BHT-ARIMA-based relay protection device fault rate prediction method
Through the BHT-ARIMA method, a reliability distribution model of the relay protection device is established and a comprehensive moving average algorithm for block Hankel tensor autoregression is solved, which is difficult to predict the failure rate in the prior art. It realizes high-precision failure rate prediction, and improves the safe and stable operation and device utilization of the power station.
Patent Information
- Application Number
- CN202410550096.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-06
- Publication Date
- 2024-11-22
AI Technical Summary
The prior art is difficult to accurately and quickly predict the failure rate of relay protection devices, resulting in difficulties in maintenance and maintenance, affecting the safe and stable operation of the power station and the utilization rate of the device.
Using a BHT-ARIMA-based method, the intrinsic features and correlations of the failure rate time series are captured by establishing the reliability distribution model of the relay protection device, using a one-way delayed embedding transformation and block Hankel tensor autoregression comprehensive moving average algorithm. extract potential information for prediction.
Accurate and fast failure rate prediction is achieved, the prediction accuracy is improved, the device utilization is improved, and the maintenance cost is reduced, which is significantly better than the traditional ARIMA method.
Smart Images

Figure CN119003968A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of inspection and maintenance of current transformers in hydropower plants, and in particular to a fault rate prediction method for relay protection devices based on BHT-ARIMA. Background Art
[0002] Relay protection is the first line of defense for the power system. The reliability of its protection device is crucial. A lot of manpower and material resources are spent on the operation and inspection of protection devices every year. The main reason is that the protection device is an independent and real-time entity, and it is difficult to evaluate and diagnose its status online. The protection device is a microcomputer protection with high reliability. However, under the influence of temperature, humidity, salinity and other environmental factors during the long-term uninterrupted operation, some chips are damaged by aging, relays have node adhesion, and the power supply voltage is unstable, which leads to abnormal exit of the protection device. In severe cases, it will cause false operation and produce bad consequences.
[0003] Therefore, accurate and rapid prediction of the failure rate of relay protection devices is of great significance to the safe and stable operation of power stations. It is also necessary for reliability-centered maintenance and overhaul strategies. It also helps to improve device utilization and reduce cost losses. Summary of the invention
[0004] In order to solve the current technical problems, the main purpose of the present invention is to provide a relay protection device failure rate prediction method based on BHT-ARIMA, which can accurately and quickly predict the failure rate of the relay protection device; in addition, this method can effectively mine and analyze the implicit features in the time series of the failure rate of the relay protection device to ensure the prediction accuracy; moreover, this method is also necessary for the reliability-centered maintenance and overhaul strategy, which can help to improve the utilization rate of the device and reduce cost losses.
[0005] In order to realize the above technical features, the object of the present invention is achieved as follows: 1. A method for predicting the failure rate of a relay protection device based on BHT-ARIMA, characterized in that it comprises the following steps:
[0006] Step 1: Establish a reliability distribution model of relay protection device based on mechanism analysis;
[0007] Step 2, the generated failure rate time series is used as the input of the block Hankel tensor autoregressive integrated moving average;
[0008] Step 3: One-way delayed embedding transform and block Hankel tensor autoregressive integrated moving average algorithm are used to obtain the predicted failure rate of relay protection.
[0009] 2. According to the BHT-ARIMA-based relay protection device failure rate prediction method of claim 1, it is characterized in that the specific establishment process of the relay protection device reliability distribution model in step 1 is:
[0010] The life distribution of the relay protection device of the hydropower station is a continuous random variable. The reliability distribution model adopts the two-parameter Weibull distribution, and the reliability function is expressed as:
[0011]
[0012] Where, t age represents time; γ represents scale parameter; δ represents shape parameter;
[0013] Then the failure rate distribution function is expressed as:
[0014]
[0015] In the formula, λ age represents the failure rate distribution function.
[0016] 3. According to the BHT-ARIMA-based relay protection device failure rate prediction method of claim 2, it is characterized in that the specific generation process of the failure rate time series in step 2 is:
[0017] One-way delay embedding transform is used to capture the intrinsic time correlation of the fault rate of relay protection devices.
[0018] 4. According to the BHT-ARIMA-based relay protection device failure rate prediction method of claim 3, it is characterized in that the specific steps in step 2 are:
[0019] Step 2.1, counting the number of relay protection devices, and then obtaining the number of relay protection devices to be monitored, assuming that the total number is K;
[0020] Step 2.2: Based on the time series data of the relay protection device failure rate obtained by formula (2) in step 1, a K×t-order failure rate time series matrix is formed, which is recorded as
[0021]
[0022] In the formula, K represents the total number of relay protection devices; t represents the time scale; represents the failure rate of the Kth relay protection device at time t;
[0023] Step 2.3: Use the one-way delayed embedding transformation method to transform the relay protection device failure rate into a high-order tensor along the time direction.
[0024]
[0025] In the formula, represents the replication matrix; H ο represents the Hankelized matrix; Fold (K,ο) Represents a one-way delayed transformation operator that transforms the input tensor Hankel transformation to higher order tensors in, In the Hankel tensor In this case, J = K, and ο represents the rank of the single-phase delay transformation;
[0026] The inverse transform is expressed as:
[0027]
[0028] In the formula, H ο -1 Represents the Hankelized inverse matrix; Unfold (I,ο) represents a one-way delayed inverse transform operator; It is the Moore-Penrose pseudo-inverse;
[0029] The fault rate matrix of relay protection device is transformed into Convert to third-order Hankel tensor in, X represents the Hankel high-rank tensor.
[0030] 5. According to the BHT-ARIMA-based relay protection device failure rate prediction method of claim 4, it is characterized in that the specific steps in step 3 are:
[0031] Step 3.1, extract the core tensor using Tucker decomposition;
[0032] Tucker decomposition can extract More valuable information will Represented as a tensor with a projection matrix:
[0033]
[0034] In the formula, is the low-rank core tensor, is the joint orthogonal factor matrix;
[0035] Step 3.2, explicit expression of the block Hankel tensor autoregressive integrated mobility model:
[0036] Assume l tis the actual data value at the tth time point, then the autoregressive moving average model is:
[0037]
[0038] In the formula, for The b-order difference of Indicates t i The prediction error when represents the u-th actual error; and is the coefficient of the autoregressive moving average model; y, b, u represent the yth order difference, bth order difference, and uth random error in the past, respectively;
[0039] The block Hankel tensor autoregressive integrated mobility model is expressed as:
[0040]
[0041] In the formula, for The b-order difference of ti Indicates t i The prediction error at the moment; ε ti-j represents the actual error of the past u observations;
[0042] Step 3.3, calculate the b-order difference of the Hankel core tensor of the relay protection device failure rate,
[0043] Will Projection to core tensor On the top, low-rank Tucker decomposition is achieved, t i Core tensor difference at time It is expressed as:
[0044]
[0045] In the formula, Respectively represent the failure rate of relay protection device t i The prediction error at the moment and the actual error of the past u observations; in practical problems, we hope that the prediction error Can drop to 0, based on this, the following optimization model is formed:
[0046]
[0047] In the formula, s = y + b + u, N = 3, I is the unit matrix, according to the above optimization model, the solution is
[0048]
[0049] In the formula, Indicates the time scale;
[0050] Step 3.4, Refactoring
[0051] get After predicting the value, we reconstruct Predicted value:
[0052]
[0053] right Perform inverse difference to obtain Finally, the inverse one-way delay embedding transform is applied to obtain the predicted value of the relay protection device failure rate time series
[0054] The present invention has the following beneficial effects:
[0055] 1. This prediction method can accurately and quickly predict the failure rate of relay protection devices; in addition, this method can effectively mine and analyze the implicit features in the time series of the failure rate of relay protection devices to ensure the prediction accuracy; moreover, this method is also necessary for reliability-centered maintenance and overhaul strategies, which can help improve device utilization and reduce cost losses.
[0056] 2. The one-way delayed embedding transform captures the intrinsic time correlation of multiple time series of historical data on the failure rate of relay protection devices. BHT-ARIMA integrates methods such as SET, Tucker decomposition and tensor ARIMA, and utilizes the advantages of Hankel tensorization. It is significantly superior to existing short time series prediction methods such as ARIMA in terms of effectiveness and efficiency.
[0057] 3. The present invention improves the prediction accuracy of the failure rate of relay protection devices by extracting and utilizing the embedded potential information between the historical data of the failure rate of relay protection devices, adopting a one-way delayed embedding transformation and a multi-series prediction model based on block Hankel tensor autoregressive integrated moving average. BRIEF DESCRIPTION OF THE DRAWINGS
[0058] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.
[0059] Figure 1 It is a flow chart of the method of the present invention.
[0060] Figure 2 This is a diagram of the prediction results of the present invention.
[0061] Figure 3 This is a diagram of the prediction process of the present invention.
[0062] Figure 4 This is a system diagram of the present invention. DETAILED DESCRIPTION
[0063] The embodiments of the present invention are further described below in conjunction with the accompanying drawings.
[0064] Embodiment 1:
[0065] Terminology Note:
[0066] Autoregressive moving average model (denoted as ARIMA);
[0067] One-way delayed embedding transform (denoted as SET);
[0068] Block Hankel Tensor Autoregressive Integrated Moving Average (denoted as BHT-ARIMA);
[0069] Tucker (remembered as Tucker).
[0070] Recurrent Neural Network (RNN).
[0071] The defects in the prior art are:
[0072] 1) There is little data on the failure rate of relay protection devices, and existing prediction models cannot effectively and accurately predict the future failure rate of relay protection devices.
[0073] 2) ARIMA is a single-input single-output forecasting method that ignores the correlation between technical features. This method cannot capture the nonlinear relationship in the time series of relay protection device failure rate.
[0074] 3) Due to the simple neuron structure of RNN, there are serious gradient vanishing and gradient exploding problems during the training process, which makes it still unable to accurately describe long-span nonlinear relationships.
[0075] Based on this, refer to Figure 1-4 This embodiment provides a method for predicting the failure rate of a relay protection device based on BHT-ARIMA, comprising the following steps:
[0076] A method for predicting the failure rate of a relay protection device based on BHT-ARIMA comprises the following steps:
[0077] Step 1: Establish a reliability distribution model of relay protection device based on mechanism analysis;
[0078] Weibull distribution has been widely used in reliability analysis and can well describe the reliability bathtub curve characteristics of products. The life of relay protection devices obeys Weibull distribution. The life distribution of hydropower station relay protection equipment is a continuous random variable. The reliability model adopts the two-parameter Weibull distribution, and the reliability function can be expressed as:
[0079]
[0080] Where, t age represents time; γ represents scale parameter; δ represents shape parameter;
[0081] Then the failure rate distribution function is expressed as:
[0082]
[0083] In the formula, λ age represents the failure rate distribution function.
[0084] Step 2, the generated failure rate time series is used as the input of the block Hankel tensor autoregressive integrated moving average;
[0085] In actual projects, power companies are more interested in the current or future failure rates of relay protection devices in order to achieve monitoring, analysis and early warning of the failure rates of relay protection devices.
[0086] The failure rate of relay protection devices is time-dependent, but the failure rate data is very limited, which makes the existing prediction model unable to effectively and accurately predict the failure rate of relay protection devices. Therefore, the focus of the present invention is to find a model that can mine the intrinsic correlation of the failure rate time series and propose a prediction method suitable for short-term prediction.
[0087] The present invention proposes a one-way delayed embedding transform (denoted as SET) and a multi-sequence prediction model based on block Hankel tensor autoregressive integrated moving average (denoted as BHT-ARIMA) to predict the failure rate of relay protection devices. The specific steps are as follows:
[0088] Step 2.1, counting the number of relay protection devices, and then obtaining the number of relay protection devices to be monitored, assuming that the total number is K;
[0089] Step 2.2: Based on the time series data of the relay protection device failure rate obtained by formula (2) in step 1, a K×t-order failure rate time series matrix is formed, which is recorded as
[0090]
[0091] In the formula, K represents the total number of relay protection devices; t represents the time scale; represents the failure rate of the Kth relay protection device at time t;
[0092] Step 2.3: Use the one-way delayed embedding transformation method to transform the relay protection device failure rate into a high-order tensor along the time direction.
[0093]
[0094] In the formula, represents the replication matrix; H ο represents the Hankelized matrix; Fold (K,ο) Represents a one-way delayed transformation operator that transforms the input tensor Hankel transformation to higher order tensors in, In the Hankel tensor In this case, J = K, and ο represents the rank of the single-phase delay transformation;
[0095] The inverse transform is expressed as:
[0096]
[0097] In the formula, H ο -1 Represents the Hankelized inverse matrix; Unfold (I,ο) represents a one-way delayed inverse transform operator; It is the Moore-Penrose pseudo-inverse;
[0098] The fault rate matrix of relay protection device is transformed into Convert to third-order Hankel tensor in, X represents the Hankel high-order tensor;
[0099] Step 3, using one-way delayed embedding transform and block Hankel tensor autoregressive integrated moving average algorithm to obtain the relay protection predicted failure rate;
[0100] Block Hankel Tensor Autoregressive Integrated Moving Average (BHT-ARIMA) based on Tucker decomposition;
[0101] Step 3.1, extract the core tensor using Tucker decomposition;
[0102] Tucker decomposition can extract More valuable information will Represented as a tensor with a projection matrix:
[0103]
[0104] In the formula, is the low-rank core tensor, is the joint orthogonal factor matrix;
[0105] Step 3.2, explicit expression of the Block Hankel Tensor Autoregressive Integrated Moving Average (BHT-ARIMA) model:
[0106] Assume l t is the actual data value at the tth time point, then the ARIMA (y, b, u) model is:
[0107]
[0108] In the formula, for The b-order difference of Indicates t i The prediction error when represents the u-th actual error; and is the ARIMA coefficient;
[0109] In the prediction of multiple short time series, ARIMA needs to predict each event sequence one by one, which leads to slow calculation speed. At the same time, it is also unable to consider the correlation between CT cable health time series, resulting in poor prediction effect. Therefore, the BHT-ARIMA model is proposed to explicitly express the continuous core tensor ARIMA model to solve the problem of calculation effect and efficiency. The BHT-ARIMA (y, b, u) model can be expressed as:
[0110]
[0111] In the formula, for The b-order difference of represents the actual error of the past u observations; Indicates t i The prediction error at the moment; y, b, u represent the past y-th order difference, b-th order difference, and past u-th random error respectively;
[0112] Step 3.3, calculate the b-order difference of the Hankel core tensor of the relay protection device failure rate,
[0113] Will Projection to core tensor On the top, low-rank Tucker decomposition is achieved, t i Core tensor difference at time It is expressed as:
[0114]
[0115] In the formula, Respectively represent the failure rate of relay protection device t i The prediction error at the moment and the actual error of the past u observations; in practical problems, we hope that the prediction error Can drop to 0, based on this, the following optimization model is formed:
[0116]
[0117] In the formula, s = y + b + u, N = 3, I is the unit matrix, according to the above optimization model, the solution is
[0118]
[0119] In the formula, Indicates the time scale;
[0120] Step 3.4, Refactoring
[0121] get After predicting the value, we reconstruct Predicted value:
[0122]
[0123] right Perform inverse difference to obtain Finally, the inverse one-way delay embedding transform is applied to obtain the predicted value of the relay protection device failure rate time series
[0124] Embodiment 2:
[0125] Use the model for validation
[0126] Taking the actual failure rate of a certain relay protection device as an example, the paper illustrates the effectiveness, feasibility and superiority of a proposed relay protection device failure rate prediction method based on BHT-ARIMA through example analysis.
[0127] Figure 2 The prediction results of the BHT-ARIMA prediction method and the ARIMA method constructed by the present invention are given to illustrate the accuracy of the method proposed by the present invention. In addition, in order to facilitate quantitative analysis and more intuitively display the accuracy of various algorithms, Table 1 gives the root mean square error values of the BHT-ARIMA prediction method and the ARIMA method constructed by the present invention.
[0128] The calculation formula of the root mean square error (denoted as RMSE) is:
[0129]
[0130] Where n represents the nth prediction time, and N represents the total number of prediction times. n , Y n ′ respectively represent the predicted nth failure rate and the nth failure rate historical data. RMSE represents the root mean square error.
[0131] It can be seen from Table 1 that the RMSE value of the proposed BHT-ARIMA method is reduced by 87.07% compared with the ARIMA prediction method. That is, the proposed method has better performance in predicting the failure rate of relay protection devices. Therefore, it can be concluded that the proposed BHT-ARIMA-based relay protection device failure rate prediction method can accurately predict the failure rate of relay protection devices.
[0132] Table 1 RMSE comparison of different prediction methods
[0133]
[0134] Embodiment 3:
[0135] The failure rate prediction method of relay protection device is:
[0136]
[0137]
[0138] Embodiment 4:
[0139] See also Figure 4 , using a display screen for display. The application client of this project is deployed in the intranet and can be accessed through the subsystem under the Three Gorges Power Plant website. The server of this project is deployed in Zone III. The failure rate of the relay protection device is obtained from the industrial Internet platform in Zone IV. To ensure network security, a firewall needs to be installed between the two zones, and customized software needs to be deployed on the big data platform side of Zone IV for data transmission.
Claims
1. A method for predicting the failure rate of relay protection devices based on BHT-ARIMA, characterized in that: The following steps are involved: Step 1: Establish a reliability distribution model of relay protection device based on mechanism analysis; Step 2: Use the generated failure rate time series as the input of the block Hankel tensor autoregressive integrated moving average; In step 3, the one-way delayed embedding transform and block Hankel tensor autoregressive integrated moving average algorithm are used to obtain the predicted failure rate of relay protection.
2. The method for predicting the failure rate of a relay protection device based on BHT-ARIMA according to claim 1, characterized in that: The specific process of establishing the reliability distribution model of the relay protection device in step 1 is as follows: The life distribution of the relay protection device of a hydropower station is a continuous random variable. The reliability distribution model adopts the two-parameter Weibull distribution, and the reliability function is expressed as: Where, t age represents time; γ represents scale parameter; δ represents shape parameter; Then the failure rate distribution function is expressed as: Where λ age represents the failure rate distribution function.
3. The method for predicting the failure rate of a relay protection device based on BHT-ARIMA according to claim 2, characterized in that: The specific generation process of the failure rate time series in step 2 is: One-way delay embedding transform is used to capture the inherent time correlation of the failure rate of relay protection devices.
4. The method for predicting the failure rate of a relay protection device based on BHT-ARIMA according to claim 3 is characterized in that: The specific steps in step 2 are: Step 2.1, count the number of relay protection devices, and then obtain the number of relay protection devices to be monitored, assuming the total number is K; Step 2.2: Based on the time series data of the relay protection device failure rate obtained by formula (2) in step 1, a K×t-order failure rate time series matrix is formed, which is recorded as Where K represents the total number of relay protection devices; t represents the time scale; represents the failure rate of the Kth relay protection device at time t; Step 2.3: Use the one-way delayed embedding transformation method to transform the relay protection device failure rate into a high-order tensor along the time direction. Where, represents the replication matrix; H ο represents the Hankelized matrix; Fold (K,ο) Represents a one-way delayed transformation operator that takes the input tensor Hankel transformation to higher-order tensors in, In the Hankel tensor In the equation, J=K, and ο represents the rank of the single-phase delay transformation; The inverse transform is expressed as: Where, Represents the Hankelized inverse matrix; Unfold (I,ο) represents the one-way delayed inverse transform operator; It is the Moore-Penrose pseudo-inverse; The failure rate matrix of relay protection device is transformed into Convert to a third-order Hankel tensor in, X represents a Hankel high-rank tensor.
5. The method for predicting the failure rate of a relay protection device based on BHT-ARIMA according to claim 4, characterized in that: The specific steps in step 3 are: Step 3.1, extract the core tensor using Tucker decomposition; Tucker decomposition can extract while maintaining high-order structure More valuable information will be Represented as a tensor with a projection matrix: Where, is the low-rank core tensor, n=1,...,N is the joint orthogonal factor matrix; Step 3.2, explicit expression of the block Hankel tensor autoregressive integrated mobility model: Assume l t is the actual data value at the t-th time point, then the autoregressive moving average model is: Where, for The b-order difference of Indicates t i The prediction error when represents the u-th actual error; and is the coefficient of the autoregressive moving average model; y, b, and u represent the past y-th order difference, b-th order difference, and past u-th random error respectively; The block Hankel tensor autoregressive integrated mobility model is expressed as: Where, for b-order difference of ; Indicates t i The prediction error at the moment; represents the actual error of the past u observations; Step 3.3, calculate the b-order difference of the Hankel core tensor of the relay protection device failure rate, Will Projection to core tensor On the top, realize low-rank Tucker decomposition, t i Core tensor difference at time Expressed as: Where, Represent the failure rate of relay protection device t i The prediction error at the moment and the actual error of the past u observations; In practical problems, we hope that the prediction error Can drop to 0, based on this, the following optimization model is formed: Where s = y + b + u, N = 3, I is the unit matrix, according to the above optimization model, we can get Where, Indicates time scale; Step 3.4, Reconstruction get After predicting the value, we reconstruct it through Tucker decomposition. Predicted value: right Perform inverse difference to obtain Finally, the inverse one-way delay embedding transform is applied to obtain the predicted value of the relay protection device failure rate time series
