A short-term power load prediction method and system based on GAT-LSTM, a storage medium and a computer

Abstract

The application relates to the technical field of power prediction, in particular to a short-term power load prediction method and system based on GAT-LSTM, a storage medium and a computer, the method comprises the following steps: downloading a power load historical data set, and performing self-adaptive normalization on the power load historical data; inputting the normalized data into a graph attention mechanism layer, and capturing potential correlation between a target residence and related residence loads through a GAT layer; inputting the data processed through the GAT layer into an LSTM layer, further acquiring time long-time dependence of the load data, and outputting a load prediction result; the short-term power load prediction based on GAT-LSTM is characterized in that the input information is historical data of the power load, first, the data set is normalized, then the spatial dependence of the input data is captured by using a graph attention network, the output result is input into a long short-term memory network, the time dependence of the data is extracted through the LSTM, and finally, a high-precision load prediction result is obtained.

Description

Technical Field

[0001] This invention relates to the field of power forecasting technology, and specifically to a short-term power load forecasting method, system, storage medium, and computer based on GAT-LSTM. Background Technology

[0002] With the rapid development of the power industry and the popularization of smart grid technology, power load forecasting plays an important role in energy planning such as power generation and distribution, and provides important technical support for the further advancement of the energy internet. Short-term load forecasting technology is the foundation for the efficient operation and analysis of power systems.

[0003] Existing load forecasting methods can be broadly categorized into statistical methods and machine learning methods. Classical statistical algorithms include exponential smoothing, autoregressive moving average (ARMA), autoregressive composite moving average (ARIMA), and fuzzy logic. However, the accuracy of statistical methods decreases when a longer forecasting timeframe is required. Machine learning methods, capable of capturing complex nonlinear relationships, have been widely applied in power load forecasting. In recent years, deep learning, as an emerging method in the field of machine learning, has become a powerful tool for processing big data, possessing strong automatic feature extraction capabilities and advantages in handling high-dimensional and nonlinear data. It has received widespread attention in the field of short-term power load forecasting. Recurrent neural networks (RNNs) and their variants, including Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRUs), have achieved success in power load forecasting.

[0004] Short-term load forecasting (STLF) is characterized by complexity, randomness, dispersion, and uncertainty, posing significant challenges to load forecasting. This is especially true in recent years, with the continuous implementation of demand-side response policies making load forecasting even more difficult. Existing load forecasting methods only consider temporal information (such as historical load values ​​and weather information), neglecting spatial information between the target residence and related residences. However, due to the influence of common conditions such as weather, seasons, and holidays, houses in the same or adjacent areas may exhibit similar electricity consumption patterns, and this information is very helpful for load forecasting of the target residence. With the continuous development of graph neural networks (GNNs), spatiotemporal concepts have been introduced into the field of energy forecasting and have gradually attracted the attention of scholars both domestically and internationally, becoming a popular research topic. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a short-term power load forecasting method, system, storage medium and computer based on GAT-LSTM.

[0006] To solve the above-mentioned technical problems, the first technical solution adopted by the present invention is as follows:

[0007] A short-term power load forecasting method based on GAT-LSTM, including

[0008] Download the historical power load dataset and perform adaptive normalization on the historical power load data;

[0009] The normalized data is input into the graph attention mechanism layer, and the potential correlation between the target residential building and related residential loads can be captured through the GAT layer.

[0010] The data processed by the GAT layer is input into the LSTM layer to further obtain the long-term time dependence of the load data and output the load forecast results.

[0011] To solve the above-mentioned technical problems, the second technical solution adopted by the present invention is as follows:

[0012] A short-term power load forecasting system based on GAT-LSTM includes a processing unit that executes the aforementioned short-term power load forecasting method based on GAT-LSTM.

[0013] To solve the above-mentioned technical problems, the third technical solution adopted by the present invention is as follows:

[0014] A storage medium storing a computer program that, when executed by a processor, implements the aforementioned short-term power load forecasting method based on GAT-LSTM.

[0015] To solve the above-mentioned technical problems, the fourth technical solution adopted by the present invention is as follows:

[0016] A computer includes at least a memory and a processor, wherein the memory stores a computer program, and the processor implements the above-described short-term power load forecasting method based on GAT-LSTM when executing the computer program in the memory.

[0017] The beneficial effects of this invention are as follows: This application is based on GAT-LSTM for short-term power load forecasting. The input information is historical power load data. First, the dataset is normalized. Then, a graph attention network (GAT) is used to capture the spatial dependencies of the input data. The output results are input into a long short-term memory network (LSTM). The temporal dependencies of the data are extracted through LSTM, thereby obtaining high-precision load forecasting results. Attached Figure Description

[0018] Figure 1 This is a schematic diagram illustrating the principle framework of a short-term power load forecasting method based on GAT-LSTM according to the present invention.

[0019] Figure 2This is a flowchart illustrating the multi-head attention mechanism calculation of a short-term power load forecasting method based on GAT-LSTM according to the present invention.

[0020] Figure 3 This is a schematic diagram of the basic unit structure of the LSTM network of the present invention. Detailed Implementation

[0021] To explain in detail the technical content, objectives, and effects of the present invention, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0022] Please refer to Figures 1 to 3 A short-term power load forecasting method based on GAT-LSTM includes downloading historical power load datasets and adaptively normalizing the historical power load data.

[0023] The normalized data is input into the graph attention mechanism layer, and the potential correlation between the target residential building and related residential loads can be captured through the GAT layer.

[0024] The data processed by the GAT layer is input into the LSTM layer to further obtain the long-term time dependence of the load data and output the load forecast results.

[0025] Furthermore, the historical power load dataset was obtained from OpenEI.

[0026] The OpenEI mentioned above refers to the Open Energy Information (OpenEI) website built by the U.S. Department of Energy.

[0027] Furthermore, adaptive normalization of historical power load data includes...

[0028] Given historical electricity load data X, after windowing and segmenting X, S consists of n sequences with a window length of L: X = {X1, X2, ..., X...} n};

[0029] Adaptive normalization is applied to S, resulting in the normalized sequence: R = {r1, r2, ... r} n}

[0030] Furthermore, the calculation formula for the graph attention mechanism layer is as follows:

[0031]

[0032] Where α(·) is an attention function.

[0033] Furthermore, the graph attention mechanism layer employs multi-head attention, and the calculation formula is further as follows:

[0034]

[0035] Where || represents the concatenation operation, α ij (k) W is the weight coefficient calculated by the k-th attention mechanism. k These are the learning parameters corresponding to the k-th attention mechanism.

[0036] Furthermore, the graph attention mechanism layer has multiple sets.

[0037] As can be seen from the above description, by employing multi-head attention and adding multiple sets of independent attention mechanisms, the learning ability of the system can be greatly improved.

[0038] Furthermore, the calculation formula for the LSTM layer includes...

[0039] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0040] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0041]

[0042]

[0043] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0044] h t =o t *tanh(C t )

[0045] The system includes several basic units, which include an input gate, a forget gate, and an output gate.

[0046] The information from the input gate, after being processed by the activation functions sigmoid and tanh, jointly determines the vector retained in the state memory unit.

[0047] The input of the forget gate and the intermediate output ht-1 of the state unit St-1 together determine the forgetting part of the state memory unit.

[0048] The output gate controls the information flowing into the current hidden state ht from the memory cell at the current moment.

[0049] Example 1

[0050] Reference Figure 1 A short-term power load forecasting method based on GAT-LSTM, including

[0051] Download the historical power load dataset and perform adaptive normalization on the historical power load data;

[0052] The normalized data is input into the graph attention mechanism layer, and the potential correlation between the target residential building and related residential loads can be captured through the GAT layer.

[0053] The data processed by the GAT layer is input into the LSTM layer to further obtain the long-term time dependence of the load data and output the load forecast results.

[0054] in,

[0055] The historical power load dataset was obtained from OpenEI.

[0056] The OpenEI mentioned above refers to the Open Energy Information (OpenEI) website built by the U.S. Department of Energy.

[0057] Adaptive normalization of historical power load data further includes

[0058] Given historical electricity load data X, after windowing and segmenting X, S consists of n sequences with a window length of L: X = {X1, X2, ..., X...} n};

[0059] Adaptive normalization is applied to S, resulting in the normalized sequence: R = {r1, r2, ... r} n}

[0060] The calculation formula for the graph attention mechanism layer is as follows:

[0061]

[0062] Where α(·) is an attention function that adaptively controls the contribution of neighboring node j to node i.

[0063] The graph attention mechanism layer employs multi-head attention. To learn the attention weights in different subspaces, the calculation formula is further defined as follows:

[0064]

[0065] Where || represents the concatenation operation, αij (k) W is the weight coefficient calculated by the k-th attention mechanism. k These are the learning parameters corresponding to the k-th attention mechanism. The calculation process for the multi-head attention mechanism is as follows: Figure 2 As shown, different colors represent the calculation process of different attention heads (i.e., the configuration of two wavy lines with a straight line in the middle. Due to the requirements of the attached figure, the color is not displayed, but this configuration represents three attention heads, i.e., K=3 in the following figure). K=3 in the figure. After the calculation is completed, the above results are spliced ​​together.

[0066] The graph attention mechanism layer has multiple sets.

[0067] The calculation formula for the LSTM layer includes

[0068] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0069] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0070]

[0071]

[0072] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0073] h t =o t *tanh(C t )

[0074] Reference Figure 3 The system includes several basic units, which include an input gate, a forget gate, and an output gate.

[0075] The information from the input gate, after being processed by the activation functions sigmoid and tanh, jointly determines the vector retained in the state memory unit.

[0076] The input of the forget gate and the intermediate output ht-1 of the state unit St-1 together determine the forgetting part of the state memory unit.

[0077] The output gate controls the information flowing into the current hidden state ht from the memory cell at the current moment.

[0078] Example 2

[0079] A short-term power load forecasting system based on GAT-LSTM includes a processing unit that executes the short-term power load forecasting method based on GAT-LSTM described in Embodiment 1.

[0080] Example 3

[0081] A storage medium storing a computer program, which, when executed by a processor, implements the short-term power load forecasting method based on GAT-LSTM as described in Embodiment 1.

[0082] Example 4

[0083] A computer includes at least a memory and a processor, wherein the memory stores a computer program, and the processor, when executing the computer program in the memory, implements the short-term power load forecasting method based on GAT-LSTM as described in Embodiment 1.

[0084] In summary, this application proposes a short-term power load forecasting method based on GAT-LSTM. The input information is historical power load data. First, the dataset is normalized. Then, a graph attention network (GAT) is used to capture the spatial dependencies of the input data. The output results are then input into a long short-term memory network (LSTM). The LSTM extracts the temporal dependencies of the data, thereby obtaining high-precision load forecasting results.

[0085] Attention mechanisms are now widely used in sequence-based tasks, with the advantage of amplifying the influence of the most important parts of the data. This property has proven useful for many tasks, and the number of models incorporating attention mechanisms is continuously increasing. Graph neural networks also benefit from this, using attention during the aggregation process to integrate the outputs of multiple models and generate random walks oriented towards important targets. In this invention, a Graph Attention Network (GAT) is used to capture the potential correlation between the target residence and the electricity load of related residences, thereby obtaining more accurate load forecasting results. A Graph Attention Network (GAT) is a spatial graph convolutional network whose attention mechanism is used to determine the weights of node neighborhoods when aggregating feature information.

[0086] This invention utilizes an LSTM network to obtain the time dependency of power load information, thereby acquiring load forecasting results. LSTM is an improved model of RNN (Recurrent Neural Network), and the LSTM network unit structure is as follows: Figure 2As shown, during training, the original RNN is prone to gradient explosion or vanishing problems as training time increases and the number of network layers increases, thus failing to capture information from earlier, long-distance data. LSTM units can learn long-term dependencies, allowing past information to re-enter later, thereby solving the gradient vanishing problem.

[0087] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A short-term power load forecasting method based on GAT-LSTM, characterized in that, include Download the historical power load dataset and perform adaptive normalization on the historical power load data; The normalized data is input into the graph attention mechanism layer, and the potential correlation between the target residential and related residential loads is captured through the GAT layer. The data processed by the GAT layer is input into the LSTM layer to further obtain the long-term dependency of the load data and output the load forecast results. The calculation formula for the graph attention mechanism layer is as follows: ； in, ( ) is an attention function; The calculation formula for the LSTM layer includes: ； The LSTM layer includes several basic units, including an input gate, a forget gate, and an output gate. The information from the input gate, after being processed by the activation functions sigmoid and tanh, jointly determines the vector retained in the state memory unit. The input of the forget gate and the intermediate input of the state unit together determine the forgetting part of the state memory unit; The output gate controls the information flowing into the current hidden state from the memory cell at the current moment.

2. The short-term power load forecasting method based on GAT-LSTM according to claim 1, characterized in that, The historical power load dataset was obtained from OpenEI.

3. The short-term power load forecasting method based on GAT-LSTM according to claim 1, characterized in that, Adaptive normalization of historical power load data further includes: Given historical electricity load data X, the set S obtained after windowing X consists of n sequences with a window length of L: X = ; Adaptive normalization is applied to set S, and the normalized sequence is: R = .

4. The short-term power load forecasting method based on GAT-LSTM according to claim 1, characterized in that, The graph attention mechanism layer employs multi-head attention, and the calculation formula is further as follows: ； In this context, || represents the concatenation operation.

5. The short-term power load forecasting method based on GAT-LSTM according to claim 4, characterized in that, The graph attention mechanism layer has multiple sets.

6. A short-term power load forecasting system based on GAT-LSTM, characterized in that, It includes a processing unit that executes the short-term power load forecasting method based on GAT-LSTM as described in any one of claims 1-5.

7. A storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the short-term power load forecasting method based on GAT-LSTM as described in any one of claims 1-5.

8. A computer, comprising at least a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program on the memory, it implements the short-term power load forecasting method based on GAT-LSTM as described in any one of claims 1-5.

Images