Medium and long term load prediction method and device in power system, medium and equipment

Abstract

The invention discloses a long-term load prediction method and device in a power system, a medium and equipment, and relates to the field of power, and the method comprises the steps: obtaining long-term load data in the power system, and determining an original time sequence; the method comprises the following steps of: decomposing an original time sequence into hour-level time sequence data, day-level time sequence data and month-level time sequence data by adopting a downsampling method, decomposing data sequences of different time scales based on a season and trend decomposition STL (Standard Template Library) algorithm, and determining a season component, a trend component and a residual component; determining a prediction result of a trend component through an LSTM model, determining a prediction result of a residual component based on a multilayer perceptron (MLP), and determining a prediction result of a season component by adopting Fourier series; the prediction results of the season component, the trend component and the residual component are fused to determine the prediction results of different time scales, and the prediction results of the different time scales are fused to accurately obtain the long-term load prediction result in the power system.

Description

Technical Field

[0001] The present invention relates to the field of electric power technology, and in particular to a method, device, medium and equipment for predicting medium- to long-term load in an electric power system. Background Art

[0002] As one of the world's largest energy consumers, my country is vigorously promoting the green transformation of its energy structure, which urgently requires the construction and planning of the power system to actively cooperate and respond. Load forecasting is the basis for the new power system to formulate power generation plans, carry out demand response, and conduct power market transactions, and plays a very important role in the power system.

[0003] At present, there are two commonly used load forecasting methods: traditional forecasting methods based on time series analysis and machine learning forecasting methods based on artificial intelligence. Traditional forecasting methods based on time series analysis: such as grayscale forecasting method, entropy method, exponential smoothing method, cumulative autoregressive moving average method, etc., but they perform poorly when dealing with non-adaptive rules; machine learning forecasting methods based on artificial intelligence: including BP neural network, support vector machine, Transformer model, long short-term memory network, etc.; conventional means, in order to consider the impact of multi-dimensional time domain characteristics on medium and long-term loads, obtain the multi-dimensional components of the load change data of the power system, perform sequence decomposition, and extract component feature information, and then design different methods according to the differences between different components for separate forecasting.

[0004] However, the medium- and long-term load changes in the power system are affected by many factors. For example, the system load is affected by many factors such as weather, holidays, population mobility, and renewable energy power generation. The changes are very complex. First, time series of different granularities reflect different information. For example, hourly data reflects load changes within a day, and daily data reflects fluctuations such as holidays and weekends; monthly data can reflect seasonal changes. This information is related and complementary. The existing technology only considers component feature information, and fails to comprehensively consider the complementarity and dependency between related factors in multidimensional component feature information at different time scales. When predicting future medium- and long-term load changes in the power system, the accuracy of medium- and long-term load prediction in the power system is poor. Summary of the invention

[0005] The present invention provides a method, device, medium and equipment for predicting medium- and medium-term load in a power system, which is used to solve the above-mentioned problem existing in the prior art, that is, how to improve the accuracy of medium- and medium-term load prediction in the power system in the prior art. The present invention provides a method for predicting medium- and medium-term load in a power system, which comprises:

[0006] Acquire long-term load data of the power system and determine the original time series of the load data; the original time series contains load change information of different granularity time series;

[0007] Decompose the original time series into data series of different time scales, including hourly time series data, daily time series data, and monthly time series data;

[0008] Based on the seasonal and trend decomposition STL algorithm, the data series of different time scales are decomposed, and the seasonal components and trend components in the data series of different time scales are determined by the inner loop structure. The remaining sequence after the trend component and seasonal component in the data series of different time scales are subtracted is taken as the residual component.

[0009] The trend components of different time scales are predicted respectively through the LSTM-based trend component prediction model to obtain the change prediction results of future trend components of different time scales;

[0010] The residual components of different time scales are predicted respectively by multi-layer perceptron MLP to obtain the change prediction results of future residual components of different time scales;

[0011] Fourier series are used to fit seasonal components of different time scales to determine the prediction results of future seasonal component changes;

[0012] By fusing the change prediction results of seasonal components, trend components and residual components at different time scales, the change prediction results at different time scales are determined, and the change prediction results at different time scales are fused to obtain the prediction results of future medium- and long-term load changes in the power system.

[0013] Optionally, the seasonal and trend decomposition STL algorithm is used to decompose data sequences of different time scales, and an inner loop structure is used to determine seasonal components and trend components in data sequences of different time scales, specifically including:

[0014] By smoothing the data points based on local weighted regression, the long-term trend of the data is estimated and the trend component is determined;

[0015] The seasonal components were determined by smoothing the position of each seasonal cycle separately and using locally weighted regression to estimate the mean of each seasonal position.

[0016] Optionally, the method of fitting seasonal components of different time scales respectively using Fourier series to determine the prediction results of future seasonal component changes specifically includes:

[0017] The seasonal time series component is obtained using the following formula:

[0018]

[0019] Among them, S p(t) is the seasonal time series component, k is the number of Fourier terms, p is the length of the seasonal period, and a j is the cosine component coefficient, which indicates the amplitude of different cosine components in the seasonal function, b j are the sinusoidal component coefficients, which represent the amplitudes of the different sinusoidal components in the seasonal function.

[0020] Optionally, the method of predicting the residual components of different time scales respectively by using a multi-layer perceptron MLP to obtain the change prediction results of the future residual components of different time scales specifically includes:

[0021] When the window time length of the input original time series is L, the input residual component feature is The feature representation after MLP learning is as follows:

[0022]

[0023] After learning the features of the residual component by stacking multiple layers of networks, the number of neurons in the last output layer is set to T to determine the predicted results of the future residual component changes in the next T time steps.

[0024] Optionally, the prediction results of changes at different time scales are integrated to obtain the prediction results of future medium- and long-term load changes in the power system, specifically including:

[0025] The prediction results of changes at different time scales are integrated. The daily and monthly data are mapped into hourly granularity sequences by using a linear layer, and the weighted summation is performed to obtain the prediction results of future long-term load changes in the power system.

[0026] The present invention provides a device for predicting medium and long-term loads in a power system, comprising:

[0027] An acquisition module is used to acquire long-term load data of the power system and determine the original time series of the load data; the original time series contains load change information of different granularity time series;

[0028] The original time series decomposition module is used to decompose the original time series into data series of different time scales, including hourly time series data, daily time series data, and monthly time series data;

[0029] The data series decomposition module of different time scales is used to decompose the data series of different time scales based on the seasonal and trend decomposition STL algorithm, and uses the inner loop structure to determine the seasonal component and trend component in the data series of different time scales, and takes the remaining sequence after subtracting the trend component and seasonal component in the data series of different time scales as the residual component;

[0030] The trend component prediction module is used to predict the trend components of different time scales respectively through the LSTM-based trend component prediction model to obtain the change prediction results of future trend components of different time scales;

[0031] The residual component prediction module is used to predict the residual components of different time scales respectively through the multi-layer perceptron MLP to obtain the change prediction results of the future residual components of different time scales;

[0032] Seasonal component prediction module, used to fit seasonal components of different time scales using Fourier series to determine the prediction results of future seasonal component changes;

[0033] The fusion and prediction module is used to determine the prediction results of changes in different time scales by fusing the change prediction results of seasonal components, trend components and residual components at different time scales, and to obtain the prediction results of medium- and long-term load changes in the future power system.

[0034] The present invention provides a computer-readable storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned method for medium- and long-term load forecasting in a power system is implemented.

[0035] The present invention provides a computer device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned method for predicting medium- and long-term load in a power system when executing the program.

[0036] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention provides a method for medium- and long-term load forecasting for an electric power system. The method takes into account certain trend characteristics and periodic fluctuation characteristics of medium- and long-term load changes, divides the original time series into hourly, daily and monthly levels by adopting a downsampling method, and then uses an STL algorithm to decompose the original medium- and long-term load of the electric power system to obtain seasonal components, trend components and residual components; by adopting an LSTM model, a multi-layer perceptron MLP and a Fourier series, components of different time granularities are predicted respectively, so that the prediction results have higher prediction accuracy, and by comprehensively considering the load change information at different time scales and the multi-dimensional time domain feature dependency, the medium- and long-term load forecast results are obtained by weighted fusion based on the prediction results of different time granularities, which effectively improves the medium- and long-term load forecasting accuracy of the electric power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0038] Figure 1 A flow chart of a method for predicting medium and long-term load in a power system provided by an embodiment of the present invention;

[0039] Figure 2 LSTM model diagram provided for an embodiment of the present invention;

[0040] Figure 3 A schematic diagram of a computer device for a method for medium- and long-term load forecasting in an electric power system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] The technical solution of the present invention and how the technical solution of the present invention solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present invention will be described below in conjunction with the accompanying drawings.

[0043] Figure 1 is a flow chart of a method for predicting medium and long-term load in a power system provided by an embodiment of the present invention. Figure 1 As shown, this embodiment shows a method for predicting medium- and long-term load in a power system, including:

[0044] S1: Acquire medium- and long-term load data of the power system and determine the original time series of the load data; the original time series contains load change information of different granularity time series.

[0045] S2: Decompose the original time series into data series of different time scales, including hourly time series data, daily time series data, and monthly time series data.

[0046] For example, based on the original time series, by adopting the downsampling method, specifically, drawing on the pooling operation in the convolutional neural network, for example, the maximum pooling or average pooling operation can be selected, and the size of the pooling template depends on the length of the period to be decomposed. After period decomposition, the hourly time series data T h , daily time series data T d and monthly time series data T m .

[0047] S3: Decompose the data series of different time scales based on the seasonal and trend decomposition STL algorithm, use the inner loop structure to determine the seasonal component and trend component in the data series of different time scales, and take the remaining sequence after subtracting the trend component and seasonal component in the data series of different time scales as the residual component.

[0048] Exemplarily, the STL algorithm is used to decompose the medium- and long-term load of the original power system load into a time series to obtain the corresponding seasonal component, trend component and residual component. The specific calculation formula is as follows:

[0049] L t =T t +S t +R t

[0050] Among them, L t is the medium and long term load of the original power system, S t is the seasonal component of the load at time step t, T t is the load trend component at time step t, R t is the load residual component at time step t.

[0051] S4: The trend components of different time scales are predicted separately through the LSTM-based trend component prediction model to obtain the change prediction results of future trend components of different time scales.

[0052] For example, the trend change of medium- and long-term load is slowly rising as a whole. To capture the current trend change, it is necessary to combine the historical overall power load trend change information for learning. Therefore, the long short-term memory network LSTM model with memory function is selected to predict the trend component, such as Figure 2 As shown in the figure, LSTM introduces three gates, namely input gate, forget gate, output gate and a cell state. These mechanisms enable LSTM to better handle long-term dependencies in sequences.

[0053] For example, the forget gate in LSTM indicates what information is discarded from the previous cell state, and its scope is the cell state C of the previous layer. t-1 On, represents how much information in the past is forgotten, and the calculation formula is as follows:

[0054] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0055] Among them, h t-1 is the hidden variable output by the previous cell, x t is the current input, ft Represents the forget gate value, W f is the coefficient matrix of the forget gate, b f represents the bias of the forget gate, and σ represents the sigmoid activation function.

[0056] For example, the input gate needs to determine what new information is stored in the cell state, and calculate the current cell state based on the hidden variable output by the previous cell and the current input. However, not all input information can be remembered, and some input information needs to be filtered out. Therefore, referring to the design of the forget gate, the input gate calculates the vector i t To control the information that needs to be remembered, and then combine the information left after the forget gate in the previous time step with i t The result of acting on the current cell state is used to update the cell state; the specific calculation formula of the input gate is as follows:

[0057] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0058] Next, the specific calculation formula for the current cell state is as follows:

[0059]

[0060] Finally, the cell state is updated by combining the information from the forget gate and the input gate:

[0061]

[0062] Among them, C t is the new cell state, representing the output of the current time step under the ideal state, W i is the weight matrix of the input gate, b i is the bias term of the input gate, h t-1 is the hidden variable output by the previous cell, x t is the current input, is the current cell state, W C is the weight matrix of the current cell state, b C is the bias term of the current cell state, i t is the output of the input gate at time t, σ represents the sigmoid activation function, and f t represents the forget gate value, C t-1 is the previous cell state.

[0063] However, in reality not all learned information can be expressed, so the final output in the output gate also needs to filter the current cell state; first, use the sigmoid function to determine which part of the cell state will be output, and the calculation process is as follows:

[0064] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0065] Next, the cell state is processed through the tangent function and multiplied with the output of the sigmoid gate to determine the final output. The calculation process is as follows:

[0066] h t =o t ×tanh(C t )

[0067] Among them, t is the output of the output gate at time t, W o is the weight matrix of the output gate, b o is the bias term of the output gate.

[0068] The LSTM network is composed of multiple LSTMs in series. The load state feature information at different times is stored in the hidden state h t Transmit and couple the load information of different time nodes.

[0069] Finally, the LSTM network weights and biases are obtained through error back propagation and adaptive gradient descent algorithm Adam training to obtain the trend component prediction model.

[0070] S5: The residual components of different time scales are predicted separately through the multi-layer perceptron MLP to obtain the change prediction results of the future residual components of different time scales.

[0071] For example, the residual sequence is the residual sequence after subtracting the trend component and seasonal component from the original time series. Compared with the previous two, the residual component is nonlinear, volatile, and irregular. Therefore, a multi-layer perceptron (MLP) is used to predict the residual component. The multi-layer perceptron consists of multiple perceptron layers (called neurons). Each neuron is a linear transformation followed by a nonlinear activation function, which can learn complex nonlinear relationships. If the window time length of the input historical sequence is L, the input residual component feature is The feature representation after MLP learning is as follows:

[0072]

[0073] After learning the features of the residual components by stacking multiple layers of networks, if we want to predict the residual changes in the next T time steps, the number of neurons in the last output layer is set to T.

[0074] S6: Use Fourier series to fit seasonal components of different time scales and determine the prediction results of future seasonal component changes.

[0075] Optionally, the trend component is determined by estimating the long-term trend of the data by smoothing the data points based on local weighted regression; the seasonal component is determined by estimating the average value of each seasonal position by smoothing the position of each seasonal cycle separately using local weighted regression.

[0076] Since electricity consumption surges every summer, the seasonal component of the time series often takes the shape of a sine or cosine function. Therefore, in this embodiment, by using Fourier series fitting, it can generate a smooth function, which is conducive to explaining and stably fitting the data.

[0077] Optionally, the seasonal component is fitted using Fourier series, and the specific calculation formula is as follows:

[0078]

[0079] where k is the number of Fourier terms, p is the length of the seasonal period, and a j is the cosine component coefficient, which indicates the amplitude of different cosine components in the seasonal function, b j are the sinusoidal component coefficients, representing the amplitudes of the different sinusoidal components in the seasonal function, and t is the time variable.

[0080] Exemplarily, when fitting the seasonal component, for hourly time series data, the seasonal cycle length p is 1, and for daily time series data, the seasonal cycle length p is 24; the number of Fourier terms k is adjusted according to experimental results.

[0081] S7: By fusing the change prediction results of seasonal components, trend components and residual components at different time scales, the change prediction results at different time scales are determined, and the change prediction results at different time scales are fused to obtain the prediction results of the medium- and long-term load changes in the future power system.

[0082] For example, the prediction results of trend component, seasonal component and residual component based on hourly, daily and monthly time series data can be expressed as T h , S h , R h , T d , S d , Rd , T m , S m , R m First, the sequence feature components at different time granularities are superimposed and fused to obtain the prediction results at each time period. Taking the daily time series data as an example, the calculated daily prediction results are as follows:

[0083] L d =T d +S d +R d

[0084] For example, since the dimensions of the time series predicted at different time scales are inconsistent, a linear layer can be used to map the daily and monthly data into sequences with hourly granularity, and then perform weighted summation on them to obtain the final prediction result.

[0085] The above is a method for predicting medium- and long-term loads in a power system provided by one or more embodiments of this specification. Based on the same idea, this specification also provides a corresponding device for predicting medium- and long-term loads in a power system, including:

[0086] An acquisition module is used to acquire long-term load data of the power system and determine the original time series of the load data; the original time series contains load change information of different granularity time series;

[0087] The original time series decomposition module is used to decompose the original time series into data series of different time scales, including hourly time series data, daily time series data, and monthly time series data;

[0088] The data series decomposition module of different time scales is used to decompose the data series of different time scales based on the seasonal and trend decomposition STL algorithm, and uses the inner loop structure to determine the seasonal component and trend component in the data series of different time scales, and takes the remaining sequence after subtracting the trend component and seasonal component in the data series of different time scales as the residual component;

[0089] The trend component prediction module is used to predict the trend components of different time scales respectively through the LSTM-based trend component prediction model to obtain the change prediction results of future trend components of different time scales;

[0090] The residual component prediction module is used to predict the residual components of different time scales respectively through the multi-layer perceptron MLP to obtain the change prediction results of the future residual components of different time scales;

[0091] Seasonal component prediction module, used to fit seasonal components of different time scales using Fourier series to determine the prediction results of future seasonal component changes;

[0092] The fusion and prediction module is used to determine the prediction results of changes in different time scales by fusing the change prediction results of seasonal components, trend components and residual components at different time scales, and to obtain the prediction results of medium- and long-term load changes in the future power system.

[0093] For the specific definition of the medium- and long-term load forecasting device for the power system, please refer to the definition of the medium- and long-term load forecasting method for the power system mentioned above, which will not be repeated here. Each module in the above-mentioned medium- and long-term load forecasting device for the power system can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0094] The present invention also provides a computer-readable storage medium, which stores a computer program. The computer program can be used to execute the above-mentioned method for predicting medium- and long-term load in a power system.

[0095] The present invention also provides Figure 3 The structural diagram of the computer device shown in FIG. Figure 3 As shown, at the hardware level, the computer device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and may also include hardware required for other services. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the medium- and long-term load forecasting method for the power system provided in the above embodiment.

[0096] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of the present invention.

Claims

1. A method for forecasting medium- and long-term load in a power system, characterized in that: include: Acquire long-term load data of the power system and determine the original time series of the load data; the original time series contains load change information of different granularity time series; Decompose the original time series into data series of different time scales, including hourly time series data, daily time series data, and monthly time series data; Based on the seasonal and trend decomposition STL algorithm, the data series of different time scales are decomposed, and the seasonal components and trend components in the data series of different time scales are determined by the inner loop structure. The remaining sequence after the trend component and seasonal component in the data series of different time scales are subtracted is taken as the residual component. The trend components of different time scales are predicted respectively through the LSTM-based trend component prediction model to obtain the change prediction results of future trend components of different time scales; The residual components of different time scales are predicted respectively by multi-layer perceptron MLP to obtain the change prediction results of future residual components of different time scales; Fourier series are used to fit seasonal components of different time scales to determine the prediction results of future seasonal component changes; By fusing the change prediction results of seasonal components, trend components and residual components at different time scales, the change prediction results at different time scales are determined, and the change prediction results at different time scales are fused to obtain the prediction results of future medium- and long-term load changes in the power system.

2. The method for medium- and long-term load forecasting in a power system according to claim 1, characterized in that: The seasonal and trend decomposition STL algorithm decomposes data sequences of different time scales, and uses an inner loop structure to determine seasonal components and trend components in data sequences of different time scales, specifically including: By smoothing the data points based on local weighted regression, the long-term trend of the data is estimated and the trend component is determined; The seasonal components were determined by smoothing the position of each seasonal cycle separately and using locally weighted regression to estimate the mean of each seasonal position.

3. The method for medium- and long-term load forecasting in a power system according to claim 1, characterized in that: The Fourier series is used to fit seasonal components of different time scales to determine the prediction results of future seasonal component changes, specifically including: The seasonal time series component is obtained using the following formula: Among them, S p (t) is the seasonal time series component, k is the number of Fourier terms, p is the length of the seasonal period, and a j is the cosine component coefficient, which indicates the amplitude of different cosine components in the seasonal function, b j are the sinusoidal component coefficients, which represent the amplitudes of the different sinusoidal components in the seasonal function.

4. The method for medium- and long-term load forecasting in a power system according to claim 1, characterized in that: The residual components of different time scales are predicted respectively by the multi-layer perceptron MLP to obtain the change prediction results of the future residual components of different time scales, specifically including: When the window time length of the input original time series is L, the input residual component feature is The feature representation after MLP learning is as follows: After learning the features of the residual component by stacking multiple layers of networks, the number of neurons in the last output layer is set to T to determine the predicted results of the future residual component changes in the next T time steps.

5. The method for medium- and long-term load forecasting in a power system according to claim 1, characterized in that: The prediction results of changes in different time scales are integrated to obtain the prediction results of future long-term load changes in the power system, specifically including: The prediction results of changes at different time scales are integrated. The daily and monthly data are mapped into hourly granularity sequences by using a linear layer, and the weighted summation is performed to obtain the prediction results of future long-term load changes in the power system.

6. A medium- and long-term load forecasting device for a power system, characterized in that: include: An acquisition module is used to acquire long-term load data of the power system and determine the original time series of the load data; the original time series contains load change information of different granularity time series; The original time series decomposition module is used to decompose the original time series into data series of different time scales, including hourly time series data, daily time series data, and monthly time series data; The data series decomposition module of different time scales is used to decompose the data series of different time scales based on the seasonal and trend decomposition STL algorithm, and uses the inner loop structure to determine the seasonal component and trend component in the data series of different time scales, and takes the remaining sequence after subtracting the trend component and seasonal component in the data series of different time scales as the residual component; The trend component prediction module is used to predict the trend components of different time scales respectively through the LSTM-based trend component prediction model to obtain the change prediction results of future trend components of different time scales; The residual component prediction module is used to predict the residual components of different time scales respectively through the multi-layer perceptron MLP to obtain the change prediction results of the future residual components of different time scales; Seasonal component prediction module, used to fit seasonal components of different time scales using Fourier series to determine the prediction results of future seasonal component changes; The fusion and prediction module is used to determine the prediction results of changes in different time scales by fusing the change prediction results of seasonal components, trend components and residual components at different time scales, and to obtain the prediction results of medium- and long-term load changes in the future power system.

7. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method for medium- and long-term load forecasting in a power system as described in any one of claims 1 to 5 above is implemented.

8. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for medium- and long-term load forecasting in an electric power system as described in any one of claims 1 to 5 is implemented.

Images