Uninterrupted power supply (UPS)-based mains supply fault prediction system

Abstract

The invention relates to the technical field of fault prediction, and particularly discloses a UPS-based commercial power fault prediction system, which comprises a fault acquisition module, a monitoring analysis module, a model construction module and a model determination module, the fault acquisition module is used for quantifying the accurate state of fault judgment, obtaining a monitoring misalignment value and judging the monitoring accuracy of the UPS equipment; the monitoring analysis module is used for acquiring a load peak working period and a UPS misalignment judgment period when monitoring that the mains supply fault is inaccurate, and calculating the period similarity; the model construction module is used for extracting commercial power parameters to construct a training set when the two periods are similar, and constructing a commercial power fault prediction model by utilizing linear regression, a decision tree, a neural network and a long-short-term memory network model algorithm; the model determining module is used for analyzing monitoring misalignment values of different models to obtain model adaptive values and determining an optimization algorithm; according to the method, the accuracy of mains supply fault prediction can be improved, and guarantee is provided for stable operation of UPS equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of fault prediction, and in particular to a mains fault prediction system based on UPS. Background Art

[0002] UPS (Uninterruptible Power Supply) is a power supply device that can provide stable power supply to load equipment when the mains power is interrupted or abnormal. However, the occurrence of mains power failure brings many challenges to the stable operation of UPS.

[0003] A Chinese patent application with announcement number CN115859058B discloses a UPS fault prediction method and system based on a width learning network, including: preprocessing historical data under different states of UPS power battery; determining the value range of characteristic nodes, enhanced nodes, and characteristic window number parameters in the width learning network; introducing a simulated annealing algorithm to improve the mutation operation in the genetic algorithm to obtain an improved genetic algorithm; optimizing the width learning parameters through the improved genetic algorithm; determining the threshold range of normal UPS power battery based on the grid search algorithm and the Ksigma model; and comprehensively using GA+BLS+K-sigma to obtain the final prediction and diagnosis model to monitor the status of the UPS power battery in real time.

[0004] The above technology adopts a width learning network, and determines the threshold range through an improved genetic algorithm optimization and a grid search algorithm. The above technology constructs a general UPS power battery status monitoring model, which does not fully consider the relationship between the AC power failure and the UPS inaccuracy judgment cycle and the load peak working cycle. If it is judged whether the UPS inaccuracy judgment cycle is similar to the load peak working cycle, if they are similar, using different model algorithms to construct a model is conducive to predicting the load peak caused by the AC power failure.

[0005] The above technology is currently used for real-time monitoring of the battery status of UPS power supplies, and the battery status is determined by the diagnostic model. There is no prediction of AC power failure, and no monitoring and evaluation system has been established. If the monitoring inaccuracy value is obtained by calculating the fault false alarm rate and the fault omission rate, it will be helpful to evaluate the monitoring accuracy of UPS equipment for AC power failures. In terms of model evaluation, the current technology lacks a quantitative evaluation of monitoring accuracy and a comprehensive evaluation system for model performance. If the model fitness value is obtained by calculating the effective cycle ratio, inaccuracy improvement ratio, etc., and the optimal algorithm is determined, it will be helpful to timely discover problems in the monitoring process and optimize the model.

[0006] To this end, the present invention provides a UPS-based mains fault prediction system. Summary of the invention

[0007] The purpose of the present invention is to provide a UPS-based mains fault prediction system to solve the above-mentioned problems.

[0008] The purpose of the present invention can be achieved through the following technical solutions:

[0009] The UPS-based mains fault prediction system includes the following modules:

[0010] Fault collection module: used to obtain the accurate status of the mains fault judgment of the UPS device, and perform quantitative analysis to obtain the monitoring inaccuracy value to determine whether the UPS device monitors the mains fault accurately;

[0011] Monitoring and analysis module: If the UPS device does not accurately monitor the mains failure, it is used to obtain the peak working cycle of the load device, build a peak cycle group, obtain the inaccuracy judgment cycle of the UPS device, build an inaccuracy cycle group, and determine whether the inaccuracy judgment cycle of the UPS device is similar to the peak working cycle of the load;

[0012] Model building module: If similar, obtain the mains parameters that the UPS device mistakenly marks as a mains failure, and use different model algorithms to build a mains failure prediction model;

[0013] Model determination module: constructs a mains fault prediction model based on different model algorithms, performs numerical analysis on the monitoring misalignment values ​​output by different model algorithms, obtains the model adaptation value, and determines the optimal algorithm for the mains fault prediction model.

[0014] As a further technical solution of the present invention: the monitoring misalignment value is obtained in the following manner:

[0015] Obtain the mains fault status monitored by the UPS device from the UPS device log;

[0016] During the monitoring period, the number of times the UPS device marks the mains power supply as a fault state is obtained to obtain the number of fault markings;

[0017] The number of UPS fault markings including the number of erroneously marked fault states is obtained, and the number of erroneously marked fault states is processed by ratio with the number of fault markings to obtain the fault false alarm rate.

[0018] Obtain the number of times when the mains power fails and the UPS fails to identify the mains power failure, and obtain the number of missed failures;

[0019] The fault omission number is processed by ratioing the actual fault number obtained in advance to obtain the fault omission rate;

[0020] Based on the fault false alarm rate and fault missed alarm rate, a weighted summation process is performed to obtain the monitoring inaccuracy value;

[0021] If the monitoring inaccuracy value of the UPS device is lower than the preset monitoring accuracy threshold, it indicates that the UPS device is not accurately monitoring the mains failure.

[0022] As a further technical solution of the present invention: the judgment method of whether the misalignment judgment cycle of the UPS device is similar to the peak working cycle of the load is:

[0023] Obtaining a monitoring period in which the UPS device monitoring misalignment value is lower than a preset monitoring accuracy threshold, and obtaining a misalignment determination period;

[0024] Obtain the misalignment determination cycles in all monitoring cycles and construct a misalignment cycle group;

[0025] Obtain the peak working cycle of the load equipment within the entire monitoring period and construct a peak cycle group;

[0026] Based on the misaligned period group and the peak period group, the period similarity is obtained by using the period matching window group and entropy calculation method;

[0027] If the cycle similarity is higher than the preset similarity threshold, the misaligned cycle group and the peak cycle group are considered to be similar in the time dimension.

[0028] As a further technical solution of the present invention: the method for obtaining the period similarity is:

[0029] By using the cycle matching window group cutting method, the time cutting window of the monitoring period is determined, the monitoring period is divided into multiple cycle matching windows, and a cycle matching window group is constructed;

[0030] Based on the misalignment period group, obtain each period matching window group, including the number of misalignment determination periods;

[0031] Based on the peak cycle group, obtain each cycle matching window group, including the number of peak working cycles;

[0032] Through the frequency formula, obtain each cycle matching window group, including the frequency of the peak working cycle and the frequency of the misalignment judgment cycle;

[0033] The frequency of the peak working cycle and the frequency of the misalignment judgment cycle are numerically calculated to obtain the cycle similarity.

[0034] As a further technical solution of the present invention: the frequency of the peak working cycle and the frequency of the misalignment determination cycle are numerically calculated in the following manner:

[0035] Based on the frequency of the peak working cycle and the frequency of the misalignment judgment cycle, the cycle entropy of the misalignment cycle group and the cycle entropy of the peak cycle group are obtained through the entropy value formula;

[0036] The cycle similarity is obtained through the ratio formula based on the cycle entropy of the misaligned cycle group and the cycle entropy of the peak cycle group.

[0037] As a further technical solution of the present invention: the construction method of the mains fault prediction model is:

[0038] Obtain the mains parameters that the UPS device mistakenly marks as a mains failure, and build a training set for the mains failure prediction model;

[0039] Based on the training set of the mains fault prediction model, a mains fault prediction model is constructed using linear regression model algorithm, decision tree model algorithm, neural network model algorithm, and long short-term memory network model algorithm.

[0040] As a further technical solution of the present invention: the method for obtaining the mains parameters of the mains fault error mark is:

[0041] If the misalignment determination cycle of the UPS device is similar to the peak working cycle of the load, the mains parameters are extracted from the UPS working log within the misalignment determination cycle of the UPS device;

[0042] Among them, the mains power parameters include: voltage, frequency, and phase;

[0043] Based on the mains parameters, the voltage fluctuation range, mains frequency deviation and voltage harmonic content are extracted within the misalignment judgment period.

[0044] As a further technical solution of the present invention: the preferred algorithm of the mains fault prediction model is determined as follows:

[0045] Obtain different model algorithms to build a mains fault prediction model and output the model fitness value;

[0046] The model fitness values ​​of different model algorithms are sorted in order from large to small, and the model algorithm with the largest model fitness value is determined as the preferred algorithm of the mains fault prediction model.

[0047] As a further technical solution of the present invention: the method for obtaining the model fitness value is:

[0048] A mains fault prediction model is constructed based on different model algorithms to obtain the monitoring mismatch value of a UPS device equipped with the mains fault prediction model during a peak working cycle, and the monitoring mismatch value of a UPS device not equipped with the mains fault prediction model during a peak working cycle;

[0049] Performing difference processing on the monitoring misalignment value of the UPS device equipped with the mains fault prediction model during the peak working period and the monitoring misalignment value of the UPS device not equipped with the mains fault prediction model during the peak working period to obtain the monitoring misalignment difference;

[0050] Obtain the monitoring misalignment of the UPS device during the peak working period and determine the positive or negative nature of the monitoring misalignment;

[0051] If the monitoring misalignment is negative, the peak duty cycle is marked as a valid optimization cycle;

[0052] Quantitatively analyze the effective optimization cycle to obtain the effective cycle ratio;

[0053] The ratio of the monitoring misalignment difference within the effective optimization period is calculated to obtain the misalignment improvement ratio;

[0054] The effective period ratio and the misalignment improvement ratio are weightedly summed to obtain the model fitness value.

[0055] As a further technical solution of the present invention: the effective period ratio is obtained as follows:

[0056] Obtaining the number of effective optimization cycles and the number of peak working cycles in multiple peak working cycles, and performing ratio processing on the number of effective optimization cycles and the number of peak working cycles to obtain an effective cycle ratio;

[0057] The misalignment improvement ratio is obtained as follows:

[0058] The monitoring misalignment error within the effective optimization period is obtained, the absolute value of the monitoring misalignment error is taken, and the ratio is processed with the monitoring misalignment value of the UPS device without the mains fault prediction model during the peak working period to obtain the misalignment improvement ratio.

[0059] Beneficial effects of the present invention:

[0060] (1) The fault collection module conducts quantitative analysis on the mains fault judgment of the UPS equipment, obtains the monitoring inaccuracy value, and determines whether the UPS's monitoring of the mains fault is accurate. If the monitoring is inaccurate, the monitoring and analysis module can construct a load peak cycle group and a UPS inaccuracy cycle group, and calculate the cycle similarity. It can not only identify the high-incidence cycle of faults, but also focus on monitoring and predicting these periods, thereby improving the pertinence of the mains fault prediction, allowing relevant personnel to take more targeted preventive measures, which is conducive to reducing equipment abnormalities caused by mains faults;

[0061] (2) When the UPS misalignment judgment cycle is similar to the load peak working cycle, the model construction module extracts the AC power parameters to construct a training set and uses a variety of model algorithms to build a AC power fault prediction model. The model determination module analyzes the monitoring misalignment values ​​of different models in the peak working cycle, calculates the model fitness value, and selects the optimal algorithm. By using the most suitable model to predict AC power faults, the reliability of UPS in predicting AC power faults is improved, ensuring that UPS equipment can operate stably when AC power fails, which is conducive to reducing the risks caused by prediction errors. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The present invention will be further described below in conjunction with the accompanying drawings.

[0063] Figure 1 It is a module diagram of the mains fault prediction system based on UPS of the present invention;

[0064] Figure 2 It is a flow chart of a method for obtaining the model adaptability value in the UPS-based mains fault prediction system of the present invention. DETAILED DESCRIPTION

[0065] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0066] Example 1

[0067] See also Figure 1 As shown, the present invention is a UPS-based mains fault prediction system, comprising the following modules:

[0068] Fault collection module: obtains the number of mains fault judgments of the UPS device, performs quantitative analysis on the accuracy of fault judgment, obtains the monitoring inaccuracy value, and determines whether the UPS device accurately monitors the mains fault;

[0069] In some embodiments, obtaining a mains power failure status monitored by a UPS device from a device log of the UPS;

[0070] It should be noted that the UPS device can detect whether the mains power is normal in real time. When the mains power fails, such as power outage, abnormal voltage, abnormal frequency, etc., the UPS will detect it in time and record the mains power failure in the device log;

[0071] During the monitoring period, the number of times the UPS device marks the mains power supply as a fault state is obtained to obtain the number of fault markings;

[0072] The number of UPS fault markings including the number of erroneously marked fault states is obtained, and the number of erroneously marked fault states is processed by ratio with the number of fault markings to obtain the fault false alarm rate.

[0073] Obtain the number of times when the mains power fails and the UPS fails to identify the mains power failure, and obtain the number of missed failures;

[0074] The fault omission number is processed by ratioing the actual fault number obtained in advance to obtain the fault omission rate;

[0075] It should be noted that the actual number of failures of the UPS device is obtained by summing the number of failures correctly marked by the UPS device and the number of failures missed, while the number of failures correctly marked by the UPS device is obtained by difference processing between the number of failure markings and the number of faults incorrectly marked;

[0076] Based on the fault false alarm rate and fault missed alarm rate, a weighted summation process is performed to obtain the monitoring inaccuracy value;

[0077] It should be noted that among the UPS fault marking times, the number of times the fault status is mistakenly marked and the number of times the fault is missed are obtained by analyzing the work log of the UPS device by professional technicians in this field;

[0078] Compare the monitoring inaccuracy value of the UPS device with the preset monitoring accuracy threshold to determine whether the UPS device accurately monitors the mains failure;

[0079] If the monitoring inaccuracy value of the UPS device is lower than the preset monitoring accuracy threshold, it indicates that the UPS device is not accurately monitoring the mains failure;

[0080] If the monitoring inaccuracy value of the UPS device is higher than the preset monitoring accuracy threshold, it indicates that the UPS device is more accurate in monitoring the mains failure;

[0081] Monitoring and analysis module: If the UPS device does not monitor the mains failure accurately enough, it is used to obtain the peak working cycle of the load device, build a peak cycle group, obtain the inaccuracy judgment cycle of the UPS device, build an inaccuracy cycle group, and determine whether the inaccuracy judgment cycle of the UPS device is similar to the peak working cycle of the load;

[0082] If the UPS device does not monitor the mains fault accurately enough, obtain the monitoring period in which the UPS device monitoring inaccuracy value is lower than the preset monitoring accuracy threshold, and obtain the inaccuracy determination period;

[0083] Obtain the misalignment determination cycles in all monitoring cycles and construct a misalignment cycle group;

[0084] In some embodiments, a peak duty cycle of a load device is obtained from monitoring data of a power grid monitoring system;

[0085] Obtain the peak working cycle of the load equipment within the entire monitoring period and construct a peak cycle group;

[0086] It should be noted that by determining the peak working cycle of the load equipment, the maximum demand period for the mains electricity of the load equipment can be grasped during the entire monitoring period. During the peak working cycle, the mains system is under greater pressure, and voltage fluctuations, frequency anomalies and other problems are prone to occur, thereby increasing the probability of mains failure. By constructing a peak cycle group, it is helpful to identify the high-incidence cycle of failures and improve the pertinence and accuracy of mains failure prediction;

[0087] Based on the misaligned period group and the peak period group, the period similarity is obtained by using the period matching window group and entropy calculation method;

[0088] The time shear window Δt of the monitoring period is determined by the period matching window group shearing method, the monitoring period is divided into multiple period matching windows, and the period matching window group Z is constructed. i ={z 1 、z 2 ,...,z n}, i is the number of the period matching window group, and the value range of i is [1, n];

[0089] Based on the misaligned period group, obtain each period matching window group Z i The number of misalignment determination cycles m 1i ;

[0090] By formula: Get each period matching window group, including the frequency f of the misalignment determination period 1i , where n is the total number of period matching window groups;

[0091] Based on the peak cycle group, get each cycle matching window group Z i , including the number of peak duty cycles m 2i ;

[0092] By formula: Get each period matching window group Z i , including the frequency f of the peak duty cycle 2i ;

[0093] By formula: Get the period entropy H of the misaligned period group 1 ;

[0094] By formula: Get the period entropy H of the peak period group 2 ;

[0095] By formula: Get the cycle similarity Xs;

[0096] It should be noted that when the cycle similarity Xs=1, |H 1-H 2 |=0, that is, the period entropy H of the misaligned period group 1 , the period entropy H of the peak period group 2 The same means that the misaligned period group and the peak period group have the highest period similarity on the period matching window group;

[0097] The cycle similarity Xs is analyzed with the preset similarity threshold to determine whether the misaligned cycle group and the peak cycle group are similar in the time dimension;

[0098] If the cycle similarity Xs is higher than the preset similarity threshold, the misaligned cycle group and the peak cycle group are considered similar in the time dimension, otherwise, they are not similar;

[0099] The technical solution of this embodiment is: obtain the number of AC power failure judgments of the UPS device, perform quantitative analysis on the accuracy of the fault judgment state, obtain the monitoring inaccuracy value, determine whether the UPS device accurately monitors the AC power failure, if the UPS device is not accurate enough in monitoring the AC power failure, obtain the peak working cycle of the load, construct a peak cycle group, obtain the inaccurate judgment cycle of the UPS device, construct an inaccurate cycle group, and determine whether the inaccurate judgment cycle of the UPS device is similar to the peak working cycle of the load. By calculating the cycle similarity, not only can the high-incidence cycles of faults be identified, but also these time periods can be monitored and predicted in a focused manner, thereby improving the pertinence of AC power failure predictions and allowing relevant personnel to take more targeted preventive measures, which is conducive to reducing equipment abnormalities caused by AC power failures.

[0100] Example 2

[0101] like Figure 1 As shown, the UPS-based mains fault prediction system also includes the following modules:

[0102] Model building module: If the misalignment judgment cycle of the UPS device is similar to the peak working cycle of the load, the mains parameters that the UPS device mistakenly marks as a mains failure are obtained, and a mains failure prediction model is built using different model algorithms;

[0103] If the misalignment determination cycle of the UPS device is similar to the peak working cycle of the load, the mains parameters are extracted from the UPS working log within the misalignment determination cycle of the UPS device;

[0104] Among them, the mains power parameters include: voltage, frequency, and phase;

[0105] Based on the mains parameters, extract the voltage fluctuation range, mains frequency deviation, and voltage harmonic content within the misalignment judgment period;

[0106] It should be noted that the voltage fluctuation range, mains frequency deviation, and voltage harmonic content are extracted from the work log of the UPS device. The UPS device has the ability to monitor and record mains parameter data;

[0107] The voltage fluctuation range refers to the difference between the maximum and minimum values ​​of the mains voltage within the misalignment determination cycle. For example, if the mains voltage reaches a maximum of 230V and a minimum of 210V within a certain misalignment determination cycle, the voltage fluctuation range is 230-210=20V.

[0108] If the mains voltage waveform is not an ideal sine wave, harmonics will be generated. The voltage harmonic content refers to the percentage of the root mean square value of each harmonic voltage to the root mean square value of the fundamental voltage.

[0109] For example, if the effective value of the fundamental voltage is 220V and the effective value of a harmonic voltage is 22V, then the voltage harmonic content of the harmonic is (22÷220)×100%=10%;

[0110] Obtain the voltage fluctuation range, mains frequency deviation, and voltage harmonic content at the time when the UPS device mistakenly marks the mains failure;

[0111] The voltage fluctuation range, mains frequency deviation, and voltage harmonic content within the mains fault mis-marking state and misjudgment period are used as the training set of the mains fault prediction model;

[0112] Based on the training set of the mains fault prediction model, a mains fault prediction model is constructed using linear regression model algorithm, decision tree model algorithm, neural network model algorithm, and long short-term memory network model algorithm;

[0113] It should be noted that the linear regression model algorithm uses the voltage fluctuation range, mains frequency deviation, and voltage harmonic content as independent variables and whether a mains fault occurs as the dependent variable to calculate the linear relationship between the independent variables and the dependent variables. The linear regression model calculates the linear combination between the mains parameters and the mains fault to predict the possibility of the mains fault.

[0114] The decision tree model algorithm recursively divides the feature space and builds a tree structure to make decisions. The decision tree will gradually divide the data into different nodes according to the different values ​​of the mains parameters, and finally determine whether a mains failure will occur;

[0115] The neural network model algorithm is composed of a stack of layers composed of multiple neurons. Through the connections and weights between neurons, the neural network learns the complex patterns and nonlinear relationships in the data. The neural network can take the mains parameters as input, and after nonlinear transformations of multiple hidden layers, it finally outputs the probability of mains failure.

[0116] The long short-term memory network model algorithm processes the time series data of the mains parameters and combines the correlation between the mains parameters at different time points to predict the occurrence of mains failures.

[0117] Specifically, the construction method of the mains fault prediction model is as follows:

[0118] S1. Mains power data collection and preprocessing;

[0119] The data set of the mains fault prediction model is preprocessed, the missing values ​​in the data set are filled using the mean filling method, and the outliers in the data set are removed using the 3 times standard deviation principle;

[0120] S2, label and divide the data set;

[0121] In the data set of the mains fault prediction model, the voltage fluctuation range, mains frequency deviation, and voltage harmonic content of the mains fault mislabeled state are used as labels;

[0122] For example, the label value of the voltage fluctuation range, mains frequency deviation, and voltage harmonic content in the mains fault mislabeling state is 0, otherwise the label value is 1;

[0123] The data set of the mains fault prediction model is divided into a training set and a test set in a ratio of 8:2;

[0124] S3, use different model algorithms for training;

[0125] The divided data set of the mains fault prediction model is input into the linear regression model algorithm, the decision tree model algorithm, the neural network model algorithm, and the long short-term memory network model algorithm for training;

[0126] Model determination module: constructs a mains fault prediction model based on different model algorithms, performs numerical analysis on the monitoring misalignment values ​​output by different model algorithms, obtains the model adaptation value, and determines the optimal algorithm for the mains fault prediction model;

[0127] A mains fault prediction model is constructed based on different model algorithms to obtain the monitoring mismatch value of a UPS device equipped with the mains fault prediction model during a peak working cycle, and the monitoring mismatch value of a UPS device not equipped with the mains fault prediction model during a peak working cycle;

[0128] like Figure 2 As shown, the monitoring misalignment value is numerically analyzed to obtain the model adaptation value;

[0129] Specifically, the model fitness value is obtained as follows:

[0130] Obtaining a monitoring mismatch value of a UPS device equipped with a mains fault prediction model during a peak working cycle and performing difference processing on a monitoring mismatch value of a UPS device not equipped with a mains fault prediction model during a peak working cycle to obtain a monitoring mismatch difference;

[0131] It should be noted that the larger the absolute value of the monitoring misalignment, the more significant the impact of the model on the monitoring accuracy. A negative misalignment with a large absolute value indicates that the model is effective in reducing monitoring misalignment. A positive misalignment with a large absolute value indicates that there are serious problems with the model, which may lead to more inaccurate monitoring results.

[0132] Obtain the monitoring misalignment of the UPS device during the peak working period and determine the positive or negative nature of the monitoring misalignment;

[0133] If the monitoring misalignment is negative, the peak working period is marked as a valid optimization period, otherwise, it is marked as an invalid optimization period;

[0134] Obtaining the number of effective optimization cycles and the number of peak working cycles in multiple peak working cycles, and performing ratio processing on the number of effective optimization cycles and the number of peak working cycles to obtain an effective cycle ratio;

[0135] Obtain the monitoring misalignment error within the effective optimization period, take the absolute value of the monitoring misalignment error, and compare it with the monitoring misalignment value of the UPS device without the mains fault prediction model during the peak working period to obtain the misalignment improvement ratio;

[0136] The effective cycle ratio and the misalignment improvement ratio are weighted and summed to obtain the model fitness value;

[0137] Obtain different model algorithms to construct a mains fault prediction model, output model fitness values, sort the model fitness values ​​of different model algorithms in descending order, and determine the model algorithm with the largest model fitness value as the preferred algorithm for the mains fault prediction model;

[0138] It should be noted that the effective cycle ratio reflects the stability and reliability of the model in practical applications. The higher the effective cycle ratio, the higher the frequency of the model playing an active role in the peak working cycle, and the greater its continuous contribution to improving monitoring accuracy;

[0139] The misalignment improvement ratio measures the degree to which different model algorithms can improve the monitoring misalignment value within the effective optimization cycle, reflecting the model's ability to improve monitoring accuracy. The larger the misalignment improvement ratio, the more obvious the effect of the model on reducing the monitoring misalignment value within the effective optimization cycle.

[0140] The model fitness value reflects the comprehensive performance of the model in terms of monitoring accuracy. If the model fitness value is higher, that is, the model algorithm has a higher value in UPS power failure prediction, it is more likely to become the preferred algorithm; on the contrary, the smaller the model fitness value, the worse the comprehensive performance of the model, and it is necessary to optimize the model algorithm or reselect other models.

[0141] The technical solution of this embodiment is: if the misalignment judgment cycle of the UPS device is similar to the peak working cycle of the load, obtain the AC parameters that the UPS device mistakenly marks as AC fault, use different model algorithms to build a AC fault prediction model, build a AC fault prediction model based on different model algorithms, perform numerical analysis on the monitoring misalignment values ​​output by different model algorithms, obtain the model fitness value, determine the optimal algorithm of the AC fault prediction model, calculate the model fitness value, select the optimal algorithm, and use the most suitable model to predict AC faults, thereby improving the reliability of UPS in predicting AC faults, ensuring that UPS equipment can operate stably when AC faults occur, and helping to reduce the risks caused by prediction errors.

[0142] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the present invention.

Claims

1. The mains failure prediction system based on UPS is characterized by: Includes the following modules: Fault collection module: used to obtain the accurate status of the mains fault judgment of the UPS device, and perform quantitative analysis to obtain the monitoring inaccuracy value to determine whether the UPS device monitors the mains fault accurately; Monitoring and analysis module: If the UPS device does not accurately monitor the mains failure, it is used to obtain the peak working cycle of the load device, build a peak cycle group, obtain the inaccuracy judgment cycle of the UPS device, build an inaccuracy cycle group, and determine whether the inaccuracy judgment cycle of the UPS device is similar to the peak working cycle of the load; Model building module: If similar, it is used to obtain the mains parameters that the UPS device mistakenly marks as a mains failure, and use different model algorithms to build a mains failure prediction model; Model determination module: Based on the mains fault prediction model constructed by different model algorithms, the monitoring misalignment values ​​output by different model algorithms are numerically analyzed to obtain the model adaptation value and determine the optimal algorithm for the mains fault prediction model.

2. The UPS-based mains fault prediction system according to claim 1, characterized in that: The monitoring misalignment value is obtained in the following manner: Obtain the mains fault status monitored by the UPS device from the UPS device log; During the monitoring period, the number of times the UPS device marks the mains power supply as a fault state is obtained to obtain the number of fault markings; The number of UPS fault markings including the number of erroneously marked fault states is obtained, and the number of erroneously marked fault states is processed by ratio with the number of fault markings to obtain the fault false alarm rate. Obtain the number of times when the mains power fails and the UPS fails to identify the mains power failure, and obtain the number of missed failures; The fault omission number is processed by ratioing the actual fault number obtained in advance to obtain the fault omission rate; Based on the fault false alarm rate and fault missed alarm rate, a weighted summation process is performed to obtain the monitoring inaccuracy value; If the monitoring inaccuracy value of the UPS device is lower than the preset monitoring accuracy threshold, it indicates that the UPS device is not accurately monitoring the mains failure.

3. The UPS-based mains fault prediction system according to claim 1, characterized in that: The judgment method of whether the misalignment judgment cycle of the UPS device is similar to the peak working cycle of the load is as follows: Obtaining a monitoring period in which the UPS device monitoring misalignment value is lower than a preset monitoring accuracy threshold, and obtaining a misalignment determination period; Obtain the misalignment determination cycles in all monitoring cycles and construct a misalignment cycle group; Obtain the peak working cycle of the load equipment within the entire monitoring period and construct a peak cycle group; Based on the misaligned period group and the peak period group, the period similarity is obtained by using the period matching window group and the period entropy calculation method; If the cycle similarity is higher than the preset similarity threshold, the misaligned cycle group and the peak cycle group are considered to be similar in the time dimension.

4. The UPS-based mains fault prediction system according to claim 3, characterized in that: The cycle similarity is obtained in the following manner: By using the cycle matching window group cutting method, the time cutting window of the monitoring period is determined, the monitoring period is divided into multiple cycle matching windows, and a cycle matching window group is constructed; Based on the misalignment period group, obtain each period matching window group, including the number of misalignment determination periods; Based on the peak cycle group, obtain each cycle matching window group, including the number of peak working cycles; Through the frequency formula, obtain each cycle matching window group, including the frequency of the peak working cycle and the frequency of the misalignment judgment cycle; The frequency of the peak working cycle and the frequency of the misalignment judgment cycle are numerically calculated to obtain the cycle similarity.

5. The UPS-based mains fault prediction system according to claim 4, characterized in that: The frequency of the peak working cycle and the frequency of the misalignment determination cycle are numerically calculated as follows: Based on the frequency of the peak working cycle and the frequency of the misalignment judgment cycle, the cycle entropy of the misalignment cycle group and the cycle entropy of the peak cycle group are obtained through the entropy value formula; The cycle similarity is obtained through the ratio formula based on the cycle entropy of the misaligned cycle group and the cycle entropy of the peak cycle group.

6. The UPS-based mains fault prediction system according to claim 1, characterized in that: The construction method of the mains fault prediction model is as follows: Obtain the mains parameters that the UPS device mistakenly marks as a mains failure, and build a training set for the mains failure prediction model; Based on the training set of the mains fault prediction model, a mains fault prediction model is constructed using linear regression model algorithm, decision tree model algorithm, neural network model algorithm, and long short-term memory network model algorithm.

7. The UPS-based mains fault prediction system according to claim 6, characterized in that: The method for obtaining the mains parameters of the mains fault error mark is: If the misalignment determination cycle of the UPS device is similar to the peak working cycle of the load, the mains parameters are extracted from the UPS working log within the misalignment determination cycle of the UPS device; Among them, the mains power parameters include: voltage, frequency, and phase; Based on the mains parameters, the voltage fluctuation range, mains frequency deviation and voltage harmonic content are extracted within the misalignment judgment period.

8. The UPS-based mains fault prediction system according to claim 1, characterized in that: The preferred algorithm of the mains fault prediction model is determined as follows: Obtain different model algorithms to build a mains fault prediction model and output the model fitness value; The model fitness values ​​of different model algorithms are sorted in descending order, and the model algorithm corresponding to the largest model fitness value is determined as the preferred algorithm of the mains fault prediction model.

9. The UPS-based mains fault prediction system according to claim 8, characterized in that: The method for obtaining the model fitness value is as follows: A mains fault prediction model is constructed based on different model algorithms to obtain the monitoring mismatch value of a UPS device equipped with the mains fault prediction model during a peak working cycle, and the monitoring mismatch value of a UPS device not equipped with the mains fault prediction model during a peak working cycle; Performing difference processing on the monitoring misalignment value of the UPS device equipped with the mains fault prediction model during the peak working period and the monitoring misalignment value of the UPS device not equipped with the mains fault prediction model during the peak working period to obtain the monitoring misalignment difference; Obtain the monitoring misalignment of the UPS device during the peak working period and determine the positive or negative nature of the monitoring misalignment; If the monitoring misalignment is negative, the peak duty cycle is marked as a valid optimization cycle; Quantitatively analyze the effective optimization cycle to obtain the effective cycle ratio; The ratio of the monitoring misalignment difference within the effective optimization period is calculated to obtain the misalignment improvement ratio; The effective period ratio and the misalignment improvement ratio are weightedly summed to obtain the model fitness value.

10. The UPS-based mains power failure prediction system according to claim 9, characterized in that: The effective cycle ratio is obtained as follows: Obtain the number of effective optimization cycles and the number of peak working cycles in multiple peak working cycles, and perform ratio processing on the number of effective optimization cycles and the number of peak working cycles to obtain an effective cycle ratio; The misalignment improvement ratio is obtained as follows: The monitoring misalignment error within the effective optimization period is obtained, the absolute value of the monitoring misalignment error is taken, and the ratio is processed with the monitoring misalignment value of the UPS device without the mains fault prediction model during the peak working period to obtain the misalignment improvement ratio.

Images