Integrated energy system state estimation method based on LSTM automatic encoder

Abstract

The invention relates to the technical field of power systems, and particularly discloses an integrated energy system state estimation method based on an LSTM automatic encoder, and the method comprises the following steps: building an IES state estimation model, and describing an energy coupling relation in the IES through a constraint equation; state estimation of the IES is carried out based on a weighted least square method; constructing an LSTM automatic encoder, and training the LSTM automatic encoder by using historical operation data of the IES as a training data set to obtain a trained LSTM automatic encoder model; inputting to-be-estimated IES system time sequence data into the trained LSTM automatic encoder model, and obtaining a state estimation result of the integrated energy system through forward propagation of the model; determining a loss function, wherein the loss function comprises prediction loss and regularization loss; the method has the advantages that the long-term dependency relationship and the nonlinear characteristics in the time sequence data can be effectively learned and captured, so that the accuracy and the reliability of state estimation are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a method for estimating the state of an integrated energy system based on an LSTM autoencoder. Background Art

[0002] In recent years, with the profound changes in the global energy structure, the Integrated Energy System (IES), as a form of interconnection of multiple energy networks such as electricity, heat, cold and gas, has increasingly become a key way to achieve efficient energy utilization, promote energy transformation and build a new power system. The emergence of IES has not only greatly improved the efficiency of energy resource integration and utilization, but also provided a new solution to the problem of new energy consumption.

[0003] However, to ensure the safe, reliable and efficient operation of the integrated energy system, accurate state estimation is particularly important. Traditional state estimation methods mainly rely on the construction of the system's physical model and historical observation data, but due to the high complexity and nonlinear characteristics of IES, traditional methods are difficult to meet the actual application needs. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a method for state estimation of an integrated energy system based on LSTM autoencoder.

[0005] The object of the present invention is achieved by the following technical scheme: a method for estimating the state of an integrated energy system based on an LSTM autoencoder, the method comprising the following steps:

[0006] Step S1, establishing an IES state estimation model, and describing the energy coupling relationship in the IES through constraint equations;

[0007] The state estimation of IES is represented by a point state estimation model, which is expressed as:

[0008]

[0009] In the formula, y is the IES state quantity; x is the IES measurement quantity; is the IES network topology and parameters; f is the functional relationship between input data and state quantity;

[0010] Step S2: Based on the historical measurement data, the IES state estimation model is solved by weighted least square method to obtain the IES state estimation result;

[0011] Step S3, determining the state estimation model loss function; constructing an LSTM autoencoder, and using the historical operation data of the IES as a training data set to train the LSTM autoencoder through the loss function to obtain a trained LSTM autoencoder state estimation model; the loss function is used to measure the difference between the calculation result of the LSTM autoencoder and the state estimation result of the IES obtained in step S2;

[0012] Step S4: input the IES time series data to be estimated into the trained LSTM autoencoder state estimation model, and obtain the state estimation result of the integrated energy system through the forward propagation of the model.

[0013] Specifically, the constraint equation is:

[0014]

[0015] F{·} and D{·} represent the real part and the imaginary part respectively, P j and Q j are the active power and reactive power injected into the node respectively, V is the voltage, G is the grid node admittance matrix, A is the node-branch association matrix, M is the natural gas flow, ψ is the weight coefficient, and λ is the Lagrange multiplier.

[0016] Specifically, the weighted least squares method is:

[0017]

[0018] In the formula, z e With z g are the measurement data vectors of the power grid and the natural gas grid respectively; m e With m g are the measurement models of the power grid and the natural gas grid respectively; x e With x g are the state vectors of the power grid and natural gas grid respectively; R e With R g are the measurement covariance matrices of the power grid and the natural gas grid respectively; c(x e ,x g ) is the zero equality constraint of the power grid and gas grid in IES.

[0019] Specifically, the LSTM autoencoder includes an encoder, a fully connected prediction network and a decoder;

[0020] The encoder is composed of multiple layers of LSTM units, the input is the input time series data obtained by IES observation, and the output is a low-dimensional latent space feature vector. For a given input variable, the data of each time step is processed in turn by the LSTM unit and the corresponding latent space feature variable is generated;

[0021] The fully connected prediction network uses the latent space feature variables obtained by the encoder as input variables to obtain the latent space state estimation result of the IES variable;

[0022] The decoder is composed of multiple layers of LSTM units, and its input is the latent space state estimation result of the IES variable output by the fully connected prediction network, which is reconstructed back to the original data space as the final output result.

[0023] Specifically, the encoder is given an input variable X = {x1, x2, ... x T}, the encoder processes the data of each time step in turn through the LSTM unit and generates the corresponding latent space feature variable h={h1,h2,…,h T}, as shown below:

[0024]

[0025] Specifically, the fully connected prediction network obtains the latent space state estimation result of the IES variable through the following formula:

[0026]

[0027] Specifically, the decoder calculates the final output result by the following formula:

[0028]

[0029] Specifically, the loss function includes prediction loss and regularization loss, and the prediction loss can be expressed as:

[0030]

[0031] Where, L p is the predicted loss; y is the true value of the IES state quantity, Estimate values ​​for the LSTM autoencoder;

[0032] The regularization loss uses the L2 regularization term:

[0033]

[0034] Where, L r is the regularization loss; λ is the regularization coefficient; θ en ,θ c ,θ de They are the parameters of encoder neural network, fully connected neural network and decoder neural network respectively.

[0035] The present invention has the following advantages:

[0036] The present invention can show significant performance advantages. By making full use of the advantages of data-driven models, this method can efficiently and accurately estimate the state of IES complex models, providing strong support for ensuring the stable operation and effective management of IES. The present invention has demonstrated superior generalization capabilities in dealing with IES's changing operating scenarios and multi-type energy systems. Traditional state estimation methods usually rely on sophisticated system physical modeling and large amounts of historical data analysis, and have low analysis accuracy when faced with IES's complex nonlinear characteristics and interactions with multiple energy networks. In contrast, the LSTM autoencoder-based method uses deep learning technology to effectively learn and capture long-term dependencies and nonlinear characteristics in time series data, thereby improving the accuracy and reliability of state estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a schematic diagram of the estimation method flow of the present invention;

[0038] Figure 2 The training flow chart of IES state estimation model based on LSTM-AE;

[0039] Figure 3 It is a schematic diagram of IES topology;

[0040] Figure 4 This is a schematic diagram of the grid node voltage amplitude estimation results;

[0041] Figure 5 Schematic diagram of gas pressure estimation results at natural gas grid nodes. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention, that is, the embodiments described are only part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0043] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0044] It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including a..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0045] The present invention is further described below in conjunction with the accompanying drawings, but the protection scope of the present invention is not limited to the following description. Figures 1 to 5 As shown, a method for state estimation of an integrated energy system based on an LSTM autoencoder is characterized in that the method comprises the following steps:

[0046] Step S1, establishing an IES state estimation model, and describing the energy coupling relationship in the IES through constraint equations;

[0047] The constraint equation is:

[0048]

[0049] F{·} and D{·} represent the real part and the imaginary part respectively, P j and Q j are the active power and reactive power injected into the node respectively, V is the voltage, G is the grid node admittance matrix, A is the node-branch association matrix, M is the natural gas flow, ψ is the weight coefficient, and λ is the Lagrange multiplier.

[0050] The point state estimation model is used to represent the state estimation of IES. The IES state estimation model is expressed as:

[0051]

[0052] In the formula, y is the IES state quantity, which mainly includes the grid node voltage, natural gas network node pressure, etc.; x is the IES measurement quantity, which mainly includes the grid node active and reactive power, load power, node voltage phase angle, natural gas network pipeline flow and other data that can be obtained by the measurement device; is the IES network topology and parameters, including pipeline length, diameter, equipment parameters, etc.; f is the functional relationship between input data and state quantity;

[0053] Step S2: Solve the IES state estimation model based on the weighted least squares method to obtain the IES state estimation result, which is specifically:

[0054]

[0055] In the formula, z e With z g are the measurement data vectors of the power grid and the natural gas grid respectively; m e With m g are the measurement models of the power grid and the natural gas grid respectively; x e With x g are the state vectors of the power grid and natural gas grid respectively; R e With R g are the measurement covariance matrices of the power grid and the natural gas grid respectively; c(x e ,x g ) is the zero equality constraint of the power grid and gas grid in IES;

[0056] Step S3, determining the state estimation model loss function; constructing an LSTM autoencoder, and using the historical operation data of the IES as a training data set to train the LSTM autoencoder through the loss function to obtain a trained LSTM autoencoder state estimation model; the loss function is used to measure the difference between the calculation result of the LSTM autoencoder and the state estimation result of the IES obtained in step S2;

[0057] LSTM is a recursive neural network that can process and predict time series data. All nodes in it contain a memory unit and three gate structures: input gate, forget gate and output gate. The input gate controls the new information that needs to be input at the current moment, the forget gate controls the information that needs to be discarded at the previous moment, and the output gate determines the information output in the memory unit. The gating mechanism and parameter update rules of LSTM are shown in the following formula:

[0058]

[0059] f t =σ(W f [H t-1 ,x t ]+b f )

[0060] i t =σ(W i [H t-1 ,x t ]+b i )

[0061]

[0062] O t=σ(W O [H t-1 ,x t ]+b O )

[0063] H t =O t tanh(C t )

[0064] Where: C t , C t-1 and is the unit state at the current moment, the unit state at the previous moment, and the unit state of the output layer; f t 、i t and O t are the output results of the forget gate, input gate and output gate respectively; σ is the Sigmoid activation function; W f , W i , W c , and W O are the weight parameter matrices corresponding to the gate; H t-1 is the hidden layer output of the previous unit; x t is the input of the current unit; b f 、b i 、b c and b O are the bias vectors associated with these gates respectively; [] represents the concatenation operation of the vectors;

[0065] AE consists of three parts: an encoder, a decoder, and a fully connected prediction neural network. The encoder maps high-dimensional input data to a low-dimensional latent space representation, and the decoder reconstructs the latent space representation prediction result obtained by LSTM into data as close to the original input as possible. By minimizing the reconstruction error, AE can extract the main features of the data and remove noise and redundant information; the LSTM autoencoder of the present invention combines the advantages of LSTM and autoencoder, and can simultaneously handle time dependency and high-dimensional feature extraction problems. The LSTM autoencoder includes an encoder, a fully connected prediction network, and a decoder;

[0066] The encoder is composed of multiple layers of LSTM units. The input is the input time series data obtained by IES observation, and the output is a low-dimensional latent space feature vector. For a given input variable X = {x1, x2, ... x T}, the data of each time step is processed in turn by the LSTM unit and the corresponding latent space feature variables h = {h1,h2,…,h T}; as follows:

[0067]

[0068] In the fully connected prediction network, the latent space feature variable h obtained by the encoder contains the temporal correlation characteristics of the input data. Taking h as the input variable of the fully connected prediction network, the latent space state estimation result of the IES variable is obtained, as shown in the following formula:

[0069]

[0070] The decoder is also composed of multiple layers of LSTM units. The input is Z, and the latent space state estimation result of the IES variable is reconstructed back to the original data space as the final output result.

[0071]

[0072] The specific expression of the loss function includes two parts: prediction loss and regularization loss. The prediction loss measures the difference between the model estimation result and the actual IES state quantity, which can be expressed as:

[0073]

[0074] Where, L p is the predicted loss; y is the true value of the IES state quantity, Estimate values ​​for the LSTM autoencoder;

[0075] The regularization loss uses the L2 regularization term:

[0076]

[0077] Where, L r is the regularization loss; λ is the regularization coefficient; θ en ,θ c ,θ de They are the parameters of encoder neural network, fully connected neural network and decoder neural network respectively.

[0078] Step S4: input the IES time series data to be estimated into the trained LSTM autoencoder state estimation model, and obtain the state estimation result of the integrated energy system through the forward propagation of the model.

[0079] LSTM can effectively capture long-range dependencies in time series data, and therefore performs well in various time series prediction tasks. However, the simple LSTM model has insufficient modeling capabilities when facing high-dimensional and complex IES state estimation data. To overcome this challenge, the autoencoder (AE), as an unsupervised learning model, can extract low-dimensional features of complex data by learning the compressed representation of data. Combining AE with LSTM can make full use of the features extracted by AE to enhance the LSTM model's ability to capture time dependencies and nonlinear features, thereby achieving a more accurate estimation of the state of the integrated energy system.

[0080] The feasibility of the method is verified by the IES consisting of an improved IEEE 9-node system and a 10-node medium-pressure gas distribution network, where the coupling nodes are located at nodes 2 and 5 in the power grid and nodes N4 and N9 in the natural gas network, all of which are connected through gas turbines. The system base power is 100MW, and the improved IES is as follows Figure 3 As shown;

[0081] The IES operating parameters were adjusted using the Model Carlo simulation method, including the active and reactive loads of the power grid and the gas load of the natural gas grid, to obtain the state data set under multiple operating states of the IES, and record the relevant observations, including the gas pressure of the relevant nodes of the natural gas grid and the node voltage with high correlation with the target node of the state estimation in the power grid as input features. The data set was divided into a training set and a test set at a ratio of 7.5:2.5, with 3000 training records and 1000 test records;

[0082] The root mean square error is used to describe the deviation between the estimated result and the true value. The specific calculation formula is:

[0083]

[0084] In the formula, R RMSE is the root mean square error calculation result, is the estimated value and y is the true value.

[0085] In order to improve the accuracy of the state estimation model, Gaussian noise with a standard deviation of 0.005 is superimposed on the input data. The estimated results of the voltage amplitude of the power grid node and the gas pressure of the natural gas network node by this method model are as follows: Figure 4 As shown;

[0086] The above is only a preferred embodiment of the present invention and does not limit the present invention in any form. Any technician familiar with the art can make many possible changes and modifications to the technical solution of the present invention by using the above-mentioned technical content without departing from the scope of the technical solution of the present invention, or modify it into an equivalent embodiment of equivalent changes. Therefore, any changes, modifications, equivalent changes and modifications made to the above embodiments based on the technology of the present invention without departing from the content of the technical solution of the present invention belong to the protection scope of the present technical solution.

Claims

1. A method for state estimation of an integrated energy system based on LSTM autoencoder, characterized in that: The method comprises the following steps: Step S1, establishing an IES state estimation model, and describing the energy coupling relationship in the IES through constraint equations; Step S2: Based on the historical measurement data, the IES state estimation model is solved by weighted least square method to obtain the IES state estimation result; Step S3, determining the state estimation model loss function; constructing an LSTM autoencoder, and using the historical operation data of the IES as a training data set to train the LSTM autoencoder through the loss function to obtain a trained LSTM autoencoder state estimation model; the loss function is used to measure the difference between the calculation result of the LSTM autoencoder and the state estimation result of the IES obtained in step S2; Step S4: input the IES time series data to be estimated into the trained LSTM autoencoder state estimation model, and obtain the state estimation result of the integrated energy system through the forward propagation of the model.

2. The method for state estimation of an integrated energy system based on an LSTM autoencoder according to claim 1, characterized in that: The IES state estimation model is expressed as: In the formula, y is the IES state quantity; x is the IES measurement quantity; is the IES network topology and parameters, including pipeline length, diameter, equipment parameters, etc.; f is the functional relationship between input data and state quantity.

3. The method for state estimation of an integrated energy system based on an LSTM autoencoder according to claim 1, characterized in that: The constraint equation is: F{·} and D{·} represent the real part and the imaginary part respectively, P j and Q j are the active power and reactive power injected into the node respectively, V is the voltage, G is the grid node admittance matrix, A is the node-branch association matrix, M is the natural gas flow, ψ is the weight coefficient, and λ is the Lagrange multiplier.

4. The method for state estimation of an integrated energy system based on an LSTM autoencoder according to claim 1, characterized in that: The method for solving the IES state estimation model by weighted least squares method is: In the formula, z e With z g are the measurement data vectors of the power grid and the natural gas grid respectively; m e With m g are the measurement models of the power grid and the natural gas grid respectively; x e With x g are the state vectors of the power grid and natural gas grid respectively; R e With R g are the measurement covariance matrices of the power grid and the natural gas grid respectively; c(x e ,x g ) is the zero equality constraint of the power grid and gas grid in IES.

5. The method for state estimation of an integrated energy system based on an LSTM autoencoder according to claim 1, characterized in that: The LSTM autoencoder includes an encoder, a fully connected prediction network and a decoder; The encoder is composed of multiple layers of LSTM units, the input is the input time series data obtained by IES observation, and the output is a low-dimensional latent space feature vector. For a given input variable, the data of each time step is processed in turn by the LSTM unit and the corresponding latent space feature variable is generated; The fully connected prediction network uses the latent space feature variables obtained by the encoder as input variables to obtain the latent space state estimation result of the IES variable; The decoder is composed of multiple layers of LSTM units, and its input is the latent space state estimation result of the IES variable output by the fully connected prediction network, which is reconstructed back to the original data space as the final output result.

6. The method for state estimation of an integrated energy system based on an LSTM autoencoder according to claim 5, characterized in that: The encoder is given an input variable X = {x1, x2, ... x T }, the encoder processes the data of each time step in turn through the LSTM unit and generates the corresponding latent space feature variable h={h1,h2,…,h T }, as shown below:

7. The method for state estimation of an integrated energy system based on LSTM autoencoder according to claim 6, characterized in that: The fully connected prediction network obtains the latent space state estimation result of the IES variable through the following formula:

8. The method for state estimation of an integrated energy system based on LSTM autoencoder according to claim 7, characterized in that: The decoder calculates the final output result by the following formula 9. The method for state estimation of an integrated energy system based on LSTM autoencoder according to claim 1, characterized in that: The loss function includes prediction loss and regularization loss, and the prediction loss can be expressed as: Where, L p is the predicted loss; y is the true value of the IES state quantity, Estimate values ​​for the LSTM autoencoder; The regularization loss uses the L2 regularization term: Where, L r is the regularization loss; λ is the regularization coefficient; θ en ,θ c ,θ de They are the parameters of encoder neural network, fully connected neural network and decoder neural network respectively.

Images