Load prediction method and system

Abstract

The invention discloses a load prediction method, which comprises the following steps of: acquiring an original load sequence, decomposing the original load sequence by utilizing first modal decomposition to obtain a modal component, and performing load reconstruction on the modal component through sample entropy aggregation to obtain a reconstructed load sequence; performing primary decomposition on a high-frequency component in the modal components by using second modal decomposition, determining the number of decomposition modes and a penalty factor, and performing secondary decomposition on a high-frequency component in the reconstructed load sequence by using third modal decomposition to obtain a modal sub-sequence; screening according to the modal sub-sequence to obtain an input feature, and analyzing the correlation between the reconstructed load sequence and the input feature by using a maximum mutual information method to obtain an input feature of the hybrid deep learning prediction model; according to the method, the actually measured load data is acquired, the actually measured load data is input into the hybrid deep learning prediction model, the prediction result is obtained, the power load can be well predicted, and the prediction result has high stability.

Description

Technical Field

[0001] The present invention relates to the technical field of load forecasting, and in particular to a load forecasting method and system. Background Art

[0002] In today's rapidly developing society, the role of energy is becoming more and more important. It is not only the basis of economic development, but also the driving force of modern social operation. With the acceleration of industrialization and urbanization and the growth of population, global energy demand shows a trend of continuous growth. However, at the same time, traditional energy resources such as coal, oil and natural gas are facing the risk of depletion, which not only leads to fluctuations in energy prices, but also has a serious impact on the environment. Therefore, how to manage and use energy efficiently and reasonably has become an urgent problem to be solved. In energy management, load forecasting is a key link. Accurate load forecasting can help power companies optimize power generation plans, reduce energy waste, improve the efficiency and reliability of power grid operations, and also help reduce operating costs and avoid insufficient or excessive power supply. However, traditional load forecasting methods, such as time series analysis and regression analysis, often rely on simplified assumptions and linear models, which are difficult to capture the complexity and nonlinear characteristics of load data. In addition, these methods are also unable to cope with large amounts of data and multiple variables.

[0003] In order to overcome these limitations, advanced technologies such as data decomposition, feature screening and deep learning have been introduced into the field of load forecasting in recent years. Deep learning models, such as convolutional neural networks (CNN), recurrent neural networks (RNN) and long short-term memory networks (LSTM), have powerful nonlinear modeling capabilities and automatic feature learning capabilities. These models can automatically learn the complex patterns and laws of load changes from the decomposed and screened data without manually setting complex mathematical formulas or model structures. This feature of deep learning models gives them significant advantages in processing nonlinear, time-varying and uncertain load data. The method of integrating data decomposition, feature screening and deep learning provides a new solution for load forecasting. Compared with traditional technologies, this fusion method can better cope with the nonlinearity, time-varying and uncertainty of load data, provide a more accurate basis for the optimal operation of energy systems, power dispatching, etc., and achieve efficient utilization and reasonable allocation of energy. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is: the present invention solves the problem that the prior art has limitations when facing complex load change patterns and large amounts of data.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a load forecasting method, comprising:

[0008] Obtaining an original load sequence, decomposing the original load sequence using a first modal decomposition to obtain modal components, and reconstructing the modal components through sample entropy aggregation to obtain a reconstructed load sequence;

[0009] Decomposing the high-frequency components in the modal components once by using the second modal decomposition to determine the number of decomposed modes and the penalty factor, and decomposing the high-frequency components in the reconstructed load sequence twice by using the third modal decomposition based on the number of decomposed modes and the penalty factor to obtain a modal subsequence;

[0010] The input features are obtained by screening the modal subsequences, and the correlation between the reconstructed load sequence and the input features is analyzed using the maximum mutual information method to obtain the input features of the hybrid deep learning prediction model.

[0011] Obtain measured load data, input the measured load data into the hybrid deep learning prediction model, and obtain a prediction result.

[0012] As a preferred solution of the load forecasting method of the present invention, wherein: decomposing the original load sequence by using the first modal decomposition comprises:

[0013] Initialize the improved complete integrated empirical mode decomposition algorithm and set key parameters;

[0014] Add adaptive noise to the original load sequence, generate Gaussian white noise each time the noise is added, standardize the Gaussian white noise, and add it to the original signal according to the set different noise amplitudes;

[0015] Perform empirical mode decomposition on the original signal with added noise to obtain several intrinsic mode functions;

[0016] All the intrinsic mode functions obtained by adding noise and decomposing are collected together, and the mean of each corresponding intrinsic mode function component is calculated to obtain the decomposition result.

[0017] As a preferred solution of the load forecasting method of the present invention, the load reconstruction of the modal component by sample entropy aggregation includes:

[0018] The sample entropy of each intrinsic mode function is calculated, and according to the sample entropy value of each intrinsic mode function, the intrinsic mode functions with similar sample entropy values ​​are aggregated, and the aggregated intrinsic mode functions are reconstructed into a new signal to obtain a reconstructed load sequence.

[0019] As a preferred solution of the load forecasting method of the present invention, wherein: using the second modal decomposition to decompose the high-frequency components in the modal components comprises:

[0020] Initialize the improved adaptive noise complete set empirical mode decomposition algorithm and set relevant parameters;

[0021] Perform empirical mode decomposition on the high frequency component and white noise to obtain the intrinsic mode function component, and select the Kth intrinsic mode function component of the noise as the noise component of the signal to be decomposed;

[0022] The improved adaptive noise complete set empirical mode decomposition is used to decompose the signal to be decomposed, and the intrinsic mode function is obtained. The mode function is analyzed to obtain the number of decomposed modes and the penalty factor.

[0023] As a preferred solution of the load forecasting method of the present invention, wherein: using the third mode decomposition to perform secondary decomposition on the high frequency components in the reconstructed load sequence includes:

[0024] Initialize the center frequency and mode function, and iterate using the initialized parameters;

[0025] For each iteration, the modal function is updated using the Wiener filtering method and the center frequency is calculated;

[0026] Determine whether the modal function meets the convergence condition, that is, whether it no longer changes within the tolerance range. If it does not meet the convergence condition, continue to iterate and update the Lagrange multiplier to meet the constraint condition;

[0027] If the convergence condition is met or the maximum number of iterations is reached, the iteration is stopped and the decomposed mode function and the corresponding center frequency are output;

[0028] The modal function of the high-frequency component is identified from the decomposed modal function, and the high-frequency component function is decomposed twice using variational modal decomposition to obtain a modal subsequence.

[0029] As a preferred solution of the load forecasting method of the present invention, the analysis of the correlation between the reconstructed load sequence and the input feature using the maximum mutual information method includes:

[0030] Calculate the mutual information value for each relevant input feature and the reconstructed load sequence;

[0031] Determine the joint probability distribution of the input feature and the reconstructed load sequence, calculate the marginal probability distribution of the input feature and the reconstructed load sequence, calculate the logarithm of the ratio of the joint probability distribution to the product of the marginal probability distribution of the input feature and the reconstructed load sequence multiplied by the joint probability distribution, and sum them to obtain the value of the mutual information of the input feature and the reconstructed load sequence;

[0032] The calculated mutual information values ​​are sorted, and features whose mutual information values ​​are greater than the first threshold are selected as input features of the hybrid deep learning prediction model.

[0033] As a preferred solution of the load forecasting method described in the present invention, the hybrid deep learning forecasting model includes:

[0034] Use convolutional neural networks to extract local features from time series data. After the convolutional neural network layer, use a bidirectional long short-term memory network layer to further process the features.

[0035] After the bidirectional long short-term memory network layer, the attention mechanism is introduced to obtain a hybrid deep learning prediction model.

[0036] In a second aspect, the present invention provides a load forecasting system, comprising:

[0037] A load reconstruction module is used to obtain an original load sequence, decompose the original load sequence using a first modal decomposition to obtain modal components, and reconstruct the modal components through sample entropy aggregation to obtain a reconstructed load sequence;

[0038] A decomposition module, used to perform a primary decomposition of the high-frequency components in the modal components by using the second modal decomposition, determine the number of decomposed modes and the penalty factor, and perform a secondary decomposition of the high-frequency components in the reconstructed load sequence by using the third modal decomposition based on the number of decomposed modes and the penalty factor to obtain a modal subsequence;

[0039] The correlation analysis module is used to obtain input features based on the modal subsequence screening, and analyze the correlation between the reconstructed load sequence and the input features using the maximum mutual information method to obtain the input features of the hybrid deep learning prediction model;

[0040] The prediction module is used to obtain measured load data, input the measured load data into the hybrid deep learning prediction model, and obtain a prediction result.

[0041] In a third aspect, the present invention provides a computing device, comprising:

[0042] Memory and processor;

[0043] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the load forecasting method are implemented.

[0044] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, which implement the steps of the load forecasting method when executed by a processor.

[0045] The beneficial effects of the present invention are as follows: through methods such as improved complete integrated empirical mode decomposition (ICEEMD) and adaptive noise complete ensemble empirical mode decomposition (ICEEMDAN), the original load sequence can be effectively decomposed and reconstructed, thereby constructing a high-quality data set, combining the maximum mutual information method (MIC) for feature screening, and a hybrid deep learning model of convolutional neural network (CNN), bidirectional long short-term memory network (BiLSTM) and attention mechanism (Attention), the accuracy of power load forecasting can be improved, and by comprehensively considering data decomposition, feature screening and deep learning technology, the power load can be better predicted, and the prediction result has high stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0047] Figure 1 An overall flow chart of a load forecasting method provided by the present invention;

[0048] Figure 2 An original load sequence diagram of a load forecasting method provided by the present invention;

[0049] Figure 3 A primary decomposition result diagram of a load forecasting method provided by the present invention;

[0050] Figure 4 A sample entropy result diagram of a load forecasting method provided by the present invention;

[0051] Figure 5 A secondary decomposition result diagram of a load forecasting method provided by the present invention;

[0052] Figure 6 A sample entropy reconstruction result diagram of a load forecasting method provided by the present invention;

[0053] Figure 7 A VMD-IMF1 correlation feature screening result diagram of a load forecasting method provided by the present invention;

[0054] Figure 8 A VMD-IMF2 correlation feature screening result diagram of a load forecasting method provided by the present invention;

[0055] Fig. 9A VMD-IMF3 correlation feature screening result diagram of a load forecasting method provided by the present invention;

[0056] Fig.10 A Co-IMF2 correlation feature screening result diagram of a load forecasting method provided by the present invention;

[0057] Fig.11 A Co-IMF3 correlation feature screening result diagram of a load forecasting method provided by the present invention;

[0058] Fig.12 A power load prediction effect diagram of a load prediction method provided by the present invention;

[0059] Fig.13 A power load forecasting error diagram of a load forecasting method provided by the present invention. DETAILED DESCRIPTION

[0060] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0061] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0062] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0063] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0064] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0065] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0066] Example 1

[0067] Reference Figure 1 , as an embodiment of the present invention, provides a load forecasting method, comprising:

[0068] S100: obtaining an original load sequence, decomposing the original load sequence by using the first modal decomposition to obtain modal components, and reconstructing the modal components by sample entropy aggregation to obtain a reconstructed load sequence;

[0069] In the embodiment of the present application, the improved complete integrated empirical mode decomposition algorithm is initialized and key parameters are set;

[0070] Add adaptive noise to the original load sequence, generate Gaussian white noise each time the noise is added, standardize the Gaussian white noise, and add it to the original signal according to the set different noise amplitudes;

[0071] Perform empirical mode decomposition on the original signal with added noise to obtain several intrinsic mode functions;

[0072] All the intrinsic mode functions obtained by adding noise and decomposing are collected together, and the mean of each corresponding intrinsic mode function component is calculated to obtain the decomposition result.

[0073] Specifically, the present invention preferably improves the complete integrated empirical mode decomposition, initializes the improved complete integrated empirical mode decomposition algorithm to set key parameters, including the amplitude of added noise, the number of sets, etc., and generates a Gaussian white noise sequence with a mean of 0 and a standard deviation of 1 in each iteration. The generated Gaussian white noise is standardized to ensure that it has the same energy level as the original signal.

[0074] It should be noted that the improved complete integrated empirical mode decomposition is an optimization of the traditional integrated empirical mode decomposition. By introducing adaptive noise, the modal aliasing problem existing in the traditional integrated empirical mode decomposition is effectively solved. During the decomposition process, the improved complete integrated empirical mode can more accurately decompose complex signals into multiple intrinsic mode functions (IMFs). Its advantage is that the decomposition results have higher accuracy and reliability, and the ability to handle non-stationary and nonlinear signals is stronger. At the same time, the noise residue is reduced, and the accuracy and efficiency of the decomposition are improved.

[0075] In an embodiment of the present application, the sample entropy of each intrinsic mode function is calculated, and according to the sample entropy value of each intrinsic mode function, the intrinsic mode functions with similar sample entropy values ​​are aggregated, and the aggregated intrinsic mode functions are reconstructed into a new signal to obtain a reconstructed load sequence.

[0076] Specifically, sample entropy quantifies the randomness and complexity of a sequence by comparing the similarity of subsequence patterns of different lengths. When calculating, there is no need to assume the distribution of the data in advance, and it is less dependent on the length of the data. Specifically, given a time series, subsequences of similar length are constructed, and the number of subsequences that meet specific similarity conditions is counted to calculate the sample entropy value. The larger the sample entropy value, the more complex and random the sequence is; the smaller the value, the more regular and stable the sequence is.

[0077] It should be noted that sample entropy is widely used in many fields such as physiological signal analysis and engineering monitoring, and can effectively reflect the dynamic changes and complexity characteristics of the system.

[0078] S200: Decomposing the high-frequency components in the modal components once by using the second modal decomposition, determining the number of decomposed modes and the penalty factor, and decomposing the high-frequency components in the reconstructed load sequence twice by using the third modal decomposition based on the number of decomposed modes and the penalty factor, to obtain a modal subsequence;

[0079] In the embodiment of the present application, the improved adaptive noise complete set empirical mode decomposition algorithm is initialized and relevant parameters are set;

[0080] Perform empirical mode decomposition on the high frequency component and white noise to obtain the intrinsic mode function component, and select the Kth intrinsic mode function component of the noise as the noise component of the signal to be decomposed;

[0081] The improved adaptive noise complete set empirical mode decomposition is used to decompose the signal to be decomposed, and the intrinsic mode function is obtained. The mode function is analyzed to obtain the number of decomposed modes and the penalty factor.

[0082] Specifically, the key parameters of the algorithm are determined, including the number of decomposition layers, the number of noise assists, and the noise amplitude. The parameters will directly affect the performance of the algorithm and the accuracy of the decomposition results. The Kth IMF of the noise is selected as the noise component of the signal to be decomposed. This is based on the assumption that noise mainly affects the high-frequency part of the signal. By selecting a specific IMF, the impact of noise can be better simulated and estimated.

[0083] In the embodiment of the present application, the center frequency and the modal function are initialized, and the initialized parameters are used for iteration;

[0084] For each iteration, the modal function is updated using the Wiener filtering method and the center frequency is calculated;

[0085] Determine whether the modal function meets the convergence condition, that is, whether it no longer changes within the tolerance range. If it does not meet the convergence condition, continue to iterate and update the Lagrange multiplier to meet the constraint condition;

[0086] If the convergence condition is met or the maximum number of iterations is reached, the iteration is stopped and the decomposed mode function and the corresponding center frequency are output;

[0087] The modal function of the high-frequency component is identified from the decomposed modal function, and the high-frequency component function is decomposed twice using variational modal decomposition to obtain a modal subsequence.

[0088] Specifically, set the number of modes, penalty parameter, tolerance, maximum number of iterations and other parameters and initialize the center frequency and modal function. Use Wiener filtering to update the modal function by minimizing the sum of the bandwidths of the modal function, calculate the power spectrum of the modal function, and estimate the center frequency based on the center of gravity of the power spectrum. Based on the estimated center frequency, use Wiener filtering to further update the modal function until the convergence condition is met. Wiener filtering provides an effective mechanism to minimize the error of signal estimation and can accurately extract the various components of the signal in the frequency domain. In this way, variational modal decomposition can achieve high-quality decomposition of the signal and extract modal functions with different frequency characteristics.

[0089] It should be noted that through multi-level modal decomposition, the high-frequency features in the signal can be captured more accurately. Multi-level decomposition helps to more accurately identify and separate different components in the signal, thereby improving the accuracy of signal processing. The modal function obtained by variational modal decomposition is more easily associated with the actual physical phenomenon or system state, thereby improving the interpretability of the signal. More accurate signal decomposition helps to build more reliable prediction models and analysis methods.

[0090] S300: The input features are obtained by screening the modal subsequences, and the correlation between the reconstructed load sequence and the input features is analyzed by using the maximum mutual information method to obtain the input features of the hybrid deep learning prediction model;

[0091] In the embodiment of the present application, a mutual information value is calculated for each relevant input feature and the reconstructed load sequence;

[0092] Determine the joint probability distribution of the input feature and the reconstructed load sequence, calculate the marginal probability distribution of the input feature and the reconstructed load sequence, calculate the logarithm of the ratio of the joint probability distribution to the product of the marginal probability distribution of the input feature and the reconstructed load sequence multiplied by the joint probability distribution, and sum them to obtain the value of the mutual information of the input feature and the reconstructed load sequence;

[0093] The calculated mutual information values ​​are sorted, and features whose mutual information values ​​are greater than the first threshold are selected as input features of the hybrid deep learning prediction model.

[0094] Specifically, the present invention prioritizes the maximum mutual information for correlation analysis, and can also use the Pearson correlation coefficient;

[0095] It should be noted that the maximum mutual information can capture the nonlinear relationship between variables, while the Pearson correlation coefficient is mainly used to measure the linear relationship between two variables. The maximum mutual information is applicable to different types of data, including continuous, discrete, and mixed data, while the Pearson correlation coefficient is usually used for continuous data. The maximum mutual information is more robust to noise and outliers in the data, while the person correlation coefficient will be affected by outliers.

[0096] S400: Obtain measured load data, input the measured load data into the hybrid deep learning prediction model, and obtain a prediction result;

[0097] In the embodiment of the present application, a convolutional neural network is used to extract local features in time series data, and after the convolutional neural network layer, a bidirectional long short-term memory network layer is used to further process the features;

[0098] After the bidirectional long short-term memory network layer, the attention mechanism is introduced to obtain a hybrid deep learning prediction model.

[0099] Specifically, the convolutional neural network is mainly composed of convolutional layers, pooling layers and full connections. The convolutional layer performs convolution operations with the input data through the convolution kernel to extract local features. The pooling layer is used to reduce the dimension of the feature map, reduce the amount of calculation and the risk of overfitting. The fully connected layer integrates the features extracted by the previous layer for final classification or prediction;

[0100] The bidirectional LSTM network consists of two LSTM layers, forward and backward. The forward LSTM layer processes from the beginning to the end of the sequence to capture past information; the backward LSTM layer processes from the end to the beginning of the sequence to obtain future information. Through this bidirectional processing, the bidirectional LSTM network can learn the contextual information in the sequence more comprehensively;

[0101] It should be noted that convolutional neural networks have the characteristics of local connection and weight sharing, which greatly reduces the number of network parameters, improves training efficiency, can automatically learn features from a large amount of data, and have strong modeling capabilities for complex data;

[0102] The bidirectional long short-term memory network can effectively handle long-term dependency problems in long sequence data and avoid gradient vanishing or exploding. The attention mechanism can accurately capture the importance of input features, that is, assign different weights to different features, assign larger weights to key factors, and assign smaller weights to weakly related features to distinguish the importance of input features, thereby improving the model's processing efficiency of input features.

[0103] The above is a schematic scheme of a load forecasting method of this embodiment. It should be noted that the technical scheme of the load forecasting device and the technical scheme of the load forecasting method described above belong to the same concept, and the details of the technical scheme of the load forecasting device in this embodiment that are not described in detail can all be referred to the description of the technical scheme of the load forecasting method described above.

[0104] The load forecasting device in this embodiment includes:

[0105] A load reconstruction module is used to obtain an original load sequence, decompose the original load sequence using a first modal decomposition to obtain modal components, and reconstruct the modal components through sample entropy aggregation to obtain a reconstructed load sequence;

[0106] A decomposition module, used to perform a primary decomposition of the high-frequency components in the modal components by using the second modal decomposition, determine the number of decomposed modes and the penalty factor, and perform a secondary decomposition of the high-frequency components in the reconstructed load sequence by using the third modal decomposition based on the number of decomposed modes and the penalty factor to obtain a modal subsequence;

[0107] The correlation analysis module is used to obtain input features based on the modal subsequence screening, and analyze the correlation between the reconstructed load sequence and the input features using the maximum mutual information method to obtain the input features of the hybrid deep learning prediction model;

[0108] The prediction module is used to obtain measured load data, input the measured load data into the hybrid deep learning prediction model, and obtain a prediction result.

[0109] This embodiment also provides a computing device, which is applicable to load forecasting, and includes:

[0110] Memory and processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the load forecasting method proposed in the above embodiment.

[0111] This embodiment further provides a storage medium on which a computer program is stored. When the program is executed by a processor, the load forecasting method proposed in the above embodiment is implemented.

[0112] The storage medium proposed in this embodiment and the method for implementing load forecasting proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0113] Through the above description of the implementation methods, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software and necessary general hardware, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ReadOnly, Memory, ROM), random access memory (RandomAccess Memory, RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform the methods of various embodiments of the present invention.

[0114] Example 2

[0115] As shown in Table 1, Figure 2-13 As shown, a load forecasting method is provided to verify and illustrate the technical effects adopted in this method.

[0116] The original load sequence is decomposed by using the improved integrated empirical mode decomposition (ICEEMD). Figure 2 and Figure 3 It can be seen that the original load has a strong follow-up property, and after one decomposition, there are 14 inherent modal components at the boundary, and the entropy values ​​of the 14 samples are relatively regular. Components 1, 2, and 3 can be selected to be aggregated into high-frequency components, components 4, 5, 6, and 7 can be aggregated into medium-frequency components, and the remaining components can be aggregated into low-frequency components, such as Figure 4 It is known;

[0117] The sample entropy aggregation method is used to reconstruct the load and reduce the complexity of the subsequent prediction sub-model. Figure 5-Figure 6It can be seen that component 1 is a high-frequency component with a higher frequency. Therefore, VMD is used to perform secondary decomposition of the high-frequency component and decompose and reconstruct it into 5 components;

[0118] The improved adaptive noise complete ensemble empirical mode decomposition (ICEEMDAN) is used to decompose the high-frequency components, determine the number of decomposed modes and the penalty factor, and use variational mode decomposition (VMD) to perform secondary decomposition on the high-frequency components in the sample entropy reconstruction sequence to further reduce the non-stationarity of the original load sequence and obtain a smooth and stable modal subsequence. Then, the maximum mutual information method (MIC) is used to conduct an in-depth analysis of the correlation between the reconstructed components and the input features to determine the final model input features. The CNN-BiLSTM-Attention hybrid deep learning prediction model is constructed by integrating the convolutional neural network (CNN), the bidirectional long short-term memory network (BiLSTM), and the attention mechanism (Attention), which reduces the impact of the input on the prediction accuracy of the output. Figure 7-13 It can be seen that the maximum temperature, minimum temperature, average temperature, relative humidity and rainfall have a great influence on the component.

[0119] Among them, the highest temperature has the greatest impact on the component, followed by the lowest temperature and average temperature. The approximate values ​​of the correlation coefficients are all greater than 0.5, while the correlation coefficients of relative humidity and rainfall are mostly less than 0.5, and the correlation effect is weak. In order to reduce the calculation amount of the model, the characteristic relative humidity and rainfall are deleted, and the prediction accuracy is obtained in Table 1;

[0120] Table 1 Prediction accuracy

[0121]

[0122] As can be seen from Table 1, the prediction accuracy of the present invention can reach 98.55%. Through the decomposition of ICEEMD and ICEEMDAN, combined with sample entropy aggregation and secondary decomposition of VMD, the non-stationarity of the original load sequence is effectively reduced, thereby constructing a high-quality data set. The maximum mutual information method (MIC) is used for feature selection, and the hybrid deep learning model constructed by combining CNN, BiLSTM and Attention mechanism can significantly improve the accuracy of load prediction.

[0123] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A load forecasting method, characterized in that: include: Obtaining an original load sequence, decomposing the original load sequence using a first modal decomposition to obtain modal components, and reconstructing the modal components through sample entropy aggregation to obtain a reconstructed load sequence; Decomposing the high-frequency components in the modal components once by using the second modal decomposition to determine the number of decomposed modes and the penalty factor, and decomposing the high-frequency components in the reconstructed load sequence twice by using the third modal decomposition based on the number of decomposed modes and the penalty factor to obtain a modal subsequence; The input features are obtained by screening the modal subsequences, and the correlation between the reconstructed load sequence and the input features is analyzed using the maximum mutual information method to obtain the input features of the hybrid deep learning prediction model. Obtain measured load data, input the measured load data into the hybrid deep learning prediction model, and obtain a prediction result.

2. The load forecasting method according to claim 1, characterized in that: Decomposing the original load sequence by using the first mode decomposition includes: Initialize the improved complete integrated empirical mode decomposition algorithm and set key parameters; Add adaptive noise to the original load sequence, generate Gaussian white noise each time the noise is added, standardize the Gaussian white noise, and add it to the original signal according to the set different noise amplitudes; Perform empirical mode decomposition on the original signal with added noise to obtain several intrinsic mode functions; All the intrinsic mode functions obtained by adding noise and decomposing are collected together, and the mean of each corresponding intrinsic mode function component is calculated to obtain the decomposition result.

3. The load forecasting method according to claim 2, characterized in that: Load reconstruction of modal components by sample entropy aggregation includes: The sample entropy of each intrinsic mode function is calculated, and according to the sample entropy value of each intrinsic mode function, the intrinsic mode functions with similar sample entropy values ​​are aggregated, and the aggregated intrinsic mode functions are reconstructed into a new signal to obtain a reconstructed load sequence.

4. The load forecasting method according to claim 3, characterized in that: Decomposing the high frequency components in the modal components by using the second modal decomposition includes: Initialize the improved adaptive noise complete set empirical mode decomposition algorithm and set relevant parameters; Perform empirical mode decomposition on the high frequency component and white noise to obtain the intrinsic mode function component, and select the Kth intrinsic mode function component of the noise as the noise component of the signal to be decomposed; The improved adaptive noise complete set empirical mode decomposition is used to decompose the signal to be decomposed, and the intrinsic mode function is obtained. The mode function is analyzed to obtain the number of decomposed modes and the penalty factor.

5. The load forecasting method according to claim 4, characterized in that: The secondary decomposition of the high-frequency components in the reconstructed load sequence using the third mode decomposition includes: Initialize the center frequency and mode function, and iterate using the initialized parameters; For each iteration, the modal function is updated using the Wiener filtering method and the center frequency is calculated; Determine whether the modal function meets the convergence condition, that is, whether it no longer changes within the tolerance range. If it does not meet the convergence condition, continue to iterate and update the Lagrange multiplier to meet the constraint condition; If the convergence condition is met or the maximum number of iterations is reached, the iteration is stopped and the decomposed mode function and the corresponding center frequency are output; The modal function of the high-frequency component is identified from the decomposed modal function, and the high-frequency component function is decomposed twice using variational modal decomposition to obtain a modal subsequence.

6. The load forecasting method according to claim 5, characterized in that: The maximum mutual information method is used to analyze the correlation between the reconstructed load sequence and the input features, including: Calculate the mutual information value for each relevant input feature and the reconstructed load sequence; Determine the joint probability distribution of the input feature and the reconstructed load sequence, calculate the marginal probability distribution of the input feature and the reconstructed load sequence, calculate the logarithm of the ratio of the joint probability distribution to the product of the marginal probability distribution of the input feature and the reconstructed load sequence multiplied by the joint probability distribution, and sum them to obtain the value of the mutual information of the input feature and the reconstructed load sequence; The calculated mutual information values ​​are sorted, and features whose mutual information values ​​are greater than the first threshold are selected as input features of the hybrid deep learning prediction model.

7. The load forecasting method according to claim 1 or 6, characterized in that: The hybrid deep learning prediction model includes: Use convolutional neural networks to extract local features from time series data. After the convolutional neural network layer, use a bidirectional long short-term memory network layer to further process the features. After the bidirectional long short-term memory network layer, the attention mechanism is introduced to obtain a hybrid deep learning prediction model.

8. A system for a load forecasting method, characterized in that: include: A load reconstruction module is used to obtain an original load sequence, decompose the original load sequence using a first modal decomposition to obtain modal components, and reconstruct the modal components through sample entropy aggregation to obtain a reconstructed load sequence; A decomposition module, used to perform a primary decomposition of the high-frequency components in the modal components by using the second modal decomposition, determine the number of decomposed modes and the penalty factor, and perform a secondary decomposition of the high-frequency components in the reconstructed load sequence by using the third modal decomposition based on the number of decomposed modes and the penalty factor to obtain a modal subsequence; The correlation analysis module is used to obtain input features based on the modal subsequence screening, and analyze the correlation between the reconstructed load sequence and the input features using the maximum mutual information method to obtain the input features of the hybrid deep learning prediction model; The prediction module is used to obtain measured load data, input the measured load data into the hybrid deep learning prediction model, and obtain a prediction result.

9. An electronic device, characterized in that: The device comprises: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images