A power distribution terminal backup power supply life prediction method, device, equipment and medium

Abstract

The present application belongs to the technical field of power distribution network, and particularly relates to a power distribution terminal backup power life prediction method, device, equipment and medium. The present application method inputs the spectrum data of the backup power of the power distribution terminal into a pre-trained life prediction model, and obtains the life prediction result of the backup power of the power distribution terminal. The training method of the life prediction model is as follows: constructing a lightweight network model; constructing a flexible loss function; wherein the flexible loss function includes two loss terms, and the weight of each loss term is determined based on a flexible coefficient; training the lightweight network model based on the training set and the flexible loss function; in the training process, the mayfly algorithm is used to optimize the parameters of the lightweight network model and the flexible coefficient; the life prediction model is obtained after the model training is completed; wherein the mayfly algorithm uses the method of extreme jumping to prevent the population from falling into local optimum, and enhances the stability and robustness of the model.

Description

Technical Field

[0001] The present invention belongs to the technical field of power distribution networks, and in particular relates to a method, device, equipment and medium for predicting the life of a backup power supply for a power distribution terminal. Background Art

[0002] Intelligent distribution network automation terminals include several types, such as DTUs (Distribution Terminal Units) and FTUs (Feeder Terminal Units). These terminals are characterized by their large number, wide distribution, and high maintenance costs, making improving their stability crucial. The instability of distribution network automation terminals primarily stems from their power supply method. They typically draw power from the primary side via three-wire, five-pole PTs (Potential Transformers). However, due to factors such as 10kV line harmonics, PT failures frequently cause power outages in the automation terminals. Therefore, renewable energy backup power sources have become the mainstream backup power supply method for automation terminals. Therefore, accurately predicting the lifespan of backup power supplies for distribution network automation terminals can significantly reduce line inspection frequency, lower maintenance costs, and enhance the intelligence and stability of the distribution network.

[0003] For the problem of backup power supply life prediction, existing technology combines the attention mechanism and ResNet for power supply remaining life prediction. This method has complex modeling and requires a large amount of preliminary battery data, which increases training time and makes the model more prone to overfitting. Summary of the Invention

[0004] The purpose of the present invention is to provide a method, device, equipment and medium for predicting the life of a distribution terminal backup power supply, so as to solve the problem in the prior art of combining the attention mechanism and ResNet for power supply remaining life prediction, which is complex in modeling and requires a large amount of preliminary battery data, resulting in increased training time and more prone to model overfitting.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] In a first aspect, the present invention provides a method for predicting the life of a backup power supply of a power distribution terminal, comprising:

[0007] Obtain backup power spectrum data of distribution terminals;

[0008] Inputting the spectrum data of the backup power supply of the power distribution terminal into a pre-trained life prediction model, the life prediction model outputting a life prediction result of the backup power supply of the power distribution terminal;

[0009] The lifespan prediction model is trained as follows:

[0010] Constructing a lightweight network model; wherein the lightweight network model adopts the Shufflenet neural network;

[0011] Get the training set for model training;

[0012] Constructing a flexible loss function; wherein the flexible loss function includes two loss terms, and the weight of each loss term is determined based on the flexibility coefficient;

[0013] The lightweight network model is trained based on a training set and a flexible loss function; during the training process, the parameters of the lightweight network model and the flexible coefficient are optimized using an ephemeral algorithm; after the model training is completed, the life prediction model is obtained; wherein, the ephemeral algorithm uses an extreme value jump-out method to prevent the population from falling into a local optimum.

[0014] Furthermore, the parameters of the lightweight network model and the flexibility coefficient are optimized using the ephemeral algorithm, including:

[0015] The parameters of the lightweight network model and the flexibility coefficient are used as parameters to be optimized, and the current position of the population is determined according to the parameters to be optimized; the population position is updated to determine whether the current population falls into a local optimum; if it falls into a local optimum, the population is moved to the next position when it falls into the local optimum. Jumped value Replace; if it does not fall into the local optimum, continue to update the position and mate the population;

[0016] After the preset conditions are met, the optimization of the parameters of the lightweight network model and the flexibility coefficient is completed.

[0017] Furthermore, the jump value It is calculated as follows:

[0018]

[0019] in, is the best position of the current iteration, The best position for the next iteration; λ 1、 λ 2 is made of β Parameters of decision; β Represents a binary number; is the current vector position; Represents a random number between -1 and 1; Indicates a random number between -0.5 and 0.5; represents the calculation coefficient; Indicates the average position of the population in the current iteration.

[0020] Furthermore, the flexible loss function is expressed as follows:

[0021]

[0022] in, is a flexible loss function; MSE is the mean square error, MAE is the mean absolute error, is the flexibility coefficient; y i is the actual value; is the predicted value; n is the sample size.

[0023] Furthermore, in the step of obtaining the spectrum data of the backup power supply of the distribution terminal, a terahertz sensor is used to collect the spectrum data of the backup power supply of the distribution terminal.

[0024] In a second aspect, the present invention provides a device for predicting the life of a backup power supply for a power distribution terminal, comprising:

[0025] A data acquisition module is used to obtain spectrum data of standby power supply of distribution terminals;

[0026] A life prediction module, configured to input the spectrum data of the distribution terminal standby power supply into a pre-trained life prediction model, wherein the life prediction model outputs a life prediction result of the distribution terminal standby power supply;

[0027] The lifespan prediction model is trained as follows:

[0028] Constructing a lightweight network model; wherein the lightweight network model adopts the Shufflenet neural network;

[0029] Get the training set for model training;

[0030] Constructing a flexible loss function; wherein the flexible loss function includes two loss terms, and the weight of each loss term is determined based on the flexibility coefficient;

[0031] The lightweight network model is trained based on a training set and a flexible loss function; during the training process, the parameters of the lightweight network model and the flexible coefficient are optimized using an ephemeral algorithm; after the model training is completed, the life prediction model is obtained; wherein, the ephemeral algorithm uses an extreme value jump-out method to prevent the population from falling into a local optimum.

[0032] Furthermore, the parameters of the lightweight network model and the flexibility coefficient are optimized using the ephemeral algorithm, including:

[0033] The parameters of the lightweight network model and the flexibility coefficient are used as parameters to be optimized, and the current position of the population is determined according to the parameters to be optimized; the population position is updated to determine whether the current population falls into a local optimum; if it falls into a local optimum, the population is moved to the next position when it falls into the local optimum. Jumped value Replace; if it does not fall into the local optimum, continue to update the position and mate the population;

[0034] After the preset conditions are met, the optimization of the parameters of the lightweight network model and the flexibility coefficient is completed.

[0035] Furthermore, the jump value It is calculated as follows:

[0036]

[0037] in, is the best position of the current iteration, The best position for the next iteration; λ 1、 λ 2 is made of β Parameters of decision; β Represents a binary number; is the current vector position; Represents a random number between -1 and 1; Indicates a random number between -0.5 and 0.5; represents the calculation coefficient; Indicates the average position of the population in the current iteration.

[0038] In a third aspect, the present invention provides an electronic device comprising a processor and a memory, wherein the processor is configured to execute a computer program stored in the memory to implement the above-mentioned method for predicting the life of a backup power supply for a distribution terminal.

[0039] In a fourth aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores at least one instruction, and when the at least one instruction is executed by a processor, the method for predicting the life of a backup power supply for a distribution terminal as described above is implemented.

[0040] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0041] This paper constructs a lightweight network model (using the ShuffleNet neural network) and combines it with a flexible loss function for training, effectively improving the accuracy of lifespan prediction for distribution network terminal backup power supplies. Furthermore, the Mayfly algorithm optimizes model parameters and flexibility coefficients, preventing overfitting and enhancing the model's generalization capabilities. This method addresses the convergence issue of the distribution network terminal backup power supply lifespan prediction model. Compared to other algorithms, it achieves a more stable convergence speed, higher accuracy, and stronger generalization capabilities.

[0042] Compared to the complex modeling and large amounts of upfront data required in existing technologies, the lightweight network model used in this paper reduces computational complexity and training costs. Through the optimization of the Mayfly algorithm, training time is further shortened and training efficiency is improved.

[0043] The extreme value escape method used in the Mayfly algorithm effectively prevents the population from falling into local optimality during training, thereby enhancing the stability and robustness of the model.

[0044] By using terahertz sensors to collect spectral data of the backup power supply at the distribution terminal, more accurate and reliable input information is provided for life prediction, further improving the accuracy of the prediction.

[0045] Accurately predicting the life of the backup power supply of distribution network automation terminals can significantly reduce the frequency of line inspections and maintenance costs, improve the intelligence and stability of the distribution network, and is of great significance for ensuring the safety of the power grid and improving power supply reliability.

[0046] The present invention provides a device for predicting the life of a backup power supply for a power distribution terminal, an electronic device, and a computer-readable storage medium, which also solve the problems raised in the background technology section. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are intended to explain the present invention and do not constitute an undue limitation of the present invention. In the accompanying drawings:

[0048] Figure 1 This is a flow chart of a method for predicting the life of a backup power supply for a power distribution terminal according to an embodiment of the present invention;

[0049] Figure 2 This is a diagram showing the training principle of the life prediction model according to an embodiment of the present invention;

[0050] Figure 3 This is a comparison diagram of the model convergence curves in the embodiment of the present invention;

[0051] Figure 4 This is a flow chart of channel shuffling in an embodiment of the present invention;

[0052] Figure 5 : This is a comparison diagram of the frequency domain spectra before and after noise reduction in an embodiment of the present invention; wherein (a) spectrum before noise reduction and (b) spectrum after noise reduction;

[0053] Figure 6 A comparison diagram of the convergence curves of the algorithm according to the embodiment of the present invention and other algorithms;

[0054] Figure 7 This is a confusion matrix diagram of the model on the test set in an embodiment of the present invention;

[0055] Figure 8 This is a structural block diagram of a device for predicting the life of a backup power supply for a power distribution terminal according to an embodiment of the present invention;

[0056] Figure 9 The figure is a structural block diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0057] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that, unless there is a conflict, the embodiments and features in the embodiments of the present application can be combined with each other.

[0058] The following detailed description is an exemplary description, which is intended to provide further detailed description of the present invention. Unless otherwise indicated, all technical terms used in the present invention have the same meaning as those generally understood by those skilled in the art. The terms used in the present invention are only for describing specific embodiments, and are not intended to limit the exemplary embodiments according to the present invention.

[0059] Terahertz (THz) refers to electromagnetic waves with frequencies between 0.1 and 10 THz (wavelengths between 0.03 and 3 mm). It combines some spectral characteristics of far-infrared and X-rays, with excellent penetration and chemical selectivity. Terahertz spectroscopy can be used to analyze the composition, structure, and dynamic properties of materials, and is widely used in material classification and defect detection.

[0060] Shufflenet is a convolutional neural network architecture designed specifically for mobile devices with very limited computing power.

[0061] The Mayfly Algorithm (MA) is an emerging swarm intelligence optimization algorithm that simulates the flight and mating behavior of mayflies, increases algorithm diversity, and combines the advantages of particle swarm, genetic, and firefly algorithms. It has strong global search capabilities and performs well in multi-objective optimization, nonlinear optimization, discrete optimization, and other problems.

[0062] Example 1

[0063] The application provides a power distribution terminal backup power supply life prediction method, which acquires spectrum data of a power distribution terminal backup power supply, inputs the spectrum data into a pre-trained life prediction model, and performs life prediction. In order to improve the prediction accuracy, a flexible loss function is constructed, and the mayfly algorithm is used to optimize the model parameters and the flexible coefficient, thereby avoiding model overfitting and enhancing the generalization ability of the model.

[0064] As shown in Figure 1 A power distribution terminal backup power supply life prediction method comprises the following specific steps:

[0065] S1, acquiring spectrum data of a power distribution terminal backup power supply.

[0066] Optionally, a terahertz sensor is used to collect the spectrum data of the power distribution terminal backup power supply.

[0067] S2, inputting the spectrum data of the power distribution terminal backup power supply into a pre-trained life prediction model, and outputting a life prediction result of the power distribution terminal backup power supply by the life prediction model.

[0068] In step S2, the life prediction model is trained in the following manner:

[0069] S21, constructing a lightweight network model; wherein the lightweight network model adopts a Shufflenet neural network;

[0070] In step S21, the lightweight network model aims to achieve the best model accuracy with limited computing resources. The core of the model architecture adopts two new operations: pointwise group convolution (PGCONV) and channel shuffle, which solve the drawbacks of traditional convolution, greatly reduce the computational load of the model while maintaining accuracy. PGCONV is a channel sparse connection method, which groups different feature maps of the input layer, and then uses different convolution kernels to convolve each group. In this way, each convolution kernel no longer processes all input channels, but only processes a part of input channels, which greatly reduces the computational load of convolution. The drawback of this operation is that after grouping, the feature maps in each group cannot communicate across groups, which indirectly reduces the model accuracy. The other core operation of the algorithm, channel shuffle, solves this problem, as shown in the operation example Figure 4 For a network model with 12 channels of power distribution terminal backup power supply spectrum, the input layer is divided into 3 groups, first reshaped into (3, 4) two dimensions, then transposed into (4, 3), and finally reshaped into one dimension, that is, only simple dimension operation and transposition are needed to realize uniform channel shuffle. Each group contains the features of other groups, solving the problem of inter-group communication.

[0071] Specifically, this solution can optimize the lightweight network from the following four aspects:

[0072] 1) Minimize memory access cost (MAC) by making the number of input and output channels equal. Assume that the number of input and output channels of the network is and , the feature map size is ,but The number of floating point operations per second (FLOPs) of convolution is ;

[0073] , according to inequality (1), When , MAC takes the minimum value.

[0074] (1)

[0075] 2) Avoid excessive use of group convolutions, which can lead to increased memory access. The FLOPs of PGCONV is , where g is the number of groups. From formula (2), we can get: fixed input When C remains unchanged, an increase in g will increase MAC.

[0076] (2)

[0077] 3) Reduce network fragmentation to improve training speed. Network fragmentation refers to the parallel and mixed branches of the algorithm. Although excessive fragmentation can slightly improve network accuracy, it will reduce model efficiency, which is not worth the cost on mobile devices with low computing power.

[0078] 4) Reduce element-level operations to improve network training speed. Experimental results show Although operations such as [1] have small FLOPs, they require large memory accesses, so try to avoid them.

[0079] S22. Obtain a training set for model training;

[0080] In step S22, the spectrum data of the distribution terminal backup power supplies of different specifications and different aging degrees and the corresponding labels are constructed into a data set, and the data set is divided into a training set and a test set.

[0081] S23. Construct a flexible loss function; wherein the flexible loss function includes two loss terms, and the weight of each loss term is determined based on the flexibility coefficient.

[0082] In step S23, the flexible loss function is expressed as follows:

[0083] (3)

[0084] in, is a flexible loss function; MSE is the mean square error, MAE is the mean absolute error, is the flexibility coefficient; y i is the actual value; is the predicted value; n is the sample size.

[0085] The applicant needs to explain that the reason for choosing MSE as the new loss function is that its calculation method makes the algorithm more penalized for large errors, which is conducive to the rapid convergence of the model and is more suitable for mobile devices with low computing power studied in this scheme. The reason for choosing MAE as the new loss function is that it is more robust to outliers, which is conducive to reducing the impact of outliers on the model and reducing overfitting of the model. As one of the dimensions of multi-objective optimization of the Mayfly algorithm, it participates in iteration together.

[0086] S24. The lightweight network model is trained based on a training set and a flexible loss function. During the training process, the parameters of the lightweight network model and the flexible coefficient are optimized using an ephemeral algorithm. After the model training is completed, the life prediction model is obtained. The ephemeral algorithm uses an extreme value jump-out method to prevent the population from falling into a local optimum.

[0087] In step S24, the parameters of the lightweight network model and the flexibility coefficient are optimized using the ephemeral algorithm, including: taking the parameters of the lightweight network model and the flexibility coefficient as the parameters to be optimized, determining the current position of the population according to the parameters to be optimized; updating the population position, judging whether the current population falls into a local optimum; if it falls into a local optimum, making the population at the next position when it falls into the local optimum Jumped value Replace; if it does not fall into the local optimum, continue to update the position and mate the population; after the preset conditions are met, complete the optimization of the parameters of the lightweight network model and the flexibility coefficient.

[0088] Specifically, before training, the model and algorithm parameters are initialized, including initializing the lightweight network model: setting the structure of the convolution layer (including convolution kernel size, number, step size, padding method, etc.), pooling layer (type, size, step size, etc.), fully connected layer (number of input and output channels, activation function, etc.), and initializing the Mayfly algorithm parameters: including population size, mating probability, mutation probability, upper limit of iterations, learning rate (for network training), flexibility coefficient (initialized to 0.5).

[0089] Specifically, the mayfly algorithm randomly generates two groups of mayflies, representing males and females respectively, and each mayfly is randomly placed in the problem space as a candidate solution. Indicates that female mayflies use express, For male mayflies exist The position at the time, For female mayflies exist The positions of male and female mayflies are expressed by formulas (4) and (5) respectively.

[0090] (4)

[0091] (5)

[0092] in, is the velocity vector of the i-th mayfly at time t+1.

[0093] In the fitness calculation, the performance of each mayfly is evaluated according to a predefined objective function.

[0094] Specifically, the algorithm optimizes the position update rule as follows: male mayflies adjust their positions based on their own experience and the experience of their neighbors, while female mayflies fly to the optimal male for mating.

[0095] The speeds of male and female mayflies are determined by formulas (6) and (7), respectively.

[0096] (6)

[0097] (7)

[0098] (8)

[0099] in, Representative Mayfly exist Dimensions The speed of time, 、 is the attraction coefficient, is the individual visibility distance; e is the base of natural logarithms; r p Represents the current individual position and distance, r g Represents the current individual position and The distance is calculated by formula (8), where xi is the current location, X i is the reference position, is the Euclidean distance between the current position and the reference position; x ij and X ij In the dimensions j The current position and reference position on n is the number of dimensions of the search space. For mayflies The individual historical optimal position of is the current optimal individual position; is the position of the male mayfly at moment t; is the position of the female mayfly at moment t; r mf is the distance between the female mayfly and the male mayfly; is the random walk coefficient, for A random number between .

[0100] Algorithmic mating process: The offspring of mayflies are produced through mating. The production method is shown in formula (9), where the best-positioned male and the best-positioned female produce the first offspring. , the second best positioned male and the second best positioned female produce a second offspring .

[0101] (9)

[0102] Among them, male is the father, female is the mother; L is a random number.

[0103] Iteration and termination of the algorithm: Repeat the above steps until the termination condition is met (the maximum number of iterations is reached or the optimal solution that meets the accuracy requirements is found).

[0104] Specifically, such as Figure 2 As shown in the figure, this scheme proposes a new extreme value escape algorithm JEV to solve the problem that the model is prone to fall into the extreme value trap. and And the corresponding judgment conditions, so that the next position of the population when it falls into the local optimum Jumped value Replace, thus jumping out of the extreme value trap, The judgment condition is shown in formula (10). is the flexibility coefficient, and its value range is .

[0105] (10)

[0106] Bounce Value The calculation method of is shown in formula (11).

[0107] (11)

[0108] in, is the best position of the current iteration, The best position for the next iteration; λ 1、 λ 2 is made of β Decision parameters. is the current vector position, , represents a random number between -1 and 1; , represents a random number between -0.5 and 0.5. ,but 1, the rest is 0. The current vector position and the current best position in the simultaneous participation and parameters The randomness of the selection makes the population more diverse, and the search process for the optimal value more flexible and efficient. represents the calculation coefficient, where T The current iteration number , MaxT is the preset maximum number of iterations; Indicates the average position of the population in the current iteration; β Represents a binary number, when , β for 0 ,otherwise β is 1, where , d It is the algorithm dimension.

[0109] The method of the present invention can perform model training and lifespan prediction on mobile devices with limited computing power. In practical applications, maintenance personnel can promptly detect the lifespan of distribution network terminal backup power supplies during line inspections, significantly reducing line inspection frequency, lowering maintenance costs, and improving the intelligence and stability of the distribution network. Analysis of the proposed algorithm's model convergence curves and prediction confusion matrices for distribution terminal backup power supplies with different parameters shows that the proposed method can accurately predict the remaining lifespan of backup power supplies at different levels of loss, with an average accuracy of 92.8%.

[0110] Figure 3 The model convergence curves of the regression analysis were compared between this method, the Levy flight method and the control group. Figure 3As shown in the figure, while the Levy flight method helps the algorithm escape local optima more than the control method, its randomness leads to more invalid attempts during the search process, which increases computation time and ultimately slows model convergence. Furthermore, due to its lack of stability, it can still get stuck in local optima in some experiments. While this method may not converge as quickly as the Levy flight method in early iterations, it achieves stable overall convergence, avoids getting stuck in local optima, and ultimately achieves high accuracy.

[0111] Below, this solution is further explained and illustrated with reference to a simulation example.

[0112] 1. Collection of initial data

[0113] The experiment uses a BT-PTS-FC portable terahertz spectrometer, which has the advantages of small size and large radio frequency range. The specific parameters are shown in Table 1.

[0114] Table 1 Instrument parameters

[0115]

[0116] The experiment selected two different specifications of distribution terminal backup power supplies, whose specifications are shown in Table 2. Ultraviolet accelerated testing was used to cause different degrees of wear and tear on the power supplies, and their remaining lifespans were used as sample labels. To improve the performance of the algorithm model, the training set samples were prepared with their remaining lifespans uniformly distributed between 0 and 15 years. To make the test set samples more realistic, the test set was prepared with samples with a remaining lifespan of 0 to 3 years accounting for 10%, samples with a remaining lifespan of 3 to 6 years accounting for 15%, samples with a remaining lifespan of 6 to 9 years accounting for 20%, samples with a remaining lifespan of 9 to 12 years accounting for 25%, and samples with a remaining lifespan of 12 to 15 years accounting for 30%. The samples were evenly distributed within each stage. The specific distribution of the samples is shown in Table 3.

[0117] Table 2 Specifications of backup power supply

[0118]

[0119] Table 3 Experimental sample distribution

[0120]

[0121] The prepared samples are sequentially used with a terahertz spectrometer to obtain their ATR spectra. The instrument probe transmits terahertz waves to the sample and collector in turn, and the ATR spectrum of the current sample is displayed on the host.

[0122] 2. Data Preprocessing and Noise Reduction

[0123] Due to human operation factors and equipment precision, etc., the collected data will appear some abnormal points, missing points and outliers. The following pretreatment method is used to remove them.

[0124] 1) Abnormal points

[0125] If , the point is identified as an abnormal point, and the method of is used to remove the abnormal points. y t represents the data value at time point t, is the average value of the data.

[0126] 2) Missing points

[0127] For missing points in the data, spline interpolation method is used to fill in. Spline interpolation method can produce similar effects as high-order polynomial interpolation, but the calculation is simpler, and the fitting effect is better. In addition, changing the shape of the fitting curve only needs to change the control amount at the node, which makes the spline interpolation more flexible.

[0128] 3) Outliers

[0129] When formula (12) or (13) occurs, the data is removed based on the method of y t .

[0130] (12)

[0131] (13)

[0132] Where, Δ y t is the change of data value of adjacent time points; is the sample standard deviation.

[0133] In order to facilitate the regression analysis of the algorithm, and at the same time, it is convenient to observe the change characteristics of the spectrum in the frequency band, the ATR spectrum is zero-padded and fast Fourier transformed to obtain its average frequency spectrum, as shown in (a) of Figure 5 . In order to retain the frequency band with rich feature information, SVD denoising method is used for denoising, and the effect after denoising is shown in (b) of Figure 5 .

[0134] III. Establishing life prediction model using the developed new algorithm and tuning

[0135] The hybrid algorithm of the present scheme is used to train 400 training set samples of DC48V, and the model convergence curve is compared with Shufflenet-HDC method and MOA-VMD method, and the results are shown in Figure 6The final prediction accuracy of the model is 96.9%, and the model confusion matrix is as shown in

[0136] The trained model is applied to the 200 test sets of DC48V for verification, and the final accuracy is 96.9%, and the model confusion matrix is as shown in Figure 7 It can be seen that the overall performance of the prediction model on the test set is stable, and there is no fluctuation in the prediction accuracy at a certain stage.

[0137] In order to detect the generalization ability of the model, the algorithm of the present scheme is respectively trained using DC48V and DC24V, and the trained model is applied to the 200 test sets of DC24V, and the Shufflenet-HDC and MOA-VMD methods are compared. The average accuracy is shown in Table 4, and it can be seen that the performance of the model trained by the algorithm of the present scheme using the DC48V training set on the DC24V test set is not much different from the performance of the model trained by the DC24V training set on the DC24V test set. The average difference is 4.3%, while the difference between the other two methods is larger, which proves that the generalization ability is not as good as the algorithm of the present scheme.

[0138] Table 4 Prediction accuracy of different algorithms on DC24V test set

[0139]

[0140] Example 2

[0141] As shown in Figure 8 Based on the same inventive concept as the above embodiment, the present application also provides a power distribution terminal backup power supply life prediction device, comprising:

[0142] A data acquisition module is configured to acquire power distribution terminal backup power supply spectrum data.

[0143] A life prediction module is configured to input the power distribution terminal backup power supply spectrum data into a pre-trained life prediction model, and the life prediction model outputs a life prediction result of the power distribution terminal backup power supply.

[0144] The life prediction model is trained in the following manner:

[0145] A lightweight network model is constructed; wherein the lightweight network model adopts a Shufflenet neural network.

[0146] A training set for model training is obtained.

[0147] Constructing a flexible loss function; wherein the flexible loss function includes two loss terms, and the weight of each loss term is determined based on the flexibility coefficient;

[0148] The lightweight network model is trained based on a training set and a flexible loss function; during the training process, the parameters of the lightweight network model and the flexible coefficient are optimized using an ephemeral algorithm; after the model training is completed, the life prediction model is obtained; wherein, the ephemeral algorithm uses an extreme value jump-out method to prevent the population from falling into a local optimum.

[0149] Specifically, the parameters of the lightweight network model and the flexibility coefficient are optimized using the ephemeral algorithm, including:

[0150] The parameters of the lightweight network model and the flexibility coefficient are used as parameters to be optimized, and the current position of the population is determined according to the parameters to be optimized; the population position is updated to determine whether the current population falls into a local optimum; if it falls into a local optimum, the population is moved to the next position when it falls into the local optimum. Jumped value Replace; if it does not fall into the local optimum, continue to update the position and mate the population;

[0151] After the preset conditions are met, the optimization of the parameters of the lightweight network model and the flexibility coefficient is completed.

[0152] Specifically, the jump value It is calculated as follows:

[0153]

[0154] in, is the best position of the current iteration, The best position for the next iteration; λ 1、 λ 2 is made of β Decision parameters. is the current vector position, , represents a random number between -1 and 1; , represents a random number between -0.5 and 0.5. ,but 1, the rest is 0. The current vector position and the current best position in the simultaneous participation and parameters The randomness of the selection makes the population more diverse, and the search process for the optimal value more flexible and efficient. represents the calculation coefficient, where T The current iteration number , MaxT is the preset maximum number of iterations; Indicates the average position of the population in the current iteration; β Represents a binary number, when , β for 0 ,otherwise β is 1, where , d It is the algorithm dimension.

[0155] Example 3

[0156] like Figure 9 As shown, the present invention also provides an electronic device 100 for implementing a method for predicting the life of a backup power supply of a power distribution terminal;

[0157] The electronic device 100 includes a memory 101 , at least one processor 102 , a computer program 103 stored in the memory 101 and executable on the at least one processor 102 , and at least one communication bus 104 .

[0158] The memory 101 can be used to store a computer program 103. The processor 102 implements the steps of a method for predicting the life of a distribution terminal backup power supply in Example 1 by running or executing the computer program stored in the memory 101 and calling the data stored in the memory 101.

[0159] The memory 101 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application program required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data (such as audio data) created according to the use of the electronic device 100. In addition, the memory 101 may include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device.

[0160] The at least one processor 102 may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The processor 102 may be a microprocessor or any conventional processor. The processor 102 is the control center of the electronic device 100 and connects various parts of the entire electronic device 100 using various interfaces and lines.

[0161] The memory 101 in the electronic device 100 stores multiple instructions to implement a method for predicting the life of a backup power supply for a power distribution terminal. The processor 102 can execute the multiple instructions to implement:

[0162] Obtain backup power spectrum data of distribution terminals;

[0163] Inputting the spectrum data of the backup power supply of the power distribution terminal into a pre-trained life prediction model, the life prediction model outputting a life prediction result of the backup power supply of the power distribution terminal;

[0164] The lifespan prediction model is trained as follows:

[0165] Constructing a lightweight network model; wherein the lightweight network model adopts the Shufflenet neural network;

[0166] Get the training set for model training;

[0167] Constructing a flexible loss function; wherein the flexible loss function includes two loss terms, and the weight of each loss term is determined based on the flexibility coefficient;

[0168] The lightweight network model is trained based on a training set and a flexible loss function; during the training process, the parameters of the lightweight network model and the flexible coefficient are optimized using an ephemeral algorithm; after the model training is completed, the life prediction model is obtained; wherein, the ephemeral algorithm uses an extreme value jump-out method to prevent the population from falling into a local optimum.

[0169] Example 4

[0170] If the module / unit integrated in the electronic device 100 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the process in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. Computer-readable media may include: any entity or device that can carry computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory and read-only memory (ROM, Read-Only Memory).

[0171] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0172] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0173] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0174] These computer program instructions can also be loaded into a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 one or more flowcharts and / or blocks

[0175] In the description of the present specification, the description of the terms "one embodiment", "an example", "a specific example" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0176] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application but not to limit it, and although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that the specific embodiments of the present application can be modified or replaced by the equivalent, without departing from the spirit and scope of the present application, and any modification or equivalent replacement should be covered in the protection scope of the claims of the present application.

Claims

1. A method for predicting the life of a backup power supply for a power distribution terminal, characterized in that: include: Obtain backup power spectrum data of distribution terminals; Inputting the spectrum data of the backup power supply of the power distribution terminal into a pre-trained life prediction model, the life prediction model outputting a life prediction result of the backup power supply of the power distribution terminal; The lifespan prediction model is trained as follows: Constructing a lightweight network model; wherein the lightweight network model adopts the Shufflenet neural network; Get the training set for model training; Constructing a flexible loss function; wherein the flexible loss function includes two loss terms, and the weight of each loss term is determined based on the flexibility coefficient; The lightweight network model is trained based on the training set and the flexible loss function; during the training process, the parameters of the lightweight network model and the flexible coefficient are optimized using the ephemeral algorithm, including: taking the parameters of the lightweight network model and the flexible coefficient as the parameters to be optimized, determining the current position of the population according to the parameters to be optimized; updating the population position, judging whether the current population falls into a local optimum according to the JEV algorithm; if it falls into a local optimum, making the population at the next position when it falls into the local optimum Jumped value Replace; if it does not fall into the local optimum, continue to update the position and mate the population; after the preset conditions are met, complete the optimization of the parameters of the lightweight network model and the flexibility coefficient; after the model training is completed, the life prediction model is obtained; wherein, the mayfly algorithm adopts the extreme value jump method of the JEV algorithm to prevent the population from falling into the local optimum; The judgment conditions are as follows: in, is the flexibility coefficient, and its value range is ; The bounce value It is calculated as follows: in, is the best position of the current iteration, The best position for the next iteration; λ 1、 λ 2 is made of β Parameters of decision; β Represents a binary number; is the current vector position; Represents a random number between -1 and 1; Indicates a random number between -0.5 and 0.5; represents the calculation coefficient; Indicates the average position of the population in the current iteration; The flexible loss function is expressed as follows: in, is a flexible loss function; MSE is the mean square error, MAE is the mean absolute error, is the flexibility coefficient; y i is the actual value; is the predicted value; n is the sample size.

2. The method for predicting the life of a backup power supply for a power distribution terminal according to claim 1, characterized in that: In the step of obtaining the spectrum data of the backup power supply of the distribution terminal, a terahertz sensor is used to collect the spectrum data of the backup power supply of the distribution terminal.

3. A device for predicting the life of a backup power supply for a power distribution terminal, characterized in that: include: A data acquisition module is used to obtain spectrum data of standby power supply of distribution terminals; A life prediction module, configured to input the spectrum data of the distribution terminal standby power supply into a pre-trained life prediction model, wherein the life prediction model outputs a life prediction result of the distribution terminal standby power supply; The lifespan prediction model is trained as follows: Constructing a lightweight network model; wherein the lightweight network model adopts the Shufflenet neural network; Get the training set for model training; Constructing a flexible loss function; wherein the flexible loss function includes two loss terms, and the weight of each loss term is determined based on the flexibility coefficient; The lightweight network model is trained based on the training set and the flexible loss function; during the training process, the parameters of the lightweight network model and the flexible coefficient are optimized using the ephemeral algorithm, including: taking the parameters of the lightweight network model and the flexible coefficient as the parameters to be optimized, determining the current position of the population according to the parameters to be optimized; updating the population position, judging whether the current population falls into a local optimum according to the JEV algorithm; if it falls into a local optimum, making the population at the next position when it falls into the local optimum Jumped value Replace; if it does not fall into the local optimum, continue to update the position and mate the population; after the preset conditions are met, complete the optimization of the parameters of the lightweight network model and the flexibility coefficient; after the model training is completed, the life prediction model is obtained; wherein, the mayfly algorithm adopts the extreme value jump method of the JEV algorithm to prevent the population from falling into the local optimum; The judgment conditions are as shown in the formula: in, is the flexibility coefficient, and its value range is ; The bounce value It is calculated as follows: in, is the best position of the current iteration, The best position for the next iteration; λ 1、 λ 2 is made of β Parameters of decision; β Represents a binary number; is the current vector position; Represents a random number between -1 and 1; Indicates a random number between -0.5 and 0.5; represents the calculation coefficient; Indicates the average position of the population in the current iteration; The flexible loss function is expressed as follows: in, is a flexible loss function; MSE is the mean square error, MAE is the mean absolute error, is the flexibility coefficient; y i is the actual value; is the predicted value; n is the sample size.

4. An electronic device, characterized in that: It comprises a processor and a memory, wherein the processor is used to execute a computer program stored in the memory to implement the method for predicting the life of a backup power supply for a distribution terminal as claimed in claim 1 or 2.

5. A computer-readable storage medium, characterized in that The computer-readable storage medium stores at least one instruction, and when the at least one instruction is executed by the processor, the method for predicting the life of the backup power supply of the distribution terminal according to claim 1 or 2 is implemented.

Images