Distribution network automation equipment group fault diagnosis method, system, equipment and medium

Abstract

The invention discloses a distribution network automation equipment group fault diagnosis method and system, equipment and a medium, and relates to the technical field of distribution network automation, and the method comprises the steps: obtaining fault data of a distribution network automation equipment group, carrying out the nonlinear feature extraction of the fault data, inputting the nonlinear feature as an input signal into an echo state network for training, and obtaining an echo state network; and an output value is obtained through a trainable activation function based on differential homeomorphism, parameter optimization is performed on the output value by adopting a tangent search algorithm based on a discrete fractional order method, a trained echo state network is obtained, and fault diagnosis is performed on the distribution network automation equipment group. According to the method, a fixed activation function is replaced by an activation function containing trainable parameters, although the trainable parameters of a hidden layer are increased, the fitting ability and learning ability of the network are greatly improved, and through adjustment of fractional calculus, the convergence speed of the algorithm can be increased, and the robustness of the algorithm is improved. And the speed and the effect of fault diagnosis of the distribution network automation equipment group are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network automation, and in particular to a distribution network automation equipment group fault diagnosis method, system, equipment and medium. Background Art

[0002] As a key component of the power grid, the distribution network directly serves the vast number of electricity users and is an important bridge connecting energy production and consumption. The distribution network is entering a new stage of development. It needs to take into account efficiency and benefits while ensuring safety and quality, and move towards a smart distribution network that is safer, more reliable, green, intelligent, friendly, interactive, and economical and efficient. In this process, distribution automation has become an indispensable part of smart grid construction, playing a vital role in monitoring the operating status of the distribution network and improving power supply reliability.

[0003] Since the distribution automation system involves a large number of terminal devices, the integration and debugging of these devices requires close cooperation between the master station and the site. Considering the complex internal structure and numerous communication lines of outdoor switchgear, it is especially necessary for both parties to maintain efficient communication to solve problems during large-scale deployment. In particular, the demand for fault detection of key components such as switch cabinets and distribution automation terminals in different application scenarios is growing, but it still mainly relies on manual inspection methods, which is not only time-consuming and labor-intensive but also costly. It may also lead to increased unplanned power outages and increased difficulty in cross-departmental coordination, which directly affects the overall power supply stability.

[0004] The fault signals of distribution network automation equipment groups are usually expressed in physical quantities such as current and voltage. These signals are important parameters reflecting the operating status of the equipment and are of great significance for timely detection and diagnosis of faults. The current and voltage data generated by these devices when a fault occurs have nonlinear characteristics, and traditional time domain feature extraction technologies, such as mean, root mean square, kurtosis and skewness, as well as frequency domain feature extraction technologies, such as power spectrum analysis, envelope spectrum analysis and fast Fourier transform, are difficult to effectively analyze such nonlinear signals. In addition, due to the complex nesting of hidden layer parameters of traditional neural network models, their loss function becomes a strongly nonlinear non-convex function, which makes global optimization extremely difficult. Usually, only the optimization method of gradient information can be relied on to find local suboptimal solutions, which will affect the learning effect of the network and lead to inaccurate fault identification and speed difference when diagnosing equipment faults. Summary of the invention

[0005] The purpose of the present invention is to provide a distribution network automation equipment group fault diagnosis method, system, equipment and medium to address the deficiencies of the above-mentioned prior art, so as to solve the problem in the prior art that due to the complex nesting of hidden layer parameters of the traditional neural network model, its loss function becomes a strongly nonlinear non-convex function, which makes global optimization extremely difficult. Usually, only the optimization method relying on gradient information can be used to find local suboptimal solutions, which will affect the learning effect of the network and lead to inaccurate fault identification when diagnosing equipment faults.

[0006] The present invention specifically provides the following technical solutions:

[0007] A method for diagnosing faults of a distribution network automation equipment group comprises the following steps:

[0008] Obtain fault data of distribution network automation equipment group and perform nonlinear feature extraction on the fault data;

[0009] The nonlinear features are input into the echo state network, and the internal state of the reserve pool at the next moment is updated through the trainable activation function DiTAC based on differential homeomorphism and the connection weights of the echo state network input layer and the internal state of the reserve pool. The fault output value is obtained through the internal state of the reserve pool at the next moment and the weight of the output layer of the echo state network. The tangent search algorithm based on the discrete fractional-order Caputo method is used to optimize the output layer weights and the trainable parameters of DiTAC. The echo state network is updated with the optimized parameters to obtain a trained echo state network.

[0010] Inputting the fault data of the distribution network automation equipment group collected in real time into the trained echo state network to obtain the optimal fault output value, and outputting the optimal fault output value as the fault category, and performing fault diagnosis on the distribution network automation equipment group, wherein the nonlinear feature extraction of the fault data includes:

[0011] The fault data is coarse-grained to obtain a coarse-grained sequence; the specific expression is:

[0012]

[0013] in, is the kth coarse-grained sequence when the scale factor of the time series is τ, 1≤k≤τ, where the time series is the time series of the fault data X={x1,x2,…,x T};

[0014] The composite multi-scale fractional reverse diffusion entropy algorithm CMFRDE is used to obtain each scale factor τ. The specific expression is:

[0015]

[0016] Among them, x is the time variable, c is the number of categories, m is the vector dimension, d is the time delay, and α is the fractional order;

[0017] The nonlinear characteristics of fault data are extracted by CMFRDE.

[0018] Preferably, the method of updating the internal state of the reserve pool at the next moment by connecting the weight of the input layer of the echo state network and the internal state of the reserve pool based on the trainable activation function DiTAC based on differential homeomorphism, and obtaining the fault output value by connecting the internal state of the reserve pool at the next moment with the weight of the output layer of the echo state network, comprises:

[0019] Get the input signal at time n+1, and update the internal state x(n+1) of the echo state network's reserve pool through the input signal; the specific expression is:

[0020] x(n+1)=f DiTAC (ω in u(n+1)+ωx(n));

[0021] The output equation is obtained through the updated internal state of the reservoir; the specific expression is:

[0022] y(n+1)=f out (ω out x(n+1));

[0023] Among them, ω in is the connection weight matrix of the echo state network input layer, with a dimension of N×K; ω is the weight connection matrix of the reserve pool, with a dimension of N×N; ω out is the weight connection matrix of the output layer of the echo state network, with a dimension of L×N, f DiTAC (·) and f out (·) are the activation functions of the output layer of the reservoir and the echo state network, respectively. The activation function of the reservoir is f DiTAC (·) is a DiTAC activation function, and the activation function of the output layer is f out (·) is the identity function, and y(n+1) is the output value.

[0024] Preferably, after obtaining the output equation through the updated internal state of the reserve pool, the method further includes:

[0025] Training adjusts the weights ω of the output layer out The DiTAC activation function can train the parameter θ to obtain the optimization objective function, which is expressed as:

[0026]

[0027] Among them, s is the number of training samples, y i The network predicts the label for the i-th training sample, is the true label of the i-th training sample;

[0028] The optimal output equation is obtained by optimizing the objective function.

[0029] Preferably, the method of optimizing the output layer weights and the trainable parameters of DiTAC using a tangent search algorithm based on a discrete fractional-order Caputo method comprises:

[0030] Generate a random initial population X0 within the boundary of the solution space of the optimization objective function; the specific expression is:

[0031] X0 = lb + (ub - lb). * rand (D);

[0032] Among them, lb and ub are the lower and upper limits of the problem, the rand function generates uniformly distributed random numbers in the range [0,1], and D is the dimension of the problem; the solution space is the potential space where the output layer weights and the trainable parameters of the DiTAC activation function exist in the optimization objective function, and the output layer weights and the trainable parameters of the DiTAC activation function are the optimal parameters solved, that is, the optimal solution;

[0033] Perform random local walks in the random initial population to obtain relevant solutions, and replace the variables in the relevant solutions with the values ​​of the corresponding variables in the current optimal solution; the specific expression is:

[0034]

[0035] The first-order difference is replaced by the discrete fractional Caputo derivative form, specifically:

[0036]

[0037] Among them, D 1 [·] is the first-order difference, is the new solution of the i-th component, is the optimal solution among all solutions obtained before and after the tth iteration, the optimal solution of the i-th component, is the backward difference of the t-j+1th iteration of the ith component, j is the number of truncation, v is the fractional order, step is the size of the move, tan(θ) is the direction of the move, and L is the number of neurons in the output layer;

[0038] When the value of each solution obtained exceeds the problem limits lb and ub, each solution obtained is repaired; the specific expression is:

[0039] X(X<lb)=rand*(ub-lb)+lb;

[0040] X(X>ub)=rand*(ub-lb)+lb;

[0041] In the random initial population, a global random walk is performed with variable step length and tangent flight. When the h value is close to 0, the solution obtained is closest to the current solution. The global and local solutions are combined by exploring the search equation to obtain the value of the corresponding variable after the combination, and this value is used as the optimal solution for parameter optimization. The specific expression is:

[0042]

[0043] Preferably, the fault categories include short circuit, open circuit, voltage abnormality, wrong wiring inside the secondary switch cabinet, and loose wiring inside the secondary switch cabinet.

[0044] The present invention provides a distribution network automation equipment group fault diagnosis system, comprising:

[0045] The acquisition module is used to acquire the fault data of the distribution network automation equipment group and perform nonlinear feature extraction on the fault data;

[0046] A training module is used to input nonlinear features into an echo state network, update the internal state of the reserve pool at the next moment through the trainable activation function DiTAC based on differential homeomorphism and the connection weight of the echo state network input layer and the internal state of the reserve pool, obtain the fault output value through the internal state of the reserve pool at the next moment and the weight of the output layer of the echo state network, and use the tangent search algorithm based on the discrete fractional Caputo method to optimize the output layer weight and the trainable parameters of DiTAC, update the echo state network through the optimized parameters, and obtain a trained echo state network;

[0047] The diagnosis module is used to input the fault data of the distribution network automation equipment group collected in real time into the trained echo state network to obtain the optimal fault output value, and output the optimal fault output value as the fault category to perform fault diagnosis on the distribution network automation equipment group.

[0048] The present invention provides a computer device, comprising a memory and a processor, wherein a program is stored in the memory, and when the program is executed by the processor, the processor executes the steps of the above-mentioned distribution network automation equipment group fault diagnosis method.

[0049] The present invention provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned method for diagnosing faults of a distribution network automation equipment group are implemented.

[0050] Compared with the prior art, the present invention has the following significant advantages:

[0051] The present invention replaces the fixed activation function in the echo state network with a trainable activation function based on differential homeomorphism, performs a series of processing on nonlinear features, obtains the output value of the echo state network, improves the fitting ability and learning ability of the network, and simultaneously optimizes the output layer weights and trainable parameters of the echo state network through a tangent search algorithm based on a discrete fractional-order Caputo method, adjusts the parameters by fractional-order calculus, improves the global search ability of the echo state network, thereby realizing the processing of fault data by an optimal echo state network, obtaining an optimal output, and improving the speed and effect of fault diagnosis of distribution network automation equipment groups. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is an overall flow chart of a method for diagnosing faults in a distribution network automation equipment group according to the present invention;

[0053] Figure 2 This is a diagram of the DESN network structure in the present invention. DETAILED DESCRIPTION

[0054] The following is a clear and complete description of the technical solutions of the embodiments of the present invention in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0055] Through artificial intelligence technology, the distribution network automation equipment group can more accurately perceive and analyze the working status and health of the distribution equipment, thereby achieving more efficient and accurate data collection and analysis, timely discovering potential faults and abnormal conditions, and improving the accuracy of fault diagnosis and the efficiency of detection and maintenance. Using data-driven methods combined with artificial intelligence technology, the data of the distribution network automation equipment group is analyzed and processed.

[0056] By establishing AI intelligent diagnosis and analysis algorithms and quantitative models, real-time monitoring, fault diagnosis and prediction of equipment operation status can be achieved. This can improve the operation and maintenance personnel's understanding of equipment status so that problems can be quickly discovered and solved, and also provide a basis for intelligent operation and maintenance. Table 1 shows common fault categories and their causes.

[0057] Table 1 Common fault categories and their causes

[0058]

[0059] The present invention proposes a composite multi-scale fractional reverse spread entropy, which is based on a complexity measurement method of fractional calculus and entropy theory, and combines the comprehensiveness of multi-scale, the flexibility of fractional calculus, historical dependence and the advantages of reverse spread entropy.

[0060] The main idea of ​​the present invention is to use the composite multi-scale fractional order reverse spread entropy algorithm to extract comprehensive features from the fault data of the distribution network automation equipment group, and then input the obtained features into the echo state network based on the differential homeomorphism trainable activation function for fault multi-classification processing, in which the optimizer of the algorithm adopts the tangent search algorithm based on the discrete fractional order Caputo method. Figure 1 shown.

[0061] like Figure 1 As shown, the present invention provides a distribution network automation equipment group fault diagnosis method to be described below, which specifically includes the following steps:

[0062] Step S1: Acquire fault data of a distribution network automation equipment group, and perform nonlinear feature extraction on the fault data.

[0063] The present invention uses a composite multi-scale fractional order reverse spread entropy algorithm to extract features from fault data, without decomposing or transforming the data, and directly measures the complexity of the data. This overcomes the shortcomings of traditional time and frequency domain feature extraction methods that often cannot effectively extract hidden fault features. In addition, this parameter-free measure can accurately estimate in a computationally efficient manner, and further considers fractional order information, and can measure the dynamic changes of fractional order time series, so that more comprehensive fault features can be extracted from fault data.

[0064] Fractional Reverse Dispersion Entropy (FRDE):

[0065] Reverse Dispersion Entropy (RDE) is a concept used in signal processing and nonlinear dynamics analysis, especially in fault diagnosis and signal complexity evaluation. As a new nonlinear dynamic analysis method, RDE is based on the theory of permutation entropy (PE), combines the advantages of dispersion entropy (DE) and reverse permutation entropy (RPE), and performs well in stability and discrimination.

[0066] For a time series X = {x(1), x(2), …, x(T)} of length T, the steps for calculating fractional RDE are as follows:

[0067] (1) Through the normal distribution function:

[0068]

[0069] Map the time series X = {x(i), i = 1, 2, ..., T} to Y = {y(i), i = 1, 2, ..., T}, where y(i) ranges from 0 to 1, where μ is the expected value and σ is the average value. 2 is the variance.

[0070] (2) Through linear transformation:

[0071]

[0072] Map the new sequence Y = {y(i), i = 1, 2, ..., T} to the range of Z = {z(i), i = 1, 2, ..., T}, where int is the rounding function, c is the number of categories, and z(i) is a positive integer from 1 to c.

[0073] (3) Phase space reconstruction:

[0074] Reconstruct Z into L embedding vectors with a delay of d and an embedding dimension of m. The matrix composed of all embedding vectors is:

[0075]

[0076] In the formula, the number of embedded vectors L is T-(m-1)d.

[0077] (4) Calculate the dispersion pattern:

[0078] Each embedding vector can be mapped into a scattering pattern π. The total number of possible scattering patterns is c m Because each embedding vector has m components, each component can be an integer between 1 and c.

[0079] (5) Calculate the probability of each scattering mode:

[0080] The relative probability of the i-th scattering mode can be expressed as

[0081]

[0082] Where: P(π i ) is the ratio of the number of the ith scatter pattern to the number of embedded vectors; Number{π i} refers to the number of embedding vectors mapped to the i-th scatter pattern.

[0083] (6) Calculate fractional-order RDE:

[0084]

[0085] Among them, Γ(·) is the gamma function, ψ(·) is the digamma function, and α is the fractional order.

[0086] Composite multi-scale fractional inverse diffusion entropy algorithm:

[0087] RDE has great advantages in analyzing the randomness and dynamic mutation behavior of time series at a single scale. In order to measure the complexity of time series at different scales, multi-scale analysis methods were born. However, multi-scale calculations based on coarse-grained process definitions depend on the length of the time series and do not consider the relationship between the new coarse-grained sequences after calculation, resulting in information loss. In response to the above problems, CMFRDE is proposed in combination with a composite multi-scale method.

[0088] Perform nonlinear feature extraction on fault data, including:

[0089] The fault data is coarse-grained to obtain a coarse-grained sequence; the specific expression is:

[0090]

[0091] in, is the kth coarse-grained sequence when the scale factor of the time series is τ, 1≤k≤τ, where the time series is the time series of the fault data X={x1,x2,…,x T}.

[0092] The composite multi-scale fractional reverse diffusion entropy algorithm CMFRDE is used to obtain each scale factor τ. The specific expression is:

[0093]

[0094] Among them, x is the time variable, c is the number of categories, m is the vector dimension, d is the delay, and α is the fractional order.

[0095] The nonlinear characteristics of fault data are extracted by CMFRDE.

[0096] From the CMFRDE algorithm, we can see that when the scale factor is τ, the original sequence is continuously divided into coarse-grained sequences of length τ according to the initial points [1,τ], and then the average value of each small sequence is calculated, and then τ new coarse-grained sequences are obtained in order. After calculating the FRDE value of the new sequence, all entropy values ​​are averaged. This not only reduces the fluctuation of entropy value with the increase of scale factor, but also has better stability than the traditional coarse-grained multi-scale process.

[0097] Step S2: Input the nonlinear features into the echo state network, update the internal state of the reserve pool at the next moment through the trainable activation function DiTAC based on differential homeomorphism and the connection weights of the echo state network input layer and the internal state of the reserve pool, and obtain the fault output value through the internal state of the reserve pool at the next moment and the output layer weights of the echo state network, and use the tangent search algorithm based on the discrete fractional-order Caputo method to optimize the output layer weights and the trainable parameters of DiTAC, update the echo state network with the optimized parameters, and obtain the trained echo state network.

[0098] DiTAC Echo State Networks (DESN) based on diffeomorphic trainable activation functions:

[0099] The echo state network is an artificial recurrent neural network. The network is characterized by the presence of feedback loops in the internal neurons, which can maintain an activation state even in the absence of input, showing a memory effect.

[0100] Diffeomorphism-based TrainableActivationFunction (DiTAC):

[0101] Activation functions (AFs) play an important role in the success of deep neural networks because they give the latter nonlinear characteristics. In fact, activation functions are crucial for the network to approximate almost arbitrarily complex functions, learn meaningful feature representations, and achieve high predictive performance. In the framework of artificial neural network neurons, the role of the activation function is to put the weighted sum into a nonlinear function for nonlinearization. Its essence is: the weighted sum can be regarded as a linear hyperplane, which divides the spatial data into two categories and can be regarded as a binary linear classifier. However, most real-world problems are nonlinear, and linear classifiers are not universally meaningful at the application level. Therefore, it is necessary to make the linear classifier nonlinear, and activation functions need to be used. The choice of nonlinear activation function in the neural network architecture is crucial and seriously affects the performance of the neural network. The ideal activation function needs to meet many conditions, such as nonlinearity, non-saturation, and almost everywhere differentiability. Existing traditional activation functions all have their own shortcomings.

[0102] In addition to their nonlinear properties, activation functions have various properties that directly affect the performance of the network. Traditional activation functions, such as Logistic Sigmoid and TanhUnit, map input values ​​to a small range, which may cause the network gradient to approach zero, thus affecting training performance. The Rectified Linear Unit (ReLU) and its variants (e.g., LReLU and PReLU) partially address this problem by mapping the input to a range that is unrestricted in one or both directions. Exponential activation functions such as ELU inherit the advantages of ReLU but also push the response of the activation function towards zero mean to improve performance. In general, fixed activation functions (AFs) have limited nonlinearity (and therefore limited expressive power) and impose a learning bias on the network. Therefore, it is challenging to tune them to different problem types and data complexities. Trainable activation functions (TAFs) such as PReLU, Swish, and PELU adjust the shape of the standard fixed activation function by adding several learnable parameters. Such functions have only achieved a minor improvement in expressive power, as the performance of these TAFs is often similar to their basic non-trainable activation functions. Maxout units provide an alternative approach to activation functions. Despite significant improvements over ReLU in solving classification tasks, the number of parameters in the Maxout layer increases as the number of neurons in the network increases. A diffeomorphism is a differentiable, reversible function that has a differentiable inverse. Diffeomorphism-based Trainable Activation function (DiTAC) is a differentiable parameterized TAF based on efficient and highly-expressive parametric diffeomorphisms (CPAB). Although DiTAC only adds negligible trainable parameters, it is still very expressive. This is in contrast to existing trainable activation functions such as PReLU, Swish, and PELU, which can typically only learn a certain shape or only learn convex functions, while DiTAC is able to learn a variety of shapes to adapt to different types of data and tasks.

[0103] Assume T θ are diffeomorphisms parameterized by θ. However, using diffeomorphisms usually involves expensive computations. The main reason for adopting CPAB transforms as the family of diffeomorphisms to be used in the DiTAC method is that they are both expressive and efficient. All CPAB transforms are assumed to be one-dimensional.

[0104] set up is a finite interval, let From Ω to These functions are also piecewise affine with respect to some fixed partition of Ω into subintervals. is a finite-dimensional linear space. set up set up represents the parameterized by θ If x∈Ω, then Defines a trajectory So that φ θ (x,0)=x and φ θ (x,t). By integration The CPAB transformation space definition and its integral equation obtained by the elements in are:

[0105]

[0106] The equivalent ordinary differential equation is:

[0107] dφ θ (x,t) / dt=v θ (φ θ (x,t)). (9);

[0108] Each is monotonically increasing and diffeomorphic.

[0109] Unlike existing TAFs, which assign a parameter to each input channel, DiTAC derives its expressive power from the expressive power of the CPAB transform. The expressive power (and differentiability) of DiTAC is apparent from the first data transformation experienced by each neuron. Importantly, the availability of closed-form expressions for the CPAB transform and its gradients allows DiTAC to be easily used as a drop-in replacement for any activation function in any neural network architecture. θ is defined on a finite interval [a, b]. Its range is also a finite interval, which may or may not overlap with Ω. Since some AFs inputs may fall outside [a, b], the mainstream version of DiTAC used in this invention is to convert T θ Combined with GELU, GELU is a widely used AFs in state-of-the-art models. GELU is given by GELU(x) = x·Φ(x), where Φ(x) is the cumulative distribution function of the standard normal distribution. The mainstream form of DiTAC is:

[0110]

[0111] in:

[0112]

[0113] In view of the superiority of the DiTAC activation function, the present invention will use it to replace the traditional activation function of the extreme learning machine and propose an echo state network based on a differential homeomorphism trainable activation function.

[0114] DESN algorithm principle:

[0115] Echo State Networks (ESN) are three-layer artificial recurrent neural networks, consisting of an input layer, a reservoir (also known as a hidden layer), and an output layer. Each layer is connected to other layers through neurons. Each connection has a weight, and all connections constitute a weight matrix. The neurons inside the reservoir are sparsely connected, have feedback loops, and have memory capabilities. Unlike traditional RNNs, once the connection weights of the input layer and the reservoir of the echo state network are initialized, they will no longer be adjusted during the entire training process. Only the connection weights of the output layer need to be adjusted. This feature greatly simplifies the training of the network and reduces the time and space overhead of training.

[0116] The structure diagram of the classic echo state network consists of an input layer, a reservoir, and an output layer. Assuming that DESN has K input layer neurons, N internal neurons in the reservoir, and L output layer neurons, then at time n, the input variable u(n), the internal state of the reservoir x(n), and the output variable y(n) are respectively expressed as follows:

[0117] u(n)=[u1(n),u2(n),…,u K (n)] T (12);

[0118] x(n)=[x1(n),x2(n),…,x N (n)] T (13);

[0119] y(n)=[y1(n),y2(n),…,y L (n)] T (14);

[0120] The internal state of the reservoir at the next moment is updated by connecting the weights of the input signal and the internal state of the reservoir based on differential homeomorphism, and the output value is obtained by the internal state of the reservoir at the next moment and the weight of the output layer, including:

[0121] Get the input signal at time n+1, and update the internal state x(n+1) of the echo state network's reserve pool through the input signal; the specific expression is:

[0122] x(n+1)=fDiTAC (ω in u(n+1)+ωx(n)) (15);

[0123] The output equation is obtained through the updated internal state of the reservoir; the specific expression is:

[0124] y(n+1)=f out (ω out x(n+1)) (16);

[0125] Among them, ω in is the connection weight matrix of the echo state network input layer, with a dimension of N×K; ω is the weight connection matrix of the reserve pool, with a dimension of N×N; ω out is the weight connection matrix of the output layer of the echo state network, with a dimension of L×N, f DiTAC (·) and f out (·) are the activation functions of the output layer of the reservoir and the echo state network, respectively. The activation function of the reservoir is f DiTAC (·) is a DiTAC activation function, and the activation function of the output layer is f out (·) is the identity function, and y(n+1) is the output value.

[0126] y(n+1)=ω out x(n+1) (17);

[0127] When the reserve pool neurons use leaky integration neurons, the performance of the echo state network is improved. At this time, the internal state of the reserve pool is updated as follows:

[0128] x(n+1)=(1-α)x(n)+αf DiTAC (ω in u(n+1)+ωx(n)) (18);

[0129] Where α is the leakage rate, which can be considered as the speed of updating the reserve pool. When α = 1, formula (18) becomes formula (15), and the leaky integration neuron becomes an ordinary analog neuron.

[0130] After obtaining the output equation through the updated internal state of the reservoir, it also includes:

[0131] After DESN is initialized, it only needs to train and adjust the weights ω of the output layer out The DiTAC activation function can train the parameters θ, while the connection weights W of the input layer in The connection weight W inside the reservoir is no longer adjusted during the training process. Therefore, we can simply solve ω out The process of and θ is considered as a multivariate linear regression process. The specific expression is:

[0132]

[0133] Among them, s is the number of training samples, y i The network predicts the label for the i-th training sample, is the true label of the i-th training sample. The label is the fault category label predicted by the entire network. In order to suppress the disadvantage that the L2 norm of the traditional multivariate linear regression objective function is easy to increase or decrease the error, the present invention replaces the L2 norm with the L1 norm. The unknown parameter ω of the optimization objective function out After the DiTAC activation function is embedded, θ is equivalent to a non-smooth non-convex optimization problem. The network structure diagram is shown in Figure 2 shown.

[0134] The optimal output equation is obtained by optimizing the objective function.

[0135] Tangent search algorithm based on discrete fractional-order Caputo method:

[0136] Discrete fractional-order Caputo method:

[0137] Caputo form of fractional backward differencing (C-FOBD):

[0138]

[0139] The finite memory window version of C-FOBD with window size can be written as:

[0140]

[0141] For 0<v<1, equation (21) can be written as:

[0142]

[0143] The tangent search algorithm based on the discrete fractional-order Caputo method is used to optimize the output layer weights and the trainable parameters of DiTAC, including:

[0144] Initialize the initial population: Like other population-based optimization algorithms, CTSA first generates a random initial population X0 within the boundary of the solution space of the optimization objective function; the specific expression is:

[0145] X0=lb+(ub-lb).*rand(D) (23);

[0146] Among them, lb and ub are the lower and upper limits of the problem, the rand function generates uniformly distributed random numbers in the range [0,1], and D is the dimension of the problem; the solution space is the potential space where the output layer weights and the trainable parameters of the DiTAC activation function exist in the optimization objective function. The output layer weights and the trainable parameters of the DiTAC activation function are the optimal parameters solved, that is, the optimal solution, which can minimize the optimization objective function (that is, the cumulative error between the true label and the predicted label).

[0147] Enhanced search: In enhanced search, CTSA first performs random local walks in a random initial population to obtain relevant solutions, and replaces the variables in the relevant solutions with the values ​​of the corresponding variables in the current optimal solution. The proportion of replaced variables is 20% for problems with dimensions greater than 4 and 50% for problems with dimensions less than or equal to 4. The specific expression is:

[0148]

[0149] Rewritten as discrete form:

[0150]

[0151] Using the discrete fractional Caputo derivative form to replace the first-order difference is equivalent to:

[0152]

[0153] Among them, D 1 [·] is the first-order difference, is the new solution of the i-th component, is the optimal solution among all solutions obtained before and after the tth iteration (the optimal solution of the i-th component), is the backward difference of the t-j+1th iteration of the ith component, j is the number of truncation, v is the fractional order, step is the size of the movement, tan(θ) is the direction of the movement, and L is the number of neurons in the output layer.

[0154]

[0155] Therefore, the solution obtained The similarity ratio with the current best solution is less than 50%, which helps to strengthen the local search around the current solution. When the value of each solution obtained exceeds the problem bounds lb and ub, each solution obtained is repaired; the specific expression is:

[0156] X(X<lb)=rand*(ub-lb)+lb (28);

[0157] X(X>ub)=rand*(ub-lb)+lb (29);

[0158] Exploration search: Unlike local search methods, population-based metaheuristics have great exploration capabilities, thanks to global random walks. In this algorithm, CTSA performs global random walks with variable step sizes and tangent flights in a random initial population, and the tangent function helps to effectively explore the search space. In fact, h values ​​close to π / 2 will increase the tangent value, so that the solution obtained will be far away from the current solution. When the h value is close to 0, the solution obtained is closest to the current solution, and the global and local are combined through the exploration search equation to obtain the value of the corresponding variable after the combination, which is used as the optimal solution for parameter optimization; the specific expression is:

[0159]

[0160] The exploratory search equation is applied to each variable with probability 1 / D, where D is the dimensionality of the problem.

[0161] And in each iteration, a search agent is randomly selected to obtain a new random solution, and the new random solution replaces the worst solution with a probability lower than the threshold; the specific expression is:

[0162]

[0163] The worst solution is the worst value among the values ​​of the corresponding variables after combination.

[0164] Step S3: input the fault data of the distribution network automation equipment group collected in real time into the trained echo state network to obtain the optimal fault output value, and output the optimal fault output value as the fault category to perform fault diagnosis on the distribution network automation equipment group.

[0165] Fault categories include equipment short circuit, open circuit, abnormal voltage, incorrect wiring inside the secondary switch cabinet, and loose wiring inside the secondary switch cabinet.

[0166] Based on the above method, the present invention provides a distribution network automation equipment group fault diagnosis system, including: a collection module, a training module and a diagnosis module.

[0167] Among them, the acquisition module is used to obtain the fault data of the distribution network automation equipment group and perform nonlinear feature extraction on the fault data; the training module is used to input the nonlinear features into the echo state network, and update the internal state of the reserve pool at the next moment through the trainable activation function DiTAC based on differential homeomorphism and the echo state network input layer connection weights and the internal state of the reserve pool, and obtain the fault output value through the internal state of the reserve pool at the next moment and the output layer weight of the echo state network, and use the tangent search algorithm based on the discrete fractional-order Caputo method to optimize the output layer weights and the trainable parameters of DiTAC, and update the echo state network with the optimized parameters to obtain a trained echo state network; the diagnosis module is used to input the fault data of the distribution network automation equipment group collected in real time into the trained echo state network to obtain the optimal fault output value, and output the optimal fault output value as the fault category to perform fault diagnosis on the distribution network automation equipment group.

[0168] The present invention also provides a computer device, including a memory and a processor, wherein a program is stored in the memory, and when the program is executed by the processor, the processor executes the steps of a distribution network automation equipment group fault diagnosis method.

[0169] In accordance with the disclosed embodiments, a computing device may communicate with one or more external devices (e.g., keyboards, pointing devices, Bluetooth communications, etc.), or with any device (e.g., routers, modems, etc.) that enables a computing device to communicate with one or more other computing devices.

[0170] The present invention also provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of a distribution network automation equipment group fault diagnosis method are implemented.

[0171] According to the disclosed embodiments, the storage medium may be a non-volatile computer-readable storage medium, such as but not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, the storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.

[0172] The above content is a further detailed description of the present invention in combination with a specific preferred embodiment. For technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as belonging to the protection scope of the present invention.

Claims

1. A method for diagnosing faults in a distribution network automation equipment group, characterized in that: include: Obtain fault data of distribution network automation equipment group and perform nonlinear feature extraction on the fault data; The nonlinear features are input into the echo state network, and the internal state of the reserve pool at the next moment is updated through the trainable activation function DiTAC based on differential homeomorphism and the connection weights of the echo state network input layer and the internal state of the reserve pool. The fault output value is obtained through the internal state of the reserve pool at the next moment and the weight of the output layer of the echo state network. The tangent search algorithm based on the discrete fractional-order Caputo method is used to optimize the output layer weights and the trainable parameters of DiTAC. The echo state network is updated with the optimized parameters to obtain a trained echo state network. The fault data of the distribution network automation equipment group collected in real time is input into the trained echo state network to obtain the optimal fault output value, and the optimal fault output value is output as the fault category to perform fault diagnosis on the distribution network automation equipment group.

2. A distribution network automation equipment group fault diagnosis method as claimed in claim 1, characterized in that: The nonlinear feature extraction of fault data includes: The fault data is coarse-grained to obtain a coarse-grained sequence; the specific expression is: in, is the kth coarse-grained sequence when the scale factor of the time series is τ, 1≤k≤τ, where the time series is the time series of the fault data X={x1,x2,…,x T }; The composite multi-scale fractional reverse diffusion entropy algorithm CMFRDE is used to obtain each scale factor τ. The specific expression is: Among them, x is the time variable, c is the number of categories, m is the vector dimension, d is the time delay, and α is the fractional order; The nonlinear characteristics of fault data are extracted by CMFRDE.

3. A distribution network automation equipment group fault diagnosis method as claimed in claim 1, characterized in that: The method of updating the internal state of the reserve pool at the next moment by connecting the weight of the input layer of the echo state network and the internal state of the reserve pool based on the differential homeomorphism, and obtaining the fault output value by connecting the internal state of the reserve pool at the next moment and the weight of the output layer of the echo state network, includes: Get the input signal at time n+1, and update the internal state x(n+1) of the echo state network's reserve pool through the input signal; the specific expression is: x(n+1)=f DiTAC (ω in u(n+1)+ωx(n)); The output equation is obtained through the updated internal state of the reservoir; the specific expression is: y(n+1)=f out (ω out x(n+1)): Among them, ω in is the connection weight matrix of the echo state network input layer, with a dimension of N×K; ω is the weight connection matrix of the reserve pool, with a dimension of N×N; ω out is the weight connection matrix of the output layer of the echo state network, with a dimension of L×N, f DiTAC (·) and f out (·) are the activation functions of the output layer of the reservoir and the echo state network, respectively. The activation function of the reservoir is f DiTAC (·) is a DiTAC activation function, and the activation function of the output layer is f out (·) is the identity function, and y(n+1) is the output value.

4. A distribution network automation equipment group fault diagnosis method as claimed in claim 3, characterized in that: After obtaining the output equation through the updated internal state of the storage pool, the method further includes: Training adjusts the weights ω of the output layer out The DiTAC activation function can train the parameter θ to obtain the optimization objective function, which is expressed as: Among them, s is the number of training samples, y i The network predicts the label for the i-th training sample, is the true label of the i-th training sample; The optimal output equation is obtained by optimizing the objective function.

5. A distribution network automation equipment group fault diagnosis method as claimed in claim 4, characterized in that: The method uses a tangent search algorithm based on a discrete fractional-order Caputo method to optimize the output layer weights and the trainable parameters of DiTAC, including: Generate a random initial population X0 within the solution space boundary of the optimization objective function; the specific expression is: X0 = lb + (ub - lb). * rand (D); Among them, lb and ub are the lower and upper limits of the problem, the rand function generates uniformly distributed random numbers in the range [0,1], and D is the dimension of the problem; the solution space is the potential space where the output layer weights and the trainable parameters of the DiTAC activation function exist in the optimization objective function, and the output layer weights and the trainable parameters of the DiTAC activation function are the optimal parameters solved, that is, the optimal solution; Perform random local walks in the random initial population to obtain relevant solutions, and replace the variables in the relevant solutions with the values ​​of the corresponding variables in the current optimal solution; the specific expression is: The first-order difference is replaced by the discrete fractional Caputo derivative form, specifically: Among them, D 1 [·] is the first-order difference, is the new solution of the i-th component, is the optimal solution among all solutions obtained before and after the tth iteration, that is, the optimal solution of the i-th component, ΔX i t-j+1 is the backward difference of the t-j+1th iteration of the ith component, j is the number of truncation, v is the fractional order, step is the size of the move, tan(θ) is the direction of the move, and L is the number of neurons in the output layer; When the value of each solution obtained exceeds the problem limits lb and ub, each solution obtained is repaired; the specific expression is: X(X<lb)=rand*(ub-lb)+lb; X(X>ub)=rand*(ub-lb)+lb; In the random initial population, a global random walk is performed with variable step length and tangent flight. When the h value is close to 0, the solution obtained is closest to the current solution. The global and local solutions are combined by exploring the search equation to obtain the value of the corresponding variable after the combination, and this value is used as the optimal solution for parameter optimization. The specific expression is:

6. A distribution network automation equipment group fault diagnosis method as claimed in claim 1, characterized in that: The fault categories include short circuit, open circuit, abnormal voltage of equipment, wrong wiring inside the secondary switch cabinet, and loose wiring inside the secondary switch cabinet.

7. A distribution network automation equipment group fault diagnosis system, characterized in that: include: The acquisition module is used to obtain the fault data of the distribution network automation equipment group and perform nonlinear feature extraction on the fault data; A training module is used to input nonlinear features into an echo state network, update the internal state of the reserve pool at the next moment through the trainable activation function DiTAC based on differential homeomorphism and the connection weight of the echo state network input layer and the internal state of the reserve pool, obtain the fault output value through the internal state of the reserve pool at the next moment and the weight of the output layer of the echo state network, and use the tangent search algorithm based on the discrete fractional Caputo method to optimize the output layer weight and the trainable parameters of DiTAC, update the echo state network through the optimized parameters, and obtain a trained echo state network; The diagnosis module is used to input the fault data of the distribution network automation equipment group collected in real time into the trained echo state network to obtain the optimal fault output value, and output the optimal fault output value as the fault category to perform fault diagnosis on the distribution network automation equipment group.

8. A computer device, characterized in that: It comprises a memory and a processor, wherein a program is stored in the memory, and when the program is executed by the processor, the processor executes the steps of a distribution network automation equipment group fault diagnosis method as claimed in any one of claims 1 to 6.

9. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a distribution network automation equipment group fault diagnosis method described in any one of claims 1 to 6 are implemented.

Images