A power dispatching method based on a prediction model

Abstract

The application discloses a power dispatching method based on a prediction model and relates to the technical field of power dispatching; the method comprises the following steps: obtaining original power load data in a target data set, preprocessing the original power load data to obtain initial power load data, performing variational modal decomposition on the initial power load data to obtain a plurality of smooth intrinsic modal functions, inputting each smooth intrinsic modal function into a target prediction model to obtain a plurality of predicted intrinsic modal functions, combining each predicted intrinsic modal function to obtain final power load prediction data, and performing power dispatching according to the final power load prediction data; through data collection preprocessing, modal decomposition, model prediction and integration reconstruction, accurate final power load prediction results can be output, strong data support is provided for power dispatching, power grid operation optimization and efficient power resource allocation are realized, and power resource consumption is effectively reduced.

Description

Technical Field

[0001] This invention belongs to the field of power dispatching technology, specifically relating to a power dispatching method based on a predictive model. Background Technology

[0002] Power dispatch is a crucial component of the power system. It involves real-time monitoring and analysis of power load demand, rationally scheduling power generation, optimizing grid operation, and ensuring the stability and reliability of power supply. Simultaneously, power dispatch aims to reduce generation and transmission costs, improving the economic efficiency of the power system. In the face of power load fluctuations or sudden faults, dispatchers rapidly adjust dispatch strategies and reallocate power resources to maintain the balance and normal operation of the power system. Power dispatch is a core means of ensuring the safe, economical, and efficient operation of the power system.

[0003] Existing predictive models for power dispatch suffer from several drawbacks. They are poorly adaptable to complex and stochastic loads, such as difficulty in accurately predicting the power generation of intermittent renewable energy sources like wind and solar, leading to prediction errors. They also lack dynamic response capabilities, causing the model to adjust its predictions lag behind during rapid load changes or sudden faults, affecting the timeliness of dispatch. Furthermore, they are sensitive to data quality; when the input data contains noise, missing data, or anomalies, the model's robustness is poor, easily resulting in erroneous predictions. These limitations hinder the efficient application of predictive models in power dispatch. Summary of the Invention

[0004] The purpose of this invention is to solve the problem of reduced power dispatch efficiency caused by inaccurate prediction of electricity load by prediction models, and to propose a power dispatch method based on prediction models.

[0005] In a first aspect of this invention, a power dispatching method based on a predictive model is first proposed, the method comprising:

[0006] Obtain the raw power load data from the target dataset, and preprocess the raw power load data to obtain initial power load data; the raw power load data is the raw power load data of any one day in the target dataset;

[0007] The initial power load data is subjected to variational mode decomposition to obtain multiple smooth intrinsic mode functions; each smooth intrinsic mode function is input into the target prediction model to obtain multiple predicted intrinsic mode functions as output.

[0008] The final power load forecast data is obtained by combining the predicted intrinsic mode functions.

[0009] Power dispatch is performed based on the final power load forecast data.

[0010] Optionally, the abnormal power load values ​​are processed to obtain initial power load data, including:

[0011] The normalized load data is obtained by performing maximum and minimum normalization on the abnormal power load values;

[0012] Normalized feature data are obtained by performing maximum and minimum normalization on multiple feature values;

[0013] The normalized load data and the normalized feature data are combined to obtain normalized data;

[0014] The normalized data is processed using a sliding window method. The window width is n, and the normalized data segments are extracted and organized into initial power load data with a window sliding step size of 1.

[0015] Optionally, variational mode decomposition is performed on the initial power load data to obtain multiple smooth intrinsic mode functions, including:

[0016] The variational mode decomposition parameters are preset, and the optimal number of modes is determined by the center frequency method; the variational mode decomposition parameters include a penalty factor, an initial center frequency, and a convergence threshold;

[0017] The iterative update based on variational mode decomposition is specifically as follows:

[0018] Construct a constrained objective function, introduce the augmented Lagrangian function as an unconstrained problem, and use the objective algorithm to update the eigenmode functions, center frequencies and Lagrangian operators. Iterate until the numerical difference between adjacent eigenmode functions is less than the convergence threshold, and output multiple smooth eigenmode functions.

[0019] Optionally, each smooth intrinsic mode function is input into the target prediction model to obtain multiple predicted intrinsic mode functions, including:

[0020] The working principle of the target model is as follows:

[0021] Obtain the smooth intrinsic mode function, input the smooth intrinsic mode function into the first convolution module and pass it through multiple 3×1 convolutions to obtain local feature vectors. After batch normalization to stabilize the feature distribution, use the Tanh activation function to filter and output effective feature vectors.

[0022] The effective features are input into the second convolution module and undergo multiple 1×3 convolutions to obtain a refined feature vector;

[0023] The carefully selected feature vector is input into the third convolutional layer and processed through multiple 1×1 convolutions to output a fused feature vector.

[0024] The fused feature vector is sequentially input into the first BiGRU module and the second BiGRU module to obtain a bidirectional temporal feature vector;

[0025] The bidirectional temporal feature vector is input into the fully connected layer to output the predicted intrinsic mode function.

[0026] Optionally, variational mode decomposition is performed on the initial power load data to obtain multiple smooth intrinsic mode functions, including:

[0027] The variational mode decomposition parameters are set, including the penalty factor, initial center frequency, and convergence threshold. The optimal number of modes is determined by the center frequency method.

[0028] The iterative update based on variational mode decomposition is specifically as follows:

[0029] Construct a constrained objective function, introduce the augmented Lagrangian function as an unconstrained problem, and use the objective algorithm to update the eigenmode functions, center frequencies and Lagrangian operators. Iterate until the numerical difference between adjacent eigenmode functions is less than the convergence threshold, and output multiple smooth eigenmode functions.

[0030] Optionally, the final power load forecast data is obtained by combining the predicted intrinsic mode functions, including:

[0031] The predicted intrinsic mode functions are normalized and summed to obtain normalized power load prediction data. The power load prediction data is then recovered by inverse normalization to obtain the final power load prediction data.

[0032] Optionally, power dispatching based on the final power load forecast data includes:

[0033] The average power load within a preset time period is determined in the final power load forecast data. If the average power load is less than the lower limit of the threshold interval, it is considered a low-end interval.

[0034] If the average power load is not less than the lower limit of the threshold interval and not greater than the upper limit of the threshold interval, then it is a flat interval.

[0035] If the average power load is not less than the upper limit of the threshold range, then it is the peak range;

[0036] Dynamic adjustments are made based on the defined low-load, flat-load, and peak-load periods of electricity load.

[0037] Power dispatch is carried out based on the adjusted power load range.

[0038] Optionally, dynamic adjustments can be made based on the defined low-load, flat-load, and peak-load intervals, including:

[0039] Based on the final power load forecast, further time series trend analysis is performed to identify power load fluctuations. The specific steps are as follows:

[0040] The power load values ​​were analyzed using the sliding window method, with a window width of 12 hours and power load data extracted in 1-hour increments.

[0041] Long Short-Term Memory (LSTM) networks are used to model the time series of power loads to capture the short-term and long-term trends of power load changes.

[0042] Based on the trend of power load changes predicted by the LSTM model, the time interval of power load is dynamically adjusted.

[0043] Optionally, power dispatching is performed based on the adjusted power load range, including:

[0044] According to the low-valley period, the power generation frequency of the power generation system is reduced and the charging operation of the energy storage system is started to store the surplus electrical energy; the output of the peak-shaving generator set is reduced or the generator set is scheduled for maintenance.

[0045] Send incentive signals to interruptible load users to encourage them to increase their electricity demand during off-peak hours; adjust the power of interconnection lines in inter-regional power grids to achieve cross-regional consumption and optimized allocation of off-peak electricity.

[0046] Based on the aforementioned flat section interval, the generator set output is maintained, the energy storage system is maintained in a low-power floating charge state, and the generator set is controlled to output power in an orderly manner according to the principle of economic dispatch.

[0047] Optionally, power dispatching is performed based on the adjusted power load range, including:

[0048] According to the peak interval control, the energy storage system performs discharge operation to release the stored electrical energy to supplement the grid supply gap;

[0049] Start the standby peak-shaving generator units to increase the overall power output on the generation side;

[0050] Send demand response instructions to large users, execute interruptible load protocols, or guide them to reduce their electricity load through price signals;

[0051] Optimize the scheduling of flexible loads and perform short-term delay processing for non-emergency power consumption tasks.

[0052] The beneficial effects of this invention are:

[0053] This invention proposes a power dispatching method based on a prediction model. Through data acquisition and preprocessing, it provides high-quality input for power load forecasting; it uses variational mode decomposition to extract key modes, simplifying complex data; the prediction model accurately models each mode, improving prediction accuracy; after dimensionality recovery, the prediction results are more practical, achieving efficient power dispatching, optimizing resource allocation, and improving the efficiency and economy of power grid operation. Attached Figure Description

[0054] The invention will now be further described with reference to the accompanying drawings.

[0055] Figure 1 A flowchart of a power dispatching method based on a prediction model provided in an embodiment of the present invention. Detailed Implementation

[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The term "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and B can represent: A alone, A and B simultaneously, and B alone. Furthermore, descriptions involving "first," "second," etc., in this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" or "second" can explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0057] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] This invention provides a power dispatching method based on a predictive model. See also... Figure 1 , Figure 1 A flowchart illustrating a power dispatching method based on a predictive model, provided as an embodiment of the present invention. The method includes the following steps:

[0059] S101, Obtain the original power load data in the target dataset, and preprocess the original power load data to obtain the initial power load data;

[0060] S102, variational mode decomposition is performed on the initial power load data to obtain multiple smooth intrinsic mode functions;

[0061] S103, input each smooth intrinsic mode function into the target prediction model to obtain multiple predicted intrinsic mode functions;

[0062] S104, combine the predicted intrinsic mode functions to obtain the final power load prediction data;

[0063] S105, power dispatch is carried out based on the final power load forecast data.

[0064] The raw power load data is the raw power load data of any one day in the target dataset;

[0065] The power dispatching method based on a prediction model provided by this invention collects and preprocesses power load data, transforms it into multiple smooth mode functions using variational mode decomposition, and then predicts and integrates these functions using a prediction model to obtain accurate final power load prediction data. This provides accurate data support for power dispatching, optimizes power grid operation and efficiently allocates resources, and reduces the consumption of power resources.

[0066] In one implementation, the target dataset is obtained by statistical analysis of historical power area data.

[0067] In one embodiment, preprocessing the raw power load data to obtain initial power load data includes:

[0068] The raw power load data is divided into preset time periods, with each time period corresponding to a power load value.

[0069] The original power load data is weighted and filled by using the original power load data of adjacent times on the same day and the original power load data of the same time on adjacent days to obtain the power load value without missing data.

[0070] Abnormal power load values ​​that exceed the normal fluctuation range are determined by comparing the power load values ​​without missing values ​​with the threshold.

[0071] The abnormal power load value is corrected by the average of adjacent time points to obtain the power load value without abnormality;

[0072] The initial power load data is obtained by processing the power load values ​​without abnormalities.

[0073] In one implementation, the raw power load data is divided according to a preset time period, for example, divided into two-hour periods. The raw power load data is then weighted and filled using raw power load data from adjacent times on the same day and raw power load data from the same time on adjacent days to obtain a power load value without missing values. The specific calculation is as follows:

[0074]

[0075] in, The original power load data without missing values ​​at time t on day d. , , and The weighting coefficients are set based on data continuity to ensure that the filled values ​​closely reflect the true trend; The load value at time t-1 on day d. The load value at time t+1 on day d. The load value at time t on day d-1. The load value at time t on day d+1.

[0076] In one implementation, the abnormal power load value is corrected using the average of adjacent time points to obtain the abnormal power load value. The process is as follows:

[0077]

[0078]

[0079] in, Let be the load value at time t on day k. The load average is replaced with the load average of adjacent time periods to replace outliers, ensuring data smoothing; and Yes: The anomaly detection threshold at time t is set based on the fluctuation range of historical data; and Let be the power load values ​​at times t-1 and t+1 on day k.

[0080] In one implementation, missing values ​​are eliminated by weighted filling, and outliers are accurately identified and corrected by threshold comparison, ultimately obtaining complete and clean initial power load data, providing high-quality input for subsequent modeling and prediction, and improving prediction accuracy and reliability.

[0081] In one embodiment, processing the power load values ​​without abnormalities to obtain initial power load data includes:

[0082] Normalized load data is obtained by performing maximum and minimum normalization on power load values ​​without abnormalities;

[0083] Normalized feature data are obtained by performing maximum and minimum normalization on multiple feature values;

[0084] The normalized load data and the normalized feature data are combined to obtain normalized data;

[0085] The normalized data is processed using a sliding window method. The window width is n, and the normalized data segments are extracted and organized into initial power load data with a window sliding step size of 1.

[0086] In one implementation, normalization eliminates the dimensional differences between power load data and feature data, improving the stability and accuracy of model training. For example, the sliding window method effectively captures short-term fluctuation patterns by setting the window width to 6 and the step size to 1, and organizes the data into a three-dimensional format suitable for deep learning models. This provides accurate and reliable input for power load forecasting, significantly enhancing the accuracy and reliability of the forecast.

[0087] In one embodiment, variational mode decomposition of the initial power load data yields multiple smoothed intrinsic mode functions, including:

[0088] The variational mode decomposition parameters are preset, and the optimal number of modes is determined by the center frequency method. The variational mode decomposition parameters include the penalty factor, the initial center frequency, and the convergence threshold.

[0089] The iterative update based on variational mode decomposition is specifically as follows:

[0090] Construct a constrained objective function, introduce the augmented Lagrangian function as an unconstrained problem, and use the objective algorithm to update the eigenmode functions, center frequencies and Lagrangian operators. Iterate until the numerical difference between adjacent eigenmode functions is less than the convergence threshold, and output multiple smooth eigenmode functions.

[0091] In one implementation, the process of constructing the constraint objective function is as follows:

[0092]

[0093]

[0094] in, Let k be the eigenmode function. The center frequency of the k-th eigenmode function; The sign for the time partial derivative. The Dirac distribution is used for the Hilbert transform; j is the imaginary unit; This is the original load time series. For constraint symbols;

[0095] Introducing the augmented Lagrangian function transforms the problem into an unconstrained one, as follows:

[0096]

[0097] in, The penalty factor is manually adjusted based on the decomposition target to ensure the accuracy of the reconstructed signal. The value of the Lagrange operator at time t is used to ensure the strictness of the constraints.

[0098] The process of updating the intrinsic mode function using the objective algorithm is as follows:

[0099]

[0100] The center frequency update process is as follows:

[0101]

[0102] The Lagrange operator update process is as follows:

[0103]

[0104] in, This is the frequency domain representation of the k-th eigenmode function in the (n+1)-th iteration; The frequency domain representation of the original data is derived from... Obtained through Fourier transform The frequency domain representation of the Lagrange multipliers is given by... Obtained through Fourier transform; To update the step size to control the convergence speed; Angular frequency represents the frequency variable in the frequency domain and is used to quantify the oscillation speed of a signal;

[0105] The calculation process for the numerical difference between adjacent intrinsic mode functions is as follows:

[0106]

[0107]

[0108] in, The difference between adjacent intrinsic mode functions. The total difference average of k intrinsic mode functions;

[0109] like If the iteration converges, multiple smooth intrinsic mode functions are output.

[0110] In one implementation, the variational mode decomposition parameters are set using the center frequency method, and then optimized iteratively using the augmented Lagrangian function to output a smooth intrinsic mode function. This process can accurately decompose the initial power load data, remove noise, and extract key frequency features, providing a more accurate and smoother mode function for subsequent power load forecasting and improving the input quality of the forecasting model.

[0111] In one embodiment, each smooth intrinsic mode function is input into the target prediction model to obtain multiple predicted intrinsic mode functions as output, including:

[0112] The working principle of the target model is as follows:

[0113] Obtain the smooth intrinsic mode function, input the smooth intrinsic mode function into the first convolution module and pass it through multiple 3×1 convolutions to obtain local feature vectors. After batch normalization to stabilize the feature distribution, use the Tanh activation function to filter and output effective feature vectors.

[0114] The effective features are input into the second convolutional module and undergo multiple 1×3 convolutions to obtain a refined feature vector;

[0115] The carefully selected feature vector is input into the third convolutional layer and processed through multiple 1×1 convolutions to output a fused feature vector.

[0116] The fused feature vector is sequentially input into the first BiGRU module and the second BiGRU module to obtain a bidirectional temporal feature vector;

[0117] The bidirectional temporal feature vector is input into the fully connected layer to predict the intrinsic mode function.

[0118] In one implementation, a multi-layer convolutional module efficiently captures local, multi-dimensional, and fused features of the signal, resulting in comprehensive and accurate feature extraction. Batch normalization and the Tanh activation function work together to stabilize feature distribution and filter effective information. The BiGRU module performs bidirectional processing to accurately grasp temporal dependencies. The fully connected layer outputs the prediction results, making the overall architecture flexible, efficient, and highly adaptable.

[0119] In one embodiment, the final power load forecast data is obtained by combining the predicted intrinsic mode functions, including:

[0120] The predicted intrinsic mode functions are normalized and summed to obtain normalized power load prediction data. The power load prediction data is then recovered by inverse normalization to obtain the final power load prediction data.

[0121] In one implementation, the normalized summation of each predicted intrinsic mode function yields the normalized total load forecast. Then, through inverse normalization, using the original maximum and minimum values ​​from the normalization process, the normalized forecast is restored to the final power load forecast data. This ensures the forecast falls within the original dimensions, improving forecast accuracy and providing a reliable basis for power dispatch.

[0122] In one embodiment, power dispatching based on final power load forecast data includes:

[0123] Determine the average power load within a preset time period in the final power load forecast data. If the average power load is less than the lower limit of the threshold range, it is considered a low-end range.

[0124] If the average power load is not less than the lower limit of the threshold interval and not greater than the upper limit of the threshold interval, then it is a flat interval.

[0125] If the average power load is not less than the upper limit of the threshold range, then it is the peak range;

[0126] Dynamic adjustments are made based on the defined low-load, flat-load, and peak-load periods of electricity load.

[0127] Power dispatch is carried out based on the adjusted power load range.

[0128] In one implementation, power load intervals are divided and power is allocated based on the final power load forecast data. This can accurately match the power demand at different times, reduce costs during off-peak periods, ensure grid stability during peak periods, and ensure reasonable resource transition during flat periods, thereby improving the overall efficiency and effectiveness of power dispatching.

[0129] In one embodiment, dynamic adjustment is performed based on the defined low-load, flat-load, and peak-load intervals, including:

[0130] Further time-series trend analysis is performed based on the final power load forecast to identify power load fluctuations. The specific steps are as follows:

[0131] The power load values ​​were analyzed using the sliding window method, with a window width of 12 hours and power load data extracted in 1-hour increments.

[0132] Long Short-Term Memory (LSTM) networks are used to model the time series of power loads to capture the short-term and long-term trends of power load changes.

[0133] Based on the trend of power load changes predicted by the LSTM model, the time interval of power load is dynamically adjusted.

[0134] One implementation uses a sliding window method combined with an LSTM model to accurately capture short-term and long-term fluctuations in power load and dynamically optimize time period division. Its 12-hour window and 1-hour step size sliding window effectively balance data detail preservation and trend smoothing. LSTM's powerful time series modeling capabilities ensure prediction accuracy, and the dynamic adjustment mechanism updates the interval in real time based on the prediction results, enabling refined and flexible allocation of power resources. This contributes to efficient grid operation and rational energy utilization, reduces management costs, and improves system stability and power supply quality for users.

[0135] In one embodiment, power dispatching based on the adjusted power load range includes:

[0136] During off-peak periods, the power generation frequency of the power generation system is reduced and the charging operation of the energy storage system is activated to store surplus electrical energy; the output of peak-shaving generator sets is reduced or generator sets are scheduled for maintenance.

[0137] Send incentive signals to interruptible load users to encourage them to increase their electricity demand during off-peak hours; adjust the power of interconnection lines in inter-regional power grids to achieve cross-regional consumption and optimized allocation of off-peak electricity.

[0138] Based on the principle of maintaining generator output in the flat section, maintaining the low-power floating charge state of the energy storage system, and controlling the generator output in an orderly manner according to the principle of economic dispatch.

[0139] In one implementation method, energy efficiency and system stability are achieved through dynamic scheduling of off-peak and peak power loads. During off-peak hours, energy storage charging, generator maintenance, user power incentivization, and inter-regional tie line adjustments can avoid power waste, optimize resource allocation, and improve economy and reliability. During peak hours, stable output is maintained, and the system operates in an orderly manner according to the principle of economic dispatch, significantly improving the flexibility and efficiency of power dispatch.

[0140] In one embodiment, power dispatching based on the adjusted power load range includes:

[0141] The energy storage system is controlled to discharge during peak periods to release stored electrical energy to supplement the grid supply gap.

[0142] Start the standby peak-shaving generator units to increase the overall power output on the generation side;

[0143] Send demand response instructions to large users, execute interruptible load protocols, or guide them to reduce their electricity load through price signals;

[0144] Optimize the scheduling of flexible loads and perform short-term delay processing for non-emergency power consumption tasks.

[0145] In one implementation, the power shortage is effectively supplemented by the discharge of the energy storage system in conjunction with the peak-shaving unit, thereby improving the power supply stability of the power grid during peak hours. At the same time, demand response and flexible load adjustment are adopted to rationally guide electricity consumption behavior, alleviate load pressure, and optimize the balance between power supply and demand.

[0146] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A power dispatching method based on a predictive model, characterized in that, The method includes: Obtain the raw power load data from the target dataset, and preprocess the raw power load data to obtain initial power load data; the raw power load data is the raw power load data of any one day in the target dataset; The initial power load data is subjected to variational mode decomposition to obtain multiple smooth intrinsic mode functions; Each smooth intrinsic mode function is input into the target prediction model to obtain multiple predicted intrinsic mode functions as output; The working principle of the target prediction model is as follows: Each smooth intrinsic mode function is obtained, and each smooth intrinsic mode function is input into the first convolution module. After multiple 3×1 convolutions, local feature vectors are obtained. After batch normalization to stabilize the feature distribution, the effective feature vectors are filtered out using the Tanh activation function. The effective feature vector is input into the second convolution module and undergoes multiple 1×3 convolutions to obtain a refined feature vector; The refined feature vector is input into the third convolutional layer and processed through multiple 1×1 convolutions to output a fused feature vector. The fused feature vector is sequentially input into the first BiGRU module and the second BiGRU module to obtain a bidirectional temporal feature vector; The bidirectional temporal feature vector is input into the fully connected layer to predict the intrinsic mode function. The final power load forecast data is obtained by combining the predicted intrinsic mode functions. The time period interval of the power load is determined based on the final power load forecast data; The power load values ​​were analyzed using the sliding window method, with a window width of 12 hours and power load data extracted in 1-hour increments. Long Short-Term Memory (LSTM) networks are used to model the time series of power loads to capture the short-term and long-term trends of power load changes. Based on the trend of power load changes predicted by the LSTM model, the time interval of power load is dynamically adjusted. Power dispatch is carried out based on the adjusted power load time periods.

2. The power dispatching method based on a predictive model according to claim 1, characterized in that, Preprocessing the raw power load data to obtain initial power load data includes: The raw power load data is divided into preset time periods, with each time period corresponding to a power load value; The original power load data is weighted and filled by using the original power load data of adjacent times on the same day and the original power load data of the same time on adjacent days to obtain the power load value without missing data. Based on the comparison between the missing power load value and the threshold, abnormal power load values ​​that exceed the normal fluctuation range are determined; The abnormal power load value is corrected by the average of adjacent time points to obtain the power load value without abnormality. The abnormal power load values ​​are processed to obtain the initial power load data.

3. The power dispatching method based on a predictive model according to claim 2, characterized in that, The initial power load data is obtained by processing the abnormal power load values, including: The normalized load data is obtained by performing maximum and minimum normalization on the abnormal power load values; Normalized feature data are obtained by performing maximum and minimum normalization on multiple feature values; The normalized load data and the normalized feature data are combined to obtain normalized data; The normalized data is processed using a sliding window method. The window width is n, and the normalized data segments are extracted and organized into initial power load data with a window sliding step size of 1.

4. The power dispatching method based on a predictive model according to claim 1, characterized in that, Variational mode decomposition of the initial power load data yields multiple smooth intrinsic mode functions, including: The variational mode decomposition parameters are preset, and the optimal number of modes is determined by the center frequency method; the variational mode decomposition parameters include a penalty factor, an initial center frequency, and a convergence threshold; The iterative update based on variational mode decomposition is specifically as follows: Construct a constrained objective function, introduce the augmented Lagrangian function as an unconstrained problem, and use the objective algorithm to update the eigenmode functions, center frequencies and Lagrangian operators. Iterate until the numerical difference between adjacent eigenmode functions is less than the convergence threshold, and output multiple smooth eigenmode functions.

5. The power dispatching method based on a predictive model according to claim 1, characterized in that, The final power load forecast data is obtained by combining the predicted intrinsic mode functions, including: The predicted intrinsic mode functions are normalized and summed to obtain normalized power load prediction data. The power load prediction data is then recovered by inverse normalization to obtain the final power load prediction data.

6. The power dispatching method based on a predictive model according to claim 1, characterized in that, The time intervals for determining the power load based on the final power load forecast data include: The average power load within a preset time period is determined in the final power load forecast data. If the average power load is less than the lower limit of the threshold interval, it is considered a low-end interval. If the average power load is not less than the lower limit of the threshold interval and not greater than the upper limit of the threshold interval, then it is a flat interval. If the average power load is not less than the upper limit of the threshold range, then it is considered a peak range.

7. The power dispatching method based on a predictive model according to claim 6, characterized in that, Power dispatch is carried out based on the adjusted power load time periods, including: According to the low-valley period, the power generation frequency of the power generation system is reduced and the charging operation of the energy storage system is started to store the surplus electrical energy; the output of the peak-shaving generator set is reduced or the generator set is scheduled for maintenance. Send incentive signals to interruptible load users to encourage them to increase their electricity demand during off-peak hours; adjust the power of interconnection lines in inter-regional power grids to achieve cross-regional consumption and optimized allocation of off-peak electricity. Based on the aforementioned flat section interval, the generator set output is maintained, the energy storage system is maintained in a low-power floating charge state, and the generator set is controlled to output power in an orderly manner according to the principle of economic dispatch.

8. The power dispatching method based on a predictive model according to claim 7, characterized in that, Power dispatch is carried out based on the adjusted power load time periods, including: According to the peak interval control, the energy storage system performs discharge operation to release the stored electrical energy to supplement the grid supply gap; Start the standby peak-shaving generator units to increase the overall power output on the generation side; Send demand response instructions to large users, execute interruptible load protocols, or guide them to reduce their electricity load through price signals; Optimize the scheduling of flexible loads and perform short-term delay processing for non-emergency power consumption tasks.

Images