A smart fusion terminal with accurate prediction of operational faults

Abstract

This invention relates to the field of data processing technology, and in particular to an intelligent fusion terminal with accurate operational fault prediction function. The system includes a processor and a memory. The processor executes a computer program in the memory to perform the following steps: obtaining the normality of the residual distribution in the residual sequence corresponding to the initial SARIMA model that has passed the Ljung-Box test; if the normality of the residual distribution is less than a preset normality threshold, dividing the residual sequence into at least two subsequences, obtaining the secondary test judgment factor of the initial SARIMA model based on the outlier distribution and trend strength in each subsequence, and performing a secondary test on the initial SARIMA model to obtain the optimal SARIMA model; and deploying the optimal SARIMA model to the intelligent fusion terminal to improve the accuracy of operational fault prediction for distribution network areas.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to an intelligent fusion terminal with accurate prediction function of operational faults. Background Technology

[0002] Intelligent fusion terminals are typically deployed in substations, switchgear, or integrated monitoring nodes to collect and centrally process various operating parameters of the distribution network area in real time, such as voltage, current, power, and equipment temperature. Because this type of operating data often exhibits periodic variations, such as load fluctuations during the day-night cycle and the phased patterns formed during equipment operation, existing technologies often employ an autoregressive integral moving average model with a seasonal term (SARIMA model) to model and analyze the monitoring data. To ensure that the SARIMA model accurately depicts the changing patterns of the monitoring data, model validation is usually required after the SARIMA model is established to guarantee the reliability of the prediction model and the accuracy of the prediction results.

[0003] The traditional approach is to perform the Ljung-Box test (white noise test) on the SARIMA model. This method calculates the autocorrelation statistics of the residual sequence at multiple lag orders to statistically test whether there is still a significant correlation in the residuals. The less significant the Ljung-Box test result (i.e., the larger the p-value), the cleaner the value is extracted from the data, and the better the SARIMA model is.

[0004] During the operation monitoring of intelligent fusion terminals, the data structure exhibited by anomalies is not continuously stable, but is closely related to the operating conditions of the equipment. During the day's operation, the equipment often experiences different load phases. For example, it maintains low load operation at night and a relatively stable workload during the day. At this time, the prediction error of the SARIMA model is relatively small, and the residuals mainly exhibit random fluctuations. However, during the high-load operation phase of a certain peak electricity consumption period, due to the sudden increase in production tasks, the equipment load gradually increases. The SARIMA model still makes predictions according to the original pattern, so the prediction error will continuously increase. However, since the duration of this continuous increase is relatively short compared to the entire historical time, it is easily masked by a large number of normal residuals in the overall statistical calculation process. Statistically, no significant autocorrelation structure is detected, leading to the Ljung-Box test result indicating that the SARIMA model has fully explained the data structure. This results in missed detections during the model diagnosis phase, affecting the accuracy of subsequent operational fault prediction for distribution network areas.

[0005] Therefore, improving the accuracy of SARIMA model verification, and thus improving the accuracy of intelligent fusion terminals in predicting operational faults in distribution network areas, has become an urgent problem to be solved. Summary of the Invention

[0006] In view of this, embodiments of the present invention provide an intelligent fusion terminal with accurate prediction function of operational faults, in order to solve the problem of how to improve the accuracy of SARIMA model verification, and thus improve the accuracy of intelligent fusion terminal in predicting operational faults in distribution network areas.

[0007] This invention provides an intelligent fusion terminal with accurate fault prediction function, including a memory, a processor, and a computer program stored in the memory and running on the processor. The processor executes the computer program to perform the following steps:

[0008] For any operating parameter monitored by the target intelligent fusion terminal, an initial SARIMA model for the operating parameter that has passed the Ljung-Box test is obtained. The training set of the operating parameter is input into the initial SARIMA model to obtain a residual sequence. Based on the deviation between the residual distribution in the residual sequence and the standard normal distribution, the normality of the residual sequence is obtained.

[0009] If the normality proximity is less than the preset normality proximity threshold, then the outliers in the residual sequence are obtained using the box plot method. Based on the preset time sliding window and the preset sliding step size, the residual sequence is divided into at least two subsequences. Based on the outlier distribution and trend strength in each subsequence, the secondary test judgment factor of the initial SARIMA model is obtained.

[0010] Based on the secondary test judgment factor, the initial SARIMA model is subjected to a secondary test to obtain the secondary test result. Based on the secondary test result, the initial SARIMA model is optimized to obtain the optimal SARIMA model for any operating parameter.

[0011] The optimal SARIMA model for each operating parameter monitored by the target intelligent fusion terminal is obtained, and the optimal SARIMA model for each operating parameter is deployed to the target intelligent fusion terminal for prediction of the data of each operating parameter, so as to achieve accurate prediction of the operating faults of the distribution network area.

[0012] Preferably, obtaining the normality approximation of the residual sequence based on the deviation between the residual distribution in the residual sequence and the standard normal distribution includes:

[0013] Obtain the cumulative distribution function of the residual sequence and the cumulative distribution function of the standard normal distribution. Calculate the absolute value function of the difference between the cumulative distribution function of the residual sequence and the cumulative distribution function of the standard normal distribution. Form an integration interval using the maximum and minimum values ​​in the residual sequence. Integrate the absolute value function of the difference within the integration interval to obtain the integration result. Use the negative of the integration result as the independent variable of the natural exponential function to obtain the normality approximation of the residual sequence.

[0014] Preferably, obtaining the secondary test judgment factor of the initial SARIMA model based on the outlier distribution and trend strength in each subsequence includes:

[0015] The number of outliers in each subsequence is obtained. Subsequences with more than 1 outliers are recorded as subsequences to be analyzed, resulting in at least one subsequence to be analyzed. For any subsequence to be analyzed, the outlier residual density is obtained based on the number of outliers in the subsequence and the time interval between two adjacent outliers.

[0016] Obtain the abnormal residual density of each of the subsequences to be analyzed, and denote the subsequence to be analyzed corresponding to the maximum abnormal residual density as the target subsequence. Based on the trend significance and trend persistence of the target subsequence, obtain the residual trend strength of the target subsequence.

[0017] The mean between the abnormal residual density and the residual trend strength of the target subsequence is calculated to obtain the secondary test judgment factor of the initial SARIMA model.

[0018] Preferably, obtaining the outlier residual density of any subsequence to be analyzed based on the number of outliers in any subsequence to be analyzed and the time interval between two adjacent outliers includes:

[0019] Calculate the proportion of the number of outliers in any subsequence to be analyzed to the total number of data in any subsequence to be analyzed. Use the proportion as the independent variable of the natural exponential function to obtain the exponential function value. Use the exponential function value as the independent variable of the hyperbolic tangent function to obtain the first density.

[0020] Calculate the time interval between any two adjacent outliers in any subsequence to be analyzed, subtract a constant 1 from each time interval to obtain the normal run length, and use the negative of the mean of all normal run lengths as the independent variable of the natural exponential function to obtain the second density.

[0021] The mean between the first density and the second density is calculated to obtain the abnormal residual density of any subsequence to be analyzed.

[0022] Preferably, obtaining the residual trend strength of the target subsequence based on the trend significance and trend persistence of the target subsequence includes:

[0023] Linear fitting is performed on all outliers in the target subsequence to obtain the fitted value of each outlier in the target subsequence. The absolute value of the difference between each outlier in the target subsequence and its fitted value is calculated. The negative of the sum of all the absolute values ​​of the difference is used as the independent variable of the natural exponential function to obtain the significance of the outlier residual trend of the target subsequence.

[0024] If the number of outliers in the target subsequence is greater than or equal to 3, then for any outlier in the target subsequence other than the first and last outlier, obtain the left and right adjacent outliers of the outlier, subtract the left adjacent outlier from the outlier to obtain the left trend value of the outlier, subtract the right adjacent outlier from the outlier to obtain the right trend value of the outlier, calculate the product between the left trend value and the right trend value, if the product is greater than 0, then set the trend persistence feature value of the outlier to 1, if the product is less than or equal to 0, then set the trend persistence feature value of the outlier to 0.

[0025] The trend persistence feature value of each outlier in the target subsequence, excluding the first and last outliers, is obtained respectively. The mean of all trend persistence feature values ​​is calculated to obtain the trend persistence of the target subsequence.

[0026] If the number of outliers in the target subsequence is less than 3, then the trend persistence of the target subsequence is set to 0.

[0027] The residual trend strength of the target subsequence is obtained by calculating the mean between the significance of the trend and the persistence of the trend.

[0028] Preferably, the initial SARIMA model is subjected to a secondary test based on the secondary test judgment factor to obtain the secondary test result, including:

[0029] If the secondary test judgment factor is less than the preset secondary test threshold, then the initial SARIMA model is determined to have passed the secondary test.

[0030] If the secondary test judgment factor is greater than or equal to the preset secondary test threshold, then the initial SARIMA model is determined to fail the secondary test.

[0031] Preferably, optimizing the initial SARIMA model based on the secondary test results to obtain the optimal SARIMA model for any operating parameter includes:

[0032] If the initial SARIMA model passes the quadratic test, then the initial SARIMA model is taken as the optimal SARIMA model for any of the operating parameters;

[0033] If the initial SARIMA model fails the quadratic test, the hyperparameters of the initial SARIMA model are optimized to obtain a new SARIMA model for any operating parameter that passes the Ljung-Box test. This new SARIMA model is then used as the initial SARIMA model. This process continues until the normality of the residual sequence corresponding to the initial SARIMA model is greater than or equal to a preset normality threshold, or the quadratic test judgment factor of the initial SARIMA model is less than a preset quadratic test threshold. Finally, the optimal SARIMA model for any operating parameter is obtained.

[0034] Preferably, if the normality proximity is greater than or equal to the pre-normality proximity threshold, the initial SARIMA model is determined to have passed the quadratic test, and the initial SARIMA model is taken as the optimal SARIMA model for any of the operating parameters.

[0035] The beneficial effects of the embodiments of the present invention compared with the prior art are as follows:

[0036] This invention first determines the normality of the residual sequence by analyzing the deviation between the residual distribution and the standard normal distribution in the residual sequence corresponding to the initial SARIMA model that has passed the Ljung-Box test. This allows for an overall statistical analysis of the residual sequence to determine whether the residuals still maintain random fluctuation characteristics. Further, based on the outlier distribution and trend strength in each subsequence of the residual sequence, the specific reasons for the deviation of the residual distribution from the normal distribution are analyzed, and a secondary test judgment factor for the initial SARIMA model is obtained. This identifies whether the deviation is caused by local structural changes not explained by the model. Based on the secondary test judgment factor, a secondary test is performed on the initial SARIMA model that has passed the Ljung-Box test, improving the sensitivity to local anomalies while ensuring the overall stability of the model. Finally, the initial SARIMA model that has passed the secondary test is deployed in an intelligent fusion terminal, enabling the intelligent fusion terminal to identify potential fault risks earlier and improve the accuracy and reliability of monitoring operational faults. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a flowchart of an implementation method for an intelligent fusion terminal with accurate prediction function of operational faults, provided in Embodiment 1 of the present invention. Detailed Implementation

[0039] Embodiments of this disclosure are described in detail below, with examples of these embodiments illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this disclosure, and should not be construed as limiting it.

[0040] It should be noted that the terms "first," "second," etc., used in this disclosure and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure.

[0041] To illustrate the technical solution of the present invention, specific embodiments are described below.

[0042] This invention provides an intelligent fusion terminal with accurate operational fault prediction function, including a processor and a memory. The processor executes a computer program stored in the memory to implement a method for implementing an intelligent fusion terminal with accurate operational fault prediction function, such as... Figure 1 As shown, the method includes the following steps:

[0043] Step S101: For any operating parameter monitored by the target intelligent fusion terminal, obtain the initial SARIMA model of the operating parameter that has passed the Ljung-Box test, input the training set of the operating parameter into the initial SARIMA model to obtain the residual sequence, and obtain the normality approximation of the residual sequence based on the deviation between the residual distribution in the residual sequence and the standard normal distribution.

[0044] Intelligent fusion terminals are typically deployed in substations, switchgear, or integrated monitoring nodes to collect and centrally process various operating parameters such as voltage, current, power, and equipment temperature in the distribution network area in real time. Since such operating data often has certain periodic variation characteristics, SARIMA models are often used to model and analyze the monitoring data in order to predict the monitoring data of various operating parameters at future times and achieve accurate prediction of operating faults in the distribution network area.

[0045] Therefore, in this embodiment of the invention, any intelligent fusion terminal at any location is designated as the target intelligent fusion terminal. The voltage, current, power, and equipment temperature at that location are used as the operating parameters monitored by the target intelligent fusion terminal. No restrictions are imposed here; implementers can set these parameters according to specific scenarios. Since the SARIMA model is a univariate time series model, a separate SARIMA model needs to be constructed for each operating parameter. Taking current as an example, a training set of current is obtained based on historical current data at that location. This training set must include at least three complete seasonal cycles. For example, if the seasonal cycle is daily, the training set must include at least three weeks of historical current data; if the seasonal cycle is monthly, the training set must include at least three months of historical current data. Furthermore, the data in the training set has undergone time synchronization processing, missing value processing, and standardization processing. Time synchronization processing, missing value processing, and standardization processing are existing technologies and will not be elaborated here. Further, the SARIMA model is trained based on the current training set to obtain the SARIMA model to be validated, and the SARIMA model to be validated is subjected to an Ljung-Box test.

[0046] The Ljung-Box test statistically examines the overall autocorrelation of the residual sequence output by the SARIMA model to be validated. It determines whether the residuals exhibit white noise by assessing the relationship between the statistic and the chi-square distribution, thus indicating whether the model still contains unexplained data structures. However, during the operation monitoring of intelligent fusion terminals, different load phases often occur, resulting in anomalies whose data structures are not consistently stable but closely related to equipment operating conditions. For example, during nighttime low-load operation and daytime relatively stable workloads, the model's prediction error is relatively small, and the residuals mainly exhibit random fluctuations. During peak electricity consumption periods with high loads, a sudden increase in production tasks leads to a gradual increase in equipment load. The model continues to predict according to its original pattern, causing the prediction error to continuously increase. However, because this continuous increase lasts relatively short compared to the entire historical period, it is easily masked by a large number of normal residuals during the overall statistical calculation, making the residual sequence appear close to white noise in the overall statistical sense. This leads to the model being deemed to have passed the test, thus affecting the accuracy of subsequent fault predictions for distribution network areas.

[0047] Based on the above, in order to further identify structural changes in the residual sequence that may be masked by the overall statistical process, in this embodiment of the invention, the SARIMA model to be verified through the Ljung-Box test is designated as the initial SARIMA model, and the current training set is input into the initial SARIMA model (the same training set is usually used repeatedly before the optimal model is obtained) to obtain the residual sequence corresponding to the initial SARIMA model. By supplementing the overall statistical form of the residual sequence corresponding to the initial SARIMA model through supplementary analysis, the initial SARIMA model is subjected to a second test to reduce the possibility that local changes may be diluted by a large number of random residuals in the overall statistical process, thereby improving the accuracy of detecting model quality and improving the accuracy of predicting monitoring data of various operating parameters at future times, thus achieving accurate prediction of operating faults in the distribution network area.

[0048] Considering that in time series modeling theory, when the model can fully explain the data structure, the residuals should usually be a sequence with a mean close to zero and fluctuations approximately random. This distribution characteristic makes the statistical distribution of the residuals close to a normal distribution. Therefore, in this embodiment of the invention, the deviation of the residual distribution in the residual sequence corresponding to the initial SARIMA model from the standard normal distribution is first used to measure the degree of difference between the overall statistical form of the residual sequence and the random fluctuation characteristics under the ideal state, so as to obtain the normality of the residual sequence and supplement the overall statistical form of the residual sequence. Then, it is further judged whether the residual sequence corresponding to the initial SARIMA model still maintains random fluctuation characteristics, which is used to finally determine whether the initial SARIMA model has reached the optimal state and complete the secondary test of the initial SARIMA model.

[0049] The method for obtaining the normality approximation of the residual sequence corresponding to the initial SARIMA model is as follows:

[0050] Using Python or MATLAB, obtain the cumulative distribution function of the residual sequence corresponding to the initial SARIMA model, as well as the cumulative distribution function of the standard normal distribution. The acquisition of the cumulative distribution function is an existing technique and will not be elaborated here. Calculate the absolute value function of the difference between the cumulative distribution function of the residual sequence corresponding to the initial SARIMA model and the cumulative distribution function of the standard normal distribution. Form an integration interval using the maximum and minimum values ​​in the residual sequence corresponding to the initial SARIMA model. Integrate the absolute value function of the difference within the integration interval to obtain the integral result. Use the negative of the integral result as the independent variable of the natural exponential function to obtain the normality approximation of the residual sequence corresponding to the initial SARIMA model.

[0051] In one implementation, the formula for calculating the normality of the residual sequence corresponding to the initial SARIMA model is:

[0052]

[0053] Where S represents the normality of the residual sequence corresponding to the initial SARIMA model. This represents the cumulative distribution function of the residual sequence corresponding to the initial SARIMA model. The cumulative distribution function represents the standard normal distribution. This represents the maximum value in the residual sequence corresponding to the initial SARIMA model. This represents the minimum value in the residual sequence corresponding to the initial SARIMA model. Represents the absolute value symbol. This represents the natural exponential function.

[0054] It should be noted that, The smaller the value of S, the closer the residual distribution is to the standard normal distribution. The residual distribution in the residual sequence corresponding to the initial SARIMA model is approximately random, exhibiting mainly random fluctuations. Therefore, the larger S is, the better the initial SARIMA model has explained the data structure in the current training set; conversely, the smaller S is, the better the residual distribution is. The larger S is, the less the residual distribution in the residual sequence corresponding to the initial SARIMA model conforms to the characteristics of random distribution. Consequently, the smaller S is, the more obvious the structural information in the residual sequence corresponding to the initial SARIMA model that was not explained by the initial SARIMA model is.

[0055] Furthermore, a preset threshold for the degree of normality is set. If the threshold is set too high, S needs to be very large to determine that the residual distribution is close to a normal distribution. This means that even though the model itself has fully explained the data, the residuals are considered to still contain structural information, leading to unnecessary parameter adjustments or secondary analyses, causing redundant operations in the model diagnostic process and increasing computational and analytical costs. If the threshold is set too low, structural information in the residuals may be missed. The residuals may still contain potential structural information, but the analysis is not triggered because the threshold is too low, increasing the risk of missed detection. Since the range of S is (0, 1], this solution compromises by using the center of the range as a reference threshold setting, that is, setting the preset threshold for the degree of normality to 0.5. There is no restriction here, and implementers can set it according to specific scenarios.

[0056] If S ≥ 0.5, it indicates that the residuals primarily exhibit random fluctuations, and the initial SARIMA model has adequately explained the data. This means the initial SARIMA model has passed the second test, and it is therefore selected as the optimal SARIMA model for the current. This optimal SARIMA model is then deployed to the target intelligent fusion terminal to predict current monitoring data at future times, thereby achieving accurate prediction of operational faults in the distribution network area. If S < 0.5, it indicates that the residual sequence corresponding to the initial SARIMA model may still contain unexplained local trends or structural fluctuations. In this case, further analysis of the residual sequence is needed to complete the second test of the initial SARIMA model and ultimately determine the optimal SARIMA model for the current.

[0057] Step S102: If the normality proximity is less than the preset normality proximity threshold, the outliers in the residual sequence are obtained using the box plot method. The residual sequence is divided into at least two subsequences according to the preset time sliding window and the preset sliding step size. The secondary test judgment factor of the initial SARIMA model is obtained according to the outlier distribution and trend intensity in each subsequence.

[0058] If the normality S is less than the preset normality threshold of 0.5, it indicates that there may still be local trends or structural fluctuations in the residual sequence corresponding to the initial SARIMA model that are not explained by the model. That is, the initial SARIMA model may not have fully explained the data structure in the current training set. At this time, it is necessary to further analyze whether the deviation of the residual distribution comes from local structural changes that are not explained by the model.

[0059] When there are local structural changes in a time series that are not fully characterized by the model, their performance in the residual series typically exhibits two characteristics: First, because the actual data continuously deviates from the statistical regularity described by the model during this stage, the model prediction error will continue to appear at multiple adjacent time points. Therefore, the residuals often do not appear in isolated forms, but rather exhibit a continuous or dense appearance within a certain period. This local density reflects the continuous prediction bias generated by the model during this stage, which is an important manifestation of local structural changes. Second, if the local structural change itself has a certain evolutionary process, such as a gradual change in operating status, a gradual increase or decrease in load, etc., the degree of deviation of the actual data from the model prediction results will also show a certain directional change. In this case, the residual series often shows a continuous shift or unidirectional change trend within this period. That is, the residuals no longer fluctuate randomly around the zero value, but gradually expand in a certain direction. This trend indicates that the prediction error is not generated by instantaneous random disturbances, but rather originates from a staged change in the data generation process itself.

[0060] Based on the above description, when determining whether the deviation of the residual distribution originates from local structural changes not explained by the initial SARIMA model, it is necessary to simultaneously pay attention to the degree of local concentration of the residuals over time and the trend characteristics of the residual changes. The former can reflect whether the residuals continue to appear within a local time period, while the latter can reflect whether the residuals have systematic evolutionary characteristics. This is to analyze the specific reasons for the deviation of the residual distribution from the normal distribution, obtain the secondary test judgment factor of the initial SARIMA model, and thus identify whether the deviation is caused by local structural changes not explained by the model, thereby realizing the secondary test of the initial SARIMA model.

[0061] The specific implementation method is as follows:

[0062] (1) Since the analysis needs to be conducted on the deviation of the residual distribution, it is first necessary to obtain the residuals of the distribution deviation. In this embodiment of the invention, the outliers in the residual sequence are obtained using the box plot method. These outliers are the residuals of the distribution deviation. The box plot method is existing technology and will not be described in detail here.

[0063] (2) Since local structural changes not explained by the model usually only occur in certain time periods, directly analyzing the entire residual sequence would result in dilution by a large number of random residuals (as mentioned in the background section). Therefore, it is necessary to construct a time sliding window of a preset size and divide the residual sequence corresponding to the initial SARIMA model into at least two subsequences according to the preset sliding step size for local analysis of the residual sequence. The size of the time sliding window needs to be set according to the data acquisition frequency and the operating status of the equipment. In this embodiment of the invention, current is used as an example. Considering that the peak load phase usually lasts for 120 minutes and the low load phase usually lasts for 180 minutes, assuming that the current is sampled once per minute, the size of the time sliding window can be set to 300 minutes to cover the local period from low load to high load. The preset sliding step size can be set to 75 minutes (1 / 4 of the time sliding window). There is no limitation here, and the implementer can set it according to the specific scenario. Considering the need to focus on the local concentration of residuals over time and the trend characteristics of residual changes, if there is only one outlier in a subsequence, or no outlier at all, there is no analytical significance. Therefore, subsequences with more than one outlier are recorded as subsequences to be analyzed, and at least one subsequence to be analyzed is obtained (i.e., if the number of subsequences to be analyzed is 0, it indicates that the setting of the time sliding window size may be unreasonable, and the size of the time sliding window needs to be appropriately increased) in order to analyze the residual deviation in the subsequence to be analyzed.

[0064] (3) Based on the number of outliers in each subsequence to be analyzed and the time interval between two adjacent outliers, obtain the outlier residual density of each subsequence to be analyzed.

[0065] Taking the i-th subsequence to be analyzed as an example, the abnormal residual density of the i-th subsequence to be analyzed is obtained as follows:

[0066] The proportion of outliers in the i-th subsequence to be analyzed is calculated as the proportion of all data in the i-th subsequence to be analyzed. Considering that outliers are low-probability events, in this embodiment of the invention, they are amplified by an exponential function, thereby amplifying the differences between different subsequences to be analyzed. That is, the proportion is used as the independent variable of the natural exponential function to obtain the exponential function value, and the exponential function value is used as the independent variable of the hyperbolic tangent function to obtain the first density.

[0067] Calculate the time interval between each pair of adjacent outliers in the i-th subsequence to be analyzed. Since the shortest time interval is 1, subtract a constant 1 from each time interval to represent the deviation of the time interval between two adjacent outliers from the minimum time interval, and obtain the normal run length. Use the negative of the mean of all normal run lengths as the independent variable of the natural exponential function to obtain the second density.

[0068] Calculate the mean between the first density and the second density to obtain the abnormal residual density of the i-th subsequence to be analyzed.

[0069] In one embodiment, the formula for calculating the abnormal residual density of the i-th subsequence to be analyzed is:

[0070]

[0071] in, This represents the anomalous residual density of the i-th subsequence to be analyzed. This represents the number of all data points in the i-th subsequence to be analyzed. This represents the number of outliers in the i-th subsequence to be analyzed. This represents the time corresponding to the (u+1)th outlier in the i-th subsequence to be analyzed. This represents the time corresponding to the u-th outlier in the i-th subsequence to be analyzed. Natural exponential function, This represents the hyperbolic tangent function.

[0072] It should be noted that, The larger the value, the higher the frequency of prediction error anomalies within a unit time range, and the more concentrated the outliers are in temporal distribution within the i-th subsequence to be analyzed. The larger the value, the lower the explanatory power of the initial SARIMA model for data changes in the time period corresponding to the i-th subsequence to be analyzed, and the more likely the residual deviation is to originate from local structural changes that are not explained by the initial SARIMA model. The smaller the value, the more likely it is that outliers are clustered continuously within the i-th subsequence to be analyzed, meaning that continuous or frequent abnormal prediction errors have occurred within a short timeframe. The larger the value, the more likely the residual deviation originates from local structural changes not explained by the initial SARIMA model, rather than random noise disturbances.

[0073] Similarly, obtain the abnormal residual density of each subsequence to be analyzed.

[0074] (4) The subsequence to be analyzed corresponding to the maximum abnormal residual density is recorded as the target subsequence. Based on the trend significance and trend persistence of the target subsequence, the residual trend strength of the target subsequence is obtained.

[0075] Specifically: Using the least squares method, a linear fit is performed on all outliers in the target subsequence to obtain the fitted value of each outlier in the target subsequence. The least squares method is an existing technology and will not be elaborated here. The absolute value of the difference between each outlier in the target subsequence and its fitted value is calculated respectively. The negative of the sum of all the absolute values ​​of the difference is used as the independent variable of the natural exponential function to obtain the significance of the outlier residual trend of the target subsequence.

[0076] If the number of outliers in the target subsequence is greater than or equal to 3, then for any outlier in the target subsequence other than the first and last outlier, obtain the left and right adjacent outliers of the outlier, subtract the left adjacent outlier from the outlier to obtain the left trend value of the outlier, subtract the right adjacent outlier from the outlier to obtain the right trend value of the outlier, calculate the product between the left trend value and the right trend value, if the product is greater than 0, then set the trend persistence feature value of the outlier to 1, if the product is less than or equal to 0, then set the trend persistence feature value of the outlier to 0.

[0077] The trend persistence feature value of each outlier in the target subsequence, excluding the first and last outliers, is obtained respectively. The mean of all trend persistence feature values ​​is calculated to obtain the trend persistence of the target subsequence.

[0078] If the number of outliers in the target subsequence is less than 3, it indicates that the residual deviation may not be due to local structural changes in the time series that were not characterized by the initial SARIMA model. Therefore, the trend persistence of the target subsequence is set to 0.

[0079] The residual trend strength of the target subsequence is obtained by calculating the mean between the significance of the trend and the persistence of the trend.

[0080] In one embodiment, the formula for calculating the residual trend strength of the target subsequence is:

[0081]

[0082]

[0083] in, This indicates the strength of the residual trend in the target subsequence. This indicates the number of outliers in the target subsequence. This represents the j-th outlier in the target subsequence. This represents the fitted value of the j-th outlier in the target subsequence. This represents the (j-1)th outlier in the target subsequence. Indicates the characteristic value of trend persistence. This represents the natural exponential function. Represents the absolute value symbol.

[0084] It should be noted that, The smaller the value, the closer the abnormal residuals in the target subsequence are to being distributed along the same trend direction, indicating that the residual deviations have strong systematic variation characteristics, and thus... The larger; When, explain , and The data changes among the three outliers show a consistent direction, and vice versa. When, explain , and The three outliers show inconsistent directions of data change. The larger the value, the more consistent the direction of outlier data changes in the target subsequence. This indicates that the trend of the outlier residuals is persistent, the direction of data change has not significantly changed, and there is high consistency. The residual deviation exhibits strong systematic variation characteristics, and thus... The larger.

[0085] (5) Calculate the mean between the abnormal residual density and the residual trend strength of the target subsequence to obtain the secondary test judgment factor of the initial SARIMA model, denoted as ,Right now , This represents the maximum anomalous residual density, which is also the anomalous residual density of the target subsequence. This indicates the strength of the residual trend in the target subsequence. The larger, and The larger the value, the more likely it is that some anomalous residuals in the residual sequence obtained from the initial SARIMA model occur locally in a concentrated manner, and continuously shift in a certain direction within the time period corresponding to the target subsequence, exhibiting high directional consistency. The residual deviation is more likely to originate from local structural changes not explained by the initial SARIMA model, and thus... The larger.

[0086] Thus, the secondary test judgment factor of the initial SARIMA model was obtained.

[0087] Step S103: Perform a secondary test on the initial SARIMA model according to the secondary test judgment factor to obtain the secondary test result. Optimize the initial SARIMA model according to the secondary test result to obtain the optimal SARIMA model for any operating parameter.

[0088] Secondary test judgment factor The larger the value, the more concentrated the anomalous residuals in the residual sequence obtained from the initial SARIMA model are in local time, and the more they continuously shift in a certain direction, exhibiting high directional consistency. This increases the likelihood that the initial SARIMA model has not fully explained the data structure in the current training set; conversely, the smaller the value, the lower the value. The smaller the value, the more discrete the distribution of some anomalous residuals in the residual sequence obtained from the initial SARIMA model is in local time, and the more chaotic the direction of data change is. Some anomalous residuals in the residual sequence obtained from the initial SARIMA model are more likely to be random noise disturbances.

[0089] Based on the above description, in this embodiment of the invention, a preset secondary test threshold is set; if the secondary test judgment factor of the initial SARIMA model... If the value is less than the preset secondary test threshold, it indicates that some abnormal residuals in the residual sequence obtained by the initial SARIMA model are caused by random noise disturbances. At this time, the initial SARIMA model is determined to have passed the secondary test, and the initial SARIMA model is taken as the optimal SARIMA model for the current. At the same time, the optimal SARIMA model is deployed to the target intelligent fusion terminal for predicting the monitoring data of the current at future times.

[0090] If the quadratic test decision factor of the initial SARIMA model If the residuals are greater than or equal to the preset second-order test threshold, it indicates that some abnormal residuals in the residual sequence obtained from the initial SARIMA model originate from local structural changes not explained by the initial SARIMA model. Therefore, the initial SARIMA model is determined to have failed the second-order test. In this case, the hyperparameters of the initial SARIMA model (such as autoregressive order, differencing order, or moving average order) need to be optimized. Simultaneously, the SARIMA model corresponding to the optimized hyperparameters is subjected to the Ljung-Box test to obtain a new SARIMA model for the current that passes the Ljung-Box test. This new SARIMA model is used as the initial SARIMA model, and the current training set is input into the initial SARIMA model (new SARIMA model). Following the method for obtaining the normality approximation in step S101, the normality approximation of the residual sequence corresponding to the initial SARIMA model (new SARIMA model) is obtained. If the normality approximation is greater than or equal to 0.5, the initial SARIMA model (new SARIMA model) is used as the optimal SARIMA model for the current; if the normality approximation is less than 0.5, the optimal SARIMA model is used as the optimal SARIMA model for the current. If the degree is less than 0.5, then according to the method of obtaining the secondary test judgment factor in step S102, the secondary test judgment factor of the initial SARIMA model (new SARIMA model) is obtained. If the secondary test judgment factor of the initial SARIMA model (new SARIMA model) is less than the preset secondary test threshold, then the initial SARIMA model (new SARIMA model) is taken as the optimal SARIMA model for the current. If the secondary test judgment factor of the initial SARIMA model (new SARIMA model) is greater than or equal to the preset secondary test threshold, then the hyperparameters of the initial SARIMA model (new SARIMA model) are optimized. This process continues until the normality of the residual sequence corresponding to the initial SARIMA model (new SARIMA model) is greater than or equal to 0.5, or the secondary test judgment factor of the initial SARIMA model (new SARIMA model) is less than the preset secondary test threshold. The optimal SARIMA model for the current is then obtained and deployed to the target intelligent fusion terminal for predicting the monitoring data of the current at future times.

[0091] Regarding the setting of the preset secondary test threshold, if the threshold is set too high, the model adjustment will only be triggered when the secondary test judgment factor reaches a very high level, thus reducing the sensitivity to structural changes; if the threshold is set too low, the model may be readjusted as soon as a small number of local abnormal residuals appear, random noise is easily misjudged, and model parameter adjustments may be frequently triggered, thereby increasing the computational burden. Therefore, in this embodiment of the invention, the preset secondary test threshold is set to 0.7. This is not a limitation and implementers can set it according to specific scenarios.

[0092] It is worth noting that the focus of this invention is to detect the quality of the SARIMA model, thereby improving the accuracy of using the SARIMA model to predict the monitoring data of various operating parameters at future times, and realizing accurate prediction of operating faults in the distribution network area. The hyperparameter optimization of the SARIMA model is an existing technology and will not be described here.

[0093] Thus, the optimal SARIMA model for the current was obtained.

[0094] Step S104: Obtain the optimal SARIMA model for each operating parameter monitored by the target intelligent fusion terminal, and deploy the optimal SARIMA model for each operating parameter to the target intelligent fusion terminal for predicting the data of each operating parameter, thereby achieving accurate prediction of operating faults in the distribution network area.

[0095] Following the method for obtaining the optimal SARIMA model of current in step S103, the optimal SARIMA models of various operating parameters monitored by the target intelligent fusion terminal are obtained, and the optimal SARIMA models of various operating parameters are deployed to the target intelligent fusion terminal. This avoids the problem of relying solely on overall statistical testing (Ljung-Box test) and ignoring local structural changes in traditional methods. While ensuring the overall stability of the model, the sensitivity to local abnormal changes is improved, enabling the model to identify potential fault risks earlier, improve the accuracy and reliability of monitoring the operating faults of the distribution network area, and achieve accurate prediction of the operating faults of the distribution network area.

[0096] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. An intelligent fusion terminal with accurate fault prediction function, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it performs the following steps: For any operating parameter monitored by the target intelligent fusion terminal, an initial SARIMA model for the operating parameter that has passed the Ljung-Box test is obtained. The training set of the operating parameter is input into the initial SARIMA model to obtain a residual sequence. Based on the deviation between the residual distribution in the residual sequence and the standard normal distribution, the normality of the residual sequence is obtained. If the normality proximity is less than the preset normality proximity threshold, then the outliers in the residual sequence are obtained using the box plot method. Based on the preset time sliding window and the preset sliding step size, the residual sequence is divided into at least two subsequences. Based on the outlier distribution and trend strength in each subsequence, the secondary test judgment factor of the initial SARIMA model is obtained. Based on the secondary test judgment factor, the initial SARIMA model is subjected to a secondary test to obtain the secondary test result. Based on the secondary test result, the initial SARIMA model is optimized to obtain the optimal SARIMA model for any operating parameter. The optimal SARIMA model for each operating parameter monitored by the target intelligent fusion terminal is obtained, and the optimal SARIMA model for each operating parameter is deployed to the target intelligent fusion terminal for prediction of the data of each operating parameter, so as to achieve accurate prediction of the operating faults of the distribution network area. The step of obtaining the secondary test judgment factor of the initial SARIMA model based on the outlier distribution and trend strength in each subsequence includes: The number of outliers in each subsequence is obtained. Subsequences with more than 1 outliers are recorded as subsequences to be analyzed, resulting in at least one subsequence to be analyzed. For any subsequence to be analyzed, the outlier residual density is obtained based on the number of outliers in the subsequence and the time interval between two adjacent outliers. Obtain the abnormal residual density of each of the subsequences to be analyzed, and denote the subsequence to be analyzed corresponding to the maximum abnormal residual density as the target subsequence. Based on the trend significance and trend persistence of the target subsequence, obtain the residual trend strength of the target subsequence. The mean between the abnormal residual density and the residual trend strength of the target subsequence is calculated to obtain the secondary test judgment factor of the initial SARIMA model.

2. The intelligent fusion terminal with accurate prediction function of operational faults according to claim 1, characterized in that, The step of obtaining the normality approximation of the residual sequence based on the deviation between the residual distribution in the residual sequence and the standard normal distribution includes: Obtain the cumulative distribution function of the residual sequence and the cumulative distribution function of the standard normal distribution. Calculate the absolute value function of the difference between the cumulative distribution function of the residual sequence and the cumulative distribution function of the standard normal distribution. Form an integration interval using the maximum and minimum values ​​in the residual sequence. Integrate the absolute value function of the difference within the integration interval to obtain the integration result. Use the negative of the integration result as the independent variable of the natural exponential function to obtain the normality approximation of the residual sequence.

3. The intelligent fusion terminal with accurate prediction function of operational faults according to claim 1, characterized in that, The step of obtaining the abnormal residual density of any subsequence to be analyzed based on the number of outliers in any subsequence to be analyzed and the time interval between two adjacent outliers includes: Calculate the proportion of the number of outliers in any subsequence to be analyzed to the total number of data in any subsequence to be analyzed. Use the proportion as the independent variable of the natural exponential function to obtain the exponential function value. Use the exponential function value as the independent variable of the hyperbolic tangent function to obtain the first density. Calculate the time interval between any two adjacent outliers in any subsequence to be analyzed, subtract a constant 1 from each time interval to obtain the normal run length, and use the negative of the mean of all normal run lengths as the independent variable of the natural exponential function to obtain the second density. The mean between the first density and the second density is calculated to obtain the abnormal residual density of any subsequence to be analyzed.

4. The intelligent fusion terminal with accurate prediction function of operational faults according to claim 1, characterized in that, The step of obtaining the residual trend strength of the target subsequence based on the trend significance and trend persistence of the target subsequence includes: Linear fitting is performed on all outliers in the target subsequence to obtain the fitted value of each outlier in the target subsequence. The absolute value of the difference between each outlier in the target subsequence and its fitted value is calculated. The negative of the sum of all the absolute values ​​of the difference is used as the independent variable of the natural exponential function to obtain the significance of the outlier residual trend of the target subsequence. If the number of outliers in the target subsequence is greater than or equal to 3, then for any outlier in the target subsequence other than the first and last outlier, obtain the left and right adjacent outliers of the outlier, subtract the left adjacent outlier from the outlier to obtain the left trend value of the outlier, subtract the right adjacent outlier from the outlier to obtain the right trend value of the outlier, calculate the product between the left trend value and the right trend value, if the product is greater than 0, then set the trend persistence feature value of the outlier to 1, if the product is less than or equal to 0, then set the trend persistence feature value of the outlier to 0. The trend persistence feature value of each outlier in the target subsequence, excluding the first and last outliers, is obtained respectively. The mean of all trend persistence feature values ​​is calculated to obtain the trend persistence of the target subsequence. If the number of outliers in the target subsequence is less than 3, then the trend persistence of the target subsequence is set to 0. The residual trend strength of the target subsequence is obtained by calculating the mean between the significance of the trend and the persistence of the trend.

5. The intelligent fusion terminal with accurate prediction function of operational faults according to claim 1, characterized in that, Based on the aforementioned secondary test judgment factor, a secondary test is performed on the initial SARIMA model to obtain the secondary test results, including: If the secondary test judgment factor is less than the preset secondary test threshold, then the initial SARIMA model is determined to have passed the secondary test. If the secondary test judgment factor is greater than or equal to the preset secondary test threshold, then the initial SARIMA model is determined to fail the secondary test.

6. The intelligent fusion terminal with accurate prediction function of operational faults according to claim 5, characterized in that, The step of optimizing the initial SARIMA model based on the secondary test results to obtain the optimal SARIMA model for any given operating parameter includes: If the initial SARIMA model passes the quadratic test, then the initial SARIMA model is taken as the optimal SARIMA model for any of the operating parameters; If the initial SARIMA model fails the quadratic test, the hyperparameters of the initial SARIMA model are optimized to obtain a new SARIMA model for any operating parameter that passes the Ljung-Box test. This new SARIMA model is then used as the initial SARIMA model. This process continues until the normality of the residual sequence corresponding to the initial SARIMA model is greater than or equal to a preset normality threshold, or the quadratic test judgment factor of the initial SARIMA model is less than a preset quadratic test threshold. Finally, the optimal SARIMA model for any operating parameter is obtained.

7. The intelligent fusion terminal with accurate prediction function of operational faults according to claim 1, characterized in that, If the normality proximity is greater than or equal to the pre-normality proximity threshold, then the initial SARIMA model is determined to have passed the quadratic test, and the initial SARIMA model is taken as the optimal SARIMA model for any of the operating parameters.

Images