Power load dynamic balancing method, system and equipment based on electricity price incentive mechanism

Abstract

The invention discloses a power load dynamic balancing method, system and device based on an electricity price incentive mechanism. Relates to the technical field of power load dynamic balance. The method is specially improved for solving the problems of large load prediction deviation, low regulation and control precision and obvious load rebound effect caused by difficulty in accurately describing time-varying characteristics, nonlinearity and multi-time scale characteristics of power consumption behaviors of users due to adoption of a static or linear response model in an existing electricity price incentive mechanism. Load response is dynamically predicted by constructing a nonlinear autoregressive neural network model, a multi-cycle closed-loop feedback mechanism is combined to optimize an electricity price excitation signal, and load balance constraints are introduced to realize adaptive regulation and control, so that the load response prediction accuracy is improved, multi-cycle dynamic regulation and control are realized, and the system operation stability is enhanced. And remarkable technical progress and beneficial effects are achieved.

Description

Technical Field

[0001] This application relates to the field of dynamic load balancing technology, and in particular to a method, system and equipment for dynamic load balancing based on electricity price incentive mechanisms. Background Technology

[0002] With the deepening of smart grid and electricity market reforms, the power system's ability to regulate load-side resources is increasingly becoming crucial for ensuring the safe and stable operation of the power grid. Electricity prices, as a core economic signal guiding user electricity consumption behavior, demonstrate enormous potential in peak shaving and valley filling, and in optimizing load curves through their dynamic adjustment mechanisms.

[0003] However, existing technologies typically employ static or linear response models to describe the relationship between electricity prices and load, neglecting the time-varying, nonlinear, and multi-timescale characteristics of user response behavior. This leads to significant discrepancies between the load reduction predicted by price adjustments and the actual response. Furthermore, existing methods often fail to incorporate mathematical constraints that quantify the load balancing effect when generating electricity price incentive signals, thus lacking closed-loop feedback. Therefore, there is an urgent need for a dynamic load balancing method that integrates multi-time-period response characteristics and embeds actual constraints to achieve high-precision and highly adaptive load regulation under electricity price incentive mechanisms. Existing technologies urgently need improvement to address these issues. Summary of the Invention

[0004] In order to at least overcome the above-mentioned deficiencies in the prior art, the purpose of this application is to provide a method, system and equipment for dynamic balancing of power load based on electricity price incentive mechanism to solve the above problems.

[0005] Firstly, this application provides a dynamic load balancing method based on electricity price incentive mechanisms, including: Acquire the first electricity price incentive signal and the corresponding first load response data within the first price adjustment period; Based on the first electricity price incentive signal and the first load response data, a first dynamic response model is obtained using a nonlinear autoregressive neural network. Based on the first load response data and the load predicted by the first dynamic response model, a time-period comparison is performed to obtain the first load deviation sequence; Based on the first load deviation sequence and the preset load balance target, a second electricity price incentive signal is obtained; Based on the second electricity price incentive signal, the second load response data within the corresponding second price adjustment cycle is obtained, and the second dynamic response model is obtained based on the first dynamic response model. When the second load response data exceeds the preset threshold, load balancing constraints are applied.

[0006] In one possible implementation, obtaining the first dynamic response model based on a nonlinear autoregressive neural network according to the first electricity price incentive signal and the first load response data includes: Based on the first electricity price incentive signal and the first load response data, historical electricity price sequences and historical load sequences are obtained through processing to construct a first training dataset; wherein, the input data of the first training dataset includes the historical electricity price sequences and the historical load sequences, and the output data is the corresponding predicted load sequences; A nonlinear autoregressive neural network model is constructed, comprising a sequential input layer, three hidden layers, and an output layer. The activation function of the hidden layer is a linear rectified function, and the input data is used as the input sequence of the nonlinear autoregressive neural network. The input sequence length is the historical electricity price sequence and historical load sequence of the previous set number of time periods, and the output is the load forecast value for the next period; The nonlinear autoregressive neural network model is trained based on the training dataset by minimizing the difference between the predicted load sequence and the actual load sequence until the nonlinear autoregressive neural network model converges. The trained nonlinear autoregressive neural network model is used as the first dynamic response model to predict the load response under a given first electricity price incentive signal.

[0007] In one possible implementation, the step of comparing the first load response data with the load predicted by the first dynamic response model on a time-by-time basis to obtain the first load deviation sequence includes: The first load response data is taken as the actual observed load; Based on the predicted load and the actual observed load, a time-by-time comparison is performed, and the deviation value for each time period is obtained through vector operations; Based on the deviation value of each time period, obtain the data label corresponding to each first load response data; The first load deviation sequence is obtained based on the deviation value of each time period and the corresponding data label.

[0008] In one possible implementation, obtaining the second electricity price incentive signal based on the first load deviation sequence and a preset load balance target includes: Based on the first load deviation sequence and the preset load balance target, the pricing objective function is obtained; Based on the pricing objective function, under constraints, the minimum value of the pricing objective function is obtained, which serves as the optimal electricity price incentive signal for the next cycle. The optimal electricity price incentive signal for the next cycle is used as the second electricity price incentive signal.

[0009] In one possible implementation, the step of obtaining second load response data within the corresponding second price adjustment period based on the second electricity price incentive signal, and obtaining a second dynamic response model based on the first dynamic response model, includes: Based on the second electricity price incentive signal, the next price adjustment cycle is obtained and used as the second price adjustment cycle; Obtain the actual response load for the second price adjustment period as the second load response data; The second electricity price incentive signal and the second load response data are added to the first training dataset to obtain the second training dataset; Based on the second training dataset, and using the first dynamic response model, incremental fine-tuning is performed to obtain the second dynamic response model.

[0010] In one possible implementation, the step of adding the second electricity price incentive signal and the second load response data to the first training dataset to obtain the second training dataset further includes: The second electricity price incentive signal and the second load response data are added to the first training dataset, and the training dataset and dynamic response model are updated simultaneously. When the training dataset and dynamic response model update action is triggered, the second electricity price incentive signal and the second load response data are added to the first training dataset. Then, through a sliding window, only the most recent historical data within a set period is retained as the second training dataset.

[0011] In one possible implementation, the step of applying load balancing constraints when the second load response data exceeds a preset threshold includes: Immediately after the end of the second price adjustment cycle, calculate the load stability index of the second load response data; Based on the second load response data, the benchmark load variance is obtained through historical load data within the second price adjustment cycle, and the load stability index corresponding to the second load response data is obtained. When the load stability index corresponding to the second load response data exceeds the preset threshold, load balancing constraints are applied, and a penalty term is added to the electricity price objective function for the next cycle. Specifically, the benchmark load variance is obtained based on historical load data within the first price adjustment cycle, and serves as a preset threshold for the load stability index.

[0012] In one possible implementation, the power load dynamic balancing method based on the electricity price incentive mechanism according to claim 1 is characterized in that the pricing objective function further includes a user response classification mechanism: Based on the deviation sequence, the time lag after the load begins to change significantly after the electricity price change is taken as the response delay, and the load change caused by the unit change in electricity price is taken as the response intensity. The standard deviation of the deviation sequence, the response delay, and the response intensity are used as user clustering features. Based on the user clustering characteristics, and using a clustering algorithm, user response categories are obtained, including high-response users, medium-response users, and low-response users. Based on the user response classification, the load tracking error minimized in the pricing objective function is amplified by a set factor.

[0013] In one possible implementation, the pricing objective function also includes a user response classification mechanism: Based on the deviation sequence, the time lag after the load begins to change significantly after the electricity price change is taken as the response delay, and the load change caused by the unit change in electricity price is taken as the response intensity. The standard deviation of the deviation sequence, the response delay, and the response intensity are used as user clustering features. Based on the user clustering characteristics, and using a clustering algorithm, user response categories are obtained, including high-response users, medium-response users, and low-response users. Based on the user response classification, the load tracking error minimized in the pricing objective function is amplified by a set factor.

[0014] Secondly, this application provides a data acquisition unit for a first price adjustment cycle, a data construction unit for a first dynamic response model, a data acquisition unit for a first load deviation sequence, a pricing unit, and a data construction unit for a second dynamic response model, all connected in sequence. The data acquisition unit for the first price adjustment cycle is configured to: acquire the first electricity price incentive signal and the corresponding first load response data within the first price adjustment cycle; The unit for constructing the first dynamic response model is configured to: obtain the first dynamic response model based on a nonlinear autoregressive neural network according to the first electricity price incentive signal and the first load response data; The unit for obtaining the first load deviation sequence is configured to: compare the first load response data with the load predicted by the first dynamic response model in a time-by-time manner to obtain the first load deviation sequence; The pricing unit is configured to: obtain a second electricity price incentive signal based on the first load deviation sequence and a preset load balance target; The unit for constructing the second dynamic response model is configured to: obtain the second load response data within the corresponding second price adjustment period based on the second electricity price incentive signal, and obtain the second dynamic response model based on the first dynamic response model; The pricing unit is also configured to perform load balancing constraints when the second load response data exceeds a preset threshold.

[0015] Thirdly, this application provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of any one of the methods for dynamic balancing of power load based on an electricity price incentive mechanism.

[0016] In summary, the beneficial effects that this application can achieve are: The power load dynamic balancing method, system, and equipment based on electricity price incentive mechanism proposed in this application solves the load forecasting error through a nonlinear autoregressive neural network, achieving a more accurate characterization of the nonlinearity, time-varying nature, and multi-period dependence of user response. By optimizing the electricity price based on actual deviation feedback in each cycle, it solves the problem of inaccurate and cumulative open-loop control, achieving a closed-loop optimization mechanism. By uniformly considering tracking accuracy, electricity price stability, and load curve smoothness in the pricing objective function, it solves the problem of multi-objective solution, enabling the search for a stable and economical electricity price incentive signal. Through multi-cycle continuous optimization and stability constraints, it solves the problem of load rebound when incentives stop, which is common in traditional methods, achieving a smooth load transition. The method of this application has significant technological advancements and beneficial effects in improving the accuracy of load response forecasting, achieving multi-cycle dynamic control, and enhancing system operational stability. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the method steps in an embodiment of this application; Figure 2 This is a schematic diagram of the system structure according to an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0019] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0020] Example 1 Please refer to the following: Figure 1 This is a schematic diagram of the steps of the dynamic load balancing method based on the electricity price incentive mechanism provided in the embodiment of the present invention. Further, the dynamic load balancing method based on the electricity price incentive mechanism may specifically include the contents described in steps S1 to S7.

[0021] Step S1: Obtain the first electricity price incentive signal and the corresponding first load response data within the first price adjustment period.

[0022] Step S2: Based on the first electricity price incentive signal and the first load response data, obtain the first dynamic response model using a nonlinear autoregressive neural network.

[0023] Step S3: Based on the first load response data and the load predicted by the first dynamic response model, compare them time-by-time to obtain the first load deviation sequence.

[0024] Step S4: Obtain the second electricity price incentive signal based on the first load deviation sequence and the preset load balance target.

[0025] Step S5: Based on the second electricity price incentive signal, obtain the second load response data within the corresponding second price adjustment period, and obtain the second dynamic response model based on the first dynamic response model.

[0026] Step S6: When the second load response data exceeds the preset threshold, load balancing constraints are applied.

[0027] Step S7: Repeat steps 4 to 6 to achieve multi-cycle continuous electricity price incentives and dynamic load balance until the load curve formed by the load response meets the preset stability and economic indicators. The preset stability and economic indicators include a daily load peak-valley difference rate less than a threshold and a daily average electricity price fluctuation standard deviation less than a threshold.

[0028] In the implementation of this application, the electricity price incentive signal is used to guide users to adjust their electricity consumption behavior; the dynamic response model refers to a predictive model that characterizes the relationship between electricity price changes and load response, specifically constructed using a nonlinear autoregressive neural network with three hidden layers, which achieves prediction by capturing nonlinear features in historical data; the load deviation sequence is a time-by-time difference set between predicted load and actual load, generated through vector subtraction, providing a basis for electricity pricing; the load balance constraint mechanism is triggered when load fluctuations exceed the allowable range, using threshold judgment, and achieving hard constraints by adding a penalty term to the pricing objective function.

[0029] In the implementation of this application embodiment, the electricity price signal and corresponding user load data of the initial period are first collected, and an initial prediction model is established using a nonlinear neural network; During model training, the input layer receives historical electricity price and load data for the previous N time periods, the hidden layer extracts nonlinear features through the ReLU activation function, and the output layer generates the load forecast value for the next time period. After each cycle, the point-by-point deviation between the predicted value and the actual value is calculated to form a deviation sequence that includes spatiotemporal characteristics; The deviation sequence and the preset balance target are input into the optimization algorithm to generate a new electricity price signal that takes into account historical deviation compensation. When the load data for a new cycle exceeds the variance threshold, the load balancing constraint mechanism is triggered, and a penalty term is added to the electricity pricing process to limit the load fluctuation range. By periodically updating the training dataset and fine-tuning the model parameters, the predictive model and user response characteristics can be dynamically updated.

[0030] This application effectively solves the problem of dynamic load balance under electricity price incentives. By predicting users' response to dynamic electricity prices, it improves the accuracy of load regulation. Through a closed-loop feedback mechanism, it corrects incentives in real time, suppresses load fluctuation amplitude, and can achieve stable operation over multiple cycles.

[0031] Example 2 Based on Example 1, please refer to the following: Figure 1Furthermore, the dynamic load balancing method based on electricity price incentive mechanisms may specifically include the following:

[0032] Step S1: Obtain the first electricity price incentive signal and the corresponding first load response data within the first price adjustment period.

[0033] Step S2: Based on the first electricity price incentive signal and the first load response data, obtain the first dynamic response model using a nonlinear autoregressive neural network.

[0034] Step S3: Based on the first load response data and the load predicted by the first dynamic response model, compare them time-by-time to obtain the first load deviation sequence.

[0035] Step S4: Obtain the second electricity price incentive signal based on the first load deviation sequence and the preset load balance target.

[0036] Step S5: Based on the second electricity price incentive signal, obtain the second load response data within the corresponding second price adjustment period, and obtain the second dynamic response model based on the first dynamic response model.

[0037] Step S6: When the second load response data exceeds the preset threshold, load balancing constraints are applied.

[0038] Step S7: Repeat steps 4 to 6 to achieve multi-cycle continuous electricity price incentives and dynamic load balance until the load curve formed by the load response meets the preset stability and economic indicators. The preset stability and economic indicators include a daily load peak-valley difference rate less than a threshold and a daily average electricity price fluctuation standard deviation less than a threshold.

[0039] Step S1: Obtain the first electricity price incentive signal and the corresponding first load response data within the first price adjustment period. The first electricity price incentive signal is generated by the power grid dispatch center based on the supply and demand status. The first load response data includes the actual electricity load value of the user in each time period within the period.

[0040] The first electricity price incentive signal is a price directly given by the power grid dispatch center based on the current market, and it also includes a timestamp and a sequence of electricity prices per unit of electricity within each time period during the first price adjustment cycle.

[0041] During the first price adjustment cycle, the first load response data is collected and aggregated synchronously to form the average power consumption or total power consumption every 15 minutes. For example, at the end of each 15-minute cycle, the instantaneous power measurement value within this time period is averaged to generate an average load value. The collected data then undergoes data cleaning, such as removing spike noise through a sliding window midpoint filter, and is then validated, for example, by comparing it with historical data or data from adjacent areas to identify outliers. The resulting first load response data sequence is represented as follows: Where t is the start time of the first price adjustment period. The data sampling interval, The total duration of the first price adjustment cycle is given. In this embodiment, the data sampling interval for both the first and second price adjustment cycles is 15 minutes, and the total duration of the adjustment cycle is set to one day. Therefore, each cycle includes 96 time periods, and the entire day is divided into 96 continuous control periods. The electricity price incentive signal is dynamically updated every 15 minutes. The 15-minute granularity is determined based on considerations of the real-time requirements of grid operation, user response capabilities, and the balance between communication bandwidth and computing resources. Shorter cycles, such as 1 minute, would increase the system burden and the difficulty of user response, while longer cycles, such as 1 hour, may lead to control lag and be unable to effectively cope with the rapid fluctuations of renewable energy.

[0042] In step S2, based on the first electricity price incentive signal and the first load response data, a first dynamic response model is obtained using a nonlinear autoregressive neural network.

[0043] In one possible implementation, based on the first electricity price incentive signal and the first load response data, a first dynamic response model is obtained using a nonlinear autoregressive neural network, including: Based on the first electricity price incentive signal and the first load response data, the historical electricity price sequence and the historical load sequence are processed respectively to construct the first training dataset; wherein, the input data of the first training dataset includes the historical electricity price sequence and the historical load sequence, and the output data is the corresponding predicted load sequence; A nonlinear autoregressive neural network model is constructed, which includes a sequential input layer, three hidden layers, and an output layer. The activation function of the hidden layer adopts a linear rectified function, and the input data is used as the input sequence of the nonlinear autoregressive neural network. The input sequence length is the historical electricity price sequence and historical load sequence for the previous set number of time periods, and the output is the load forecast value for the next period; The nonlinear autoregressive neural network model is trained based on the training dataset by minimizing the difference between the predicted load sequence and the actual load sequence until the nonlinear autoregressive neural network model converges. The trained nonlinear autoregressive neural network model is used as the first dynamic response model to predict the load response under a given first electricity price incentive signal.

[0044] In the implementation of this application embodiment, a nonlinear autoregressive neural network is used to enable our dynamic response model to make full use of historical data to predict future load behavior.

[0045] First, based on the first electricity price incentive signal and the first load response data obtained in step S1, the system continuously accumulates and processes historical data. This historical data includes electricity price sequences and actual load sequences for at least N consecutive 15-minute control periods, with historical electricity price sequences... and historical load sequence This constitutes the first training dataset for training the nonlinear autoregressive neural network; where N is the preset historical data window length. For example, N can be set to 96, which covers the continuous electricity price and load data of the previous 24 hours to capture daily periodicity; if weekly periodicity needs to be captured, N can be extended to 672, which means collecting data for a 7-day period.

[0046] The input data for the first training dataset is constructed using a sliding window approach, with the input vector for each training sample i being... Including the former Historical electricity value and previous periods The historical load values ​​for each time period are represented as follows:

[0047] Where k represents the corresponding time point, and the actual load at the corresponding time point k is denoted as . .

[0048] In this embodiment, It can be set to 4, the electricity price for the first hour. It can be set to 16, the load for the first 4 hours, and contextual features such as day of the week, hour, and ambient temperature can be encoded and concatenated as additional inputs to further improve the model's prediction accuracy.

[0049] Then, a nonlinear autoregressive neural network model is constructed. In this embodiment, an input layer, three sequentially connected hidden layers, and an output layer are used.

[0050] The input layer receives the input data sequence. Its number of neurons is ,in Encode other features, such as the number of hours, days of the week, temperatures, etc. For example, if Including the one-hot encoding of hours (0-23) and days of the week (0-6), there are a total of 24 + 7 = 31 features. =31, the total number of neurons in the input layer can reach 4+16+31=51.

[0051] Three hidden layers are responsible for nonlinear transformation and feature extraction of the input data. To achieve effective feature learning and nonlinear mapping, in this embodiment, the first hidden layer includes 128 neurons, the second hidden layer includes 64 neurons, and the third hidden layer includes 32 neurons. The activation function of all hidden layers is the ReLU activation function, which is computationally efficient and helps the model to be effectively trained in deep networks, accelerating convergence.

[0052] The output layer consists of a single neuron that directly outputs the load forecast for the next time step, such as time step k. Because load forecasting is a regression task, the output layer typically does not use an activation function to directly output the predicted continuous value. This predicted load is denoted as... .

[0053] Next, based on the first training dataset, the nonlinear autoregressive neural network model is trained. The model's internal parameters, including the weight matrix and bias vector between neurons in each layer, are iteratively optimized to minimize the difference between the predicted load sequence and the actual load sequence. Then, the model minimizing the difference is obtained by using the mean squared error as the loss function. The loss function is expressed as:

[0054] Where M is the number of training samples. It is the actual load value of the j-th sample. It is the predicted load value of the j-th sample. This loss function quantifies the average squared deviation between the predicted value and the true value, allowing the model to learn the mapping relationship.

[0055] During training, the existing Adam optimizer is used, and its gradient descent optimization algorithm is employed to adaptively adjust the learning rate of each parameter, thereby accelerating convergence and reducing training time.

[0056] Finally, the trained and validated nonlinear autoregressive neural network model is used as the first dynamic response model. The first dynamic response model can receive the electricity price incentive signal at future time and historical load data as input, and output the corresponding load forecast value. It can capture the nonlinear response behavior of users under different electricity price strategies, including the sensitivity to electricity price changes and the lag of the response.

[0057] In step S3, the first load deviation sequence is obtained by comparing the first load response data with the load predicted by the first dynamic response model on a time-by-time basis.

[0058] In one possible implementation, a first load deviation sequence is obtained by comparing the first load response data with the load predicted by the first dynamic response model on a time-by-time basis, including: The first load response data is used as the actual observed load; Based on the predicted load and the actual observed load, a time-by-time comparison is performed, and the deviation value for each time period is obtained through vector operations; Based on the deviation value for each time period, obtain the data label corresponding to each first load response data; The first load deviation sequence is obtained based on the deviation value and corresponding data label for each time period; The data labels include user ID, weather conditions, and date type. In the first load deviation sequence, the period when the absolute value of the deviation is greater than a preset threshold is marked as a high uncertainty period and given higher weight in subsequent model training.

[0059] When implemented in this application embodiment, the first dynamic response model predicts the accuracy of the first price adjustment cycle and identifies uncertainties in user response behavior, providing key feedback for subsequent electricity price optimization and model updates.

[0060] First, the first load response data obtained in step S1 is used as the actual observed load. ,in This represents the i-th 15-minute control period within the first price adjustment cycle. Simultaneously, using the first dynamic response model obtained in step S2, and inputting the first electricity price incentive signal within this first price adjustment cycle, along with corresponding historical load data and relevant contextual features, the load response within this cycle is predicted to obtain the predicted load. Then, in this forecasting process, we execute at the beginning of each 15-minute cycle to generate the forecasted load curve for that cycle.

[0061] Next, based on the predicted load Compared with actual observed load By performing time-period comparisons and using vector operations, the deviation value for each 15-minute period is precisely calculated. Deviation value The calculation formula is:

[0062] in, Is Actual load value for the time period Is Forecast load values ​​for the specified time period. When Deviation value When the deviation is positive, it indicates that the actual load during that period is higher than the model-predicted load, which may mean that users are not responding enough to the electricity price incentives or that there are other unmodeled electricity demands. When the deviation is negative, it indicates that the actual load is lower than the predicted load, which may mean that users are over-responding to the electricity price incentives or that there is load reduction.

[0063] Then, while calculating the deviation value for each time period, data tags corresponding to each first load response data point are also obtained, including user ID, weather conditions, and date type. The user ID is used to uniquely identify a specific electricity user or user group, and the user ID can be the meter number for each household; weather conditions include ambient temperature, light intensity, wind speed, etc., which can be used to prepare for expanding power sources; date type includes weekdays, weekends, and statutory holidays, and these classifications help to distinguish the electricity consumption characteristics under different daily behavior patterns.

[0064] Based on the deviation value and corresponding data label for each time period, a first load deviation sequence is constructed. This is a multi-dimensional time series, which includes not only the deviation amount for each time period, but also the contextual information related to these deviations.

[0065] Among them, the absolute value of the deviation in the first load deviation sequence is greater than a preset threshold. The time period is marked as an uncertain period. Preset threshold. It is set based on the mean absolute error or root mean square error of load forecasting from historical data, or by expert experience, for example, This can be set to 5% of the average load of a specific user group. In subsequent model training and updates, samples from these uncertain periods are given higher weight. This means that when optimizing model parameters, the model will pay more attention to these difficult-to-predict periods, such as atypical user responses to stimulus signals or sudden surges in electricity consumption, thereby enabling the model to learn how to better cope with complex scenarios that are difficult to predict and improving the model's adaptability.

[0066] In step S4, a second electricity price incentive signal is obtained based on the first load deviation sequence and the preset load balance target.

[0067] In one possible implementation, a second electricity price incentive signal is obtained based on a first load deviation sequence and a preset load balance target, including: Based on the first load deviation sequence and the preset load balance target, the pricing objective function is obtained; Based on the pricing objective function, under constraints, the minimum value of the pricing objective function is obtained, which serves as the optimal electricity price incentive signal for the next cycle. The optimal electricity price incentive signal for the next cycle will be used as the second electricity price incentive signal.

[0068] In the implementation of this application embodiment, the key link to achieve dynamic balance is to utilize the feedback information of the actual user response during the first price adjustment cycle (reflected by the first load deviation sequence) and combine it with the load balance target of the power grid in the next cycle (i.e. the second price adjustment cycle) to dynamically optimize and generate a more targeted and accurate electricity price incentive signal.

[0069] First, a pricing objective function is constructed based on the first load deviation sequence and the preset load balance target provided by the power grid dispatch center.

[0070] Load balancing target This refers to the 15-minute control periods of the power grid during the next second price adjustment cycle. The desired ideal load level, or target curve, is typically set by the power grid dispatch center based on actual power generation and transmission and distribution network constraints. For example, during peak renewable energy output periods, the target load may be set higher to absorb excess power and avoid wind and solar curtailment; conversely, during peak consumption periods, the target load may be lowered to alleviate grid pressure. The pricing objective function j is a multi-objective optimization function that aims to minimize the deviation between the predicted load and the target load, while also considering price stability and guiding user behavior, and incorporating load stability constraints.

[0071] Therefore, we design the pricing objective function J as follows:

[0072] in, This indicates the second price adjustment cycle, which in this embodiment is the same length as the first price adjustment cycle. It is predicted using the current first dynamic response model, which is the updated second dynamic response model, when the electricity price incentive signal is... Time period The load response is given here, and a square term is used so that the subsequent penalty increases nonlinearly with the increase of the deviation. During the time period Pre-set load balancing targets; During the time period The second electricity price incentive signal; It is the electricity price incentive signal from the previous regulation period, such as the electricity price released in the previous 15-minute cycle. It uses the square term instead of the absolute value, making the electricity price smoother and easier to gradient decrease. It is a user weighting factor determined by the user response classification mechanism, which will be detailed in the following implementation method, and is used to apply different incentive intensities to users with different response types. This is a penalty term introduced by the load balancing constraint mechanism, which will be detailed in the following implementation. Its function is to increase the value of the optimization objective function when the load stability does not meet the standard, so as to prompt the system to adjust the electricity price to restore the load stability. λ1, λ2, and λ3 are preset weight coefficients. They are positive real numbers used to balance the importance of load tracking accuracy, electricity price fluctuation stability, and other constraints. A typical weight setting can be that λ1 emphasizes load tracking, λ2 emphasizes electricity price stability, and λ3 ensures load stability.

[0073] The first term in the pricing objective function To minimize the squared deviation between the predicted load and the target load, thereby ensuring that the load curve effectively tracks the grid's preset target curve, a weighting factor is introduced. This enables the model to apply differentiated incentives to users with different responsiveness and enthusiasm, thereby improving the accuracy and fairness of regulation; Second item This is to smooth out changes in electricity prices between adjacent time periods, avoid drastic fluctuations in electricity prices, and thus increase user acceptance of electricity pricing policies; Third item This mechanism is used to incentivize users to adjust electricity prices to save electricity when load stability requirements are not met, thereby restoring load stability.

[0074] Furthermore, based on the pricing objective function, and under the premise of satisfying a series of constraints, the minimum value of the pricing objective function is obtained through numerical optimization. The corresponding electricity price incentive sequence at this point is the optimal electricity price incentive signal for the next period. The constraints include: electricity price range constraints, electricity price change rate constraints, and load balance constraints. The electricity price range is within the preset lower and upper limits of the electricity price. The electricity price change rate further ensures the stability of the electricity price by limiting the maximum change in electricity prices between adjacent periods, avoiding excessive price shocks to users. The load balance constraint ensures that the total load predicted by the method of this invention is roughly balanced with the actual total power output throughout the entire second price adjustment period, thereby maintaining the supply and demand balance of the power grid and preventing large-scale power shortages or surpluses.

[0075] Then, through an iterative algorithm, the optimal solution is approximated, and the pricing objective function J that minimizes the constraints is obtained, which serves as the optimal electricity price incentive signal. .

[0076] Finally, the optimal electricity price incentive signal for the next cycle will be given. This serves as a second electricity price incentive signal. The second electricity price incentive signal will be released to electricity users during the subsequent second price adjustment cycle to guide their electricity consumption behavior, thereby achieving the expected load regulation effect.

[0077] In step S5, based on the second electricity price incentive signal, the second load response data within the corresponding second price adjustment period is obtained, and based on the first dynamic response model, the second dynamic response model is obtained.

[0078] In one possible implementation, based on the second electricity price incentive signal, second load response data within the corresponding second price adjustment period is obtained; and based on the first dynamic response model, a second dynamic response model is obtained, including: Based on the second electricity price incentive signal, the next price adjustment cycle is obtained and used as the second price adjustment cycle; Obtain the actual response load for the second price adjustment period as the second load response data; The second electricity price incentive signal and the second load response data are added to the first training dataset to obtain the second training dataset, and at the same time, the training dataset and the dynamic response model are updated. When the training dataset and dynamic response model update action are triggered. Based on the second training dataset, a sliding window is used to retain only the data from the most recent set period as the second training dataset; Based on the second training dataset, the first dynamic response model is incrementally fine-tuned to obtain the second dynamic response model.

[0079] In one possible implementation, the second electricity price incentive signal and the second load response data are added to the first training dataset to obtain the second training dataset, and the implementation further includes: The second electricity price incentive signal and the second load response data are added to the first training dataset, and the training dataset and dynamic response model are updated simultaneously. When the training dataset and dynamic response model update action is triggered, the second electricity price incentive signal and the second load response data are added to the first training dataset. Then, through a sliding window, only the most recent historical data within a set period is retained as the second training dataset for model updates, avoiding model overfitting caused by historical data redundancy.

[0080] In the implementation of this application embodiment, after the second electricity price incentive signal is released and actually applied to electricity users within the second price adjustment cycle according to the optimal strategy generated in step S4, the power grid dispatch center will continuously collect the actual load response data of users within the second price adjustment cycle in real time, forming second load response data. The collection method and granularity of the second load response data are consistent with those of the first load response data, that is, it is a sequence of actual electricity load values ​​within the second price adjustment cycle, denoted as... .

[0081] The second electricity price incentive signal and the second load response data are added to the first training dataset. Simultaneously, the training dataset and the dynamic response model are updated. The second electricity price incentive signal is used as new input data, and the second load response data is used as new output data. These latest real-time observations are added to the existing training dataset and managed through a sliding window mechanism. Only data from the most recently set number of periods is retained as the second training dataset, such as 96 15-minute control periods from the past 24 hours or 672 15-minute control periods from the past 7 days. When new 15-minute electricity price and load data are added to the sliding window, the oldest 15-minute data is automatically removed. This sliding window mechanism ensures that the model always learns based on the latest and most relevant data, while avoiding overfitting and excessive consumption of computational resources caused by historical data redundancy. This enhances the model's timeliness and generalization ability, enabling it to quickly adapt to seasonal, routine, and sudden changes in user behavior patterns.

[0082] Then, when the training dataset and dynamic response model update action is triggered, incremental fine-tuning is performed based on the second training dataset, the existing first dynamic response model or the updated model from the previous moment, to obtain the second dynamic response model.

[0083] The second dynamic response model therefore inherits the knowledge accumulated from the first dynamic response model and is optimized by incorporating the latest user response data. It represents the system's latest understanding of user load response behavior, thus enabling more accurate prediction of load response at subsequent times, such as the load response in the third price adjustment cycle, providing a more reliable input for the next electricity price optimization.

[0084] In step S6, when the second load response data exceeds a preset threshold, load balancing constraints are applied, and a penalty term is added to the electricity pricing objective function for the next cycle. In one possible implementation, when the second load response data exceeds a preset threshold, load balancing constraints are applied, including: Immediately after the end of the second price adjustment cycle, calculate the load stability index of the second load response data; Based on the second load response data, the benchmark load variance is obtained through historical load data within the second price adjustment cycle, and the load stability index corresponding to the second load response data is obtained. When the load stability index corresponding to the second load response data exceeds the preset threshold, load balancing constraints are applied, and a penalty term is added to the electricity price objective function for the next cycle. Specifically, the benchmark load variance is obtained based on historical load data during the first price adjustment cycle under the condition of no incentive intervention, and is used as the preset threshold for the load stability index.

[0085] In the implementation of this application, the stability of power grid operation is monitored and risk assessed in real time. When abnormal fluctuations in load response are detected, intervention is made by proactively adjusting the electricity price incentive strategy to prevent the load rebound effect from further aggravating or the spread of system instability. This mechanism is an important line of defense for ensuring the safe and stable operation of the power grid.

[0086] First, once the second price adjustment cycle ends, i.e. all the second load response data has been collected, the load stability index of the second load response data is immediately calculated. This load stability index is used to quantify the degree of fluctuation in user load response during this cycle.

[0087] Then, to provide an objective reference for stability, the baseline load variance will be obtained based on the second load response data and historical load data prior to the second price adjustment cycle, expressed as:

[0088] in, The variance of the load during this period is used as an indicator of load stability. The larger the variance, the greater the load fluctuation during the price adjustment cycle, the worse the stability of the load curve, and the more likely there will be severe peak-and-trough phenomena or frequent changes; conversely, the smaller the variance, the smoother the load curve and the more stable the operation. It is the load value of the k-th 15-minute control period in the second load response data; It is the average of all k load values ​​within the cycle; K is the number of 15-minute control periods within the cycle.

[0089] When the load stability index corresponding to the second load response data Exceeding the preset threshold Load balancing constraints are applied at this time. Preset threshold. It is determined based on the historical load variance average during the first price adjustment period before penalties are introduced, under conditions of no incentive intervention (i.e., at baseline operating status). For example, It can be set to the 95th percentile of the daily load variance over the past month, or an empirical value.

[0090] When the load stability index exceeds this threshold, it indicates that the current electricity price incentive strategy may have led to excessive concentration or dispersion of user load, thereby causing new imbalances, such as unexpected load spikes at one time or load collapses at another time. This is usually a manifestation of the load rebound effect.

[0091] The load balancing constraint mechanism adds a penalty term to the electricity pricing objective function in the next cycle, such as the third price adjustment cycle, which is defined in step S4 as J. To achieve this, it is represented as:

[0092] Where α is a preset parameter that adjusts the intensity of the penalty to be greater than zero, when When the load stability index exceeds the threshold, the penalty term is a positive value, and its magnitude is quadratic with the degree to which the load stability index exceeds the threshold. That is, the more it exceeds the threshold, the heavier the penalty. This penalty term will be directly added to the pricing objective function in step S4 as its third component.

[0093] Step S7: Repeat steps 4 to 6 until the load curve formed by the load response meets the preset stability and economic indicators.

[0094] This invention is a closed-loop feedback control mechanism that achieves dynamic balance of power load by cyclically executing electricity price incentives and load responses, and then updating the model.

[0095] The cyclical execution process continues with a minimum time granularity of 15 minutes. At the end of each 15-minute control period, the system calculates the load deviation sequence and assesses load stability based on the latest acquired load response data and the updated dynamic response model (step S5) (steps S3 and S6). Then, based on this real-time feedback information and the latest load balance target of the power grid, it optimizes and generates the electricity price incentive signal for the next 15-minute control period (step S4), and publishes the obtained electricity price incentive signal for this stage to electricity users in real time. This cycle repeats, and the electricity price incentive signal is dynamically updated every 15 minutes, continuously guiding users to adjust their electricity consumption behavior. This high-frequency iteration ensures that the system can quickly respond to changes in power grid supply and demand and effectively manage load fluctuations.

[0096] The cycle will continue until the load curve formed by the load response meets the preset stability and economic indicators. The preset stability and economic indicators include: the daily load peak-valley difference rate is less than the preset threshold and the daily average electricity price fluctuation standard deviation is less than the preset threshold. Then, when the preset conditions of stability and economic indicators are met simultaneously for several consecutive days, it can be determined that the desired dynamic load balance has been achieved.

[0097] In one possible implementation, the pricing objective function also includes a user response classification mechanism: Based on the deviation sequence, the time lag after the load begins to change significantly after the electricity price change is taken as the response delay, and the load change caused by the unit change in electricity price is taken as the response intensity. The standard deviation of the deviation sequence, the response delay, and the response intensity are used as user clustering features. Based on user clustering characteristics and using a clustering algorithm, user response categories are obtained, including high-response users, medium-response users, and low-response users. Based on the user response classification, the load tracking error in the pricing objective function is increased by a set factor.

[0098] A stronger incentive is applied to low-response users by multiplying their electricity price elasticity coefficient by 1.5 times and multiplying the weight of their corresponding load tracking error term by 1.5; for high-response users, the incentive is multiplied by 0.5, thus moderately reducing the incentive intensity; for medium-response users, the incentive remains unchanged. This approach better incentivizes resource conservation. A stronger incentive signal is applied to low-response users, i.e., a heavier penalty is imposed on their load deviation, to encourage them to adjust their electricity consumption behavior more actively, thereby avoiding resource waste caused by their insensitivity to prices. Over-incentivizing high-response users is also avoided, thus achieving load balance while also considering user experience.

[0099] Example 3 This is the third embodiment of the present invention. Based on embodiments 1 and 2, please refer to the following references. Figure 2 This embodiment provides a dynamic load balancing system based on an electricity price incentive mechanism, including a sequentially connected unit for collecting data of a first price adjustment cycle, a unit for constructing a first dynamic response model, a unit for obtaining a first load deviation sequence, a pricing unit, and a unit for constructing a second dynamic response model.

[0100] The data acquisition unit for the first price adjustment cycle is configured to: acquire the first electricity price incentive signal and the corresponding first load response data within the first price adjustment cycle; The first dynamic response model unit is configured to: obtain the first dynamic response model based on a nonlinear autoregressive neural network according to the first electricity price incentive signal and the first load response data; The unit for obtaining the first load deviation sequence is configured to: compare the first load response data with the load predicted by the first dynamic response model on a time-by-time basis to obtain the first load deviation sequence; The pricing unit is configured to obtain a second electricity price incentive signal based on a first load deviation sequence and a preset load balance target; The second dynamic response model unit is configured to: obtain the second load response data within the corresponding second price adjustment cycle based on the second electricity price incentive signal, and obtain the second dynamic response model based on the first dynamic response model; The pricing unit is also configured to perform load balancing constraints when the second load response data exceeds a preset threshold.

[0101] In the implementation of this application embodiment, the data acquisition unit for the first price adjustment cycle refers to the unit used to acquire the electricity price signal and user electricity load data released by the power grid dispatch center in real time; the unit for constructing the first dynamic response model refers to the unit that establishes the mapping relationship between electricity price and load response based on the neural network algorithm and forms the first dynamic response model by training historical data; the unit for acquiring the first load deviation sequence refers to the computational unit that performs the difference analysis between the predicted value and the measured value and generates the deviation data sequence by comparing point by point; the pricing unit determines the optimal electricity price by solving the pricing objective function; the unit for constructing the second dynamic response model is to continuously optimize the prediction model by integrating new data in the new price cycle, update the first dynamic response model, and enable the dynamic response model to achieve self-adaptation through incremental learning algorithm.

[0102] In this embodiment, the system acquires the electricity price incentive signal released by the power grid and user electricity consumption data in real time through the acquisition unit, builds an initial load response model using a neural network, generates a deviation sequence by comparing the predicted and measured loads through the deviation analysis unit, calculates the optimal electricity price incentive based on the deviation sequence and the balance target, and dynamically adjusts the prediction model according to the new cycle data. When the load response is detected to exceed the threshold, the system automatically triggers a constraint mechanism to add a penalty term to the pricing process to enhance the control effect. The whole process forms a closed-loop feedback, and the units are sequentially connected through data streams to achieve multi-cycle continuous optimization.

[0103] Compared to existing technologies, traditional load control systems often employ fixed models and unidirectional control mechanisms, which cannot adapt to dynamically changing user response characteristics. This system, by embedding a dynamic response model and a closed-loop feedback architecture, achieves real-time updates of model parameters and adaptive adjustments to the incentive strategy, solving the problem of accumulated prediction bias in static models. Simultaneously, the introduction of constraint mechanisms effectively avoids excessive fluctuations during load control, improving the stability of system operation.

[0104] Through the above technical solutions, this application effectively overcomes the problems of insufficient modeling of user response time-varying characteristics and lack of closed-loop correction in the control process in the prior art; through continuous optimization of the dynamic response model, the matching accuracy between the electricity price incentive strategy and the actual load response is improved; through iterative training of multi-period data, the model's adaptability to electricity consumption characteristics in different time periods is enhanced; and through the constraint mechanism, the risk of load imbalance caused by extreme electricity price fluctuations is prevented, thus realizing precise control of the dynamic balance between power supply and demand.

[0105] Example 4 The fourth embodiment of the present invention differs from the previous embodiments in that: Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0106] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, or may be electrical, mechanical, or other forms of connection.

[0107] The units described as separate components may or may not be physically separate. As will be apparent to those skilled in the art, the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0108] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0109] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or grid device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0110] The above specific embodiments further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A dynamic load balancing method based on electricity price incentive mechanisms, characterized in that, include: Acquire the first electricity price incentive signal and the corresponding first load response data within the first price adjustment period; Based on the first electricity price incentive signal and the first load response data, a first dynamic response model is obtained using a nonlinear autoregressive neural network. Based on the first load response data and the load predicted by the first dynamic response model, a time-period comparison is performed to obtain the first load deviation sequence; Based on the first load deviation sequence and the preset load balance target, a second electricity price incentive signal is obtained; Based on the second electricity price incentive signal, the second load response data within the corresponding second price adjustment cycle is obtained, and the second dynamic response model is obtained based on the first dynamic response model. When the second load response data exceeds the preset threshold, load balancing constraints are applied.

2. The method for dynamic power load balancing based on electricity price incentive mechanism according to claim 1, characterized in that, The step of obtaining a first dynamic response model based on the first electricity price incentive signal and the first load response data, using a nonlinear autoregressive neural network, includes: Based on the first electricity price incentive signal and the first load response data, historical electricity price sequences and historical load sequences are obtained by processing them respectively to construct a first training dataset; wherein, the input data of the first training dataset includes the historical electricity price sequence and the historical load sequence, and the output data is the corresponding predicted load sequence; A nonlinear autoregressive neural network model is constructed, comprising a sequentially connected input layer, three hidden layers, and an output layer. The activation function of the hidden layer is a linear rectified function, and the input data is used as the input sequence of the nonlinear autoregressive neural network. The input sequence length is the historical electricity price sequence and historical load sequence of the previous set number of time periods, and the output is the load forecast value for the next period; The nonlinear autoregressive neural network model is trained based on the training dataset by minimizing the difference between the predicted load sequence and the actual load sequence until the nonlinear autoregressive neural network model converges. The trained nonlinear autoregressive neural network model is used as the first dynamic response model to predict the load response under a given first electricity price incentive signal.

3. The method for dynamic power load balancing based on electricity price incentive mechanism according to claim 1, characterized in that, The step of comparing the first load response data with the load predicted by the first dynamic response model in a time-period manner to obtain the first load deviation sequence includes: The first load response data is taken as the actual observed load; Based on the predicted load and the actual observed load, a time-by-time comparison is performed, and the deviation value for each time period is obtained through vector operations; Based on the deviation value of each time period, obtain the data label corresponding to each first load response data; The first load deviation sequence is obtained based on the deviation value of each time period and the corresponding data label.

4. The method for dynamic power load balancing based on electricity price incentive mechanism according to claim 1, characterized in that, The step of obtaining the second electricity price incentive signal based on the first load deviation sequence and the preset load balance target includes: Based on the first load deviation sequence and the preset load balance target, the pricing objective function is obtained; Based on the pricing objective function, under constraints, the minimum value of the pricing objective function is obtained, which serves as the optimal electricity price incentive signal for the next cycle. The optimal electricity price incentive signal for the next cycle is used as the second electricity price incentive signal.

5. The dynamic load balancing method based on electricity price incentive mechanism according to any one of claims 2 or 4, characterized in that, The step of obtaining the second load response data within the corresponding second price adjustment period based on the second electricity price incentive signal, and obtaining the second dynamic response model based on the first dynamic response model, includes: Based on the second electricity price incentive signal, the next price adjustment cycle is obtained and used as the second price adjustment cycle; Obtain the actual response load for the second price adjustment period as the second load response data; The second electricity price incentive signal and the second load response data are added to the first training dataset to obtain the second training dataset; Based on the second training dataset, and using the first dynamic response model, incremental fine-tuning is performed to obtain the second dynamic response model.

6. The method for dynamic power load balancing based on electricity price incentive mechanism according to claim 5, characterized in that, The step of adding the second electricity price incentive signal and the second load response data to the first training dataset to obtain the second training dataset further includes: The second electricity price incentive signal and the second load response data are added to the first training dataset, and the training dataset and dynamic response model are updated simultaneously. When the training dataset and dynamic response model update action is triggered, the second electricity price incentive signal and the second load response data are added to the first training dataset. Then, through a sliding window, only the most recent historical data within a set period is retained as the second training dataset.

7. The method for dynamic power load balancing based on electricity price incentive mechanism according to claim 1, characterized in that, The step of applying load balancing constraints when the second load response data exceeds a preset threshold includes: Immediately after the end of the second price adjustment cycle, calculate the load stability index of the second load response data; Based on the second load response data, the benchmark load variance is obtained through historical load data within the second price adjustment cycle, and the load stability index corresponding to the second load response data is obtained. When the load stability index corresponding to the second load response data exceeds the preset threshold, load balancing constraints are applied, and a penalty term is added to the electricity price objective function for the next cycle. Specifically, the benchmark load variance is obtained based on historical load data within the first price adjustment cycle, and serves as a preset threshold for the load stability index.

8. The dynamic load balancing method based on electricity price incentive mechanism according to claim 4, characterized in that, The pricing objective function also includes a user response classification mechanism: Based on the deviation sequence, the time lag after the load begins to change significantly after the electricity price change is taken as the response delay, and the load change caused by the unit change in electricity price is taken as the response intensity. The standard deviation of the deviation sequence, the response delay, and the response intensity are used as user clustering features. Based on the user clustering characteristics, user response classification is obtained using a clustering algorithm; Based on the user response classification, the load tracking error term in the pricing objective function is increased by a set factor.

9. A dynamic load balancing system based on electricity price incentive mechanisms, characterized in that, The system includes a sequential electrical connection unit for acquiring data from the first price adjustment cycle, a unit for constructing a first dynamic response model, a unit for obtaining a first load deviation sequence, a pricing unit, and a unit for constructing a second dynamic response model. The data acquisition unit for the first price adjustment cycle is configured to: acquire the first electricity price incentive signal and the corresponding first load response data within the first price adjustment cycle; The unit for constructing the first dynamic response model is configured to: obtain the first dynamic response model based on a nonlinear autoregressive neural network according to the first electricity price incentive signal and the first load response data; The unit for obtaining the first load deviation sequence is configured to: compare the first load response data with the load predicted by the first dynamic response model in a time-by-time manner to obtain the first load deviation sequence; The pricing unit is configured to: obtain a second electricity price incentive signal based on the first load deviation sequence and a preset load balance target; The unit for constructing the second dynamic response model is configured to: obtain the second load response data within the corresponding second price adjustment period based on the second electricity price incentive signal, and obtain the second dynamic response model based on the first dynamic response model; The pricing unit is also configured to perform load balancing constraints when the second load response data exceeds a preset threshold.

10. A computer system device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 8.

Images