Electric energy meter failure rate prediction method, system, equipment and medium

Abstract

The invention provides an electric energy meter failure rate prediction method, system and device and a medium, three-parameter Weibull distribution and a reliability function of an electric energy meter are constructed, and parameters in the three-parameter Weibull distribution of the electric energy meter are optimized and solved by using a genetic algorithm. Obtaining an electric energy meter failure rate-multi-stress-failure time relation hypersurface failure rate diagram under three-parameter Weibull distribution; based on the trained cumulative artificial neural network model, obtaining an electric energy meter failure rate-multi-stress-failure time relation hypersurface failure rate diagram under the neural network; performing mutual verification on the two obtained prediction results by using a Bayesian optimization method; drawing a failure rate-multi-stress-failure time hypersurface failure rate diagram of the electric energy meter by using a non-uniform rational B spline; the method provided by the invention has the advantages of high failure rate prediction precision, wide application environment range, mutual verification of prediction results and the like.

Description

Technical Field

[0001] The present invention belongs to the technical field related to electric energy meters, and in particular relates to a method, system, device and medium for predicting failure rate of electric energy meters. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Today's electric energy meters not only have the electric energy consumption measurement and overload protection functions of traditional electric energy meters, but also integrate intelligent functions such as power consumption detection, load management, remote communication and real-time data analysis, providing strong support for the development of smart grids. The introduction of these new functions has put forward higher requirements for the working reliability of electric energy meters. However, due to the complex and changeable use environment of electric energy meters, especially under harsh conditions such as high temperature and high humidity, electric energy meters are prone to problems such as appearance damage, display failure, communication interruption and abnormal power performance. These failures often occur due to the failure of some components or even key components of the electric energy meter, which directly affects the normal electricity consumption measurement and user experience.

[0004] However, most manufacturers currently lack awareness of the importance of reliability design for smart energy meters, and few manufacturers conduct reliability prediction. In addition, relevant reliability data is very limited, which makes it difficult for smart energy meters to ensure long-term stable operation during actual use. The reliability of smart energy meters is not only directly related to the safe and stable operation of the power grid, but also closely related to the vital interests of users. Accurate prediction of the failure rate of energy meters helps to fully grasp the reliability level of energy meters in the design stage, thereby ensuring their stability in operation. This has important reference value for the subsequent procurement, use and technical upgrade of energy meters.

[0005] Therefore, studying the failure rate of electric energy meters and related analysis techniques can effectively support the timely maintenance and rotation of electric energy meters, reduce economic losses caused by equipment failure, and ensure that the interests of power grid departments and power users are not harmed. This research direction is not only the key to ensuring the efficient operation of smart electric energy meters, but will also lay a solid foundation for the further development of smart grids. However, the current prediction of electric energy meter failure rate based on traditional Weibull distribution has the problem of predicting initial failure, and the current single method for predicting electric energy meter failure rate also has the problem of low reliability. The traditional statistical method for predicting failure rate has problems such as incomplete consideration of factors and nonlinear fitting relationship of multiple factors.

[0006] In summary, improving the reliability and accuracy of the prediction of the failure rate of electric energy meters under multiple stress conditions, reducing the economic losses caused by equipment failure, and ensuring that the interests of power grid departments and power users are not harmed are the problems that need to be solved at present. Summary of the invention

[0007] In order to overcome the above-mentioned deficiencies of the prior art, the present invention provides a method, system, device and medium for predicting the failure rate of an electric energy meter, which has the advantages of high failure rate prediction accuracy, a wide range of applicable environments, and prediction results that can verify each other.

[0008] In order to achieve the above object, the present invention adopts the following technical solution:

[0009] In a first aspect, the present invention provides a method for predicting failure rate of an electric energy meter, comprising:

[0010] Based on the accelerated degradation test data of electric energy meters under different stress conditions and different working hours, the three-parameter Weibull distribution and reliability function of electric energy meters are constructed. The parameters in the three-parameter Weibull distribution of electric energy meters are optimized and solved using genetic algorithms. The failure rate hypersurface failure rate diagram of the electric energy meter failure rate-multiple stress-failure time relationship under the three-parameter Weibull distribution is constructed.

[0011] The different stress conditions of the electric energy meter and the corresponding electric energy failure rates are used as training data to train the cumulative artificial neural network to obtain a trained prediction model. The trained prediction model is used to predict the failure rate of the electric energy meter under different stress conditions and different working times. The predicted failure rate results are used to construct the electric energy meter failure rate under the cumulative artificial neural network - multi-stress-failure time relationship hypersurface failure rate diagram;

[0012] Based on the Bayesian optimization algorithm, the failure rate diagram of the electric energy meter failure rate-multiple stress-failure time relationship hypersurface under the three-parameter Weibull distribution and the failure rate diagram of the electric energy meter failure rate-multiple stress-failure time relationship hypersurface under the cumulative artificial neural network are cross-checked to obtain the optimal solution for the prediction of the electric energy meter failure rate under different stress conditions and different working times.

[0013] For the optimal solution of the failure rate prediction of the electric energy meter under different stress conditions and different working times, the final electric energy meter failure rate-multi-stress-failure time hypersurface failure rate diagram is drawn using non-uniform rational B-spline, and the failure rate prediction result of the electric energy meter to be predicted is obtained according to the drawn failure rate-multi-stress-failure time hypersurface failure rate diagram.

[0014] In a second aspect, the present invention provides a system for predicting failure rate of electric energy meters, comprising:

[0015] The first prediction module is configured to: construct a three-parameter Weibull distribution and a reliability function of the electric energy meter based on the accelerated degradation test data of the electric energy meter under different stress conditions and different working times, optimize and solve the parameters in the three-parameter Weibull distribution of the electric energy meter by using a genetic algorithm, and construct a hypersurface failure rate diagram of the electric energy meter failure rate-multiple stress-failure time relationship under the three-parameter Weibull distribution;

[0016] The second prediction module is configured to: use different stress conditions of the electric energy meter and the corresponding electric energy failure rates as training data to train the cumulative artificial neural network to obtain a trained prediction model, use the trained prediction model to predict the failure rates of the electric energy meter under different stress conditions and different working times, and use the predicted failure rate results to construct a failure rate diagram of the electric energy meter failure rate-multi-stress-failure time relationship hypersurface under the cumulative artificial neural network;

[0017] A mutual verification module is configured to: perform mutual verification on the failure rate diagram of the electric energy meter failure rate-multiple stress-failure time relationship hypersurface under the three-parameter Weibull distribution and the failure rate diagram of the electric energy meter failure rate-multiple stress-failure time relationship hypersurface under the cumulative artificial neural network based on the Bayesian optimization algorithm, and obtain the optimal solution for the prediction of the failure rate of the electric energy meter under different stress conditions and different working times;

[0018] The third prediction module is configured to: for the optimal solution of the failure rate prediction of the electric energy meter under different stress conditions and different working times, use non-uniform rational B-splines to draw the final electric energy meter failure rate-multi-stress-failure time hypersurface failure rate diagram, and obtain the failure rate prediction result of the electric energy meter to be predicted according to the drawn failure rate-multi-stress-failure time hypersurface failure rate diagram.

[0019] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in the first aspect is performed.

[0020] In a fourth aspect, the present invention provides a computer-readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the method described in the first aspect is performed.

[0021] One or more of the above technical solutions have the following beneficial effects:

[0022] In the present invention, by constructing the three-parameter Weibull distribution and reliability function of the electric energy meter, and optimizing and solving the parameters in the three-parameter Weibull distribution of the electric energy meter using a genetic algorithm, the electric energy meter failure rate-multiple stress-failure time relationship hypersurface failure rate diagram under the three-parameter Weibull distribution is obtained; based on the trained neural network model, the electric energy meter failure rate-multiple stress-failure time relationship hypersurface failure rate diagram under the neural network is obtained; the two prediction results obtained above are cross-checked by the Bayesian optimization method to obtain the optimal solution for the prediction of the electric energy meter failure rate under different stress conditions and different working conditions; and the electric energy meter failure rate-multiple stress-failure time hypersurface failure rate diagram is drawn using non-uniform rational B-splines. The method proposed in the present invention has the advantages of high failure rate prediction accuracy, wide applicable environment range, and prediction results that can be mutually verified, and effectively solves the problems of early failure of electric energy meter failure rate prediction based on traditional Weibull distribution, low reliability of single method prediction of electric energy meter failure rate, and incomplete consideration of factors for failure rate prediction by traditional statistical methods.

[0023] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0025] Figure 1 This is a flow chart of electric energy meter failure rate prediction in Embodiment 1 of the present invention;

[0026] Figure 2 This is a comparison chart of the prediction results of the failure rate of electric energy meters based on the three-parameter Weibull distribution and the cumulative artificial neural network method in the first embodiment of the present invention;

[0027] Figure 3 Schematic diagram of mean absolute percentage error of electric energy meter failure rate prediction using different methods in Embodiment 1 of the present invention. DETAILED DESCRIPTION

[0028] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0029] It should be noted that the terms used herein are for describing specific embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0030] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0031] Embodiment 1

[0032] This embodiment discloses a method for predicting failure rate of an electric energy meter, including:

[0033] Based on the accelerated degradation test data of electric energy meters under different stress conditions and different working hours, the three-parameter Weibull distribution and reliability function of electric energy meters are constructed. The parameters in the three-parameter Weibull distribution of electric energy meters are optimized and solved using genetic algorithms. The failure rate hypersurface failure rate diagram of the electric energy meter failure rate-multiple stress-failure time relationship under the three-parameter Weibull distribution is constructed.

[0034] The different stress conditions of the electric energy meter and the corresponding electric energy failure rates are used as training data to train the cumulative artificial neural network to obtain a trained prediction model. The trained prediction model is used to predict the failure rate of the electric energy meter under different stress conditions and different working times. Then, the obtained failure rate results can be used to construct a failure rate hypersurface failure rate diagram of the electric energy meter failure rate under the cumulative artificial neural network-multi-stress-failure time relationship.

[0035] Based on the Bayesian optimization algorithm, the failure rate diagram of the electric energy meter failure rate-multiple stress-failure time relationship hypersurface under the three-parameter Weibull distribution and the failure rate diagram of the electric energy meter failure rate-multiple stress-failure time relationship hypersurface under the cumulative artificial neural network are cross-checked to obtain the optimal solution for the prediction of the electric energy meter failure rate under different stress conditions and different working times.

[0036] For the optimal solution of the failure rate prediction of the electric energy meter under different stress conditions and different working times, the final electric energy meter failure rate-multi-stress-failure time hypersurface failure rate diagram is drawn using non-uniform rational B-spline, and the failure rate prediction result of the electric energy meter to be predicted is obtained according to the drawn failure rate-multi-stress-failure time hypersurface failure rate diagram.

[0037] In this embodiment, after obtaining the test data of failure rate and failure time of various types of electric energy meters, a three-parameter Weibull distribution and reliability function model of electric energy meter failure rate is constructed, and the parameters of the three-parameter Weibull distribution are optimized by using a genetic algorithm to obtain the electric energy meter failure rate-multiple stress-failure time (FSN) result under multi-stress conditions based on the three-parameter Weibull distribution model; then, the failure rate of the electric energy meter under multi-stress conditions is predicted by using the adaptive learning of the neural network, and then the electric energy meter failure rate-multiple stress-failure time result based on the cumulative artificial neural network is obtained; secondly, according to the Bayesian optimization The method for mutual verification of the failure rate of electric energy meters is proposed, which realizes mutual verification of the failure rate-multiple stress-failure time results of electric energy meters based on the three-parameter Weibull distribution and the cumulative artificial neural network method, and obtains the optimal solution of the failure rate of a certain type of electric energy meter at different times and under different stress conditions, thereby improving the reliability and accuracy of the prediction; finally, a failure rate hypersurface drawing based on the non-uniform rational B-spline method is proposed, and the failure rate-stress-failure time surface of the electric energy meter under different stress conditions is drawn by using the optimal solution obtained by Bayesian optimization, and the failure rate diagram of the electric energy meter failure rate-multiple stress-failure time hypersurface is drawn according to the stress type. The method proposed in this embodiment has the advantages of high failure rate prediction accuracy, wide applicable environment range, and prediction results that can be mutually verified. It effectively solves the problems of early failure prediction of electric energy meter failure rate based on traditional Weibull distribution, low reliability of single method prediction of electric energy meter failure rate, incomplete consideration of factors for failure rate prediction by traditional statistical methods, and nonlinear fitting relationship of multiple factors, and has a good engineering application effect.

[0038] Combine the following Figure 1 A method for predicting the failure rate of an electric energy meter proposed in this embodiment is described in detail, specifically including:

[0039] Step 1: After obtaining the experimental data of failure rate and failure time of various types of electric energy meters, a three-parameter Weibull distribution and reliability function model of electric energy meter failure rate is constructed, and the parameters of the three-parameter Weibull distribution are optimized by using a genetic algorithm to obtain the failure rate-multi-stress-failure time (FSN) results of electric energy meters under multi-stress conditions based on the three-parameter Weibull distribution model.

[0040] Step 11: Obtain test data through accelerated degradation testing.

[0041] Since the amount of data on failure of electric energy meters under normal stress conditions is small, or the failure time data obtained due to untimely discovery is unreliable, in order to reliably and accurately predict the failure rate of a certain type of electric energy meter, the test data of the failure rates of different types of electric energy meters under multiple stress conditions are obtained based on accelerated degradation tests.

[0042] One of the methods of accelerated degradation test is to achieve accelerated degradation by increasing the ambient temperature of the energy meter under normal working conditions. The temperature acceleration factor is calculated by the Arrhenius model:

[0043]

[0044] Among them, T TAF is the temperature acceleration factor; k is the Boltzmann constant, k = 8.62 × 10 -5 eV / K,E a is the activation energy, in eV; T n is the absolute temperature of room temperature, T h is the absolute temperature of the high temperature of the test condition.

[0045] By extrapolating formula (1), we can obtain the remaining normal working time N of the electric energy meter under normal temperature stress conditions under a certain failure rate requirement: rwh Failure time N of electric energy meter under accelerated degradation test conditions fwt The relationship between:

[0046] N rwh =T TAK ·N fwt -N cwh (2)

[0047] Where N cwh It is the current working time of the electric energy meter under normal stress working conditions.

[0048] Step 12: Construct the three-parameter Weibull distribution and reliability function.

[0049] Assume that the distribution of X obeys the three-parameter Weibull distribution, that is, X~W(η,γ,β), then the density function and cumulative distribution function of X are:

[0050]

[0051] Among them, η is the scale parameter; γ is the location parameter; β is the shape parameter; and γ≥0, η>0, β>0. The scale parameter η can reflect the dispersion of the random variable, the location parameter γ can reflect the minimum value of the random variable, and the shape parameter β can control the shape of the density curve. At this time, F(x|η,γ,β) can also be regarded as the cumulative distribution of failures, that is, the failure rate of the electric energy meter.

[0052] After that, the failure time N and stress condition S are introduced and logarithmically processed, where the failure time N refers to the time when the electric energy meter loses its original function under different stress conditions and different reliability by performing group tests according to the same test method, the stress conditions and the minimum number of samples specified. Finally, the correction parameters b and c are added, and the sample function x is:

[0053] x=N'=(lnN-b)(lnS-c) x≥γ (5)

[0054] Where N′ represents the failure time of the electric energy meter under the condition of unchanged stress conditions.

[0055] Substituting formula (5) into formula (4) and introducing the reliability R(N,S), we can obtain the RSN function based on the three-parameter Weibull model:

[0056]

[0057] In the formula, R(N,S) is the reliability, which refers to the probability that the electric energy meter will not fail within the failure time N under specific stress conditions S.

[0058] Step 13: Use genetic algorithm to solve the parameters of the three-parameter Weibull distribution.

[0059] In order to reduce the mean square error between the model prediction and the actual data, the objective function f(η,γ,β) is set to perform nonlinear constrained linear programming to fit the unknown parameters η, γ and β, and the definition is:

[0060]

[0061] Among them, m S is the number of different stress conditions, n j is the number of samples under the jth stress condition, N ij is the failure time of the ith sample under the jth stress condition, R(N,S) ij is the reliability of the ith sample under the jth stress condition.

[0062] When f(η,γ,β) takes the minimum value, the optimal parameters η, γ and β can be obtained. Genetic algorithms can effectively search the parameter space and find the parameter combination that minimizes the objective function.

[0063] When using genetic functions for parameter optimization iterations, it is necessary to input the initial values ​​of η, γ, and β. The initial values ​​can be obtained by analyzing the failure rate data under different stress conditions.

[0064] By substituting the failure rate data under three different stress conditions into the following formula:

[0065]

[0066] The initial value can be obtained by solving the equation set, and then the genetic algorithm is used to optimize the model parameters and obtain the optimal parameters η, γ and β, where n S is the number of cases with different stress conditions.

[0067] Step 14: Failure rate prediction of electric energy meter under multi-stress influence based on three-parameter Weibull distribution.

[0068] Substituting the optimal parameters η, γ and β obtained by the genetic algorithm into formula (4), the failure rate-failure time relationship of the electric energy meter based on the three-parameter Weibull model under the accelerated degradation test condition is obtained:

[0069]

[0070] Then, using formula (2), we can get the failure rate-failure time result of the electric energy meter under normal temperature stress conditions.

[0071] At this point, the failure rate-failure time result of the electric energy meter under a certain stress condition based on the three-parameter Weibull distribution has been obtained. However, the failure of the electric energy meter is generally caused by the failure of the unit. Therefore, in this embodiment, the units that cause the failure of the electric energy meter are divided into six categories: metering unit, power supply unit, clock unit, display unit, communication unit and other units. Through the test data and the instruction manuals of each unit, different failure unit tables are constructed for these six types of failure units. Based on different stress types such as temperature, humidity, electrical stress, etc., it is proposed to construct a failure rate-failure time model of the electric energy meter under multiple stress conditions:

[0072]

[0073] Where: F i,3-weibull (N) is the failure rate-failure time relationship of the electric energy meter under the i-th stress condition, m B is the type of the failed unit table, μ j is the influence coefficient of the jth failure unit under the i-th stress condition, R(N,S) j is the reliability of the jth failure unit under the i-th stress condition.

[0074] By using formula (10), we can get the failure rate-failure time relationship F of a certain type of electric energy meter under multi-stress conditions: i,3-weibull (N), and then the relevant results can be used to draw the failure rate diagram of electric energy meter failure rate-multiple stress-failure time hypersurface based on three-parameter Weibull distribution.

[0075] Step 15: Degradation data simulation and model construction.

[0076] The above-mentioned failure rate of the whole meter level is obtained by weighted summing the failure rate of each module based on the influence factors of different stress types on the failure unit. However, since the influence factors of various parameters on the failure of the electric energy meter are different, the influence factors determined by some test data and instruction manuals will cause a large deviation in the failure rate of the electric energy meter. In addition, due to the complex internal circuit structure of the electric energy meter and the functional interrelationship between modules, the failure of a single component may cause the failure of multiple modules; and the failure of components such as indicator lights and buttons will not cause the electric energy meter to fail immediately, but these component failures may be caused by changes in the operating environment of the electric energy meter, such as increased ambient temperature, increased electrical stress, etc., so they should also be considered in the calculation. When the electric energy meter has not failed or it is difficult to obtain field failure data, the above method is used to directly predict the failure rate of the electric energy meter, which obviously has a large error.

[0077] Therefore, this embodiment proposes a method for predicting the failure rate of electric energy meters based on a cumulative artificial neural network method.

[0078] Step 2: Predict the failure rate of electricity meters based on the cumulative artificial neural network method.

[0079] Firstly, the degradation data of the mechanical components of the energy meter over time are simulated according to the degradation law, so as to build a model for predicting the failure rate of the energy meter based on the cumulative artificial neural network method in the next step.

[0080] In general, the functional form of the average degradation of the components of the energy meter is as follows:

[0081] x(t)=E(Z(t)) (11)

[0082] Where Z(t) represents the performance degradation of the electric energy meter components as time t changes, and E(Z(t)) represents the average degradation of the performance of the electric energy meter within time t.

[0083] Among them, if the average degradation degree of a module x(t) of the electric energy meter gradually decreases with the passage of time t, and remains unchanged at a certain value γ0 after a short time, the average degradation of the module can be modeled as:

[0084]

[0085] In the formula, γ0, γ1, and γ2 are all constants.

[0086] This embodiment predicts the failure rate of the electric energy meter based on the cumulative artificial neural network method. The proposed cumulative artificial neural network consists of an input layer, a hidden layer and an output layer. The temperature T (℃), humidity H (%), atmospheric pressure Pa (mmHg), load size P (W), heat power consumption Q (W) and working time t (h) are used as the input parameters x of the input layer after the average degradation modeling of each module of the electric energy meter.i (i=1,2,3…7), the failure rate of electric energy meters is used as the output parameter F of the output layer, the sum of squared errors (SSE) is used as the loss function and the gradient descent (SGD) is used as the optimizer to construct a complete cumulative artificial neural network model.

[0087] The basic computing unit of the cumulative artificial neural network is the neuron, which receives T, H, Pa, P, Q and t, and uses the parameters of the average degradation modeling using formula (12) as input x i , and for each input x i Each is given an adaptive weight ω i After the weighted summation of the input in the neuron, the output is generated through the activation function. The output F of each neuron i The calculation formula is:

[0088]

[0089] In the formula, m x is the number of inputs, x i is the input, ω i For input x i The corresponding adaptive weight, a is the bias, and f(x) is the activation function. The activation function uses the Softmax function, which can convert the adaptive weight ω i The input x i Transformed into probability distribution f(z i ), whose formula is:

[0090]

[0091] After getting the output F of each neuron i After that, the constructed cumulative artificial neural network is used to utilize the input x i To predict the failure rate of the electric energy meter, it is also necessary to calculate the current output F i Perform iterative optimization. In this embodiment, the error square sum SSE is selected as the loss function, and its formula is as follows:

[0092]

[0093] Where n p is the number of training times, n m is the number of iterations, o p,m Output the actual value of the failure rate of the energy meter for a certain model operation, which will be directly given in the training set; d p,m is the failure rate of the electric energy meter output in the mth iteration i The pth training value of .

[0094] Through the iterative operation of the cumulative artificial neural network, the output failure rate of the electric energy meter is continuously updated in the direction of the minimum loss function value, and then when the loss function value is minimized, the trained cumulative artificial neural network is obtained; based on the trained cumulative artificial neural network, the electric energy meter under different stress conditions and different working times is predicted to obtain the corresponding failure rate prediction results, and then the failure rate-failure time relationship F of the electric energy meter based on the cumulative artificial neural network under different stress conditions and different working times is obtained. i,A-ANN (N), the relevant results can be used to draw the failure rate diagram of electric energy meter failure rate-multiple stress-failure time hypersurface based on cumulative artificial neural network.

[0095] Step 3: Mutual verification of electric energy meter failure rates based on Bayesian optimization.

[0096] Through step 1 and step 2 of this embodiment, two sets of electric energy meter failure rate-multiple stress-failure time results based on the three-parameter Weibull distribution and cumulative artificial neural network methods can be obtained. However, since the electric energy meter failure rate is calculated based on the three-parameter Weibull distribution method, the influence of the actual working environment of the electric energy meter on the failure rate is ignored to a certain extent; the electric energy meter failure rate is calculated based on the cumulative artificial neural network method, which may cause certain errors in the failure rate results due to the distortion of the training set data and the small number of training samples. In order to mutually verify the electric energy meter failure rates obtained by the above two methods and obtain the optimal solution of the electric energy meter failure rate after working for a period of time under multiple stress conditions, this embodiment proposes a method for mutually verifying the electric energy meter failure rate-multiple stress-failure time results based on the three-parameter Weibull distribution and the cumulative artificial neural network method based on the Bayesian optimization method, and then obtain the optimal solution of the electric energy meter failure rate at different working times under various stress conditions.

[0097] In this embodiment, the Bayesian optimization method uses the Gaussian process GP as a proxy model to find the optimal solution F for the failure rate of the electric energy meter under different working hours under various stress conditions. opt (S,N) for prediction. Among them, F opt (S,N) can be modeled by a Gaussian process:

[0098] F opt (S,N)~GP(m(F i (N)), k(F i,3-weibull (N),F i,A-ANN (N))) (16)

[0099] In the formula, F i (N) is the failure rate of the electric energy meter under different working hours under certain stress conditions, F i,3-weibull (N) is the failure rate of the electric energy meter under different working hours under certain stress conditions obtained based on the three-parameter Weibull distribution method in step 1, F i,A-ANN(N) is the failure rate of the electric energy meter under different working hours under certain stress conditions obtained based on the cumulative artificial neural network method in step 2, m(F i (N)) is the mean function. k(x,y) is the kernel function. The kernel function in this embodiment is the radial basis function, and its formula is:

[0100]

[0101] Where ||xy|| is the Euclidean distance between x and y, specifically, F i,3-weibull (N) and F i,A-ANN (N), l is the length parameter, σ 2 is the variance of the kernel function.

[0102] Bayesian optimization searches for the optimal solution F of the failure rate of the electric energy meter under different working hours under various stress conditions in each iteration. opt In the (S,N) process, the next sampling point will be effectively selected by maximizing the expected improvement to obtain a more reliable result of the electricity meter failure rate.

[0103] The optimization method is optimized by maximizing the expected improvement method, and the formula is:

[0104]

[0105] In the formula, Δ n (N) is the expected distance, F i,3-weibull (N) and F i,A-ANN (N) The difference between the predicted failure rate and the theoretical failure rate at time N; σ n (N) is F i,3-weibull (N) and F i,A-ANN (N) is the standard deviation of the prediction at time N, indicating the uncertainty of the prediction; φ(x) is the probability density function of the standard normal distribution, and Φ(x) is the cumulative distribution function of the standard normal distribution.

[0106] In the above process of predicting the optimal solution of the failure rate of electric energy meters based on Bayesian optimization, for any given time N, the Gaussian process model is used to predict F opt The posterior distribution of (S,N) can be used to obtain its confidence interval, where S represents stress, N represents time, and the subscript opt ​​represents the optimal solution. opt The posterior distribution of (S,N) is expressed as:

[0107]

[0108] In the formula, μ(N) and σ 2 (N) are the mean and standard deviation of the Gaussian process prediction, which can be obtained through the Gaussian process prediction:

[0109]

[0110] In the formula, K(x,X 1:t ) and K(X 1:t ,X 1:t ) is the kernel function shown in formula (17).

[0111] Through the above Bayesian optimization prediction process, the optimal solution F of the failure rate of the electric energy meter under different working hours under various stress conditions can be obtained by mutual verification based on the three-parameter Weibull distribution and the cumulative artificial neural network method. opt (S, N) and the confidence interval of the failure rate prediction of the electric energy meter with confidence level α [μ(N)-ασ(N),μ(N)+ασ(N)].

[0112] Step 4: Failure rate hypersurface drawing based on non-uniform rational B-spline method.

[0113] The optimal solution F of the failure rate of the electric energy meter under different working hours under various stress conditions based on the Bayesian optimization method is obtained through mutual verification. opt (S, N), since the result is still a discrete point in the multidimensional space, in order to use a surface diagram that conforms to the actual continuous change trend as much as possible, this embodiment proposes a failure rate hypersurface drawing method based on the non-uniform rational B-spline method. The known F opt The (S,N) scatter results use control points, node vectors and weights to fit the failure rate-multi-stress-failure time hypersurface failure rate diagram of a certain type of electric energy meter.

[0114] This embodiment proposes an optimal solution F for the failure rate of electric energy meters under different working hours under various stress conditions obtained by Bayesian optimization based on non-uniform rational B-splines. opt (S,N) Draw the failure rate-multi-stress-failure time hypersurface failure rate diagram of a certain type of electric energy meter.

[0115] Non-uniform rational B-spline consists of control points, node vectors and weights, and its mathematical expression is:

[0116]

[0117] Where S(u,v) is the point on the curve, i.e., the optimal solution F for the failure rate of the energy meter under different working hours under various stress conditions. opt (S,N); m and n are the number of nodes in the u and v directions respectively, P i,j is a control point that will affect the final shape of the hypersurface; w i,j is the weight associated with the control point, which can be adjusted by i,j To change the closeness of the hypersurface to the control point; Ni,p (u), N j,p (v) is defined as the non-periodic node vector U = [u0,u1...u n+k+1 ]、V=[v0,v1...v n+k+1 ] on the B-spline basis function, where N i,p The recursive formula for (u) is:

[0118]

[0119] The optimal solution of the failure rate of the electric energy meter under different working hours under various stress conditions is obtained by opt Substituting (S, N) into equation (21), we can obtain F after solving the parameters. opt The non-uniform rational B-spline expression of (S, N) is obtained, that is, the failure rate-multi-stress-failure time expression of a certain type of electric energy meter, and then the failure rate-multi-stress-failure time hypersurface failure rate diagram of the electric energy meter of this model is drawn.

[0120] Based on the method of this embodiment, temperature accelerated stress tests are first carried out on various types of electric energy meters. After obtaining the test data, the failure rate-failure time relationship of the electric energy meter under multi-stress conditions is obtained based on the three-parameter Weibull distribution, and then the electric energy meter failure rate-multi-stress-failure time hypersurface failure rate diagram is drawn; at the same time, the relevant environment and working parameters of these types of electric energy meters under normal working conditions are used to perform calculations based on the cumulative artificial neural network method and obtain the failure rate-failure time relationship of each electric energy meter under multi-stress conditions, and then the electric energy meter failure rate-multi-stress-failure time hypersurface failure rate diagram is drawn; then, based on the Bayesian optimization method, the two groups of failure rate-multi-stress-failure time results obtained by the above two methods are iteratively optimized to obtain the optimal solution of the electric energy meter failure rate at different working times under multi-stress conditions; finally, based on the non-uniform rational B-spline method, the failure rate-multi-stress-failure time hypersurface failure rate diagrams of each type of electric energy meter based on the three-parameter Weibull distribution and the cumulative artificial neural network mutual verification method are drawn to improve the reliability and accuracy of the final prediction.

[0121] The specific experimental verification steps are as follows:

[0122] 1) Steps to calculate the failure rate of electric energy meters based on three-parameter Weibull distribution:

[0123] ① Divide 90 electric energy meter samples of three models A, B and C into three groups, 30 in each group, and place them in adjustable temperature test boxes respectively. Connect them in series on a single-phase AC line and share a 2000W electric heater as a load.

[0124] ② Adjust the three adjustable temperature test chambers and maintain their temperatures at 65℃, 75℃ and 85℃ respectively, then turn on the power and start timing with a time accuracy of hours, and start the temperature accelerated stress test;

[0125] ③Inspect each group of samples once every three days. During the inspection, take them out of the temperature-adjustable test chamber in groups to minimize the impact of the intermediate inspection on the temperature of the test chamber;

[0126] ④ If the electric energy meter is found to be unable to function normally during a certain inspection, such as no display on the screen, cracked case, abnormal flashing of indicator light, etc., the electric energy meter is considered to be failed, and the current cumulative number of failed electric energy meters n and failure time N are recorded;

[0127] ⑤ When at least half of the energy meters in each group of energy meter samples fail, stop the test;

[0128] ⑥ Using the above test data, the initial values ​​are obtained and the parameters of the three-parameter Weibull distribution are optimized using a genetic algorithm to obtain the relationship between the failure rate and failure time F(N|η,γ,β)~N of the electric energy meter based on the three-parameter Weibull model;

[0129] ⑦ Substitute the test temperature of the accelerated degradation test into formula (1) to obtain the respective temperature acceleration factors T AFKi , and then extrapolate to obtain the failure time N of the electric energy meter under normal stress conditions, and then obtain the failure rate-failure time relationship of the electric energy meter under single stress conditions based on the three-parameter Weibull distribution under normal working conditions.

[0130] ⑧ Repeat the above test steps, but change the stress variables, for example, keep the acceleration temperature unchanged at 85℃ but change the load to 1000W, 2000W and 3000W respectively. Finally, the failure rate-failure time relationship of a certain type of electric energy meter under multiple stress conditions based on three-parameter Weibull distribution is obtained, and the failure rate-multiple stress-failure time hypersurface failure rate diagram of this type of electric energy meter based on the three-parameter Weibull distribution method is drawn.

[0131] 2) Steps for calculating the failure rate of electric energy meters based on cumulative artificial neural networks:

[0132] ① Model the degradation data of multiple electric energy meters of three models A, B, and C, and obtain the temperature T (℃), humidity H (%), atmospheric pressure Pa (mmHg), load size P (W), heat power consumption Q (W), and working time t (h) of each meter under normal working environment by consulting the electric energy meter manual and actual measurement;

[0133] ②Preprocess the input data in ① and use the sum of squared errors to measure the features;

[0134] ③ Select part of the data as the training set, and select and process the required network design, activation basis, and training algorithm, and import the cumulative artificial neural network for training;

[0135] ④ The trained cumulative artificial neural network is used to test the test set to obtain the failure rate-failure time relationship of the electric energy meter under different working times under multi-stress conditions based on the cumulative artificial neural network, and then the failure rate diagram of the electric energy meter failure rate-multi-stress-failure time hypersurface is drawn.

[0136] For different working hours under a certain stress condition, the failure rate of A-type electric energy meter predicted based on three-parameter Weibull distribution and the failure rate of A-type electric energy meter predicted by cumulative artificial neural network are plotted as horizontal and vertical coordinates. Figure 2 The comparison chart of the prediction results of the two methods is shown in Figure 2.

[0137] 3) Electricity meter failure rate prediction method based on three-parameter Weibull distribution and cumulative artificial neural network mutual verification:

[0138] ① Based on the Bayesian optimization method, the failure rate-multiple stress-failure time results of the electric energy meter obtained based on the three-parameter Weibull distribution and the cumulative artificial neural network method are cross-checked to obtain the optimal solution F for the failure rate of the electric energy meter. opt (S,N) and confidence intervals.

[0139] ② Based on the non-uniform rational B-spline method, the optimal solution F of the failure rate of the electric energy meter under different working hours under various stress conditions is obtained. opt The non-uniform rational B-spline expression of (S,N) is proposed, and the failure rate-multi-stress-failure time hypersurface failure rate diagram of this type of electric energy meter based on the three-parameter Weibull distribution and the cumulative artificial neural network mutual verification method is drawn.

[0140] Using three methods, namely, the failure rate of electric energy meters based on three-parameter Weibull distribution, the failure rate of electric energy meters based on cumulative artificial neural network, and the failure rate of electric energy meters based on three-parameter Weibull distribution and cumulative artificial neural network mutual verification, the failure rates of electric energy meters of three models A, B, and C at different working times under certain stress conditions are predicted, and their respective mean absolute percentage errors (MAPE) are calculated. The results are shown in the figure. Figure 3 shown.

[0141] The present embodiment is a method for predicting the failure rate of electric energy meters based on three-parameter Weibull distribution and cumulative artificial neural network mutual verification, and the specific implementation steps are as follows:

[0142] Step 1: It is proposed to carry out accelerated degradation test on a certain type of electric energy meter under different external stress S conditions, and obtain the failure rate-failure time test data F(N|η,γ,β) of the electric energy meter under multiple stress conditions. ij ~N ij .

[0143] Step 2: Propose a three-parameter Weibull distribution F(N|η,γ,β) and reliability function R(N,S) model for the failure rate of electricity meters.

[0144] Step 3: According to the set objective function f(η,γ,β), it is proposed to use genetic algorithm to optimize the parameters η, γ and β of the three-parameter Weibull distribution.

[0145] Step 4: Propose a model based on accelerated degradation test to obtain the failure rate-failure time relationship F of the electric energy meter under multi-stress conditions based on the three-parameter Weibull model under normal working environment. i,3-weibull (N), and draw the failure rate diagram of the electric energy meter failure rate-multiple stress-failure time hypersurface based on the three-parameter Weibull distribution of this model of electric energy meter.

[0146] Step 5: Propose a degradation model for predicting the failure rate of electric energy meters under normal working conditions and preprocess the data set required for training the cumulative artificial neural network.

[0147] Step 6: It is proposed to train and evaluate the model to be trained based on the constructed model and data set, and then use the trained cumulative artificial neural network method to predict the failure rate of the electric energy meter at different working times under various stress conditions, and obtain the failure rate-failure time relationship F of the electric energy meter under various stress conditions based on the cumulative artificial neural network method. i,A-ANN (N), and draw the failure rate-multi-stress-failure time hypersurface failure rate diagram of this type of electric energy meter based on the cumulative artificial neural network method.

[0148] Step 7: A Bayesian optimization method is proposed to realize the mutual verification of the failure rate-multiple stress-failure time results of the electric energy meter based on the three-parameter Weibull distribution and the cumulative artificial neural network method, and then the optimal solution F of the failure rate of the electric energy meter is obtained. opt (S,N) and confidence intervals.

[0149] Step 8: Based on the non-uniform rational B-spline method, the optimal solution F of the failure rate of the electric energy meter under different working hours under various stress conditions is proposed. opt The non-uniform rational B-spline expression of (S,N) is proposed, and the failure rate-multi-stress-failure time hypersurface failure rate diagram of this type of electric energy meter based on the three-parameter Weibull distribution and the cumulative artificial neural network mutual verification method is drawn.

[0150] The method proposed in this embodiment achieves the purpose of reliably predicting the failure rate of an electric energy meter under normal stress conditions under different stress conditions. It has the advantages of high failure rate prediction accuracy, a wide range of applicable environments, and prediction results that can be mutually verified. It effectively solves the problems of early failure in traditional electric energy meter failure rate prediction based on Weibull distribution and the inability of traditional machine learning algorithms to calculate the failure rate of electric energy meters under abnormal working conditions, and has a good engineering application effect.

[0151] Embodiment 2

[0152] The purpose of this embodiment is to provide a system for predicting the failure rate of an electric energy meter, including:

[0153] The first prediction module is configured to: construct a three-parameter Weibull distribution and a reliability function of the electric energy meter based on the accelerated degradation test data of the electric energy meter under different stress conditions and different working times, optimize and solve the parameters in the three-parameter Weibull distribution of the electric energy meter by using a genetic algorithm, and construct a hypersurface failure rate diagram of the electric energy meter failure rate-multiple stress-failure time relationship under the three-parameter Weibull distribution;

[0154] The second prediction module is configured to: use different stress conditions of the electric energy meter and the corresponding electric energy failure rates as training data to train the cumulative artificial neural network to obtain a trained prediction model, use the trained prediction model to predict the failure rates of the electric energy meter under different stress conditions and different working times, and use the predicted failure rate results to construct a failure rate diagram of the electric energy meter failure rate-multi-stress-failure time relationship hypersurface under the cumulative artificial neural network;

[0155] A mutual verification module is configured to: perform mutual verification on the failure rate diagram of the electric energy meter failure rate-multiple stress-failure time relationship hypersurface under the three-parameter Weibull distribution and the failure rate diagram of the electric energy meter failure rate-multiple stress-failure time relationship hypersurface under the cumulative artificial neural network based on the Bayesian optimization algorithm, and obtain the optimal solution for the prediction of the failure rate of the electric energy meter under different stress conditions and different working times;

[0156] The third prediction module is configured to: for the optimal solution of the failure rate prediction of the electric energy meter under different stress conditions and different working times, use non-uniform rational B-splines to draw the final electric energy meter failure rate-multi-stress-failure time hypersurface failure rate diagram, and obtain the failure rate prediction result of the electric energy meter to be predicted according to the drawn failure rate-multi-stress-failure time hypersurface failure rate diagram.

[0157] In the first prediction module, it specifically includes: after obtaining the experimental data of failure rate and failure time of various types of electricity meters, by constructing a three-parameter Weibull distribution and reliability function model of the electricity meter failure rate, and using a genetic algorithm to optimize the parameters of the three-parameter Weibull distribution, the failure rate-multi-stress-failure time (FSN) results of the electricity meter under multi-stress conditions based on the three-parameter Weibull distribution model are obtained.

[0158] Test data acquisition subunit: obtain test data through accelerated degradation tests.

[0159] Since the amount of data on failure of electric energy meters under normal stress conditions is small, or the failure time data obtained due to untimely discovery is unreliable, in order to reliably and accurately predict the failure rate of a certain type of electric energy meter, the test data of the failure rates of different types of electric energy meters under multiple stress conditions are obtained based on accelerated degradation tests.

[0160] One of the methods of accelerated degradation test is to achieve accelerated degradation by increasing the ambient temperature of the energy meter under normal working conditions. The temperature acceleration factor is calculated by the Arrhenius model:

[0161]

[0162] Among them, T TAF is the temperature acceleration factor; k is the Boltzmann constant, k = 8.62 × 10 -5 eV / K,E a is the activation energy, in eV; T n is the absolute temperature of room temperature, T h is the absolute temperature of the high temperature of the test condition.

[0163] By extrapolating formula (1), we can obtain the remaining normal working time N of the electric energy meter under normal temperature stress conditions under a certain failure rate requirement: rwh Failure time N of electric energy meter under accelerated degradation test conditions fwt The relationship between:

[0164] N rwh =T TAK ·N fwt -N cwh (2)

[0165] Where N cwh It is the current working time of the electric energy meter under normal stress working conditions.

[0166] Construct subunits: Construct three-parameter Weibull distribution and reliability function.

[0167] Assume that the distribution of X obeys the three-parameter Weibull distribution, that is, X~W(η,γ,β), then the density function and cumulative distribution function of X are:

[0168]

[0169] Among them, η is the scale parameter; γ is the location parameter; β is the shape parameter; and γ≥0, η>0, β>0. The scale parameter η can reflect the dispersion of the random variable, the location parameter γ can reflect the minimum value of the random variable, and the shape parameter β can control the shape of the density curve. At this time, F(x|η,γ,β) can also be regarded as the cumulative distribution of failures, that is, the failure rate of the electric energy meter.

[0170] After that, the failure time N and stress condition S are introduced and logarithmically processed, where the failure time N refers to the time when the electric energy meter loses its original function under different stress conditions and different reliability by performing group tests according to the same test method, the stress conditions and the minimum number of samples specified. Finally, the correction parameters b and c are added, and the sample function x is:

[0171] x=N'=(lnN-b)(lnS-c) x≥γ (5)

[0172] Where N′ represents the failure time of the electric energy meter under the condition of unchanged stress conditions.

[0173] Substituting formula (5) into formula (4) and introducing the reliability R(N,S), we can obtain the RSN function based on the three-parameter Weibull model:

[0174]

[0175] In the formula, R(N,S) is the reliability, which refers to the probability that the electric energy meter will not fail within the failure time N under specific stress conditions S.

[0176] Parameter solving subunit: Use genetic algorithm to solve the parameters of three-parameter Weibull distribution.

[0177] In order to reduce the mean square error between the model prediction and the actual data, the objective function f(η,γ,β) is set to perform nonlinear constrained linear programming to fit the unknown parameters η, γ and β, and the definition is:

[0178]

[0179] Among them, m S is the number of different stress conditions, n j is the number of samples under the jth stress condition, N ij is the failure time of the ith sample under the jth stress condition, R(N,S) ij is the reliability of the ith sample under the jth stress condition.

[0180] When f(η,γ,β) takes the minimum value, the optimal parameters η, γ and β can be obtained. Genetic algorithms can effectively search the parameter space and find the parameter combination that minimizes the objective function.

[0181] When using genetic functions for parameter optimization iterations, it is necessary to input the initial values ​​of η, γ, and β. The initial values ​​can be obtained by analyzing the failure rate data under different stress conditions.

[0182] By substituting the failure rate data under three different stress conditions into the following formula:

[0183]

[0184] The initial value can be obtained by solving the equation set, and then the genetic algorithm is used to optimize the model parameters and obtain the optimal parameters η, γ and β, where n S is the number of cases with different stress conditions.

[0185] Prediction subunit: Failure rate prediction of electric energy meter under multiple stress influence based on three-parameter Weibull distribution.

[0186] Substituting the optimal parameters η, γ and β obtained by the genetic algorithm into formula (4), the failure rate-failure time relationship of the electric energy meter based on the three-parameter Weibull model under the accelerated degradation test condition is obtained:

[0187]

[0188] Then, using formula (2), we can get the failure rate-failure time result of the electric energy meter under normal temperature stress conditions.

[0189] At this point, the failure rate-failure time result of the electric energy meter under a certain stress condition based on the three-parameter Weibull distribution has been obtained. However, the failure of the electric energy meter is generally caused by the failure of the unit. Therefore, in this embodiment, the units that cause the failure of the electric energy meter are divided into six categories: metering unit, power supply unit, clock unit, display unit, communication unit and other units. Through the test data and the instruction manuals of each unit, different failure unit tables are constructed for these six types of failure units. Based on different stress types such as temperature, humidity, electrical stress, etc., it is proposed to construct a failure rate-failure time model of the electric energy meter under multiple stress conditions:

[0190]

[0191] Where: F i,3-weibull (N) is the failure rate-failure time relationship of the electric energy meter under the i-th stress condition, m B is the type of the failed unit table, μ j is the influence coefficient of the jth failure unit under the i-th stress condition, R(N,S) jis the reliability of the jth failure unit under the i-th stress condition.

[0192] By using formula (10), we can get the failure rate-failure time relationship F of a certain type of electric energy meter under multi-stress conditions: i,3-weibull (N), and then the relevant results can be used to draw the failure rate diagram of electric energy meter failure rate-multiple stress-failure time hypersurface based on three-parameter Weibull distribution.

[0193] Simulation subunit: degradation data simulation and model building.

[0194] The above-mentioned failure rate of the whole meter level is obtained by weighted summing the failure rate of each module based on the influence factors of different stress types on the failure unit. However, since the influence factors of various parameters on the failure of the electric energy meter are different, the influence factors determined by some test data and instruction manuals will cause a large deviation in the failure rate of the electric energy meter. In addition, due to the complex internal circuit structure of the electric energy meter and the functional interrelationship between modules, the failure of a single component may cause the failure of multiple modules; and the failure of components such as indicator lights and buttons will not cause the electric energy meter to fail immediately, but these component failures may be caused by changes in the operating environment of the electric energy meter, such as increased ambient temperature, increased electrical stress, etc., so they should also be considered in the calculation. When the electric energy meter has not failed or it is difficult to obtain field failure data, the above method is used to directly predict the failure rate of the electric energy meter, which obviously has a large error.

[0195] In the second prediction module of this embodiment, the failure rate of the electric energy meter is predicted based on the cumulative artificial neural network method.

[0196] Firstly, the degradation data of the mechanical components of the energy meter over time are simulated according to the degradation law, so as to build a model for predicting the failure rate of the energy meter based on the cumulative artificial neural network method in the next step.

[0197] In general, the functional form of the average degradation of the components of the energy meter is as follows:

[0198] x(t)=E(Z(t)) (11)

[0199] Where Z(t) represents the performance degradation of the electric energy meter components as time t changes, and E(Z(t)) represents the average degradation of the performance of the electric energy meter within time t.

[0200] Among them, if the average degradation degree of a module x(t) of the electric energy meter gradually decreases with the passage of time t, and remains unchanged at a certain value γ0 after a short time, the average degradation of the module can be modeled as:

[0201]

[0202] In the formula, γ0, γ1, and γ2 are all constants.

[0203] This embodiment predicts the failure rate of the electric energy meter based on the cumulative artificial neural network method. The proposed cumulative artificial neural network consists of an input layer, a hidden layer and an output layer. The temperature T (℃), humidity H (%), atmospheric pressure Pa (mmHg), load size P (W), heat power consumption Q (W) and working time t (h) are used as the input parameters x of the input layer after the average degradation modeling of each module of the electric energy meter. i (i=1,2,3…7), the failure rate of electric energy meters is used as the output parameter F of the output layer, the sum of squared errors (SSE) is used as the loss function and the gradient descent (SGD) is used as the optimizer to construct a complete cumulative artificial neural network model.

[0204] The basic computing unit of the cumulative artificial neural network is the neuron, which receives T, H, Pa, P, Q and t, and uses the parameters of the average degradation modeling using formula (12) as input x i , and for each input x i Each is given an adaptive weight ω i After the weighted summation of the input in the neuron, the output is generated through the activation function. The output F of each neuron i The calculation formula is:

[0205]

[0206] In the formula, m x is the number of inputs, x i is the input, ω i For input x i The corresponding adaptive weight, a is the bias, and f(x) is the activation function. The activation function uses the Softmax function, which can convert the adaptive weight ω i The input x i Transformed into probability distribution f(z i ), whose formula is:

[0207]

[0208] After getting the output F of each neuron i After that, the constructed cumulative artificial neural network is used to utilize the input x i To predict the failure rate of the electric energy meter, it is also necessary to calculate the current output F i Perform iterative optimization. In this embodiment, the square sum of error SSE is selected as the loss function, and its formula is as follows:

[0209]

[0210] Where np is the number of training times, n m is the number of iterations, o p,m Output the actual value of the failure rate of the energy meter for a certain model operation, which will be directly given in the training set; d p,m is the failure rate of the electric energy meter output in the mth iteration i The pth training value of .

[0211] Through the iterative operation of the cumulative artificial neural network, the output failure rate of the electric energy meter is continuously updated in the direction of the minimum loss function value, and then when the loss function value is minimized, the trained cumulative artificial neural network is obtained; based on the trained cumulative artificial neural network, the electric energy meter under different stress conditions and different working times is predicted to obtain the corresponding failure rate prediction results, and then the failure rate-failure time relationship F of the electric energy meter based on the cumulative artificial neural network under different stress conditions and different working times is obtained. i,A-ANN (N), the relevant results can be used to draw the failure rate diagram of electric energy meter failure rate-multiple stress-failure time hypersurface based on cumulative artificial neural network.

[0212] In the mutual verification module of this embodiment, mutual verification of electric energy meter failure rates is performed based on Bayesian optimization.

[0213] Through the first prediction module and the second prediction module of this embodiment, two sets of electric energy meter failure rate-multiple stress-failure time results based on the three-parameter Weibull distribution and cumulative artificial neural network methods can be obtained. However, since the failure rate of the electric energy meter is calculated based on the three-parameter Weibull distribution method, the influence of the actual working environment of the electric energy meter on the failure rate is ignored to a certain extent; the failure rate of the electric energy meter calculated based on the cumulative artificial neural network method may cause certain errors in the failure rate results due to the distortion of the training set data and the small number of training samples. In order to mutually verify the failure rates of the electric energy meters obtained by the above two methods and obtain the optimal solution of the failure rate of the electric energy meter after working for a period of time under multiple stress conditions, this embodiment proposes a method for mutually verifying the failure rate-multiple stress-failure time results of the electric energy meter calculated based on the three-parameter Weibull distribution and the cumulative artificial neural network method based on the Bayesian optimization method, and then obtain the optimal solution of the failure rate of the electric energy meter at different working times under each stress condition.

[0214] In this embodiment, the Bayesian optimization method uses the Gaussian process GP as a proxy model to find the optimal solution F for the failure rate of the electric energy meter under different working hours under various stress conditions. opt (S,N) for prediction. Among them, F opt (S,N) can be modeled by a Gaussian process:

[0215] F opt (S,N)~GP(m(F i (N)), k(Fi,3-weibull (N),F i,A-ANN (N)))(16)

[0216] In the formula, F i (N) is the failure rate of the electric energy meter under different working hours under certain stress conditions, F i,3-weibull (N) is the failure rate of the electric energy meter under different working hours under certain stress conditions obtained based on the three-parameter Weibull distribution method in step 1, F i,A-ANN (N) is the failure rate of the electric energy meter under different working hours under certain stress conditions obtained based on the cumulative artificial neural network method in step 2, m(F i (N)) is the mean function. k(x,y) is the kernel function. The kernel function in this embodiment is the radial basis function, and its formula is:

[0217]

[0218] Where ||xy|| is the Euclidean distance between x and y, specifically, F i,3-weibull (N) and F i,A-ANN (N), l is the length parameter, σ 2 is the variance of the kernel function.

[0219] Bayesian optimization searches for the optimal solution F of the failure rate of the electric energy meter under different working hours under various stress conditions in each iteration. opt In the (S,N) process, the next sampling point will be effectively selected by maximizing the expected improvement to obtain a more reliable result of the electricity meter failure rate.

[0220] The optimization method is optimized by maximizing the expected improvement method, and the formula is:

[0221]

[0222] In the formula, Δ n (N) is the expected distance, F i,3-weibull (N) and F i,A-ANN (N) The difference between the predicted failure rate and the theoretical failure rate at time N; σ n (N) is F i,3-weibull (N) and F i,A-ANN (N) is the standard deviation of the prediction at time N, indicating the uncertainty of the prediction; φ(x) is the probability density function of the standard normal distribution, and Φ(x) is the cumulative distribution function of the standard normal distribution.

[0223] In the above process of predicting the optimal solution of the failure rate of electric energy meters based on Bayesian optimization, for any given time N, the Gaussian process model is used to predict F optThe posterior distribution of (S,N) can be used to obtain its confidence interval, where S represents stress, N represents time, and the subscript opt ​​represents the optimal solution. opt The posterior distribution of (S,N) is expressed as:

[0224]

[0225] In the formula, μ(N) and σ 2 (N) are the mean and standard deviation of the Gaussian process prediction, which can be obtained through the Gaussian process prediction:

[0226]

[0227] In the formula, K(x,X 1:t ) and K(X 1:t ,X 1:t ) is the kernel function shown in formula (17).

[0228] Through the above Bayesian optimization prediction process, the optimal solution F of the failure rate of the electric energy meter under different working hours under various stress conditions can be obtained by mutual verification based on the three-parameter Weibull distribution and the cumulative artificial neural network method. opt (S, N) and the confidence interval of the failure rate prediction of the electric energy meter with confidence level α [μ(N)-ασ(N),μ(N)+ασ(N)].

[0229] In the third prediction module of this embodiment, a failure rate hypersurface based on the non-uniform rational B-spline method is drawn.

[0230] The optimal solution F of the failure rate of the electric energy meter under different working hours under various stress conditions based on the Bayesian optimization method is obtained through mutual verification. opt (S, N), since the result is still a discrete point in the multidimensional space, in order to use a surface diagram that conforms to the actual continuous change trend as much as possible, this embodiment proposes a failure rate hypersurface drawing method based on the non-uniform rational B-spline method. The known F opt The (S,N) scatter results use control points, node vectors and weights to fit the failure rate-multi-stress-failure time hypersurface failure rate diagram of a certain type of electric energy meter.

[0231] This embodiment proposes an optimal solution F for the failure rate of electric energy meters under different working hours under various stress conditions obtained by Bayesian optimization based on non-uniform rational B-splines. opt (S,N) Draw the failure rate-multi-stress-failure time hypersurface failure rate diagram of a certain type of electric energy meter.

[0232] Non-uniform rational B-spline consists of control points, node vectors and weights, and its mathematical expression is:

[0233]

[0234] Where S(u,v) is the point on the curve, i.e., the optimal solution F for the failure rate of the energy meter under different working hours under various stress conditions. opt (S,N); m and n are the number of nodes in the u and v directions respectively, P i,j is a control point that will affect the final shape of the hypersurface; w i,j is the weight associated with the control point, which can be adjusted by i,j To change the closeness of the hypersurface to the control point; N i,p (u), N j,p (v) is defined as the non-periodic node vector U = [u0,u1...u n+k+1 ]、V=[v0,v1...v n+k+1 ] on the B-spline basis function, where N i,p The recursive formula for (u) is:

[0235]

[0236] The optimal solution of the failure rate of the electric energy meter under different working hours under various stress conditions is obtained by opt Substituting (S, N) into equation (21), we can obtain F after solving the parameters. opt The non-uniform rational B-spline expression of (S, N) is obtained, that is, the failure rate-multi-stress-failure time expression of a certain type of electric energy meter, and then the failure rate-multi-stress-failure time hypersurface failure rate diagram of the electric energy meter of this model is drawn.

[0237] In further embodiments, there is also provided:

[0238] An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in Embodiment 1 is performed. For the sake of brevity, no further description is given here.

[0239] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0240] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0241] A computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the method described in embodiment 1 is completed.

[0242] The method in the first embodiment can be directly embodied as a hardware processor, or a combination of hardware and software modules in the processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware. To avoid repetition, it will not be described in detail here.

[0243] A computer program product includes a computer program, and when the computer program is executed by a processor, the method described in the first embodiment is implemented.

[0244] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer executable instructions, such as instructions included in a program module, which are executed in a device on a real or virtual processor of the target to perform the process / method as described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc. that perform specific tasks or implement specific abstract data types. In various embodiments, the functions of program modules can be combined or divided between program modules as needed. Machine executable instructions for program modules can be executed in local or distributed devices. In distributed devices, program modules can be located in local and remote storage media.

[0245] The computer program code for implementing the method of the present invention can be written in one or more programming languages. These computer program codes can be provided to the processor of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the computer or other programmable data processing device, causes the function / operation specified in the flow chart and / or block diagram to be implemented. The program code can be executed completely on a computer, partially on a computer, as an independent software package, partially on a computer and partially on a remote computer or completely on a remote computer or server.

[0246] In the context of the present invention, computer program codes or related data may be carried by any appropriate carrier to enable a device, apparatus or processor to perform the various processes and operations described above. Examples of carriers include signals, computer readable media, etc. Examples of signals may include electrical, optical, radio, acoustic or other forms of propagation signals, such as carrier waves, infrared signals, etc.

[0247] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0248] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A method for predicting failure rate of electric energy meters, characterized in that: include: Based on the accelerated degradation test data of electric energy meters under different stress conditions and different working hours, the three-parameter Weibull distribution and reliability function of electric energy meters are constructed. The parameters in the three-parameter Weibull distribution of electric energy meters are optimized and solved using genetic algorithms. The failure rate hypersurface failure rate diagram of the electric energy meter failure rate-multiple stress-failure time relationship under the three-parameter Weibull distribution is constructed. The different stress conditions of the electric energy meter and the corresponding electric energy failure rate are used as training data to train the cumulative artificial neural network to obtain a trained prediction model. The trained prediction model is used to predict the failure rate of the electric energy meter under different stress conditions and different working times, and a hypersurface failure rate diagram of the electric energy meter failure rate-multiple stress-failure time relationship under the cumulative artificial neural network is constructed. Based on the Bayesian optimization algorithm, the failure rate diagram of the electric energy meter failure rate-multiple stress-failure time relationship hypersurface under the three-parameter Weibull distribution and the failure rate diagram of the electric energy meter failure rate-multiple stress-failure time relationship hypersurface under the cumulative artificial neural network are cross-checked to obtain the optimal solution for the prediction of the electric energy meter failure rate under different stress conditions and different working times. For the optimal solution of the failure rate prediction of the electric energy meter under different stress conditions and different working times, the final electric energy meter failure rate-multi-stress-failure time hypersurface failure rate diagram is drawn using non-uniform rational B-spline, and the failure rate prediction result of the electric energy meter to be predicted is obtained according to the drawn failure rate-multi-stress-failure time hypersurface failure rate diagram.

2. The method for predicting failure rate of electric energy meters according to claim 1, characterized in that: The construction of the failure rate diagram of the electric energy meter failure rate-multiple stress-failure time relationship hypersurface under the three-parameter Weibull distribution is specifically as follows: Based on the accelerated degradation test, the stress condition and failure time corresponding to the electric energy meter are obtained, a sample function is constructed based on the stress condition and failure time, and the constructed sample function is substituted into the three-parameter Weibull distribution function to obtain the failure rate-failure time relationship under the three-parameter Weibull distribution of the electric energy meter; According to the relationship between the remaining normal working time of the electric energy meter under the set failure rate requirement under normal temperature stress conditions and the failure time of the electric energy meter under accelerated degradation test conditions, combined with the failure rate-failure time relationship under the three-parameter Weibull distribution of the electric energy meter, a hypersurface failure rate diagram of the failure rate-multi-stress-failure time relationship of the electric energy meter under normal conditions is obtained.

3. The method for predicting failure rate of electric energy meters according to claim 1, characterized in that: Genetic algorithm is used to optimize the parameters in the three-parameter Weibull distribution of the electric energy meter, specifically: According to the reliability of different electric energy meters under different stress conditions and the three-parameter Weibull distribution corresponding to the failure time of different electric energy meters under different stress conditions, the objective function is constructed; Taking the minimum constructed objective function as the optimization goal, the three parameters are optimized and solved using genetic algorithm.

4. The method for predicting failure rate of electric energy meters according to claim 1, characterized in that: Based on the Bayesian optimization algorithm, the failure rate diagram of the electric energy meter failure rate-multiple stress-failure time relationship hypersurface under the three-parameter Weibull distribution and the failure rate diagram of the electric energy meter failure rate-multiple stress-failure time relationship hypersurface under the cumulative artificial neural network are cross-checked to obtain the optimal solution for the prediction of the failure rate of the electric energy meter under different stress conditions and different working times, which is as follows: Based on the failure rate diagram of the electric energy meter failure rate-multiple stress-failure time relationship hypersurface under the three-parameter Weibull distribution and the failure rate diagram of the electric energy meter failure rate-multiple stress-failure time relationship hypersurface under the cumulative artificial neural network, a model is built using the Gaussian process as a proxy model; The optimization method of maximizing expectation is adopted to obtain the optimal solution for the failure rate prediction of electric energy meters under different stress conditions and different working conditions.

5. The method for predicting failure rate of electric energy meters according to claim 1, characterized in that: The different environmental data of the electric energy meter and the corresponding electric energy failure rate are used as training data to train the cumulative artificial neural network and obtain a trained prediction model, which is as follows: According to the different temperatures, humidity, atmospheric pressure, load size, heating power, working time and the corresponding failure rate of the electricity meter as training data, the sum of squared errors is used as the loss function, and the cumulative artificial neural network is continuously iteratively trained to obtain a trained prediction model.

6. The method for predicting failure rate of electric energy meters according to claim 3, characterized in that: The failure part of the electric energy meter is divided into different units. Based on different stress types and the reliability of different failure units under different stress conditions, a failure rate diagram of the electric energy meter failure rate-multiple stress-failure time hypersurface based on the three-parameter Weibull distribution is drawn.

7. The method for predicting failure rate of electric energy meters according to claim 1, characterized in that: The electric energy meter failure rate-multiple stress-failure time relationship hypersurface failure rate diagram under the three-parameter Weibull distribution and the electric energy meter failure rate-multiple stress-failure time relationship hypersurface failure rate diagram under the cumulative artificial neural network are drawn using non-uniform rational B-splines respectively.

8. A power meter failure rate prediction system, characterized in that: include: The first prediction module is configured to: construct a three-parameter Weibull distribution and a reliability function of the electric energy meter based on the accelerated degradation test data of the electric energy meter under different stress conditions and different working times, optimize and solve the parameters in the three-parameter Weibull distribution of the electric energy meter by using a genetic algorithm, and construct a hypersurface failure rate diagram of the electric energy meter failure rate-multiple stress-failure time relationship under the three-parameter Weibull distribution; The second prediction module is configured to: use different stress conditions of the electric energy meter and the corresponding electric energy failure rates as training data to train the cumulative artificial neural network to obtain a trained prediction model, use the trained prediction model to predict the failure rates of the electric energy meter under different stress conditions and different working times, and construct a failure rate diagram of the electric energy meter failure rate-multi-stress-failure time relationship hypersurface under the cumulative artificial neural network; A mutual verification module is configured to: perform mutual verification on the failure rate diagram of the electric energy meter failure rate-multiple stress-failure time relationship hypersurface under the three-parameter Weibull distribution and the failure rate diagram of the electric energy meter failure rate-multiple stress-failure time relationship hypersurface under the cumulative artificial neural network based on the Bayesian optimization algorithm, and obtain the optimal solution for the prediction of the failure rate of the electric energy meter under different stress conditions and different working times; The third prediction module is configured to: for the optimal solution of the failure rate prediction of the electric energy meter under different stress conditions and different working times, use non-uniform rational B-splines to draw the final electric energy meter failure rate-multi-stress-failure time hypersurface failure rate diagram, and obtain the failure rate prediction result of the electric energy meter to be predicted according to the drawn failure rate-multi-stress-failure time hypersurface failure rate diagram.

9. An electronic device, characterized in that: The method comprises a memory and a processor and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method according to any one of claims 1 to 7 is completed.

10. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the method described in any one of claims 1 to 7.

Images