Adjustable load identification method and system, terminal and medium

Abstract

The invention discloses an adjustable load identification method and system, a terminal and a medium, and relates to the technical field of electric power data analysis, and the technical scheme is characterized in that adaptive decomposition is carried out on adjustable load data to be identified by adopting an empirical wavelet transform method to obtain a plurality of decomposition signals; forming a training set by the user historical load data and the user types, and training a circulating LSTM network according to the training set; and inputting the plurality of decomposed signals as multivariate time sequence data into a cyclic LSTM network, and outputting an adjustable load type identification result by the cyclic LSTM network. According to the invention, the EWT can adaptively decompose signals, and decompose a complex load curve into a plurality of signal components with independent characteristics, so that the originally complex and interlaced load signals are decomposed to obtain a clearer characteristic level; on the basis, the LSTM network further performs feature extraction and recognition by capturing the time dependence of the decomposed load signals.

Description

Technical Field

[0001] The present invention relates to the technical field of power data analysis, and more specifically, to an adjustable load identification method, system, terminal and medium. Background Art

[0002] Adjustable loads can adjust their own power loads according to the supply and demand situation in the electricity market. Therefore, adjustable load identification is of great significance for improving the operating efficiency of the power system, promoting the innovative development of the power market, supporting the connection of new energy to the grid, improving the intelligence level of demand response resource management, and improving the reliability of the power supply system.

[0003] Existing adjustable load identification methods include load identification methods based on traditional wavelet transform, load identification methods based on standard LSTM neural network, and load identification methods based on effect price difference. (1) The traditional wavelet transform method decomposes the signal into components of different frequency bands. By selecting appropriate wavelet functions, the load data is decomposed into signals at multiple scales, and then the load signal is identified by analyzing the information at each scale. However, its frequency band division is fixed and cannot be adjusted adaptively. For complex and variable adjustable load data, it is difficult to accurately decompose and maintain the characteristics of the signal, especially when processing non-stationary signals. In addition, the traditional wavelet transform method has limited noise resistance and is easily affected by signal noise. (2) LSTM (Long Short-Term Memory Network) is a recurrent neural network (RNN) commonly used in time series analysis. The standard LSTM network maintains long-term dependency information through memory cells (cell states) and determines how to update and output hidden states through input gates, forget gates, and output gates. The load data is input into the LSTM network to identify the load type by capturing the time dependency of the data. However, when processing long time series, standard LSTM is prone to gradient vanishing or exploding problems, resulting in unstable model training. In addition, its structure is relatively fixed and it is difficult to flexibly adapt to different types of adjustable loads, and its performance is limited when processing complex signals. (3) The method based on effect price difference identifies the type of adjustable load by monitoring the relationship between electricity price and load. In this method, the data of power load is mainly determined by economic analysis and model fitting of the user's electricity consumption behavior to determine the adjustability of the load. This method is highly dependent on electricity price fluctuations and is difficult to accurately identify load elasticity when electricity price fluctuations are insufficient. In addition, due to the complex and changeable user behavior, the electricity price-based model is easily affected by data noise and external interference, making it difficult to finely identify the load. At the same time, these methods perform poorly when dealing with load data with strong nonlinearity and time-varying properties, making it difficult to obtain high-precision load identification results in practical applications.

[0004] Therefore, how to research and design an adjustable load identification method, system, terminal and medium that can overcome the above-mentioned defects is a problem that we urgently need to solve. Summary of the invention

[0005] In order to solve the deficiencies in the prior art, the purpose of the present invention is to provide an adjustable load identification method, system, terminal and medium, which combine signal decomposition and signal recognition, and use the adjustable load data as input signal for adaptive decomposition and supervised learning, and have extremely high adjustable load identification accuracy.

[0006] The above technical objectives of the present invention are achieved through the following technical solutions:

[0007] In a first aspect, a method for identifying an adjustable load is provided, comprising the following steps:

[0008] The empirical wavelet transform method is used to adaptively decompose the adjustable load data to be identified, and multiple decomposition signals are obtained;

[0009] The user historical load data and user types form a training set, and the recurrent LSTM network is trained based on the training set;

[0010] Multiple decomposed signals are input into the recurrent LSTM network as multivariate time series data, and the recurrent LSTM network outputs adjustable load type identification results.

[0011] Furthermore, the process of decomposing the adjustable load data to be identified by using the empirical wavelet transform is specifically as follows:

[0012] Perform Fourier transform on the adjustable load data to be identified to obtain the corresponding frequency spectrum;

[0013] Normalize the frequencies in the spectrum and divide the normalized spectrum into multiple frequency ranges according to the Shannon criterion;

[0014] The adjustable load data to be identified which is divided into multiple frequency ranges is decomposed by empirical scaling functions and empirical wavelet functions to obtain multiple decomposed signals.

[0015] Furthermore, the user types include adjustable industrial users, adjustable commercial users, adjustable multi-type users, new energy users, energy storage users and bidirectional users.

[0016] Furthermore, the recurrent LSTM network uses a variable learning rate for gradient clipping.

[0017] Furthermore, the recurrent LSTM network consists of an input layer, a recurrent LSTM layer, a fully connected layer, a judgment layer and an output layer.

[0018] Furthermore, the input layer is used to receive an input signal and input the input signal to the recurrent LSTM layer;

[0019] The recurrent LSTM layer is used to learn the features of the input signal and input the learned features into the fully connected layer;

[0020] The fully connected layer is used to digitize the learned features and input the digitized features into the judgment layer;

[0021] The judgment layer is used to judge the adjustable load type of the input signal according to the digitized characteristics, and transmit the adjustable load type to the output layer;

[0022] The output layer is used to output the adjustable load type identification result.

[0023] In a second aspect, an adjustable load identification system is provided, the system is used to implement an adjustable load identification method as described in any one of the first aspects, including:

[0024] A data decomposition module, used for adaptively decomposing the adjustable load data to be identified by using an empirical wavelet transform method to obtain multiple decomposition signals;

[0025] The network training module is used to form a training set with user historical load data and user types, and train the recurrent LSTM network based on the training set;

[0026] The load identification module is used to input multiple decomposed signals as multivariate time series data into a recurrent LSTM network, and the recurrent LSTM network outputs an adjustable load type identification result.

[0027] Furthermore, the data decomposition module includes:

[0028] A Fourier transform unit, used for performing Fourier transform on the adjustable load data to be identified to obtain a corresponding frequency spectrum;

[0029] A normalization processing unit, used for normalizing the frequencies in the spectrum and dividing the normalized spectrum into multiple frequency ranges according to the Shannon criterion;

[0030] The empirical wavelet transform unit is used to decompose the adjustable load data to be identified which is divided into multiple frequency ranges through the empirical scaling function and the empirical wavelet function to obtain multiple decomposed signals.

[0031] In a third aspect, a computer terminal is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, an adjustable load identification method as described in any one of the first aspects is implemented.

[0032] In a fourth aspect, a computer-readable medium is provided, on which a computer program is stored, and the computer program is executed by a processor to implement an adjustable load identification method as described in any one of the first aspects.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] 1. The adjustable load identification method provided by the present invention combines the empirical wavelet transform (EWT) with the recurrent LSTM neural network, and shows remarkable effects in processing non-stationary signals and complex adjustable load data; EWT can adaptively decompose the signal and decompose the complex load curve into multiple signal components with independent characteristics, which makes the originally complex and intertwined load signal obtain a clearer feature hierarchy through decomposition; on this basis, the LSTM network further extracts and identifies features by capturing the time dependency of these decomposed load signals;

[0035] 2. The synergistic effect of adaptive decomposition and time feature capture in the present invention: The adaptive decomposition capability of EWT can effectively extract the characteristics of load signals at different time scales, while LSTM, by taking advantage of its advantages in time series data, can extract richer time-dependent information from these multi-scale signals and achieve more accurate load identification;

[0036] 3. Improved robustness and accuracy in the present invention: The noise resistance and adaptability of EWT combined with the long short-term memory mechanism of LSTM make the whole system more robust when facing non-stationary signals and high-noise data; while the gating mechanism of LSTM can avoid gradient disappearance and explosion while retaining the contextual information of the data. This combination significantly improves the accuracy of load identification compared with a single method;

[0037] 4. The present invention can flexibly handle complex loads: the load signal decomposed by EWT can effectively deal with the superposition effect of complex load signals, and LSTM further analyzes and processes these decomposed signals in depth. This decomposition-processing process enables the model to flexibly deal with a variety of different types of load signals. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0039] Figure 1 is a flow chart of Embodiment 1 of the present invention;

[0040] Figure 2 A schematic diagram of the structure of a cyclic LSTM network in Example 1 of the present invention;

[0041] Figure 3 It is a logical diagram of the LSTM layer structure in Example 1 of the present invention;

[0042] Figure 4 This is a diagram showing the training and testing effects in Example 1 of the present invention;

[0043] Figure 5 is the original load curve diagram in Example 1 of the present invention;

[0044] Figure 6 is a component curve diagram in Example 1 of the present invention;

[0045] Figure 7 It is a system block diagram in Example 2 of the present invention. DETAILED DESCRIPTION

[0046] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0047] Embodiment 1: A method for identifying an adjustable load, such as Figure 1 As shown, the following steps are included:

[0048] S1: Adopting the empirical wavelet transform method to adaptively decompose the adjustable load data to be identified, and obtain multiple decomposition signals;

[0049] S2: The user historical load data and user types are combined into a training set, and the recurrent LSTM network is trained based on the training set;

[0050] S3: Multiple decomposed signals are input into the recurrent LSTM network as multivariate time series data, and the recurrent LSTM network outputs adjustable load type identification results.

[0051] The present invention takes into account that the actual adjustable load data curve is composed of multiple load signals superimposed, and each load signal has certain curve characteristics. Therefore, the empirical wavelet transform is first used to perform adaptive signal decomposition on the adjustable load data curve. The adjustable load data curve is decomposed into multiple independent load signal curves that are not superimposed with other load signal curves, and then each load signal curve is input into the recurrent LSTM neural network to complete the identification of each signal.

[0052] In step S1, the adjustable load data to be identified is defined as x(t), t∈[1,2,…,T], where T is the length of the data. Then perform Fourier transform on it to obtain its spectrum:

[0053]

[0054] The frequencies in the spectrum are then normalized:

[0055]

[0056] Among them, ω∈[0,2π] is the normalized frequency, and k is the index value in the spectrum. According to the Shannon criterion, only the spectrum within the range of [0,π] needs to be considered in the analysis. In this way, the entire spectrum can be decomposed into N segments, and the range of each segment is: Λ n =[ω n-1 ,ω n ], where ω 0 =0,ω T =π. n As the center, there is a transition phase, the length of the transition phase is 2τ n .

[0057] Define the empirical scaling function and empirical wavelet function:

[0058]

[0059] Among them, β(x) has: β(x) = x 4 (35-84x+70x 2 -20x 3 )

[0060] Set the proportionality factor γ = τ n / ω n , the empirical coefficients of wavelet transform can be obtained as:

[0061]

[0062] At the same time, its approximate coefficient is obtained:

[0063]

[0064] Among them, f is the signal, that is, the adjustable load data, so f = x(t). Then, the decomposed signal is:

[0065]

[0066] Among them, k=1,2,….

[0067] Through the above empirical wavelet decomposition method, the adjustable load data x(t) can be decomposed into K signals: 0 (t),f 1 (t),…,f K-1(t). The present invention improves the empirical wavelet transform, making it more suitable for decomposing adjustable load curves, and has adaptive decomposition capability, and can adaptively decompose load curves and various interferences without specifying the number of decomposed curves in advance.

[0068] In step S2, the user historical load data and user types constitute a training set to train the recurrent LSTM network. The user type X is initially divided into six types: 1. adjustable industrial users, 2. adjustable commercial users, 3. adjustable multi-type users, 4. new energy users, 5. energy storage users, and 6. bidirectional users. The user historical load data is decomposed, and the decomposition curve and user type are input into the recurrent LSTM network as a set of training samples.

[0069] The present invention adopts a recurrent LSTM neural network, which can capture the long-term dependencies of each decomposed load data, integrate the upstream and downstream information of the load data to be identified, and solve the gradient vanishing and gradient exploding problems in ordinary LSTM neural networks. It can be applied to various types of adjustable load identification and has extremely strong robustness and anti-interference capabilities.

[0070] like Figure 2 As shown in the figure, the recurrent LSTM network consists of an input layer, a recurrent LSTM layer, a fully connected layer, a judgment layer, and an output layer. The input layer is used to receive input signals and input the input signals to the recurrent LSTM layer; the recurrent LSTM layer is used to learn the features of the input signals and input the learned features to the fully connected layer; the fully connected layer is used to digitize the learned features and input the digitized features to the judgment layer; the judgment layer is used to determine the adjustable load type of the input signal based on the digitized features and transmit the adjustable load type to the output layer; the output layer is used to output the identification result of the adjustable load type.

[0071] In this embodiment, the recurrent LSTM network uses a variable learning rate for gradient clipping.

[0072] like Figure 3 As shown, the LSTM layer structure in the network can adopt the existing structure. For example, the formula of its input gate is: t =σ(W i [h t-1 ,x t ]+b i ); the formula of the forget gate is: f t =σ(W f [h t-1 ,x t ]+b f ); the formula of the output gate is: t =σ(W o [h t-1 ,xt ]+b o ). Where σ is the sigmoid activation function, W i,f,o is the training weight of the three gates, b i,f,o is the training offset of the three gates. Define the cell state: The update formula for new cells is: The update formula of the hidden state is: t =o t ⊙tanh(c t ).

[0073] In step S3, the adjustable load data to be identified is decomposed according to step S1, and the multiple components obtained after decomposition are input into the input layer of the network in step S2 as multivariate time series data, and the output result of the cyclic LSTM network is the final adjustable load type identification result.

[0074] The adjustable load curve data in the jurisdiction of a power supply station in a county in Sichuan was used as a training set, with a total of 240 groups; 60 groups were taken from the jurisdiction as test sets. When training the recurrent LSTM neural network used in the present invention, the empirical wavelet decomposition method was used to pre-process the training set, and the 1-dimensional training set was converted into multi-dimensional data with labels and input into the recurrent LSTM neural network to train the network. The training was carried out for 60 generations. The specific training and test results are shown in the figure. Figure 4 shown.

[0075] Take a certain adjustable load under the jurisdiction as the load to be identified, and its original load curve is as follows: Figure 5 shown.

[0076] Then the empirical wavelet transform method is used to decompose the data to obtain component curves such as Figure 6 As shown in the figure, three signals are decomposed and all three signals are input into the recurrent LSTM network to obtain the adjustable load identification result. The load type is adjustable commercial load, which is consistent with the actual situation.

[0077] In addition, 70 adjustable load users were randomly selected from other supply areas in the county, and their historical load data were input for load identification. On the basis of the same data, the load identification method based on traditional wavelet transform, the load identification method based on standard LSTM neural network, and the load identification method based on effect price difference were compared year-on-year. The comparison results are shown in Table 1:

[0078] Table 1 Comparison results

[0079]

[0080]

[0081] It can be seen from Table 1 that the accuracy and reliability of the above method described in the present invention are verified.

[0082] Embodiment 2: An adjustable load identification system, the system is used to implement an adjustable load identification method as described in Embodiment 1, such as Figure 7 As shown, it includes a data decomposition module, a network training module and a load identification module.

[0083] Among them, the data decomposition module is used to adaptively decompose the adjustable load data to be identified by using the empirical wavelet transform method to obtain multiple decomposition signals; the network training module is used to form a training set with user historical load data and user type, and train the recurrent LSTM network based on the training set; the load identification module is used to input multiple decomposition signals as multivariate time series data into the recurrent LSTM network, and the recurrent LSTM network outputs the adjustable load type identification result.

[0084] In this embodiment, the data decomposition module includes a Fourier transform unit, a normalization processing unit and an empirical wavelet transform unit.

[0085] Among them, the Fourier transform unit is used to perform Fourier transform on the adjustable load data to be identified to obtain the corresponding spectrum; the normalization processing unit is used to normalize the frequency in the spectrum and divide the normalized spectrum into multiple frequency ranges according to the Shannon criterion; the empirical wavelet transform unit is used to decompose the adjustable load data to be identified divided into multiple frequency ranges through empirical scaling functions and empirical wavelet functions to obtain multiple decomposed signals.

[0086] The present invention also describes a computer terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, an adjustable load identification method as described in Example 1 is implemented.

[0087] The present invention also describes a computer-readable medium on which a computer program is stored. The computer program is executed by a processor to implement an adjustable load identification method as described in Example 1.

[0088] Working principle: The present invention combines the empirical wavelet transform (EWT) with the recurrent LSTM neural network, which shows remarkable effects in processing non-stationary signals and complex adjustable load data. EWT can adaptively decompose the signal and decompose the complex load curve into multiple signal components with independent characteristics, which makes the originally complex and intertwined load signal obtain a clearer feature hierarchy through decomposition. On this basis, the LSTM network further performs feature extraction and recognition by capturing the time dependence of these decomposed load signals.

[0089] The synergistic effect of adaptive decomposition and time feature capture in the present invention: the adaptive decomposition capability of EWT can effectively extract the characteristics of load signals at different time scales, while LSTM, by taking advantage of its advantages in time series data, can extract richer time-dependent information from these multi-scale signals and achieve more accurate load identification.

[0090] The robustness and accuracy of the present invention are improved: the noise resistance and adaptability of EWT are combined with the long short-term memory mechanism of LSTM, making the whole system more robust when facing non-stationary signals and high-noise data; while the gating mechanism of LSTM can retain the contextual information of the data while avoiding gradient disappearance and explosion. This combination significantly improves the accuracy of load identification compared with a single method.

[0091] The present invention flexibly handles complex loads: the load signal decomposed by EWT can effectively deal with the superposition effect of complex load signals, and LSTM further performs in-depth analysis and processing on these decomposed signals. This decomposition-processing process enables the model to flexibly deal with a variety of different types of load signals.

[0092] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0093] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0094] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0095] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0096] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for identifying an adjustable load, characterized in that: The following steps are involved: The empirical wavelet transform method is used to adaptively decompose the adjustable load data to be identified, and multiple decomposition signals are obtained; The user historical load data and user types form a training set, and the recurrent LSTM network is trained based on the training set; Multiple decomposed signals are input into the recurrent LSTM network as multivariate time series data, and the recurrent LSTM network outputs adjustable load type identification results.

2. The adjustable load identification method according to claim 1, characterized in that: The process of decomposing the adjustable load data to be identified by using the empirical wavelet transform is specifically as follows: Perform Fourier transform on the adjustable load data to be identified to obtain the corresponding frequency spectrum; Normalize the frequencies in the spectrum and divide the normalized spectrum into multiple frequency ranges according to the Shannon criterion; The adjustable load data to be identified which is divided into multiple frequency ranges is decomposed by empirical scaling functions and empirical wavelet functions to obtain multiple decomposed signals.

3. The adjustable load identification method according to claim 1, characterized in that: The user types include adjustable industrial users, adjustable commercial users, adjustable multi-type users, new energy users, energy storage users and bidirectional users.

4. The adjustable load identification method according to claim 1, characterized in that: The recurrent LSTM network uses a variable learning rate for gradient clipping.

5. The adjustable load identification method according to claim 1, characterized in that: The recurrent LSTM network consists of an input layer, a recurrent LSTM layer, a fully connected layer, a judgment layer and an output layer.

6. The adjustable load identification method according to claim 5, characterized in that: The input layer is used to receive an input signal and input the input signal to the recurrent LSTM layer; The recurrent LSTM layer is used to learn the features of the input signal and input the learned features into the fully connected layer; The fully connected layer is used to digitize the learned features and input the digitized features into the judgment layer; The judgment layer is used to judge the adjustable load type of the input signal according to the digitized characteristics, and transmit the adjustable load type to the output layer; The output layer is used to output the adjustable load type identification result.

7. An adjustable load identification system, characterized in that: The system is used to implement an adjustable load identification method as described in any one of claims 1 to 6, comprising: A data decomposition module, used for adaptively decomposing the adjustable load data to be identified by using an empirical wavelet transform method to obtain multiple decomposition signals; The network training module is used to form a training set with user historical load data and user types, and train the recurrent LSTM network based on the training set; The load identification module is used to input multiple decomposed signals as multivariate time series data into a recurrent LSTM network, and the recurrent LSTM network outputs an adjustable load type identification result.

8. The adjustable load identification system according to claim 7, characterized in that: The data decomposition module comprises: A Fourier transform unit, used for performing Fourier transform on the adjustable load data to be identified to obtain a corresponding frequency spectrum; A normalization processing unit, used for normalizing the frequencies in the spectrum and dividing the normalized spectrum into multiple frequency ranges according to the Shannon criterion; The empirical wavelet transform unit is used to decompose the adjustable load data to be identified which is divided into multiple frequency ranges through the empirical scaling function and the empirical wavelet function to obtain multiple decomposed signals.

9. A computer terminal comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, an adjustable load identification method as described in any one of claims 1-7 is implemented.

10. A computer readable medium having a computer program stored thereon, characterized in that: The computer program is executed by a processor to implement an adjustable load identification method as described in any one of claims 1 to 7.

Images