Short-term power load prediction method and device, terminal equipment and storage medium

Abstract

The invention discloses a short-term power load prediction method and device, terminal equipment and a storage medium, and belongs to the field of power grid load prediction.The method comprises the steps that a power load time sequence in a preset interval is obtained, the power load time sequence is processed through metamorphosis mode decomposition, a plurality of mode components are obtained, and the center frequency of each mode component is calculated; inputting the modal component of which the center frequency is smaller than or equal to a preset threshold value into a bidirectional gate control unit to obtain a first prediction result; inputting the modal component of which the center frequency is greater than a preset threshold value into a preset long-short-term memory network to obtain a second prediction result; a memory enhancement unit is introduced into the preset long-short-term memory network; the memory enhancement unit is used for updating and storing historical information; and fusing the first prediction result and the second prediction result to obtain a load prediction value. Therefore, the technical problem that in the prior art, an effective load time sequence data processing mechanism cannot be constructed, and the long-term dependency feature of the sequence cannot be fully captured can be solved.

Description

Technical Field

[0001] The present application relates to the field of power grid load forecasting, and in particular to a short-term power load forecasting method, apparatus, terminal equipment, and storage medium. Background Art

[0002] Short-term power load forecasting is an important basis for the safe and stable operation and economic dispatch of power systems. It has important guiding significance for the formulation of unit start-up and shutdown plans, system spare capacity management, and operation mode optimization.

[0003] At present, researchers mainly use traditional statistical methods and deep learning methods to conduct power load forecasting research, but these methods still have the following problems: on the one hand, traditional statistical methods are difficult to effectively handle nonlinear and non-stationary load time series data, and the prediction accuracy is limited; on the other hand, a single deep learning model is prone to problems such as insufficient feature extraction and insufficient capture of long-term dependencies when processing complex load data.

[0004] Therefore, how to build an effective load time series data processing mechanism and fully capture the long-term dependency characteristics of the sequence is a technical problem that needs to be solved in the current short-term power load forecasting process. Summary of the Invention

[0005] The present application provides a short-term power load forecasting method, apparatus, terminal device and storage medium, which can solve the technical problem of building an effective load time series data processing mechanism and fully capturing the long-term dependency characteristics of the sequence.

[0006] This application provides a short-term power load forecasting method, comprising:

[0007] Obtaining a power load time series within a preset interval, processing the power load time series by variant modal decomposition to obtain a plurality of modal components, and calculating a center frequency of each of the modal components;

[0008] Inputting the modal component whose center frequency is less than or equal to a preset threshold into a bidirectional gate control unit to obtain a first prediction result;

[0009] Inputting the modal component having a center frequency greater than a preset threshold into a preset long short-term memory network to obtain a second prediction result; the preset long short-term memory network introduces a memory enhancement unit; the memory enhancement unit is used to update and store historical information;

[0010] The first prediction result and the second prediction result are integrated to obtain a load prediction value.

[0011] Compared with the prior art, the embodiments of the present application have the following beneficial effects: the power load time series is decomposed at multiple scales through a modified modal decomposition algorithm, and the modal components obtained by the decomposition are divided into two modal components with different frequencies based on the center frequency, effectively converting the complex original load time series into a characteristic sequence with significant regularity, thereby realizing deep mining of the information contained in the time series; in view of the limitations of a single model in load forecasting, the present application uses a bidirectional gate control unit suitable for processing low-frequency sequences to predict low-frequency modal components, and further introduces an additional memory enhancement unit that can store and update historical information into the traditional long-short-term memory network, so that the long-short-term memory network can store and utilize historical information over a longer time span, thereby improving the long-short-term memory network's ability to capture complex time series patterns. In particular, when facing high-frequency sequence prediction tasks, the improved long-short-term memory network can better handle long-term dependencies and short-term fluctuations, and has stronger robustness. The prediction accuracy is improved by complementing the advantages of the bidirectional gate control unit and the improved long-short-term memory network.

[0012] Furthermore, the power load time series is processed by deformed modal decomposition to obtain several modal components, including:

[0013] Construct the variational problem corresponding to the power load time series decomposition problem;

[0014] The variational problem is iteratively solved by an alternating direction multiplier method until a preset iteration termination condition is met, thereby obtaining the plurality of modal components.

[0015] Compared with the existing technology, the above embodiment has the following beneficial effects: since the abnormal modal decomposition method constructs a variational problem and further uses the alternating direction multiplier method to process the power load time series, compared with the traditional empirical mode decomposition, the abnormal modal decomposition has higher computational efficiency and more stable decomposition results. Especially when facing the problem of power load forecasting, the abnormal modal decomposition can effectively separate the periodic fluctuations, trend changes and random disturbances in the load data from the complex power load time series, thereby improving the accuracy of subsequent model predictions.

[0016] Furthermore, the variational problem corresponding to the power load time series decomposition problem includes:

[0017] The mathematical modeling of the variational problem is specifically as follows:

[0018]

[0019] Where f(t) represents the original load signal; δ(t) is the Dirac distribution function; t represents time; u k With w kare the modal components and corresponding center frequencies of the kth decomposition respectively; K is the number of modal decompositions; j is the imaginary unit; is the time derivative operator; π is the circumference of a circle; is the square of the L2 norm.

[0020] Furthermore, the calculating the center frequency of each modal component includes:

[0021] Processing each of the modal components by fast Fourier transform to obtain a frequency spectrum corresponding to each of the modal components;

[0022] The frequency with the maximum amplitude corresponding to each of the frequency spectra is taken as the center frequency of the modal component corresponding to the frequency spectrum.

[0023] Compared with the prior art, the above embodiment has the following beneficial effects: since different modal components have different center frequencies, after the original load data is decomposed at multiple scales, not only can characteristic information such as periodic fluctuations, trend changes, and random disturbances in the load sequence be effectively separated, but also the subsequences can be divided into high-frequency and low-frequency sequences according to the center frequency, and appropriate models can be selected to process sequences of different frequencies, effectively extracting more accurate prediction information from the sequence data and improving the final prediction accuracy.

[0024] Furthermore, the preset long short-term memory network includes: a forget gate, an input gate, a memory unit state, a candidate unit state and the memory enhancement unit;

[0025] Wherein, the memory enhancement unit includes: a state transition operator and a dynamic memory variable;

[0026] The dynamic memory variable of the current time step is determined by the dynamic memory variable, the memory unit state and the output of the previous time step;

[0027] The state transition operator is used to update the memory unit state according to the dynamic memory variable, the forget gate, the input gate and the candidate unit state.

[0028] Compared to existing technologies, the above embodiment has the following beneficial effects: Traditional long-term short-term memory networks rely on a fixed forget gate mechanism, which makes long-term information susceptible to decay over time. By introducing a memory enhancement unit into a traditional long-term short-term memory network, the model can more flexibly retain key long-term features by adaptively adjusting the storage weight of historical information through dynamic memory variables. Furthermore, the memory enhancement unit integrates the memory state and output of the previous moment, combined with the control of the current input gate and forget gate, to form a multi-dimensional information interaction mechanism, improving the model's ability to represent complex temporal patterns and capture long-term dependencies.

[0029] Furthermore, the dynamic memory variable of the current time step is determined by the dynamic memory variable, the memory unit state and the output of the previous time step, including:

[0030]

[0031] Among them, d (t) is the dynamic memory variable of the current time step t; W dd 、W dc and W dy represents the weight matrix; σ(*) is the Sigmoid activation function; d (t-1) Represents the dynamic memory variable of the previous time step t-1; c (t-1) Represents the memory cell state at the previous time step t-1; y (t-1) represents the output of the previous time step t-1; b d is the bias term.

[0032] Compared with the existing technology, the above embodiment has the following beneficial effects: by associating historical dynamic memory, memory state and output through the weight matrix, information integration across time scales is achieved, thereby capturing multi-level patterns in time series, thereby improving the accuracy of subsequent forget gate operations on historical information retention and update.

[0033] Furthermore, the state transition operator is used to update the memory unit state according to the dynamic memory variable, the forget gate, the input gate and the candidate unit state, including:

[0034]

[0035] Where I is the identity matrix; B is the state transition operator; f (t) is the forget gate of the current time step t; d (t) is the dynamic memory variable of the current time step t; c (t) is the memory unit state at the current time step t; i (t) is the input gate of the current time step t; is the candidate unit state at the current time step t; ⊙ is the dot product operator.

[0036] Compared with the existing technology, the above embodiment has the following beneficial effects: by introducing a dynamic memory variable to adjust the degree of effect of the forget gate, the model can achieve a dynamic balance between forgetting redundant information and retaining key features; further, the introduction of the state transition operator expands the state update dimension of the traditional long short-term memory network, and captures the complex timing dependencies in the load sequence through matrix operations.

[0037] Another embodiment of the present application further provides a short-term power load forecasting device, comprising: a variant mode decomposition module, a first prediction result acquisition module, a second prediction result acquisition module, and a result fusion module;

[0038] The abnormal modal decomposition module is used to obtain a power load time series within a preset interval, process the power load time series through abnormal modal decomposition, obtain a number of modal components, and calculate the center frequency of each modal component;

[0039] The first prediction result acquisition module is configured to input the modal component having a center frequency less than or equal to a preset threshold into the bidirectional gate control unit to obtain a first prediction result;

[0040] The second prediction result acquisition module is configured to input the modal component having a center frequency greater than a preset threshold into a preset long short-term memory network to obtain a second prediction result; the preset long short-term memory network introduces a memory enhancement unit; the memory enhancement unit is configured to update and store historical information;

[0041] The result fusion module is used to fuse the first prediction result and the second prediction result to obtain a load prediction value.

[0042] Another embodiment of the present application also provides a terminal device, including: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the steps of the short-term power load forecasting method of the present application are implemented.

[0043] Another embodiment of the present application further provides a computer-readable storage medium item, comprising: a stored computer program, which controls the device where the computer-readable storage medium is located to execute the steps of the short-term power load forecasting method of the present application when the computer program is running. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solution of the present application, the following is a brief introduction to the drawings required for use in the implementation. Obviously, the drawings described below are only some implementation methods of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0045] Figure 1 A schematic flow chart of a short-term power load forecasting method provided in some embodiments of the present application;

[0046] Figure 2 This is a structural schematic diagram of a short-term power load forecasting device provided in some embodiments of the present application. DETAILED DESCRIPTION

[0047] To make the objectives, technical solutions, and advantages of this application more clear, the technical solutions in this application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of this application.

[0048] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs; the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned figure descriptions are intended to cover non-exclusive inclusions.

[0049] In the description of the embodiments of this application, the technical terms "first" and "second" are used only to distinguish different objects and should not be understood to indicate or imply relative importance or implicitly specify the quantity, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, the meaning of "plurality" is more than two, unless otherwise clearly and specifically defined.

[0050] References to "embodiments" herein mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0051] In the description of the embodiments of this application, the term "and / or" is simply a description of the association relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent the following three situations: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in this document generally indicates that the associated objects are in an "or" relationship.

[0052] In the description of the embodiments of the present application, the term "multiple" refers to more than two (including two). Similarly, "multiple groups" refers to more than two groups (including two groups), and "multiple pieces" refers to more than two pieces (including two pieces).

[0053] In the description of the embodiments of the present application, unless otherwise expressly specified or limited, technical terms such as "installed," "connected," "connected," and "fixed" should be understood in a broad sense. For example, they can refer to fixed connections, detachable connections, or integration; mechanical connections or electrical connections; direct connections or indirect connections through an intermediate medium; internal connections between two components or interactions between two components. Those skilled in the art can understand the specific meanings of the above terms in the embodiments of the present application based on specific circumstances.

[0054] At present, researchers mainly use traditional statistical methods and deep learning methods to conduct power load forecasting research, but these methods still have the following problems: on the one hand, traditional statistical methods are difficult to effectively handle nonlinear and non-stationary load time series data, and the prediction accuracy is limited; on the other hand, a single deep learning model is prone to problems such as insufficient feature extraction and insufficient capture of long-term dependencies when processing complex load data.

[0055] Please refer to Figure 1 To solve the technical problem in the prior art that it is impossible to build an effective load time series data processing mechanism and fully capture the long-term dependency characteristics of the sequence, the embodiment of the present application provides a short-term power load forecasting method, including S101 to S104, specifically:

[0056] S101: Obtain a power load time series within a preset interval, process the power load time series through modified modal decomposition to obtain a plurality of modal components, and calculate the center frequency of each modal component.

[0057] Furthermore, in some embodiments of the present application, the processing of the power load time series by modified modal decomposition to obtain a plurality of modal components includes:

[0058] Construct the variational problem corresponding to the power load time series decomposition problem;

[0059] The variational problem is iteratively solved by an alternating direction multiplier method until a preset iteration termination condition is met, thereby obtaining the plurality of modal components.

[0060] Variational Mode Decomposition (VMD) is an adaptive signal processing method based on variational theory. By constructing a variational problem and solving it using the alternating direction multiplier method, this method can decompose complex non-stationary signals into intrinsic mode functions with specific center frequencies and limited bandwidths. Compared with traditional empirical mode decomposition, VMD technology has the advantages of a solid theoretical basis, high computational efficiency, strong noise resistance, and stable decomposition results. In power load forecasting, VMD technology can decompose complex load time series into multiple IMF (Intrinsic Mode Function) subsequences (i.e., modal components) with different frequency characteristics, effectively separating features such as periodic fluctuations, trend changes, and random disturbances in load data, providing more effective feature inputs for subsequent deep learning models, thereby improving forecasting accuracy.

[0061] Furthermore, in some embodiments of the present application, the mathematical modeling of the VMD constrained variational problem is expressed as follows:

[0062]

[0063] Where f(t) represents the original load signal; δ(t) is the Dirac distribution function; t represents time; u k With w k are the modal components and corresponding center frequencies of the kth decomposition respectively; K is the number of modal decompositions; j is the imaginary unit; is the time derivative operator; π is the circumference of a circle; is the square of the L2 norm.

[0064] Furthermore, VMD adaptively decomposes complex load time series to obtain multiple IMF subsequences with different frequency characteristics. Compared with traditional signal decomposition methods, VMD is computationally efficient and produces stable decomposition results. By performing multi-scale decomposition on the original load data, it not only effectively separates characteristic information such as periodic fluctuations, trend changes, and random disturbances in the load series, but also divides the IMF subsequences into high-frequency and low-frequency sequences based on their center frequencies, providing a foundation for differentiated predictions using subsequent deep learning models.

[0065] Furthermore, in some embodiments of the present application, since different IMF subsequences have different center frequencies, and the center frequency reflects the main oscillation characteristics of the sequence, the classification of different IMF subsequences is determined by the center frequency, wherein the center frequency calculation step of the IMF subsequence includes:

[0066] Processing each of the modal components by fast Fourier transform to obtain a frequency spectrum corresponding to each of the modal components;

[0067] The frequency with the maximum amplitude corresponding to each of the frequency spectra is taken as the center frequency of the modal component corresponding to the frequency spectrum.

[0068] Specifically, in some embodiments of the present application, the center frequency calculation formula of the IMF subsequence is specifically:

[0069] U k (f)=FFT(u k (t))

[0070]

[0071] Among them, u k (t) is the time series of the kth modal component (i.e., IMF subsequence); FFT(*) is the fast Fourier transform operator; f represents frequency; |U k (f)| is the modal component u k (t) The spectrum amplitude at frequency f; argmax means finding the amplitude |U k (f)|maximum frequency f; f center,k Represents the center frequency of the kth modal component.

[0072] S102: Inputting the modal component with a center frequency less than or equal to a preset threshold into a bidirectional gate control unit to obtain a first prediction result.

[0073] Furthermore, this application classifies IMF subsequences into high- and low-frequency sequences based on their center frequencies: IMF subsequences with center frequencies greater than a preset threshold are classified as high-frequency sequences, while those with center frequencies less than or equal to the threshold are classified as low-frequency sequences. This frequency-based sequence classification provides a basis for subsequent targeted prediction using different deep learning models.

[0074] Furthermore, when the center frequency of the modal component is less than or equal to the preset threshold, since the modal component is low-frequency sequence data at this time, the temporal characteristics of the low-frequency sequence data are relatively simple and change relatively slowly. The bidirectional gate control unit (Bi-GRU neural network) can simultaneously consider the forward and backward information of the sequence, and can effectively capture the long-term dependencies of the low-frequency data.

[0075] Furthermore, in some embodiments of the present application, the structure of the Bi-GRU neural network includes: a convolutional layer, a bidirectional GRU (Gated Recurrent Unit) layer, and a fully connected layer; wherein the convolutional layer contains 16 filters of size 4. This layer uses the ReLU activation function to complete the feature extraction of the input data; the bidirectional GRU layer contains 64 and 128 GRU units respectively. Each layer uses the ReLU activation function to simultaneously learn the forward and backward information of the data; the fully connected layer specifically includes two fully connected layers, the first layer uses the ReLU activation function, and the second layer uses the linear activation function to ensure that the power consumption sequence of the target load can be accurately output.

[0076] The Bi-GRU neural network is a special recurrent neural network structure consisting of two GRU layers operating in opposite directions, capable of extracting feature information from both the forward and backward directions of a sequence. The GRU uses reset and update gates to control information transmission. Compared to traditional RNNs (Recurrent Neural Networks), it effectively alleviates the vanishing gradient problem, while the bidirectional structure fully utilizes contextual information.

[0077] S103: Input the modal component whose center frequency is greater than a preset threshold into a preset long short-term memory network to obtain a second prediction result; the preset long short-term memory network introduces a memory enhancement unit; the memory enhancement unit is used to update and store historical information.

[0078] For high-frequency sequence data, due to its strong volatility and obvious nonlinear characteristics, the long-short-term memory network preset in this application enhances the memory capacity of the model by introducing additional memory units, enabling the model to better store and utilize historical information, and is particularly suitable for processing complex temporal dependencies and mutation characteristics in high-frequency sequences.

[0079] Furthermore, in some embodiments of the present application, the preset long short-term memory network includes: a forget gate, an input gate, a memory unit state, a candidate unit state, and the memory enhancement unit;

[0080] Wherein, the memory enhancement unit includes: a state transition operator and a dynamic memory variable;

[0081] The dynamic memory variable of the current time step is determined by the dynamic memory variable, the memory unit state and the output of the previous time step;

[0082] The state transition operator is used to update the memory unit state according to the dynamic memory variable, the forget gate, the input gate and the candidate unit state.

[0083] Traditional LSTM networks rely on a fixed forget gate mechanism, which makes long-term information susceptible to decay over time. By introducing a memory-enhancing unit (MAU) into traditional LSTM networks, the dynamic memory variables adaptively adjust the storage weights of historical information, enabling the model to more flexibly retain key long-term features. The MAU also integrates the memory state and output of the previous moment, combined with the control of the current input gate and forget gate, to form a multi-dimensional information interaction mechanism, improving the model's ability to represent complex temporal patterns and capture long-term dependencies.

[0084] In order to more conveniently refer to the preset long short-term memory network, MAGULSTM (Memory-Augmented Gated Unit Long Short-Term Memory) will be used to refer to it.

[0085] The MAGULSTM neural network used in this application is an enhanced long-term and short-term memory network structure that expands the functionality of the traditional long-term and short-term memory network (LSTM) by introducing additional memory enhancement units. The network's notable feature is its unique memory enhancement mechanism, which enhances the model's ability to capture long-term dependencies by designing specialized memory units to store and update important historical information. Compared with traditional LSTM, MAGULSTM not only retains the original input gate, forget gate, and output gate structure of LSTM, but also provides a larger memory capacity and a more flexible information processing mechanism through memory enhancement units, which enables the model to better process data sequences with complex temporal dependencies, and is particularly suitable for processing high-frequency time series data prediction tasks containing long-term dependencies.

[0086] Furthermore, in some embodiments of the present application, the equation for MAGULS™ to be updated at each time step is as follows:

[0087] y (t) =g(W yh h (t) +b y )

[0088] h (t) =o (t) ⊙tanh(c (t) )

[0089]

[0090] i (t) =σ(W ih h (t-1) +W iy y (t-1) +b i )

[0091] o (t) =σ(W oh h (t-1) +W oy y (t-1) +b o )

[0092] f (t) =σ(W fh h (t-1) +W fy y (t-1) +b f )

[0093] Where y (t) is the output of the current time step; h (t) is the hidden state of the current time step; c (t) is the memory cell state at the current time step; f (t) is the forget gate, which controls the state transition operator; B is the state transition operator, which dynamically adjusts the memory state; d (t) It is a dynamic memory variable that controls the intensity of state transition; is the candidate memory state, calculated by current input and historical information; i (t) is the input gate, which controls the introduction of new memory information; (t) is the output gate, which controls the output of the hidden state; W is the weight matrix; b is the bias term; σ(*) is the Sigmoid activation function; tanh is the hyperbolic tangent activation function; ⊙ is the point multiplication operator.

[0094] Furthermore, in some embodiments of the present application, combined with the equations of MAGULS™ at each time step, the steps of MAGULS™ at each time step are specifically as follows:

[0095] First, the forget gate f (t) Combined with the hidden state h of the previous time step (t-1) and historical output y (t-1) , the Sigmoid activation function is used to determine whether the current time step retains or discards historical memory information. (t) When it is close to 0, the memory is partially discarded; when f (t) When it is close to 1, the historical memory is retained. Secondly, the input gate i (t) and candidate memory states Together they determine the introduction of new information. The input gate controls the proportion of the current input information, while the candidate memory state is obtained by the hidden state h of the previous time step. (t-1) and historical output y (t-1) Calculated. Again, dynamic memory variable d (t) Through the dynamic memory variable d of the previous time step (t-1), memory cell state c (t-1) and output y (t-1) The dynamic memory variable d is calculated. (t) The Sigmoid activation function is limited to the range of (0,1) to control the strength of the state transition operator. Next, the state transition operator B is introduced and combined with the dynamic variable d (t) and forget gate f (t) For the memory cell state c (t) Perform dynamic adjustment and update the memory unit c(t)c^{(t)}c(t). This process removes the memory state c of the previous time step in an exponential form (t-1) Direct dependency of , achieving more flexible memory modeling. Then, the output gate o (t) By the hidden state h of the previous time step (t-1) and historical output y (t-1) Determines which information in the current hidden state will be output. The result of the output gate is combined with the memory cell state c (t) The activation value of the current hidden state h (t) Finally, the current hidden state h (t) The final output y of the current time step is generated through the weight matrix and activation function mapping of the output layer (t) .

[0096] The above examples show that during the MAGULSTM update process, this application integrates information across time scales by associating historical dynamic memory, memory state, and output through a weight matrix, thereby capturing multi-level patterns in time series and improving the accuracy of subsequent forget gate operations on historical information retention and update. Furthermore, by introducing dynamic memory variables to adjust the degree of effect of the forget gate, the model achieves a dynamic balance between forgetting redundant information and retaining key features. Furthermore, the introduction of state transition operators expands the state update dimension of traditional long-short-term memory networks, capturing complex temporal dependencies in load sequences through matrix operations.

[0097] Therefore, the aforementioned MAGULSTM introduces external memory units, enabling the network to store and utilize historical information over a longer time span, thereby improving its ability to capture complex temporal patterns. In particular, in high-frequency sequence prediction tasks, MAGULSTM is able to better handle long-term dependencies and short-term fluctuations, resulting in greater robustness.

[0098] S104: Fusing the first prediction result and the second prediction result to obtain a load prediction value.

[0099] Furthermore, in some embodiments of the present application, the fusing of the first prediction result and the second prediction result to obtain the load prediction value includes: summing the prediction results of the high-frequency and low-frequency sequences to generate a final load prediction value.

[0100] In summary, compared with the prior art, the short-term power load forecasting method provided in the embodiment of the present application has the following beneficial effects: the power load time series is decomposed at multiple scales by a variant modal decomposition algorithm, and the modal components obtained by decomposition are divided into two modal components with different frequencies based on the center frequency, effectively converting the complex original load time series into a characteristic sequence with significant regularity, thereby realizing deep mining of the information contained in the time series; in view of the limitations of a single model in load forecasting, the present application uses a bidirectional gate control unit suitable for processing low-frequency sequences to predict low-frequency modal components, and further introduces an additional memory enhancement unit that can store and update historical information into the traditional long and short-term memory network, so that the long and short-term memory network can store and utilize historical information over a longer time span, thereby improving the long and short-term memory network's ability to capture complex time series patterns, especially in the face of high-frequency sequence prediction tasks. The improved long and short-term memory network can better handle long-term dependencies and short-term fluctuations, and has stronger robustness. The prediction accuracy is improved by the complementary advantages between the bidirectional gate control unit and the improved long and short-term memory network.

[0101] like Figure 2 As shown, based on the above method embodiment, a corresponding device embodiment is provided;

[0102] An embodiment of the present invention provides a short-term power load forecasting device, including: a modified mode decomposition module 201 , a first prediction result acquisition module 202 , a second prediction result acquisition module 203 , and a result fusion module 204 .

[0103] Furthermore, in some embodiments of the present application, the abnormal modal decomposition module 201 is used to obtain the power load time series within a preset interval, process the power load time series through abnormal modal decomposition, obtain several modal components, and calculate the center frequency of each modal component; the first prediction result acquisition module 202 is used to input the modal component with a center frequency less than or equal to a preset threshold into the bidirectional gate control unit to obtain a first prediction result; the second prediction result acquisition module 203 is used to input the modal component with a center frequency greater than the preset threshold into a preset long short-term memory network to obtain a second prediction result; the preset long short-term memory network introduces a memory enhancement unit; the memory enhancement unit is used to update and store historical information; the result fusion module 204 is used to fuse the first prediction result and the second prediction result to obtain a load prediction value.

[0104] Furthermore, in some embodiments of the present application, the processing of the power load time series by variant modal decomposition to obtain a number of modal components includes: constructing a variational problem corresponding to the power load time series decomposition problem; iteratively solving the variational problem by an alternating direction multiplier method until a preset iteration termination condition is met, thereby obtaining the number of modal components.

[0105] Furthermore, in some embodiments of the present application, constructing a variational problem corresponding to the power load time series decomposition problem includes:

[0106] The mathematical modeling of the variational problem is specifically as follows:

[0107]

[0108] Where f(t) represents the original load signal; δ(t) is the Dirac distribution function; t represents time; u k With w k are the modal components and corresponding center frequencies of the kth decomposition respectively; K is the number of modal decompositions; j is the imaginary unit; is the time derivative operator; π is the circumference of a circle; is the square of the L2 norm.

[0109] Furthermore, in some embodiments of the present application, the calculation of the center frequency of each of the modal components includes: processing each of the modal components through fast Fourier transform to obtain the frequency spectrum corresponding to each of the modal components; and taking the frequency with the largest amplitude corresponding to each of the spectra as the center frequency of the modal component corresponding to the spectrum.

[0110] Furthermore, in some embodiments of the present application, the preset long short-term memory network includes: a forget gate, an input gate, a memory unit state, a candidate unit state and the memory enhancement unit; wherein, the memory enhancement unit includes: a state transition operator and a dynamic memory variable; the dynamic memory variable of the current time step is determined by the dynamic memory variable, the memory unit state and the output of the previous time step; the state transition operator is used to update the memory unit state according to the dynamic memory variable, the forget gate, the input gate and the candidate unit state.

[0111] Furthermore, in some embodiments of the present application, the dynamic memory variable of the current time step is determined by the dynamic memory variable, the memory unit state, and the output of the previous time step, including:

[0112]

[0113] Among them, d (t)is the dynamic memory variable of the current time step t; W dd 、W dc and W dy represents the weight matrix; σ(*) is the Sigmoid activation function; d (t-1) Represents the dynamic memory variable of the previous time step t-1; c (t-1) Represents the memory cell state at the previous time step t-1; y (t-1) represents the output of the previous time step t-1; b d is the bias term.

[0114] Furthermore, in some embodiments of the present application, the state transition operator is used to update the memory unit state according to the dynamic memory variable, the forget gate, the input gate, and the candidate unit state, including:

[0115]

[0116] Where I is the identity matrix; B is the state transition operator; f (t) is the forget gate of the current time step t; d (t) is the dynamic memory variable of the current time step t; c (t) is the memory cell state at the current time step t; i (t) is the input gate of the current time step t; is the candidate unit state at the current time step t; ⊙ is the dot product operator.

[0117] It can be understood that the above-mentioned device embodiment corresponds to the method embodiment of the present invention, and can implement any one of the above-mentioned method embodiments of the present invention to provide a short-term power load forecasting method.

[0118] In summary, compared with the prior art, the short-term power load forecasting device provided in the embodiment of the present application has the following beneficial effects: the power load time series is decomposed at multiple scales by a variant modal decomposition algorithm, and the modal components obtained by decomposition are divided into two modal components with different frequencies based on the center frequency, effectively converting the complex original load time series into a characteristic sequence with significant regularity, thereby realizing deep mining of the information contained in the time series; in view of the limitations of a single model in load forecasting, the present application uses a bidirectional gate control unit suitable for processing low-frequency sequences to predict low-frequency modal components, and further introduces an additional memory enhancement unit that can store and update historical information into the traditional long and short-term memory network, so that the long and short-term memory network can store and utilize historical information over a longer time span, thereby improving the long and short-term memory network's ability to capture complex time series patterns, especially in the face of high-frequency sequence prediction tasks. The improved long and short-term memory network can better handle long-term dependencies and short-term fluctuations, and has stronger robustness. The prediction accuracy is improved by the complementary advantages between the bidirectional gate control unit and the improved long and short-term memory network.

[0119] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. In addition, in the drawings of the device embodiments provided herein, the connection relationship between the modules indicates that there is a communication connection between them, which may be implemented as one or more communication buses or signal lines. Those skilled in the art may understand and implement the present invention without inventive effort.

[0120] Based on the above-mentioned embodiment of the short-term power load forecasting method, another embodiment of the present application provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the short-term power load forecasting method of any embodiment of the present application is implemented.

[0121] For example, in this embodiment, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present application. The one or more module elements may be a series of computer program instruction segments capable of completing specific functions, and the instruction segments are used to describe the execution process of the computer program in the terminal device.

[0122] The terminal device may be a computing device such as a desktop computer, a notebook computer, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0123] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, connecting various parts of the entire terminal device using various interfaces and lines.

[0124] Based on the above-mentioned method embodiments, another embodiment of the present application provides a computer-readable storage medium, including a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the short-term power load forecasting method described in any of the above-mentioned method embodiments of the present application.

[0125] Wherein, the module / unit integrated in the device / terminal equipment, if implemented in the form of a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the process in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program, when executed by the processor, can implement the steps of the above-mentioned various method embodiments. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device that can carry the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunications signal, and a software distribution medium, etc.

Claims

1. A short-term power load forecasting method, characterized in that: include: Obtaining a power load time series within a preset interval, processing the power load time series by variant modal decomposition to obtain a plurality of modal components, and calculating a center frequency of each of the modal components; Inputting the modal component whose center frequency is less than or equal to a preset threshold into a bidirectional gate control unit to obtain a first prediction result; Inputting the modal component having a center frequency greater than a preset threshold into a preset long short-term memory network to obtain a second prediction result; the preset long short-term memory network introduces a memory enhancement unit; The memory enhancement unit is used to update and store historical information; The first prediction result and the second prediction result are integrated to obtain a load prediction value.

2. A short-term power load forecasting method according to claim 1, characterized in that: The power load time series is processed by modified modal decomposition to obtain several modal components, including: Construct the variational problem corresponding to the power load time series decomposition problem; The variational problem is iteratively solved by an alternating direction multiplier method until a preset iteration termination condition is met, thereby obtaining the plurality of modal components.

3. A short-term power load forecasting method according to claim 2, characterized in that: The variational problem corresponding to the power load time series decomposition problem includes: The mathematical modeling of the variational problem is specifically as follows: Where f(t) represents the original load signal; δ(t) is the Dirac distribution function; t represents time; u k With w k are the modal components and corresponding center frequencies of the kth decomposition respectively; K is the number of modal decompositions; j is the imaginary unit; is the time derivative operator; π is the circumference of a circle; is the square of the L2 norm.

4. A short-term power load forecasting method according to claim 1, characterized in that: The calculating the center frequency of each modal component includes: Processing each of the modal components by fast Fourier transform to obtain a frequency spectrum corresponding to each of the modal components; The frequency with the maximum amplitude corresponding to each of the frequency spectra is taken as the center frequency of the modal component corresponding to the frequency spectrum.

5. A short-term power load forecasting method according to claim 1, characterized in that: The preset long short-term memory network includes: a forget gate, an input gate, a memory unit state, a candidate unit state and the memory enhancement unit; Wherein, the memory enhancement unit includes: a state transition operator and a dynamic memory variable; The dynamic memory variable of the current time step is determined by the dynamic memory variable, the memory unit state and the output of the previous time step; The state transition operator is used to update the memory unit state according to the dynamic memory variable, the forget gate, the input gate and the candidate unit state.

6. A short-term power load forecasting method according to claim 5, characterized in that: The dynamic memory variable of the current time step is determined by the dynamic memory variable, the memory unit state and the output of the previous time step, including: Among them, d (t) is the dynamic memory variable of the current time step t; W dd 、W dc and W dy represents the weight matrix; σ(*) is the Sigmoid activation function; d (t-1) Represents the dynamic memory variable of the previous time step t-1; c (t-1) Represents the memory cell state at the previous time step t-1; y (t-1) represents the output of the previous time step t-1; b d is the bias term.

7. A short-term power load forecasting method according to claim 6, characterized in that: The state transition operator is used to update the memory unit state according to the dynamic memory variable, the forget gate, the input gate, and the candidate unit state, including: Where I is the identity matrix; B is the state transition operator; f (t) is the forget gate of the current time step t; d (t) is the dynamic memory variable of the current time step t; c (t) is the memory unit state at the current time step t; i (t) is the input gate of the current time step t; is the candidate unit state at the current time step t; ⊙ is the dot product operator.

8. A short-term power load forecasting device, characterized in that: include: A modified mode decomposition module, a first prediction result acquisition module, a second prediction result acquisition module, and a result fusion module; The abnormal modal decomposition module is used to obtain a power load time series within a preset interval, process the power load time series through abnormal modal decomposition, obtain a number of modal components, and calculate the center frequency of each modal component; The first prediction result acquisition module is configured to input the modal component having a center frequency less than or equal to a preset threshold into the bidirectional gate control unit to obtain a first prediction result; The second prediction result acquisition module is configured to input the modal component having a center frequency greater than a preset threshold into a preset long short-term memory network to obtain a second prediction result; the preset long short-term memory network introduces a memory enhancement unit; The memory enhancement unit is used to update and store historical information; The result fusion module is used to fuse the first prediction result and the second prediction result to obtain a load prediction value.

9. A terminal device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, a short-term power load forecasting method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute a short-term power load forecasting method according to any one of claims 1 to 7.

Images