicted load demand, a load dispatching model based on game theory is used to obtain a load dispatching plan; The load scheduling model includes a local game layer and a global game layer. In each round of the game, the load scheduling model determines the load decision weight of the user end according to the fault tolerance factor, and adjusts the load decision of the user end based on the fault tolerance factor. The fault tolerance factor is determined based on the load variation range, equipment health status and external disturbance factors; The calculation formula of the load decision weight of the user terminal is: in, is the load decision weight of the i-th user terminal, and is the utility function of the i-th user terminal and the j-th user terminal, and is the fault tolerance factor of the i-th user terminal and the j-th user terminal, and N is the total number of user terminals in the system; The adjusting the load decision of the user terminal based on the fault tolerance factor includes: The load decision of the i-th user terminal in the t+1 round of the game is calculated according to the following formula: : in, The load decision of the i-th user terminal in the t-th round of the game, To adjust the step size, is the utility function Load decision making The gradient, To adjust the parameters, is the fault tolerance factor of the i-th user terminal.

2. The intelligent load dispatching method for a virtual power plant user terminal according to claim 1, characterized in that: The extracting features based on the original load data to obtain a feature vector includes: Performing noise filtering and outlier removal on the original load data to obtain first load data; performing data synchronization and standardization processing on the first load data to obtain second load data; performing data processing on the second load data using an adaptive fuzzy rule to obtain third load data; According to the third load data, a variational mode decomposition method is used to extract features to obtain a feature vector; the feature vector includes a long-term trend component and its dynamic variation, a periodic component and its dynamic variation, and a random fluctuation component.

3. The intelligent load dispatching method for a virtual power plant user terminal according to claim 2, characterized in that: The adopting the adaptive fuzzy rule to process the second load data to obtain third load data includes: Based on a preset fuzzy rule, according to the fluctuation range of the second load data, the parameters of the fuzzy rule are adjusted to obtain an adjusted rule width and a fuzzy set center; calculating the membership degree of each fuzzy rule according to the second load data, the adjusted rule width and the fuzzy set center; The membership degrees of the fuzzy rules are weighted averaged to calculate and obtain the third load data.

4. The intelligent load dispatching method for a virtual power plant user terminal according to claim 3, characterized in that: The method of extracting features using a variational mode decomposition method based on the third load data to obtain a feature vector includes: determining the number of decomposition layers of variational modal decomposition according to the fluctuation amplitude of the third load data; the greater the fluctuation amplitude, the more the number of decomposition layers; performing variational mode decomposition on the third load data according to the number of decomposition levels to obtain a long-term trend component, a periodic component, and a random fluctuation component corresponding to the third load data; A characteristic vector is determined based on the long-term trend component, the periodic component, the random fluctuation component and the corresponding characteristic component weights; wherein the characteristic component weights are dynamically adjusted according to the fluctuation amplitude of each component.

5. The intelligent load dispatching method for a virtual power plant user terminal according to claim 4, characterized in that: The step of performing load demand prediction according to the characteristic vector to obtain the predicted load demand includes: According to the characteristic vector, a long short-term memory network model is used to predict the load demand to obtain the predicted load demand; Before load demand forecasting is performed, a weighting coefficient corresponding to the current eigenvector is calculated based on the attenuation coefficient, fluctuation amplitude and time difference of the current eigenvector, and the parameters of the long short-term memory network model are updated based on the weighting coefficient.

6. The intelligent load dispatching method for a virtual power plant user terminal according to claim 4, characterized in that: The calculation formula of the feature component weight is as follows: in, For the The weight of the feature components, is the weighted attenuation coefficient, is the fluctuation amplitude of the characteristic component.

7. An intelligent load dispatching system for a virtual power plant user, characterized in that: include: Data acquisition module, used to obtain the original load data of the virtual power plant user end; A feature extraction module, configured to extract features based on the original load data to obtain a feature vector; A load forecasting module, configured to forecast load demand based on the characteristic vector to obtain a predicted load demand; A load scheduling module, configured to obtain a load scheduling plan based on the predicted load demand and a load scheduling model based on game theory; The load scheduling model includes a local game layer and a global game layer. In each round of the game, the load scheduling model determines the load decision weight of the user end according to the fault tolerance factor, and adjusts the load decision of the user end based on the fault tolerance factor. The fault tolerance factor is determined based on the load variation range, equipment health status and external disturbance factors; The calculation formula of the load decision weight of the user terminal is: in, is the load decision weight of the i-th user terminal, and is the utility function of the i-th user terminal and the j-th user terminal, and is the fault tolerance factor of the i-th user terminal and the j-th user terminal, and N is the total number of user terminals in the system; The adjusting the load decision of the user terminal based on the fault tolerance factor includes: The load decision of the i-th user terminal in the t+1 round of the game is calculated according to the following formula: : in, The load decision of the i-th user terminal in the t-th round of the game, To adjust the step size, is the utility function Load decision making The gradient, To adjust the parameters, is the fault tolerance factor of the i-th user terminal.

Images