Power load prediction method and system based on dynamic driving

Abstract

The invention relates to the field of power load fluctuation prediction, and provides a power load prediction method and system based on dynamic driving. The power load prediction method based on dynamic driving comprises the following steps: performing dynamic modeling on an obtained historical power load time training sequence by adopting a deterministic learning algorithm, and extracting internal dynamic information of data; on the basis of dynamic information, a multi-step forward iteration predictor based on a fourth-order Runge-Kutta model is adopted, and a power load prediction sequence composed of prediction values from the step 1 to the step 2 is obtained; a corrector is adopted to perform periodic correction on a power load prediction sequence formed by the prediction values from the step 1 to the step 2, and a power load prediction sequence correction value of the step 1 is obtained; and obtaining a power load prediction result based on the prediction value and the correction value. And the power load prediction precision is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power load fluctuation prediction, and in particular to a power load prediction method and system based on dynamic drive. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Power load series, a typical type of chaotic time series, are special time series data generated by the complex nonlinear dynamics of the power system. Their dynamic behavior exhibits complex, irregular, and seemingly random characteristics. Because power systems are extremely sensitive to initial conditions—even slight differences in initial load states can be exponentially amplified over time—the long-term evolution of power loads is difficult to accurately predict. Therefore, accurate power load forecasting has always been a major challenge in power system operation and scheduling. However, power load forecasting plays an irreplaceable and important role in the power industry. Its results are directly related to the safe and stable operation of the power system, the rational formulation of power generation plans, orderly transactions in the power market, and the efficient utilization of energy resources. For example, accurate short-term load forecasting provides a basis for optimizing daily unit commitments, while medium- and long-term load forecasting is a key reference for grid planning and power generation construction. Effective power load forecasting can provide critical decision support for practical scenarios such as economic dispatch of power systems, fault warning, and renewable energy integration, and has important theoretical research value and significant practical benefits. Currently, power load forecasting methods are primarily categorized into two main categories: traditional statistical modeling and intelligent modeling. Traditional statistical modeling methods, such as the autoregressive moving average (ARIMA) and autoregressive conditional heteroskedasticity (ARCH), perform forecasting by establishing mathematical models under linear or stationary assumptions. However, these methods struggle to adapt to the highly nonlinear and nonstationary characteristics of power load series, which can be influenced by user behavior, meteorological factors, and social events. Consequently, their forecasting accuracy is limited, making it difficult to meet the refined scheduling requirements of modern power systems.

[0004] With the development of machine learning technology, intelligent model methods have gradually become the mainstream means of power load forecasting. Artificial neural networks, with their powerful nonlinear approximation capabilities, have been widely used. These include feedforward neural networks such as multilayer perceptrons (MLPs) and radial basis functions (RBFs), as well as recurrent neural networks such as long short-term memory (LSTM) and echo state networks (ESNs). Furthermore, the attention-based Transformer model has also been introduced into the field of power load forecasting and has demonstrated excellent performance in some scenarios. Although intelligent model methods have promoted improvements in power load forecasting accuracy to a certain extent, the following key issues still exist in practical applications: 1. Existing intelligent models mostly use a data-driven static mapping approach, focusing only on the superficial correlation between historical and future load data. They ignore the inherent dynamic evolution of power load sequences as chaotic systems. As a result, the models cannot fully explore the underlying driving mechanisms of load changes, resulting in insufficient overall prediction potential. 2. In multi-step iterative forecasting scenarios, existing models generally lack targeted error correction mechanisms and are unable to effectively suppress the error diffusion effect unique to chaotic systems. This leads to a sharp decline in the accuracy of medium- and long-term power load forecasts, making it difficult to support medium- and long-term planning decisions for the power grid. 3. Most mainstream neural network models are "black box" structures. Their prediction process relies on implicit fitting of a large number of parameters, which cannot clearly reveal the dynamic mechanism of power load sequence evolution, resulting in a lack of interpretability of prediction results. This shortcoming is particularly prominent in power systems with extremely high safety and reliability requirements, and seriously restricts the application of these models in critical power dispatch and control scenarios. Summary of the Invention In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides a power load forecasting method and system based on dynamic driving. The present invention can integrate chaotic dynamic characteristics, has error suppression capabilities and strong interpretability, thereby improving the accuracy and practicality of power load forecasting, and providing more powerful technical support for the safe, economical and efficient operation of the power system.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: A first aspect of the present invention provides a power load forecasting method based on dynamic driving.

[0006] A power load forecasting method based on dynamic driving, comprising: A deterministic learning algorithm is used to perform dynamic modeling on the acquired historical power load time training sequence to extract the inherent dynamic information of the data; Based on the dynamic information, a multi-step forward iterative predictor based on the fourth-order Runge-Kutta model is used to obtain the Step to The power load forecast sequence composed of step forecast values; The error corrector is used to Step to The power load forecast sequence composed of the first step forecast value is periodically corrected to obtain the first The power load forecast sequence correction value of the first step; The first predicted value and the The power load forecasting result is obtained by correcting the power load forecasting sequence of the first step.

[0007] Furthermore, after obtaining the historical power load time training sequence, data cleaning is performed on the historical power load time training sequence, and after cleaning, the historical power load time training sequence is standardized or normalized.

[0008] Furthermore, the method of using a deterministic learning algorithm to perform dynamic modeling on the acquired historical power load time training sequence includes: A dynamic identifier is constructed using an RBF neural network, and a weight update law based on Lyapunov stability design is adopted to accurately model the dynamics of the unknown system along the system sampling trajectory:

[0009] in, represents the unknown system dynamics, is the constant weight of the RBF neural network after training convergence, is the RBF neural network regression vector, is the modeling error.

[0010] Furthermore, the multi-step forward iterative predictor based on the fourth-order Runge-Kutta model is expressed by the following formula:

[0011] in, For the The predicted value of the step, is the sampling time, is the constant weight of the embedded neural network, that is, the learned dynamics knowledge, Indicates the The instantaneous dynamics of the step, Indicates based on The first ( )-step prediction dynamics, Indicates based on The first ( ) step to improve the dynamics, Indicates based on The first ( )-step prediction dynamics, to Need to calculate in sequence and finally get the The predicted value of the step ; when When , the multi-step forward iterative predictor predicts the value of the first future time point based on the initial input; when When , the multi-step forward iterative forecaster takes the forecast value of the previous step as a new input and repeats the iterative process to generate a power load forecast sequence containing multiple future time points.

[0012] Furthermore, the process of training the error corrector includes: slicing the historical power load training time series using a sliding window to generate a length of Several subsequences of Taking the starting point of the subsequence as the initial value, a multi-step forward iterative predictor based on the fourth-order Runge-Kutta model is used to generate the corresponding predicted subsequence for the subsequence ; Construct data pairs of predicted values ​​and true values , train the error corrector.

[0013] Furthermore, in the process of training the error corrector, a confidence factor is added to the original loss function. Regularization term, optimizes the error corrector; among them, the improved loss function is:

[0014]

[0015] in, is the MLP network weight, for Regularization coefficient, is the confidence factor, represents mean probability normalization, Represents the original loss function.

[0016] Further, after the error corrector training is completed, the error corrector is used to obtain the first Correction value of the step-by-step power load forecast sequence , then based on the The first predicted value and the The step correction value is obtained to obtain the power load forecast result; it is expressed by the following formula:

[0017] in, Indicates the The power load forecast results are: is the weighting coefficient, Indicated by The power load forecast sequence composed of the step forecast values, For the The predicted value of the step Represents an error corrector.

[0018] Furthermore, the error corrector adopts an MLP neural network.

[0019] A second aspect of the present invention provides a power load forecasting system based on dynamic driving.

[0020] A power load forecasting system based on dynamic drive, comprising: A dynamic modeling module is configured to: perform dynamic modeling on the acquired historical power load time training sequence using a deterministic learning algorithm to extract the inherent dynamic information of the data; The multi-step prediction module is configured to: based on the dynamic information, use a multi-step forward iterative predictor based on the fourth-order Runge-Kutta model to obtain Step to The power load forecast sequence composed of step forecast values; The error correction module is configured to: use an error corrector to correct the Step to The power load forecast sequence composed of the first step forecast value is periodically corrected to obtain the first The power load forecast sequence correction value of the first step; The first predicted value and the The power load forecasting result is obtained by correcting the power load forecasting sequence of the first step.

[0021] A third aspect of the present invention provides a computer device, comprising: a processor adapted to execute a computer program; A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the power load forecasting method based on dynamic drive as described in the first aspect above are implemented.

[0022] Compared with the prior art, the present invention has the following beneficial effects: The present invention accurately models the nonlinear dynamic laws inherent in the chaotic time series of power load by determining the learning algorithm. Different from the limitation of traditional machine learning methods that only perform surface static mapping of "historical load data-future load data", the present invention can deeply capture the essential evolution characteristics of the power load series affected by the coupling of multiple factors such as user behavior, meteorological conditions, and social activities, thereby significantly improving the model's prediction accuracy for short-term, medium-term and long-term power loads, and providing a more reliable decision-making basis for the refined scheduling and economic operation of the power system. To address the problem of error diffusion in multi-step iterative forecasting of power load, this paper designs a prediction-correction mechanism that considers the changing patterns of prediction errors. By introducing a confidence factor based on the trend of prediction errors into the corrector network, the corrector neural network can more accurately identify and focus on reliable information in the input sequence, effectively reducing the cumulative error during the iterative forecasting process and significantly improving the performance of medium- and long-term power load forecasting. The resulting predictions can better support medium- and long-term decision-making scenarios such as grid planning and power generation construction.

[0023] The prediction mechanism of this invention relies heavily on the inherent physical information contained in power load time series data (such as the dynamic correlation between load and power consumption period, seasonal cycles, and temperature). This allows it to reveal the dynamic mechanisms of power load sequence evolution to a certain extent, and offers excellent physical interpretability. This feature effectively addresses the limited application of traditional "black box" models in critical power system scenarios, significantly enhancing the model's potential for application in safety-critical areas such as power system security and stability control and power supply assurance for critical users, providing more robust technical support for the safe operation of power systems. In summary, the present invention effectively solves the problems existing in existing power load forecasting methods, such as insufficient utilization of the intrinsic dynamic characteristics of data, large cumulative errors in multi-step forecasts, poor medium- and long-term forecasting performance, and lack of physical interpretability of prediction results, by integrating the chaotic dynamic characteristics of power load sequences, optimizing the error correction mechanism, and enhancing the physical interpretability of prediction results. It significantly improves the accuracy, reliability, and practicality of power load forecasting. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0025] Figure 1 is a flow chart of a method for predicting power load based on dynamic driving according to an embodiment of the present invention; Figure 2 is a flow chart of another embodiment of a method for power load forecasting based on dynamic driving according to an embodiment of the present invention; Figure 3 1 is a structural diagram of a power load forecasting system based on dynamic drive according to an embodiment of the present invention; Figure 4 It is a structural diagram of a computer device shown in an embodiment of the present invention. DETAILED DESCRIPTION

[0026] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0027] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0028] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0029] In order to facilitate understanding of the technical solution of the present invention, some technical terms involved in the present invention are introduced below.

[0030] Deterministic learning algorithms are an established machine learning approach. As machine learning technology continues to evolve, they have emerged as an effective learning method for nonlinear systems in dynamic environments, and have emerged in numerous research and applications. This approach includes the following core elements: (1) The use of radial basis function (RBF) neural networks provides powerful function approximation capabilities. RBF neural networks, with their unique structure, are able to efficiently fit complex nonlinear functions. This feature is particularly advantageous when processing data with complex nonlinear characteristics, such as power loads. In previous research on related machine learning applications, such as in the modeling and control scenarios of some complex industrial processes, RBF neural networks have been successfully applied to describe the dynamic characteristics of the system, demonstrating good adaptability and accuracy.

[0031] (2) The weight adjustment law based on Lyapunov stability theory ensures the stability of the algorithm during the learning process. Lyapunov stability theory provides a solid mathematical foundation for the stability analysis of dynamic systems. By introducing it into the weight adjustment process, it ensures that deterministic learning maintains overall stability while continuously approaching the true dynamic characteristics.

[0032] (3) Accurately modeling the intrinsic dynamics of time series data along the system trajectory and storing this characteristic in the form of neural network constant weights enables the deterministic learning algorithm to effectively capture the essential dynamic laws of the system. In the power load forecasting scenario, traditional intelligent models often only focus on the surface correlation between historical data and future data. However, the deterministic learning algorithm explores the intrinsic dynamic evolution laws contained in the power load sequence as a chaotic system and fixes these laws in the form of neural network constant weights, thereby achieving a deeper understanding and prediction of power load changes.

[0033] As described in the background, existing power load forecasting suffers from technical issues such as insufficient utilization of the inherent dynamic characteristics of data, large cumulative errors in multi-step forecasts, poor medium- and long-term forecasting performance, and a lack of physical interpretability of forecast results. The present invention provides a method and system for power load forecasting based on dynamics. The present invention is described in detail below through several examples.

[0034] Figure 1 is a flow chart of a method for predicting power load based on dynamic driving according to an embodiment of the present invention; Figure 1 , the method comprising: A deterministic learning algorithm is used to perform dynamic modeling on the acquired historical power load time training sequence to extract the inherent dynamic information of the data; Based on the dynamic information, a multi-step forward iterative predictor based on the fourth-order Runge-Kutta model is used to obtain the Step to The power load forecast sequence composed of step forecast values; The error corrector is used to Step to The power load forecast sequence composed of the first step forecast value is periodically corrected to obtain the first The power load forecast sequence correction value of the first step; The first predicted value and the The power load forecasting result is obtained by correcting the power load forecasting sequence of the first step.

[0035] The present invention significantly improves the short-term prediction accuracy of power load time series and improves the stability of medium- and long-term predictions by exploring the intrinsic dynamic characteristics of power load time series and constructing an effective prediction-correction mechanism.

[0036] Figure 2 is a flow chart of another embodiment of a method for power load forecasting based on dynamic drive according to an embodiment of the present invention; Figure 2 The power load forecasting method based on dynamic driving comprises the following steps: Step (1): Obtain historical power load time series data and preprocess the data.

[0037] In this embodiment, the data is preprocessed, including: first cleaning the historical power load time series data, then standardizing or normalizing the historical power load time series data, and finally dividing the power load time training sequence and the power load time test sequence according to a 7:3 ratio.

[0038] Among them, data cleaning includes: removing outliers, filling missing values, etc.

[0039] Step (2): Use a deterministic learning algorithm to perform dynamic modeling on the preprocessed power load time training sequence, extract the inherent dynamic information of the data, and store it in the form of constant weights of the neural network.

[0040] The power load time training sequence can be regarded as a multivariate time series data generated by a nonlinear dynamic system, which can be represented by the following differential equation: (1) in, is the system state variable, represents the rate of change of the system state, is the system dimension, is the parameter vector of the system, is the unknown dynamics of the system. The power load time series data is obtained by Dimensional state variables Sampling obtained.

[0041] In some embodiments, a deterministic learning method is used to perform dynamic modeling on the power load time series data. The specific process is as follows: a dynamic identifier is constructed using an RBF neural network, and a weight update law based on Lyapunov stability design is adopted to realize the dynamics of the unknown system along the system sampling trajectory. Accurate modeling of: (2) in, is the constant weight of the neural network after training convergence, that is, the dynamic knowledge to be stored, is the RBF neural network regression vector, is the modeling error.

[0042] The present invention accurately models the nonlinear dynamic laws inherent in the chaotic time series of power load by determining the learning algorithm. Different from the limitation of traditional machine learning methods that only perform surface static mapping of "historical load data-future load data", the present invention can deeply capture the essential evolution characteristics of the power load series affected by the coupling of multiple factors such as user behavior, meteorological conditions, and social activities, thereby significantly improving the model's prediction accuracy for short-term, medium-term and long-term power loads, and providing a more reliable decision-making basis for the refined scheduling and economic operation of the power system.

[0043] Step (3): Using the obtained dynamic information, a multi-step forward iterative predictor based on the fourth-order Runge-Kutta model is constructed and multi-step iterative prediction is implemented.

[0044] The 4th-order Runge-Kutta model iterative predictor embedded with dynamic knowledge is expressed by the following formula: (3) in, For the The predicted value of the step, is the sampling time, is the constant weight of the embedded neural network, that is, the learned dynamics knowledge. Indicates the The instantaneous dynamics of the step, Indicates based on The first ( )-step prediction dynamics, Indicates based on The first ( ) step to improve the dynamics, Indicates based on The first ( )-step prediction dynamics, to Need to calculate in sequence and finally get the The predicted value of the step .

[0045] In this embodiment, when When , the predictor predicts the value of the first future time point based on the initial input; when When , the predictor takes the predicted value of the previous step as a new input and repeats the iterative process to generate a prediction sequence containing multiple future time points. In this architecture, the model does not predict all future steps at once, but makes predictions step by step.

[0046] Step (4): Construct a neural network-based error corrector to perform periodic real-time correction on the predictor output.

[0047] The goal of the error corrector is to correct the predicted value with error to the true value as much as possible. The corrector is built based on the MLP neural network. The input of the network is from the first Step to A sequence of step-by-step forecast values , the output of the network is Step correction value .

[0048] In an embodiment, the training strategy of the error corrector is: first, use a sliding window to slice the power load time training sequence data to generate a length of Several subsequences of ; Then, using the starting point of the subsequence as the initial value, the predictor is used to generate the corresponding predicted subsequence for the subsequence ; Finally, construct the "predicted value-true value" data pair , used to train the MLP neural network.

[0049] It should be noted that considering the evolutionary trend that the prediction error of the iterative model is often small in the early stage and gradually increases with the increase of the number of iterations, the inference neural network should pay more attention to the front part of the sequence when processing the input sequence, because the information in these parts is relatively more reliable. Therefore, the confidence factor is introduced into the loss function. The regularization term makes the neural network pay more attention to the prediction information with higher confidence in the input sequence. The modified loss function is as follows: (4) in, Represents the original loss function, which can be the cross entropy loss function or the mean square error loss function; is the MLP network weight, for Regularization coefficient, is the confidence factor, which is obtained by normalizing the average value of the forecast error sequence by probability to reflect the changing trend of the error: (5) After the corrector training is completed, the trained corrector outputs the correction value, which is weighted and fused with the predicted value output by the predictor to obtain the final correction prediction result: (6) in, is the weighting coefficient, Indicated by A sequence of step-by-step forecast values, represents the trained calibrator, Indicates the output value of the corrector.

[0050] In this embodiment, periodic correction, that is, every time steps ( ) is corrected once, where The step size is set according to the degree of chaos of the time series data: for time series data with a lower degree of chaos, a larger correction step size can be used , thereby reducing the frequency of correction and avoiding unnecessary fluctuation interference caused by too frequent correction.

[0051] To address the problem of error diffusion in multi-step iterative forecasting of power load, this paper designs a prediction-correction mechanism that considers the changing patterns of prediction errors. By introducing a confidence factor based on the trend of prediction errors into the corrector network, the corrector neural network can more accurately identify and focus on reliable information in the input sequence, effectively reducing the cumulative error during the iterative forecasting process and significantly improving the performance of medium- and long-term power load forecasting. The resulting predictions can better support medium- and long-term decision-making scenarios such as grid planning and power generation construction.

[0052] Step (5): Determine whether the iteration is completed. After the iteration is completed, output the power load forecast result.

[0053] By integrating the chaotic dynamic characteristics of the power load sequence, optimizing the error correction mechanism and enhancing the physical interpretability of the prediction results, the present invention effectively solves the problems existing in the existing power load forecasting methods, such as insufficient utilization of the intrinsic dynamic characteristics of the data, large cumulative errors in multi-step predictions, poor medium- and long-term prediction performance, and lack of physical interpretability of the prediction results. It significantly improves the accuracy, reliability and practicality of power load forecasting.

[0054] Combination of the above Figure 1 The power load forecasting method based on dynamic driving provided by the embodiment of the present invention is introduced in detail. Next, the power load forecasting system based on dynamic driving provided by the embodiment of the present invention will be introduced with reference to the accompanying drawings.

[0055] Figure 3 This is a schematic diagram of the structure of the power load forecasting system based on dynamic drive according to an embodiment of the present invention. Figure 3 , the system of the present invention comprises: The data acquisition module is configured to collect power load time series data. This data is preprocessed, including but not limited to data cleaning (removing outliers and filling missing values), data standardization, and data partitioning (dividing the processed data into a power load time training sequence and a power load time test sequence, with the power load time training sequence used for dynamic modeling and error correction training). After preprocessing, the organized power load time training sequence is transmitted to the dynamic modeling module.

[0056] The dynamic modeling module is configured to receive the preprocessed power load time training sequence output by the data acquisition module and model the inherent dynamics of the power load time training sequence based on a deterministic learning algorithm. Specifically, this process involves constructing a radial basis function (RBF) neural network dynamic identifier, using a weight update law designed based on Lyapunov stability to approximate the nonlinear dynamic function implied by the time series, and storing constant weights. These stored constant weights are then transferred to the multi-step prediction module, providing dynamic knowledge for predictor construction.

[0057] The multi-step prediction module is configured to receive the dynamic information in the form of constant weights output by the dynamic modeling module and construct a multi-step forward iterative predictor based on the fourth-order Runge-Kutta model. This predictor uses the first state of the power load time test sequence as its initial input and iteratively predicts based on the fourth-order Runge-Kutta model, which incorporates dynamic knowledge, to generate predicted values ​​for multiple future time steps. The generated prediction sequence is transmitted in real time to the error correction module for correction processing.

[0058] The error correction module is configured to include an error corrector based on a neural network (e.g., a multi-layer perceptron (MLP)). This corrector is pre-trained using a historical power load training sequence. It extracts subsequences from this historical power load training sequence using a sliding window. The predictor in the multi-step prediction module generates predicted values ​​for these subsequences, constructing "predicted value-true value" data pairs. The corrector is trained to learn the changing patterns of prediction errors, and a confidence factor based on the prediction error trend is introduced into the training loss function to prioritize reliable prediction information. During the prediction phase, the error corrector receives the prediction sequence output by the multi-step prediction module and performs real-time corrections on the predicted values ​​according to a preset period. The power load forecast result is obtained based on the prediction sequence output by the multi-step prediction module and the corrected power load forecast sequence.

[0059] According to an embodiment of the present invention, the power load forecasting system based on dynamic drive can correspond to the method described in the embodiment of the present invention, and the above and other operations and / or functions of each module of the power load forecasting system based on dynamic drive are respectively to achieve Figure 1 For the sake of brevity, the corresponding processes of each method in are not repeated here.

[0060] See also Figure 4The computer device shown in the figure has a structure diagram, which includes a processor, a communication interface, and a computer-readable storage medium. The processor, communication interface, and computer-readable storage medium can be connected via a bus or other means. The communication interface is used to receive and send data. The computer-readable storage medium can be stored in the computer device's memory, and the computer-readable storage medium is used to store a computer program, which includes program instructions. The processor is used to execute the program instructions stored in the computer-readable storage medium. The processor (also known as the CPU (Central Processing Unit)) is the computing and control core of the computer device and is suitable for implementing one or more instructions, specifically loading and executing one or more instructions to implement the corresponding steps in the embodiment of the dynamic-driven power load forecasting method.

[0061] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage and optical storage, etc.) containing computer-usable program code.

[0062] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0063] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0064] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0065] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0066] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A power load forecasting method based on dynamic drive, characterized in that: include: A deterministic learning algorithm is used to perform dynamic modeling on the acquired historical power load time training sequence to extract the inherent dynamic information of the data; Based on the dynamic information, a multi-step forward iterative predictor based on the fourth-order Runge-Kutta model is used to obtain the Step to The power load forecast sequence composed of step forecast values; The error corrector is used to Step to The power load forecast sequence composed of the first step forecast value is periodically corrected to obtain the first The power load forecast sequence correction value of the first step; The first predicted value and the The power load forecasting result is obtained by correcting the power load forecasting sequence of the first step.

2. The power load forecasting method based on dynamic drive according to claim 1 is characterized in that: After obtaining the historical power load time training sequence, data cleaning is performed on the historical power load time training sequence, and after cleaning, standardization or normalization processing is performed on the historical power load time training sequence.

3. The power load forecasting method based on dynamic drive according to claim 1, characterized in that: The method of using a deterministic learning algorithm to perform dynamic modeling on the acquired historical power load time training sequence includes: A dynamic identifier is constructed using an RBF neural network, and a weight update law based on Lyapunov stability design is adopted to accurately model the dynamics of the unknown system along the system sampling trajectory: in, represents the unknown system dynamics, is the constant weight of the RBF neural network after training convergence, is the RBF neural network regression vector, is the modeling error.

4. The power load forecasting method based on dynamic drive according to claim 1, characterized in that: The multi-step forward iterative predictor based on the fourth-order Runge-Kutta model is expressed by the following formula: in, For the The predicted value of the step, is the sampling time, is the constant weight of the embedded neural network, that is, the learned dynamics knowledge, Indicates the The instantaneous dynamics of the step, Indicates based on The first ( )-step prediction dynamics, Indicates based on The first ( ) step to improve the dynamics, Indicates based on The first ( )-step prediction dynamics, to Need to calculate in sequence and finally get the The predicted value of the step ; when When , the multi-step forward iterative predictor predicts the value of the first future time point based on the initial input; when When , the multi-step forward iterative forecaster takes the forecast value of the previous step as a new input and repeats the iterative process to generate a power load forecast sequence containing multiple future time points.

5. The power load forecasting method based on dynamic drive according to claim 1, characterized in that: The process of training the error corrector includes: slicing the historical power load time series using a sliding window to generate a length of Several subsequences of Taking the starting point of the subsequence as the initial value, a multi-step forward iterative predictor based on the fourth-order Runge-Kutta model is used to generate the corresponding predicted subsequence for the subsequence ; Construct data pairs of predicted values ​​and true values , train the error corrector.

6. The power load forecasting method based on dynamic drive according to claim 5 is characterized in that: In the process of training the error corrector, the original loss function is added with a confidence factor Regularization term, optimizes the error corrector; among them, the improved loss function is: in, is the MLP network weight, for Regularization coefficient, is the confidence factor, represents mean probability normalization, Represents the original loss function.

7. The power load forecasting method based on dynamic drive according to claim 1, characterized in that: After the error corrector training is completed, use the error corrector to get the Correction value of the step-by-step power load forecast sequence , then based on the The first predicted value and the Step correction value is obtained to obtain the power load forecast result; It is expressed by the following formula: in, Indicates the The power load forecast results are: is the weighting coefficient, Indicated by Power load forecast sequence composed of step-by-step forecast values For the The predicted value of the step, Represents an error corrector.

8. The power load forecasting method based on dynamic drive according to claim 1, characterized in that: The error corrector adopts MLP neural network.

9. A power load forecasting system based on dynamic drive, characterized in that: include: A dynamic modeling module is configured to: perform dynamic modeling on the acquired historical power load time training sequence using a deterministic learning algorithm to extract the inherent dynamic information of the data; The multi-step prediction module is configured to: based on the dynamic information, use a multi-step forward iterative predictor based on the fourth-order Runge-Kutta model to obtain Step to The power load forecast sequence composed of step forecast values; The error correction module is configured to: use an error corrector to correct the Step to The power load forecast sequence composed of the first step forecast value is periodically corrected to obtain the first The correction value of the power load forecast sequence of the step; Based on the The first predicted value and the The power load forecasting result is obtained by correcting the power load forecasting sequence of the first step.

10. A computer device, characterized in that: a processor adapted to execute a computer program; A computer-readable storage medium having a computer program stored therein, wherein the computer program, when executed by the processor, implements the steps of the power load forecasting method based on dynamic drive according to any one of claims 1 to 8.

Images