Substation maintenance method and system based on deep learning

Abstract

The invention provides a substation maintenance method and system based on deep learning, and the method comprises the steps: S1, collecting and cleaning sensor data and equipment state data of a substation, and obtaining current, voltage, power and frequency data; s2, carrying out feature data extraction on the current, voltage, power and frequency data by adopting a Pearson correlation coefficient method to obtain transformer substation feature data; s3, performing dimensionality reduction on the characteristic data of the transformer substation by adopting a principal component analysis method; s4, constructing a transformer substation fault diagnosis model by adopting a recurrent neural network; s5, evaluating the transformer substation fault diagnosis model by using the verification set; and S6, performing fault diagnosis prediction on the test data by using the trained deep learning model. The invention aims to improve the maintenance efficiency and accuracy through an intelligent means, can effectively reduce the requirement of manual intervention, improves the maintenance efficiency, and reduces the operation cost.

Description

Technical Field

[0001] The present invention relates to the technical field of substation maintenance, and in particular to a substation maintenance method and system based on deep learning. Background Art

[0002] In the power system, the generation, transmission, distribution and use of electric energy are an inseparable whole. As an important part of the power system, the substation is responsible for the important task of power transformation and distribution. With the rapid development of the national economy and the continuous improvement of people's living standards, the power equipment manufacturing industry and the power grid structure have developed rapidly, the capacity of the power system has increased rapidly, and the system structure has become more complex, which has posed a huge challenge to the operation and inspection of power transmission and transformation equipment, especially the new intelligent substation. In order to accelerate the construction of a strong smart grid, save human resource costs, and improve work efficiency, the research on the operation and maintenance mode of intelligent substations has become an urgent problem to be solved.

[0003] However, at present, the operation and management mode of intelligent substations is still the same as that of traditional integrated automated substations, which is mainly an experience-based and periodic mode. Nowadays, the intelligent technology of substations is developing rapidly and the content of equipment operation is changing with each passing day, but the management mode of substation operation and inspection has not been innovated with the advancement of technology. The operation and inspection level of substations depends to a large extent on the working mode of operation and maintenance personnel, which leads to the lack of systematicity, scientificity, flexibility and adaptability of this mode. Summary of the invention

[0004] In order to achieve the above object, the present invention adopts the following technical solutions:

[0005] A substation maintenance method based on deep learning includes the following steps:

[0006] S1: Collect and clean substation sensor data and equipment status data to obtain current, voltage, power, and frequency data;

[0007] S2: Use the Pearson correlation coefficient method to extract characteristic data from current, voltage, power, and frequency data to obtain substation characteristic data;

[0008] S3: Use principal component analysis to reduce the dimension of substation characteristic data;

[0009] S4: Using recurrent neural network to build a substation fault diagnosis model;

[0010] S5: Use the validation set to evaluate the substation fault diagnosis model;

[0011] S6: Use the trained deep learning model to perform fault diagnosis prediction on the test data.

[0012] As a further improvement of the present technical solution, step S1 comprises:

[0013] The collected substation sensor data is standardized and normalized. For the i-th group of current, voltage, power, and frequency data x i , standardized as,

[0014]

[0015] Among them, μ i is the mean of the i-th group of data, σ i is the standard deviation of the i-th group of data, i represents the index number of the i-th group of data, x i ′ represents the data after standardization;

[0016] For the i-th group of data x i , the normalized formula is,

[0017]

[0018] in, Represents standardized and normalized data; max() represents the maximum value function; min() represents the minimum value function.

[0019] As a further improvement of the technical solution, step S2 includes:

[0020] The linear correlation between two continuous variables is measured by the Pearson correlation coefficient, and the value range is [-1,1]. The larger the absolute value, the stronger the correlation. The calculation formula of the Pearson correlation coefficient is expressed as follows:

[0021]

[0022] Among them, x s i and s i are the data variable and target variable value of the i-th sample after standardization and normalization, respectively. and are the means of the data variable and the target variable, respectively, and n represents the total number of samples;

[0023] According to the calculated Pearson correlation coefficient, the top three data variables with the Pearson correlation coefficient are selected as feature data.

[0024] As a further improvement of the present technical solution, step S3 includes:

[0025] According to the feature data, a feature data matrix X[n×m] is constructed, where each row represents a sample and each column represents a variable. The standardized data is recorded as Xstd , let X std is an n×m matrix, where m is the number of variables. The elements of the covariance matrix M are:

[0026]

[0027] Among them, X T std Yes X std The transposed matrix of

[0028] Then perform eigenvalue decomposition on the covariance matrix M to obtain the eigenvalue λ i and the corresponding unit eigenvector v i , the unit eigenvector v i Represents the new principal component direction; the first k unit eigenvectors v1, v2, ...v k The projection matrix V = [v1, v2, ... v k ], the first k unit eigenvectors constitute the transformation matrix of the data projected into the new space; and k≤n;

[0029] The standardized data X std Multiply by the projection matrix V to get the reduced dimension data Y:

[0030] Y=X std V

[0031] Among them, Y is an n×k matrix, and each row represents the projection of a sample in the new k-dimensional space.

[0032] As a further improvement of the present technical solution, step S4 includes:

[0033] 1) Define a state space and an action space, wherein the state space is the sensor data or equipment state data of the substation, and the action space is the current state decision based on the current state data;

[0034] 2) Constructing RNN model: LSTM is used as the RNN model to process the sensor data and equipment status data of the substation. The hidden state update of the LSTM is expressed as:

[0035] h t =LSTM(r t ,h t-1 )

[0036] Among them, r t is the input of the tth time step in the input sequence, h t is the hidden state at the tth time step, h t-1 represents the hidden state at the t-1th time step;

[0037] At the tth time step, the feature data is represented as X t , then the hidden state h t and the hidden state h at the previous time step t-1 The relationship between them is as follows:

[0038] h t =δ(ω1X t +ω2h t-1 +b)

[0039] Among them, the activation function δ uses the hyperbolic tangent function tanh, and ω1 represents the current input X t The weight matrix related to the previous hidden state h t-1 The relevant weight matrix; b represents the bias vector;

[0040] The output O of the hidden layer at the tth time step t and the predicted output Z t for:

[0041]

[0042] Among them, the activation function σ adopts the Softmax function, and ω3 represents the hidden state h t The relevant weight matrix; c represents the bias vector.

[0043] As a further improvement of the technical solution, the step S4 further includes: the loss function of the recurrent neural network is:

[0044] The predicted output Z at time step t using the actual variables and samples t The mean square error between them is used as the loss function L t , quantify the performance of the current model, as shown in the following formula:

[0045]

[0046] Where N is the number of output data; represents the predicted value of the variable in the i-th calculation cycle at time t; z t i Represents the actual value of the variable in the i-th calculation cycle at time t.

[0047] As a further improvement of the technical solution, step S6 includes: constructing a substation fault diagnosis model through the established recurrent neural network for evaluation:

[0048] The accuracy index is selected as the evaluation standard, and the specific calculation method is as follows:

[0049]

[0050] Where P represents the diagnosed fault, Indicates the real fault to which the given data belongs; the precision index p re It refers to the proportion of correctly diagnosed faults in all substations. The value range is 0 to 1. The larger the value, the better the diagnosis result.

[0051] The present invention also discloses a substation maintenance system based on deep learning, comprising:

[0052] Data acquisition module: used to collect current, voltage, power and frequency data of substation;

[0053] Data preprocessing module: cleans, denoises, standardizes and normalizes the collected current, voltage, power and frequency data;

[0054] Feature extraction module: extract feature data from the preprocessed data to obtain substation feature data;

[0055] Fault detection and diagnosis module: Fault detection is performed based on recurrent neural networks, and the substation fault diagnosis model is evaluated using a validation set;

[0056] Compared with the existing technology, the beneficial effects of this paper are as follows: the traditional substation maintenance method has problems such as high work intensity, low efficiency, and inaccurate fault diagnosis. This paper proposes a substation maintenance method and system based on deep learning, aiming to improve the maintenance efficiency and accuracy through intelligent means. First, the operation data of various equipment in the substation are collected in real time; then, these large-scale, multi-dimensional monitoring data are processed and analyzed using neural network algorithms; finally, a substation fault diagnosis model is constructed through a recurrent neural network. The simulation shows that the proposed substation maintenance method and system based on deep learning can achieve a diagnostic accuracy of 94%, which can effectively reduce the need for manual intervention, improve the efficiency of maintenance work, and reduce operating costs.

[0057] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention and implement it according to the contents of the specification, the following is a detailed description of the preferred embodiments of the present invention in conjunction with the accompanying drawings. The specific implementation of the present invention is given in detail by the following embodiments and their accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 A flowchart of a substation maintenance method and system based on deep learning studied in the present invention;

[0059] Figure 2 This is the fault diagnosis accuracy of different methods studied in this paper. DETAILED DESCRIPTION

[0060] The principles and features of the present invention are described below in conjunction with the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention. The present invention is described in more detail by way of example with reference to the accompanying drawings in the following paragraphs. The advantages and features of the present invention will become clearer according to the following description and claims. It should be noted that the drawings are all in a very simplified form and are not in precise proportions, and are only used to conveniently and clearly assist in explaining the purpose of the embodiments of the present invention.

[0061] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0062] Example 1: Please refer to the attached Figure 1 , a substation maintenance method based on deep learning, comprising the following steps:

[0063] S1: Collect substation sensor data and equipment status data and clean them to obtain current, voltage, power and frequency data; including:

[0064] The collected substation sensor data is standardized and normalized. For the i-th group of current, voltage, power, and frequency data x i , standardized as,

[0065]

[0066] Among them, μ i is the mean of the i-th group of data, σ i is the standard deviation of the i-th group of data, i represents the index number of the i-th group of data, x i ′ represents the data after standardization;

[0067] For the i-th group of data x i , the normalized formula is,

[0068]

[0069] in, Represents standardized and normalized data; max() represents the maximum value function; min() represents the minimum value function.

[0070] S2: Use the Pearson correlation coefficient method to extract characteristic data from current, voltage, power, and frequency data to obtain substation characteristic data; including:

[0071] The linear correlation between two continuous variables is measured by the Pearson correlation coefficient, and the value range is [-1,1]. The larger the absolute value, the stronger the correlation. The calculation formula of the Pearson correlation coefficient is expressed as follows:

[0072]

[0073] Among them, x s i and s i are the data variable and target variable value of the i-th sample after standardization and normalization, respectively. and are the means of the data variable and the target variable, respectively, and n represents the total number of samples;

[0074] According to the calculated Pearson correlation coefficient, the top three data variables with the Pearson correlation coefficient are selected as feature data.

[0075] S3: Use principal component analysis to reduce the dimension of substation characteristic data; including:

[0076] According to the feature data, a feature data matrix X[n×m] is constructed, where each row represents a sample and each column represents a variable. The standardized data is recorded as X std , let X std is an n×m matrix, where m is the number of variables. The elements of the covariance matrix M are:

[0077]

[0078] Among them, X T std Yes X std The transposed matrix of

[0079] Then perform eigenvalue decomposition on the covariance matrix M to obtain the eigenvalue λ i and the corresponding unit eigenvector v i , the unit eigenvector v i Represents the new principal component direction; the first k unit eigenvectors v1, v2, ...v k The projection matrix V = [v1, v2, ... v k ], the first k unit eigenvectors constitute the transformation matrix of the data projected into the new space; and k≤n;

[0080] The standardized data X std Multiply by the projection matrix V to get the reduced dimension data Y:

[0081] Y=X std V

[0082] Among them, Y is an n×k matrix, and each row represents the projection of a sample in the new k-dimensional space.

[0083] S4: Use recurrent neural network to build a substation fault diagnosis model; including:

[0084] 1) Define a state space and an action space, wherein the state space is the sensor data or equipment state data of the substation, and the action space is the current state decision based on the current state data;

[0085] 2) Constructing RNN model: LSTM is used as the RNN model to process the sensor data and equipment status data of the substation. The hidden state update of the LSTM is expressed as:

[0086] h t =LSTM(r t ,h t-1 )

[0087] Among them, r t is the input of the tth time step in the input sequence, h t is the hidden state at the tth time step, h t-1 represents the hidden state at the t-1th time step;

[0088] At the tth time step, the feature data is represented as X t , then the hidden state h t and the hidden state h at the previous time step t-1 The relationship between them is as follows:

[0089] h t =δ(ω1X t +ω2h t-1 +b)

[0090] Among them, the activation function δ uses the hyperbolic tangent function tanh, and ω1 represents the current input X t The weight matrix related to the previous hidden state h t-1 The relevant weight matrix; b represents the bias vector;

[0091] The output O of the hidden layer at the tth time step t and the predicted output Z t for:

[0092]

[0093] Among them, the activation function σ adopts the Softmax function, and ω3 represents the hidden state h t The relevant weight matrix; c represents the bias vector.

[0094] S5: Use the validation set to evaluate the substation fault diagnosis model;

[0095] The predicted output Z at time step t using the actual variables and samples t The mean square error between them is used as the loss function L t , quantify the performance of the current model, as shown in the following formula:

[0096]

[0097] Where N is the number of output data; represents the predicted value of the variable in the i-th calculation cycle at time t; z t i Represents the actual value of the variable in the i-th calculation cycle at time t.

[0098] Since there is an error at each position in the propagation sequence, the final target loss function is:

[0099]

[0100] Where T is the number of cycles; θ is the network parameter;

[0101] Use a neural network to estimate the Q value function and select the optimal action. The Q value function update formula is:

[0102] Q(p t ,a t )←Q(p t ,a t )+α(f t+1 +γmax a Q(p t+1 ,a)-Q(p t ,a t ))

[0103] Among them, Q(p t , a t ) is in state s t Execute action a t Q value, f t+1 Is performing action a t The immediate reward after , γ is the discount factor, and α is the learning rate;

[0104] The deep Q network approximation function uses the Q function Q(p, a; θ) containing the network parameters θ. When the number of iterations is j, the loss function is:

[0105] L j (θ j )=E (p,a,f,p′) [(f+γmaxQ(p′,a′;θ * )-Q(p,a;θj )) 2 ]

[0106] Among them, p, a and f are the current state, action and reward value, p' and a' are the next state and action, γ is the discount factor, θ j represents the network parameters in the learning process, and j represents the jth type of loss. After a period of learning, the new θ * Will be updated. The specific learning process is as follows:

[0107]

[0108] in, is the symbol used to calculate the gradient.

[0109] S6: Use the trained deep learning model to perform fault diagnosis prediction on the test data;

[0110] The substation fault diagnosis model is constructed through the established recurrent neural network for evaluation:

[0111] The accuracy index is selected as the evaluation standard, and the specific calculation method is as follows:

[0112]

[0113] Where P represents the diagnosed fault, Indicates the real fault to which the given data belongs; the precision index p re It refers to the proportion of correctly diagnosed faults in all substations. The value range is 0 to 1. The larger the value, the better the diagnosis result.

[0114] The fault diagnosis performance of the proposed method for the secondary system is compared with the particle swarm algorithm, the improved particle swarm algorithm and the transfer learning method. Figure 2 The relative error and maximum error between predicted fault and actual fault are introduced as evaluation indicators. Figure 2It can be seen that the fault diagnosis effect of the present invention is better, with the average error and maximum error being 0.008 and 0.601 respectively, which can better meet the actual needs on site. Based on the dimensionality reduction of the principal component analysis method, the present invention uses the deep Q network algorithm to optimize the parameters of the recurrent neural network model to achieve fault diagnosis, thereby further reducing the diagnostic error. At the same time, the improved particle swarm algorithm can also realize fault detection and achieve better diagnostic effect. However, compared with the RNN-DQN diagnostic model of the proposed method, its diagnostic error is improved. The fault diagnosis models of the particle swarm algorithm and the transfer learning algorithm are relatively single, so the diagnostic error is large. In summary, the present invention can improve the fault diagnosis accuracy of the substation, and with the refinement of the fault type and fault characteristics, it has more obvious advantages in processing massive multidimensional data.

[0115] Embodiment 2: The present invention also discloses a substation maintenance system based on deep learning, comprising:

[0116] Data acquisition module: used to collect current, voltage, power and frequency data of substation;

[0117] Data preprocessing module: cleans, denoises, standardizes and normalizes the collected current, voltage, power and frequency data;

[0118] Feature extraction module: extract feature data from the preprocessed data to obtain substation feature data;

[0119] Fault detection and diagnosis module: Fault detection is performed based on recurrent neural networks, and the substation fault diagnosis model is evaluated using a validation set.

[0120] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Any ordinary technician in the industry can smoothly implement the present invention as shown in the drawings and described above. However, any equivalent changes, modifications and evolutions made by technicians familiar with the profession without departing from the scope of the technical solution of the present invention using the technical content disclosed above are all equivalent embodiments of the present invention. At the same time, any equivalent changes, modifications and evolutions made to the above embodiments based on the essential technology of the present invention are still within the protection scope of the technical solution of the present invention.

Claims

1. A substation maintenance method based on deep learning, characterized in that: The following steps are involved: S1: Collect substation sensor data and equipment status data and clean them to obtain current, voltage, power and frequency data; S2: Use the Pearson correlation coefficient method to extract characteristic data from current, voltage, power, and frequency data to obtain substation characteristic data; S3: Use principal component analysis to reduce the dimension of substation characteristic data; S4: Using recurrent neural network to build a substation fault diagnosis model; S5: Use the validation set to evaluate the substation fault diagnosis model; S6: Use the trained deep learning model to perform fault diagnosis prediction on the test data.

2. The substation maintenance method based on deep learning according to claim 1 is characterized in that: The step S1 comprises: The collected substation sensor data is standardized and normalized. For the i-th group of current, voltage, power, and frequency data x i , standardized as, Among them, μ i is the mean of the i-th group of data, σ i is the standard deviation of the i-th group of data, i represents the index number of the i-th group of data, x i ′ represents the data after standardization; For the i-th group of data x i , the normalized formula is, in, Represents standardized and normalized data; max() represents the maximum value function; min() represents the minimum value function.

3. The substation maintenance method based on deep learning according to claim 1 is characterized in that: The step S2 comprises: The linear correlation between two continuous variables is measured by the Pearson correlation coefficient, and the value range is [-1,1]. The larger the absolute value, the stronger the correlation. The calculation formula of the Pearson correlation coefficient is expressed as follows: Among them, x s i and s i are the data variable and target variable value of the i-th sample after standardization and normalization, respectively. and are the means of the data variable and the target variable, respectively, and n represents the total number of samples; According to the calculated Pearson correlation coefficient, the top three data variables with the Pearson correlation coefficient are selected as feature data.

4. The substation maintenance method based on deep learning according to claim 3 is characterized in that: The step S3 comprises: According to the feature data, a feature data matrix X[n×m] is constructed, where each row represents a sample and each column represents a variable. The standardized data is recorded as X std , let X std is an n×m matrix, where m is the number of variables. The elements of the covariance matrix M are: in, Yes X std The transposed matrix of Then perform eigenvalue decomposition on the covariance matrix M to obtain the eigenvalue λ i and the corresponding unit eigenvector v i , the unit eigenvector v i Represents the new principal component direction; the first k unit eigenvectors v1, v2, ...v k The projection matrix V = [v1, v2, ... v k ], the first k unit eigenvectors constitute the transformation matrix of the data projected into the new space; and k≤n; The standardized data X std Multiply by the projection matrix V to get the reduced dimension data Y: Y=X std V Among them, Y is an n×k matrix, and each row represents the projection of a sample in the new k-dimensional space.

5. The substation maintenance method based on deep learning according to claim 4 is characterized in that: The step S4 comprises: 1) Define a state space and an action space, wherein the state space is the sensor data or equipment state data of the substation, and the action space is the current state decision based on the current state data; 2) Constructing RNN model: LSTM is used as the RNN model to process the sensor data and equipment status data of the substation. The hidden state update of the LSTM is expressed as: h t =LSTM(r t ,h t-1 ) Among them, r t is the input of the tth time step in the input sequence, h t is the hidden state at the tth time step, h t-1 represents the hidden state at the t-1th time step; At the tth time step, the feature data is represented as X t , then the hidden state h t and the hidden state h at the previous time step t-1 The relationship between them is as follows: h t =δ(ω1X t +ω2h t-1 +b) Among them, the activation function δ uses the hyperbolic tangent function tanh, and ω1 represents the current input X t The weight matrix related to the previous hidden state h t-1 The relevant weight matrix; b represents the bias vector; The output O of the hidden layer at the tth time step t and the predicted output Z t for: Among them, the activation function σ adopts the Softmax function, and ω3 represents the hidden state h t The relevant weight matrix; c represents the bias vector.

6. The substation maintenance method based on deep learning according to claim 5 is characterized in that: The step S4 also includes: the loss function of the recurrent neural network is: The predicted output Z at time step t using the actual variables and samples t The mean square error between them is used as the loss function L t , quantify the performance of the current model, as shown in the following formula: Where N is the number of output data; Z t i represents the predicted value of the variable in the i-th calculation cycle at time t; z t i Represents the actual value of the variable in the i-th calculation cycle at time t.

7. The substation maintenance method based on deep learning according to claim 6 is characterized in that: The step S6 comprises: constructing a substation fault diagnosis model through the established recurrent neural network for evaluation: The accuracy index is selected as the evaluation standard, and the specific calculation method is as follows: Where P represents the diagnosed fault, Indicates the real fault to which the given data belongs; the precision index p re It refers to the proportion of correctly diagnosed faults in all substations. The value range is 0 to 1. The larger the value, the better the diagnosis result.

8. A substation maintenance system based on deep learning, characterized in that: include: Data acquisition module: used to collect current, voltage, power and frequency data of substation; Data preprocessing module: cleans, denoises, standardizes and normalizes the collected current, voltage, power and frequency data; Feature extraction module: extract feature data from the preprocessed data to obtain substation feature data; Fault detection and diagnosis module: Fault detection is performed based on recurrent neural networks, and the substation fault diagnosis model is evaluated using a validation set; To realize a substation maintenance method based on deep learning as described in any one of claims 1-7.

Images