Improved power fault data identification method and system based on L-M algorithm

Abstract

The invention discloses an improved power failure data identification method and system based on an L-M algorithm, and the method comprises the steps: data normalization processing: collecting electrical quantity data in an existing relay protection device, extracting a target characteristic quantity in the electrical quantity data, and carrying out the normalization processing; model construction and training: constructing a power failure data recognition model, and performing model training based on the normalized target characteristic quantity and an L-M algorithm, namely, enabling an error function of the power failure data recognition model to adopt corresponding convergence methods at different stages by changing a damping coefficient; electric quantity data in the target relay protection device are collected, target characteristic quantities in the electric quantity data are extracted and subjected to normalization processing, and then the trained electric power fault data recognition model is input. According to the method, the neural network model optimized based on the L-M algorithm is established, and the abnormal operation condition of the relay protection device can be effectively identified.

Description

Technical Field

[0001] The present invention relates to the technical field of relay protection, and in particular to an improved method and system for identifying power fault data based on LM algorithm. Background Art

[0002] Power data mining technology has been a key research topic in the power industry in recent years. This technology mines hidden information from a large amount of power data, which plays an important role in the safe operation of relay protection devices. With the continuous updating of power company technology, the popularization and application of smart devices, smart meters and online monitoring systems, a large amount of device operation data has been collected and saved, and the scale of power big data is getting larger and larger. These data provide a good data foundation for power data mining. Therefore, studying power fault data from the data is a direction worth exploring.

[0003] In the research on power failures at home and abroad, most of them still use traditional machine learning algorithms, such as cluster analysis, support vector machines, decision trees, random forests, etc. The advantage of this type of algorithm is that the construction of the model is relatively simple and convenient, but the operating efficiency of the model is relatively general, and the accuracy is not high; and there are relatively few cases of the application of deep learning algorithms and neural network algorithms in data mining algorithms. Research in this field is still in its initial stage and needs further research. It is a direction worthy of research to use the historical power consumption data of relay protection devices as the basis and to use data mining technology to mine the information in these data. Summary of the invention

[0004] In order to solve the above problems, the present invention proposes an improved power fault data identification method and system based on the LM algorithm, which can detect the operation status of the relay protection device from the historical data.

[0005] The technical solution adopted by the present invention is as follows:

[0006] A method for identifying power fault data based on an improved LM algorithm, comprising:

[0007] Data normalization processing: Collect electrical quantity data from existing relay protection devices, extract target feature quantities from the electrical quantity data and perform normalization processing. The target feature quantities include voltage, current, active power and reactive power of three-phase electricity A, B and C;

[0008] Model construction and training: Construct a power fault data recognition model, and perform model training based on the normalized target feature quantity and LM algorithm, that is, by changing the damping coefficient, the error function of the power fault data recognition model adopts the corresponding convergence method at different stages;

[0009] Power fault data recognition: Collect electrical quantity data from the target relay protection device, extract target feature quantities from the electrical quantity data and perform normalization processing, and then input the trained power fault data recognition model to identify the abnormal operation of the target relay protection device.

[0010] Furthermore, the training process of the power fault data identification model includes forward operation and reverse derivation. After the normalized target feature quantity is input, forward weighted summation is first performed to the output layer; if the result obtained by the output layer does not meet the expected value, reverse derivation is performed to modify the connection weights and thresholds between the connection layers, and forward operation is performed again after obtaining new parameters until the error between the result of the output layer and the expected value meets the requirements.

[0011] Furthermore, the error function of the power fault data identification model adopts corresponding convergence methods at different stages by changing the damping coefficient, including: when the connection weight and threshold of the power fault data identification model are far away from the optimal value, the steepest descent method is used for convergence.

[0012] Furthermore, the method of changing the damping coefficient so that the error function of the power fault data identification model adopts corresponding convergence methods at different stages also includes: when the connection weight and threshold of the power fault data identification model are close to the optimal value, using the Gauss-Newton method for convergence.

[0013] Furthermore, the transfer function adopted by the intermediate layer of the power fault data identification model includes a Sigmoid function, and the transfer function adopted by the output layer includes a linear function.

[0014] An improved power fault data identification system based on LM algorithm includes:

[0015] A data normalization processing module is configured to collect electrical quantity data in an existing relay protection device, extract target characteristic quantities in the electrical quantity data and perform normalization processing, wherein the target characteristic quantities include voltage, current, active power and reactive power of three-phase electricity A, B and C;

[0016] The model building and training module is configured to build a power fault data recognition model and perform model training based on the normalized target feature quantity and the LM algorithm, that is, by changing the damping coefficient, the error function of the power fault data recognition model adopts a corresponding convergence method at different stages;

[0017] The power fault data recognition module is configured to collect electrical quantity data from the target relay protection device, extract target feature quantities from the electrical quantity data and perform normalization processing, and then input the trained power fault data recognition model to identify the abnormal operation of the target relay protection device.

[0018] Furthermore, the training process of the power fault data identification model includes forward operation and reverse derivation. After the normalized target feature quantity is input, forward weighted summation is first performed to the output layer; if the result obtained by the output layer does not meet the expected value, reverse derivation is performed to modify the connection weights and thresholds between the connection layers, and forward operation is performed again after obtaining new parameters until the error between the result of the output layer and the expected value meets the requirements.

[0019] Furthermore, the error function of the power fault data identification model adopts corresponding convergence methods at different stages by changing the damping coefficient, including: when the connection weight and threshold of the power fault data identification model are far away from the optimal value, the steepest descent method is used for convergence.

[0020] Furthermore, the method of changing the damping coefficient so that the error function of the power fault data identification model adopts corresponding convergence methods at different stages also includes: when the connection weight and threshold of the power fault data identification model are close to the optimal value, using the Gauss-Newton method for convergence.

[0021] Furthermore, the transfer function adopted by the intermediate layer of the power fault data identification model includes a Sigmoid function, and the transfer function adopted by the output layer includes a linear function.

[0022] The beneficial effects of the present invention are:

[0023] The present invention establishes a neural network model based on LM algorithm optimization, which can be used to identify abnormal operation of relay protection devices; provides an overall algorithm flow of the neural network and a neural network optimization algorithm based on LM algorithm optimization; verifies the feasibility of the power fault data recognition model improved based on LM algorithm through simulation analysis, and compares the improved model with the conventional neural network model, and the running speed and accuracy of the running results are better than those of the conventional network model. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 4 is a graph showing the relationship between the number of hidden layer nodes and the training error in Example 1 of the present invention.

[0025] Figure 2 This is a flow chart of the training of the power fault data identification model of Example 1 of the present invention.

[0026] Figure 3 This is a training error diagram of the power fault data recognition model of Example 1 of the present invention.

[0027] Figure 4 This is a diagram of the training process of the power fault data identification model of Example 1 of the present invention. DETAILED DESCRIPTION

[0028] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, the specific implementation methods of the present invention are now described. 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 the embodiments. 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.

[0029] Example 1

[0030] like Figure 1 As shown, this embodiment provides a method for identifying power fault data based on an improved LM algorithm, including:

[0031] Data normalization processing: Collect electrical quantity data from existing relay protection devices, extract target characteristic quantities from the electrical quantity data and perform normalization processing. The target characteristic quantities include voltage, current, active power and reactive power of three-phase electricity A, B and C;

[0032] Model construction and training: Construct a power fault data recognition model, and perform model training based on the normalized target feature quantity and LM algorithm, that is, by changing the damping coefficient, the error function of the power fault data recognition model adopts the corresponding convergence method at different stages;

[0033] Power fault data recognition: Collect electrical quantity data from the target relay protection device, extract target feature quantities from the electrical quantity data and perform normalization processing, and then input the trained power fault data recognition model to identify the abnormal operation of the target relay protection device.

[0034] Specifically, a three-layer neural network model is now taken as an example to explain in detail the improved power fault data identification method based on the LM algorithm in this embodiment. The design of the neural network mainly includes data normalization processing, the number of network layers, the number of input nodes, the number of intermediate layer nodes, the number of output layer nodes and the transfer function, the training method, the setting of training parameters, etc., which are specifically described as follows.

[0035] 1) Data normalization

[0036] The normalization of neural network data is to map the data of the model training set to the data interval of (0-1). Data normalization can accelerate the convergence speed of the subsequent model algorithm and improve the accuracy of the model to a certain extent. The 12 characteristic quantities of voltage, current, active power and reactive power of the three phases of users A, B and C are normalized. For the input sample data x p Where p = 1, 2, 3…12; let x p The maximum and minimum values ​​of x are max、x min The processed data is:

[0037]

[0038] The normalized data can be directly used as the input and output data of the neural network.

[0039] 2) Number of network layers

[0040] A neural network consists of an input layer, an intermediate layer, and an output layer. The number of neurons in the intermediate layer can be determined according to the specific situation. Generally, a single-layer network can solve most nonlinear problems by appropriately increasing the number of neurons. Too many intermediate layers will increase the model training time. Therefore, for most applications, a single hidden layer can meet the needs. According to actual needs, a three-layer neural network is selected, that is, the number of layers of the input layer, intermediate layer, and output layer are all one layer.

[0041] 3) Number of input layer nodes

[0042] The dimension of the model data set determines the number of nodes in the neural network input layer. 12 characteristic quantities of voltage, current, active power, and reactive power of the three phases A, B, and C of the power grid users are selected as model input quantities, so the number of model input nodes is 12.

[0043] 4) Number of nodes in the middle layer

[0044] The number of nodes in the middle layer plays a key role in the convergence of the model error function. Too many or too few nodes in the middle layer will affect the performance of the model. The general number of hidden layer nodes can be determined using the following formula.

[0045]

[0046] In (3-2), M is the number of nodes in the input layer, J is the number of nodes in the output layer, and C is a constant between [1, 10]. The input layer of the model is 12 layers, and the output layer is 2 layers. From the above formula, we can know that the number of nodes in the middle layer is 5-14. After MATLAB simulation experiments, it is verified that when the middle layer is 9 layers, the function converges to the objective function the fastest. Finally, the number of layers in the middle layer is determined to be 9. Figure 1 shown.

[0047] 5) Number of neurons in the output layer

[0048] The needs of practical problems also determine the number of neurons in the output layer of the neural network, so the output layer nodes are fixed to 2 layers.

[0049] 6) Selection of transfer function

[0050] The model uses the Sigmoid function as the transfer function of the hidden layer. The Sigmoid function is as follows.

[0051]

[0052] The output layer uses a linear function:

[0053] Y=kx+b

[0054] The training process of the model is divided into forward operation and reverse derivation. After the network inputs data, it first performs forward weighted summation to the output layer. If the result obtained by the output layer does not meet the expected value, the network performs reverse derivation to modify the weights and thresholds between the connection layers. After obtaining the new parameters, the forward operation is performed again until the error between the result of the output layer and the expected value meets the requirements. The specific process is as follows Figure 2 shown.

[0055] Assume that the model has an input layer of M, an intermediate layer of I, an output layer of J, and the input data is a group. The neurons in the input layer are denoted by m, the neurons in the hidden layer are denoted by i, and the neurons in the output layer are denoted by j (m, i, j are integers). The connection weight from the input layer to the intermediate layer is ω mi , the connection weight from the middle layer to the output layer is ω ij , the threshold of the middle layer is θ i , the threshold of the output layer is α j . u and v are the input and output of each layer respectively, such as Interpreted as the input to the i-th neuron in layer I.

[0056] Assume that the expected function is: d(n) = [d1, d2, ..., d J ], the input signal is: x(n).

[0057]

[0058] The Sigmoid excitation function is introduced in the hidden layer and recorded as:

[0059]

[0060] Then the output of the i-th neuron in the hidden layer is equal to:

[0061]

[0062] The input of the jth neuron in the output layer is equal to Weighted sum:

[0063]

[0064] The error function of the output layer is f(ω mi ,ω ij ,θi ,α j ), then the error of the jth neuron in the output layer is:

[0065]

[0066] Take the total error function of the network as the quadratic error function and record it as F(ω mi ,ω ij ,θ i ,α j ):

[0067]

[0068] 7) Adjustment of weights

[0069] Generally, the weights and thresholds are adjusted using the gradient descent method, that is, the weights and thresholds are updated along the direction of the gradient. mi ,ω ij Updated as follows:

[0070]

[0071] Threshold θ i , α j Updated as follows:

[0072]

[0073] Using the gradient descent method to modify weights and thresholds will cause the error function to fall into a local optimal solution, and the training speed is very slow. In order to solve the shortcomings of the gradient descent algorithm, the network is optimized as follows:

[0074] In general, second-order convergence is better than first-order convergence. The error function is converted from first-order convergence to second-order convergence to find the optimal solution. This method can significantly speed up the convergence speed and accuracy of the objective function. For the convenience of calculation, let the total error function:

[0075] F(ω mi ,ω ij ,θ i ,α j )=Z(x n )

[0076] (x n is a variable), then in high dimensions Z(x n ) is:

[0077]

[0078] Where H[Z(x n )] is Z(x n )’s Hessian matrix:

[0079]

[0080] when When , we can find the approximate value of Δ:

[0081]

[0082] The optimization algorithm of the above formula is Newton's method. Since the calculation amount of this method is large, the Jacobian matrix is ​​used instead of the Hessian matrix. The function Z(x n ) is denoted as J Z(xn) :

[0083]

[0084] The Δ in the above formula is updated as:

[0085]

[0086] The optimization algorithm that uses the Jacobian matrix instead of the Hessian matrix is ​​called the Gauss-Newton method, where J Z(xn) T J Z(xn) is a Gaussian term. Since the Jacobian matrix is ​​a first-order matrix, the amount of calculation is small, so the convergence speed of the Gauss-Newton method is better than that of the Newton method. Introducing the damping coefficient μ into the above formula, the original formula becomes:

[0087]

[0088] I is the unit matrix. When μ is very large, the Gaussian term J in the formula Z(xn) T J Z(xn) can be ignored, and the original formula becomes the gradient descent method. At this time:

[0089]

[0090] When μ is very small, the original formula is transformed into a second-order optimization problem. At this time:

[0091]

[0092] In the initial stage of error function optimization, the optimization speed of gradient descent method is very fast. When the function is about to converge, the speed of second-order optimization is much faster than gradient descent method. Therefore, the damping coefficient μ can be changed to make the function use the optimal algorithm at different stages. This optimization algorithm is LM (Levenberg-Marquardt) method.

[0093] The weight update formula of the neural network optimized by the LM algorithm is:

[0094]

[0095] The threshold update formula is:

[0096]

[0097] like Figure 3 The figure shows the training error of the power fault data recognition model in this embodiment. Figure 4 The figure shows the training process of the power fault data recognition model of this embodiment. Table 1 shows the calculation results of the power fault data recognition model of this embodiment:

[0098] Table 1. Operation results of L-M algorithm model

[0099]

[0100]

[0101] It can be seen that the power fault data recognition model improved based on the LM algorithm in this embodiment is feasible and can effectively identify the abnormal operation of the relay protection device. Compared with the conventional neural network model, the power fault data recognition model in this embodiment has better operation speed and accuracy of operation results than the conventional neural network model.

[0102] Example 2

[0103] This embodiment is based on embodiment 1:

[0104] This embodiment provides a power fault data identification system based on an improved LM algorithm, including:

[0105] The data normalization processing module is configured to collect electrical quantity data in the existing relay protection device, extract target characteristic quantities in the electrical quantity data and perform normalization processing, wherein the target characteristic quantities include voltage, current, active power and reactive power of the three-phase electricity of A, B and C;

[0106] The model building and training module is configured to build a power fault data recognition model and perform model training based on the normalized target feature quantity and the LM algorithm, that is, by changing the damping coefficient, the error function of the power fault data recognition model adopts a corresponding convergence method at different stages;

[0107] The power fault data recognition module is configured to collect electrical quantity data from the target relay protection device, extract target feature quantities from the electrical quantity data and perform normalization processing, and then input the trained power fault data recognition model to identify the abnormal operation of the target relay protection device.

[0108] Preferably, the training process of the power fault data identification model includes forward operation and reverse derivation. After the normalized target feature quantity is input, a forward weighted sum is first performed to the output layer; if the result obtained by the output layer does not meet the expected value, a reverse derivation is performed to modify the connection weights and thresholds between the connection layers, and a forward operation is performed again after obtaining the new parameters until the error between the result of the output layer and the expected value meets the requirements.

[0109] Preferably, the error function of the power fault data identification model adopts corresponding convergence methods at different stages by changing the damping coefficient, including: when the connection weights and thresholds of the power fault data identification model are far away from the optimal values, the steepest descent method is used for convergence.

[0110] Preferably, the error function of the power fault data identification model adopts corresponding convergence methods at different stages by changing the damping coefficient, and also includes: when the connection weight and threshold of the power fault data identification model are close to the optimal value, the Gauss-Newton method is used for convergence.

[0111] Preferably, the transfer function adopted by the intermediate layer of the power fault data identification model includes a Sigmoid function, and the transfer function adopted by the output layer includes a linear function.

[0112] Example 3

[0113] This embodiment is based on embodiment 1:

[0114] This embodiment provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the improved power fault data identification method based on the LM algorithm of Embodiment 1 are implemented. The computer program may be in source code form, object code form, executable file, or some intermediate form.

[0115] Example 4

[0116] This embodiment is based on embodiment 1:

[0117] This embodiment provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the improved power fault data identification method based on the LM algorithm of Embodiment 1. The computer program may be in source code form, object code form, executable file, or some intermediate form, etc. The storage medium includes: any entity or device capable of carrying computer program code, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content contained in the storage medium may be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the storage medium does not include electric carrier signals and telecommunication signals.

[0118] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity of description, they are expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the order of the actions described, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

Claims

1. A method for identifying power fault data based on an improved LM algorithm, characterized in that: include: Data normalization processing: Collect electrical quantity data from existing relay protection devices, extract target feature quantities from the electrical quantity data and perform normalization processing. The target feature quantities include voltage, current, active power and reactive power of three-phase electricity A, B and C; Model construction and training: Construct a power fault data recognition model, and perform model training based on the normalized target feature quantity and LM algorithm, that is, by changing the damping coefficient, the error function of the power fault data recognition model adopts the corresponding convergence method at different stages; Power fault data recognition: Collect electrical quantity data from the target relay protection device, extract target feature quantities from the electrical quantity data and perform normalization processing, and then input the trained power fault data recognition model to identify the abnormal operation of the target relay protection device.

2. According to claim 1, a method for identifying power fault data based on an improved LM algorithm is characterized in that: The training process of the power fault data recognition model includes forward operation and reverse derivation. After the normalized target feature quantity is input, a forward weighted sum is first performed to the output layer; if the result obtained by the output layer does not meet the expected value, a reverse derivation is performed to modify the connection weights and thresholds between the connection layers, and a forward operation is performed again after obtaining new parameters until the error between the result of the output layer and the expected value meets the requirements.

3. The method for identifying power fault data based on the improved LM algorithm according to claim 1 is characterized in that: The error function of the power fault data identification model adopts corresponding convergence methods at different stages by changing the damping coefficient, including: when the connection weight and threshold of the power fault data identification model are far away from the optimal value, the steepest descent method is used for convergence.

4. The method for identifying power fault data based on the improved LM algorithm according to claim 3 is characterized in that: The method of changing the damping coefficient so that the error function of the power fault data identification model adopts corresponding convergence methods at different stages also includes: when the connection weight and threshold of the power fault data identification model are close to the optimal value, using the Gauss-Newton method for convergence.

5. The method for identifying power fault data based on the improved LM algorithm according to claim 1 is characterized in that: The transfer function adopted by the intermediate layer of the power fault data identification model includes a Sigmoid function, and the transfer function adopted by the output layer includes a linear function.

6. An improved power fault data identification system based on LM algorithm, characterized in that: include: A data normalization processing module is configured to collect electrical quantity data in an existing relay protection device, extract target characteristic quantities in the electrical quantity data and perform normalization processing, wherein the target characteristic quantities include voltage, current, active power and reactive power of three-phase electricity A, B and C; The model building and training module is configured to build a power fault data recognition model and perform model training based on the normalized target feature quantity and the LM algorithm, that is, by changing the damping coefficient, the error function of the power fault data recognition model adopts a corresponding convergence method at different stages; The power fault data identification module is configured to collect electrical quantity data from the target relay protection device, extract target feature quantities from the electrical quantity data and perform normalization processing, and then input the trained power fault data identification model to identify the abnormal operation of the target relay protection device.

7. The power fault data identification system based on the improved LM algorithm according to claim 6 is characterized in that: The training process of the power fault data recognition model includes forward operation and reverse derivation. After the normalized target feature quantity is input, a forward weighted sum is first performed to the output layer; if the result obtained by the output layer does not meet the expected value, a reverse derivation is performed to modify the connection weights and thresholds between the connection layers, and a forward operation is performed again after obtaining new parameters until the error between the result of the output layer and the expected value meets the requirements.

8. The power fault data identification system based on the improved LM algorithm according to claim 6 is characterized in that: The error function of the power fault data identification model adopts corresponding convergence methods at different stages by changing the damping coefficient, including: when the connection weight and threshold of the power fault data identification model are far away from the optimal value, the steepest descent method is used for convergence.

9. The power fault data identification system based on the improved LM algorithm according to claim 8 is characterized in that: The method of changing the damping coefficient so that the error function of the power fault data identification model adopts corresponding convergence methods at different stages also includes: when the connection weight and threshold of the power fault data identification model are close to the optimal value, using the Gauss-Newton method for convergence.

10. The power fault data identification system based on the improved LM algorithm according to claim 6, characterized in that: The transfer function adopted by the intermediate layer of the power fault data identification model includes a Sigmoid function, and the transfer function adopted by the output layer includes a linear function.

Images