Remote fault monitoring system of power equipment

Abstract

The invention discloses a remote fault monitoring system for power equipment, and relates to the technical field of fault prediction.The remote fault monitoring system periodically extracts stored monitoring data through an equipment management module to generate input data, and a fault prediction unit periodically performs fault prediction on target equipment based on the input data; the data analysis unit analyzes the periodic input data and the prediction data, in the analysis process, by reducing the influence of the data volume of different monitoring parameters in the input data on the prediction probability in batches, a configuration combination capable of reducing the maximum data volume is selected as an updating strategy of target equipment in a certain allowable prediction error sub-enclosure; through the mode, under the condition that the prediction probability is not greatly influenced, the data capacity of the transmitted input data is saved, the transmission rate is improved, meanwhile, the data volume input into the fault prediction model is reduced, and the prediction efficiency of the fault prediction model is further improved.

Description

Technical Field

[0001] This invention relates to the field of fault prediction technology, specifically to a remote fault monitoring system for power equipment. Background Technology

[0002] In the power system operation and maintenance system, the stable operation of power equipment is directly related to the continuity and security of power supply. Remote fault monitoring systems, as a core technical support for operation and maintenance management, integrate data acquisition, transmission, and analysis functions to predict potential equipment faults in advance, becoming a key means to reduce operation and maintenance costs and minimize unplanned power outages. Currently, mainstream remote fault monitoring systems generally incorporate fault prediction models. By collecting multi-dimensional monitoring parameters of the equipment and inputting them into the model, they calculate the probability of fault occurrence, providing management personnel with a basis for maintenance decisions. To achieve accurate fault prediction, existing technologies generally employ long-term extraction strategies for the monitoring parameters input into fault prediction models. This involves continuously collecting time-series data over a relatively long period to ensure data integrity and comprehensiveness, avoiding prediction bias caused by insufficient sample size or data fragmentation. However, long-term data extraction leads to a surge in data transmission and storage volume, increasing hardware resource consumption and prolonging data preprocessing and model computation time, thus affecting the response speed of fault warnings. Furthermore, from the perspective of actual operation and maintenance needs, fault prediction probability is not used to accurately determine whether a fault will inevitably occur, but only as a reference for managers to make rough judgments on maintenance duration and priority. For example, when the fault prediction probability output by the model reaches a certain threshold, managers need to initiate a maintenance plan. At this point, the core role of the prediction probability is to help distinguish between potential faults that require attention and normal fluctuations that do not require urgent handling. There is no need to pursue ultra-high precision probability values. The accuracy improvement brought by excessively long parameter extraction time does not provide substantial benefits to actual maintenance decisions; on the contrary, it leads to low prediction efficiency due to data redundancy. To address the above problems, this invention proposes a solution. Summary of the Invention

[0003] The purpose of this invention is to provide a remote fault monitoring system for power equipment in order to solve the problems mentioned in the background art.

[0004] This invention provides a remote fault monitoring system for power equipment, comprising: The device management module is used to store the monitoring data received from the target device. The monitoring data contains the monitoring values ​​of several monitoring parameters. The device management module is also used to extract and generate the input data of the target device for the prediction period from the stored monitoring data of the target device according to the extraction strategy of the target device at each prediction period. The extraction strategy includes the extraction interval duration, the static extraction duration of several monitoring parameters, and the dynamic extraction duration. The remote fault prediction platform includes a fault prediction unit and a data analysis unit. The fault prediction unit receives input data from the target device for each prediction cycle and inputs it into the fault prediction model. The fault prediction model outputs the prediction data for the target device for the prediction cycle. The prediction data includes a prediction signal, which can only be selected from the numbers 1 and 0. When the prediction signal is selected as 1, it indicates that the target device is predicted to fail within a preset fixed time period in the prediction cycle. At this time, the prediction data also includes the fault type and prediction probability. When the prediction signal is 1, the fault prediction unit determines whether the fault prediction of the target device in the prediction cycle needs intervention based on the intervention index of the target device stored in it and the prediction probability contained in the prediction data. The data analysis unit is used to store the input data and prediction data received from the target device for each prediction cycle. The data analysis unit is also used to analyze the target device and generate an update strategy for the target device after the amount of input data and prediction data of the stored target device reaches a preset fixed amount. The device management module is also used to update the extraction policy of the target device after receiving the target device update policy; The fault prediction unit is also used to replace and update the intervention indicators of the target device stored in the fault prediction unit after receiving the updated strategy of the target device transmitted.

[0005] Furthermore, it also includes an equipment monitoring module, which is used to collect monitoring data of the target equipment in real time based on the preset monitoring information of the target equipment. The monitoring information contains several monitoring parameters.

[0006] Furthermore, the steps for analyzing and generating the update strategy for the target device are as follows: S11: Extract all fault types from all predicted data of the target device stored in the data analysis unit and remove duplicates. Label all remaining fault types after deduplication as A1, A2, ..., Aa, where a≥1; S12: From all the predicted data of the target device stored in the data analysis unit, the predicted data containing fault type A1 are marked as B1, B2, ..., Bb, b≥1 respectively, and all the monitoring parameters contained in the monitoring information of the target device are marked as C1, C2, ..., Cc, c≥1 respectively; S13: Based on the preset number of segments P1 of the monitoring parameter C1, create a numerical list F1 of the monitoring parameter C1 relative to the predicted data B1, and determine the optional subsets G1, G2, ..., Gg of the monitoring parameter C1 based on the numerical list F1; wherein, the elements in the numerical list F1 are arranged in consecutive positive integer order from left to right, namely 1, 2, ..., P1; The optional subsets G1, G2, ..., Gg are {1}, {1,2}, {1,2,3}, ..., {1,2,..., P2}, respectively, where P2 is the maximum optional threshold of the preset monitoring parameter C1, and P2 is the maximum optional threshold of the preset monitoring parameter C1. <P1; S14: Obtain the input data corresponding to the prediction data B1. First, extract the set of monitoring values ​​of monitoring parameter C1 from the input data. Then, perform the binding operation between the elements in the number list F1 and the set of monitoring values. The set of monitoring values ​​contains all the monitoring values ​​of monitoring parameter C1 extracted from the input data. S15: According to S14, create numerical lists F2, F3, ..., Fc of monitoring parameters C2, C3, ..., Cc relative to the predicted data B1, and determine all optional subsets of monitoring parameters C2, C3, ..., Cc based on them, and perform binding operations between the elements in the numerical lists F2, F3, ..., Fc and the corresponding sets of monitoring values. S16: Determine all possible combinations of the predicted data B1 based on the list of numbers F1, F2, ..., Fc and all possible subsets of the monitoring parameters C1, C2, ..., Cc; S17: For each combination list of the predicted data B1, remove the data bound to numbers in each optional subset of the combination list from the input data corresponding to the predicted data B1 to obtain the analysis data of the combination list, and calculate the total data capacity of the data bound to numbers in each optional subset of the combination list, and use the total as the optimized data volume of the analysis data; S18: Select several combination lists from all combination lists of predicted data B1 according to the preset filtering rules as standard combinations of predicted data B1, and obtain its optimized data volume and probability error volume at the same time. S19: Obtain all standard combinations of the prediction data B2, B3, ..., Bb, along with their optimized data volume and probability error, in sequence from S12 to S18; S110: Select several standard combinations that meet the first selection condition from all standard combinations of predicted data B1, B2, ..., Bb as all original combinations of fault type A1, and calculate their optimization amount and fault tolerance amount. The first selection criterion is as follows: Before any original combination is selected, it is one of all standard combinations of the prediction data B1, B2, ..., Bb; For any native combination, obtain its optimized data quantity when predicting the standard combination of data B1, B2, ..., Bb. Use a discrete point filtering algorithm to process all the obtained optimized data quantities and calculate the average value of all remaining optimized data quantities after data processing. Define the average value as the optimization quantity of the native combination. Similarly, the probability error quantity when predicting the standard combination of data B1, B2, ..., Bb can be obtained using a discrete point filtering algorithm. S111: Obtain all original combinations of fault types A2, A3, ..., Aa, along with their optimization and fault tolerance quantities, according to S11 to S110 respectively; S112: Select several native combinations that meet the second selection condition from all native combinations of fault types A1, A2, ..., Aa as all configuration combinations, and calculate their optimization amount and fault tolerance amount; The second selection condition is as follows: any original combination is a combination of fault types A1, A2, ..., Aa before it is selected; For any configuration combination, obtain its optimization amount when it is the original combination of fault types A1, A2, ..., Aa. Use a discrete point filtering algorithm to process all the obtained optimization amounts and calculate the average value of all remaining optimization amounts after data processing. Use the average value as the optimization amount of the configuration combination. Similarly, its fault tolerance amount when it is the original combination of fault types A1, A2, ..., Aa can be obtained by using a discrete point filtering algorithm. S113: Label all configuration combinations as J1, J2, ..., Jj, j≥1. Calculate the analysis and optimization index M1 of configuration combination J1 using the formula M1=K1×ɑ1+L1×ɑ2. In the formula, K1 and L1 are the optimization amount and fault tolerance amount of configuration combination J1 after normalization, respectively, and ɑ1 and ɑ2 are the preset weights of the optimization amount and fault tolerance amount. Similarly, the analysis and optimization indices of configuration combinations J2, J3, ..., Jj are calculated and obtained in sequence. Based on the configuration combination and its fault tolerance corresponding to the analysis and optimization indices with the largest values ​​among the configuration combinations J1, J2, ..., Jj, the update strategy for the target device is generated.

[0007] Furthermore, in S14, the binding operation is performed as follows: Based on the extraction order from the furthest to the nearest time from the current time, several monitoring values ​​are extracted from the set of monitoring values ​​in batches. The total collection time of each batch of extracted monitoring values ​​is equal to the cutting time E1. A total of P1 batches of data can be extracted. The formula for E1 is D1 / P1, where D1 is the dynamic extraction time of monitoring parameter C1 in the extraction strategy of the target device. The extracted batch P1 data is bound to elements 1, 2, ..., P1 in the number list F1 in the order of extraction. At this time, elements 1, 2, ..., P1 in the number list F1 correspond to a batch of data.

[0008] Furthermore, in S16, the following content is determined: S161: Calculate the number of optional subsets for monitoring parameters C1, C2, ..., Cc respectively, and take the monitoring parameter with the most optional subsets as the benchmark parameter for predicted data B1; S162: For each optional subset of the reference parameters, perform the following steps on the c-1 lists of numbers: There are 1≤f≤c-1 selection operations. Each selection operation is as follows: for each number list in the c-1 number list, select several elements from the number list, summarize all the selected elements with the selected elements in the optional set and the optional subset, and put them into an empty list to obtain a combination list of the predicted data B1. Here, if represents the number of optional subsets corresponding to each number list in the c-1 number lists.

[0009] Furthermore, for S18, the filtering rules are as follows: For any combination list of predicted data B1, the analysis data of the combination list is input into the fault prediction model to obtain the predicted data of the corresponding combination list output by the fault prediction model. The fault type and prediction probability are then extracted from the predicted data. The extracted fault types are compared with the fault types in the predicted data B1. If the extracted fault types match the fault types in the predicted data B1, the difference between the extracted predicted probability and the predicted probability in the predicted data B1 is calculated. If the absolute value of the difference is less than or equal to P3, the combination list is selected as the standard combination for the predicted data B1. The optimized data volume of the analysis data in the combination list is used as the optimized data volume of the standard combination, and the difference is used as the probability error of the standard combination. If the extracted fault types do not match the fault types in the predicted data B1, no processing is performed. If the absolute value of the difference is greater than P3, no processing is performed. P3 is a preset acceptable error threshold, where the difference is the predicted probability in the predicted data B1 minus the extracted predicted probability.

[0010] Furthermore, the determination of whether intervention is needed for the fault prediction of the target equipment in the prediction cycle is as follows: Obtain the intervention index of the stored target device, add and subtract the intervention index from the prediction probability in the prediction data. If the value after either the addition or subtraction is greater than or equal to P5, it is determined that the fault prediction of the target device in the prediction period needs intervention. P5 is a preset intervention critical probability threshold, and the fault prediction of the target device in the prediction period is intervened according to the preset intervention rules. If the value after addition or subtraction is less than P5, it is determined that the fault prediction of the target device in the prediction cycle does not require intervention, and the prediction data of the target device in the prediction cycle is directly transmitted to the electronic equipment of the management personnel.

[0011] Furthermore, the intervention rules are as follows: An intervention instruction for the target device in the prediction period is generated and transmitted to the device management module. After receiving the intervention instruction, the device management module obtains the extraction interval duration and the static extraction duration and dynamic extraction duration of all monitoring parameters from the extraction strategy of the target device stored in its internal storage, and extracts the input intervention data of the target device in the prediction period from the stored monitoring data of the target device based on it, and transmits the input intervention data to the fault prediction unit. After receiving the input intervention data of the target device in the prediction period, the fault prediction unit inputs it together with the received input data of the target device in the prediction period into the fault prediction model. The fault prediction model outputs the prediction data of the target device after intervention in the prediction period and transmits the prediction data to the electronic device of the management personnel.

[0012] Compared with existing technologies, it has the following advantages: This invention acquires real-time monitoring data of the target device through a device monitoring module, periodically extracts and generates input data from the stored monitoring data through a device management module, periodically predicts faults in the target device based on the input data through a fault prediction unit, and analyzes the periodic input data and the predicted data after fault prediction through a data analysis unit. During the analysis, the impact of the amount of data of different monitoring parameters in the input data on the prediction probability is reduced in batches. Within a certain allowable range of prediction error, the configuration combination that reduces the amount of data the most is selected as the update strategy for the target device. In this way, without significantly affecting the prediction probability, the data capacity of the transmitted input data is saved, the transmission rate is improved, and the amount of data input into the fault prediction model is reduced, thereby further improving the prediction efficiency of the fault prediction model. In the analysis process, this invention determines the fault tolerance based on the configuration combination. After the fault diagnosis model outputs the predicted data, it determines whether to intervene in the fault probability within the predicted data. If intervention is required, the amount of data reduced from the input data is added back into the input data, and the fault prediction model re-predicts it. This avoids the problem of loss of prediction probability caused by reducing the amount of input data, which could lead to inaccurate judgment of high-risk predicted faults by management personnel. Attached Figure Description

[0013] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] Please see Figure 1 This application provides a remote fault monitoring system for power equipment, including an equipment monitoring module, an equipment management module, and a remote fault prediction module; The equipment monitoring module is used to collect monitoring data of the target equipment in real time based on the preset monitoring information of the target equipment and transmit it to the equipment management module. The monitoring information contains several monitoring parameters, and the monitoring data contains the monitoring values ​​of several monitoring parameters. The selection of the monitoring parameters is based on reflecting the health status of the target equipment and is determined by the management personnel in combination with actual needs. These parameters include, but are not limited to, current, voltage, and temperature. In this application, the target equipment refers to the power equipment of the target power station. The equipment management module is used to manage the monitoring data of the target equipment; for any target equipment in the target power plant, the equipment management module stores the real-time collected monitoring data of the target equipment. The device management module pre-stores preset extraction strategies for target devices within the target power station. The extraction strategy for any target device includes the extraction interval duration, the static extraction duration of several monitoring parameters, and the dynamic extraction duration. For any target device, the device management module extracts a number of monitoring data from all the monitoring data of the target device stored at each prediction cycle according to the extraction strategy of the target device to generate the input data of the target device for the prediction cycle, and transmits the input data to the remote fault prediction platform. It should be noted that the interval of a prediction cycle is numerically equal to the extraction interval within the extraction strategy of the target device. Taking the input data of the target device in the current prediction period as an example, the content of the input data of the target device in the current prediction period is generated as follows: The dynamic extraction duration of all monitoring parameters is obtained from the extraction strategy of the target device. For any monitoring parameter, the start time is taken as the time corresponding to the start point of the current prediction period. The end time of the monitoring parameter in the current prediction period is calculated by combining the dynamic extraction duration of the monitoring parameter. The formula for calculating the end time is the start time + the dynamic extraction duration. All monitoring values ​​of the monitoring parameter between the start time and the end time are extracted to obtain the extraction data of the monitoring parameter in the current prediction period. Among them, the extraction interval and static extraction duration of several monitoring parameters in the extraction strategy of any target device are preset by the management personnel, and the dynamic extraction duration of any monitoring parameter is initially equal to its static extraction duration in value. The input data for the target device in the current prediction period is generated based on the extracted data of all monitoring parameters obtained in the current prediction period. The remote fault prediction platform is used to periodically predict faults in several target devices of the target power plant. The remote fault prediction platform includes a fault prediction unit and a data analysis unit. For any target device, the remote fault prediction platform transmits the input data of the target device to the fault prediction unit after receiving it for each prediction cycle. The fault prediction unit stores a pre-trained fault prediction model and intervention indicators for the target device. The fault prediction model is based on a deep learning algorithm, which can identify potential fault trends behind minute changes and make accurate predictions. Through time series analysis, it can predict the fault type and fault probability, thereby providing maintenance personnel with effective early warning information, enabling them to perform maintenance and repairs in advance, and avoiding sudden equipment failures. For any target device, after receiving the input data of the target device for each prediction period, the fault prediction unit inputs it into the fault prediction model. The fault prediction model outputs the prediction data of the target device for the prediction period. The prediction data includes a prediction signal, which can only be selected from the number 1 or 0. When the prediction signal is selected as the number 1, it means that the target device is predicted to fail within a preset fixed time period in the prediction period. At this time, the prediction data also includes the fault type and the prediction probability. The fault type refers to the type of fault that the target device is predicted to fail within the preset fixed time period in the prediction period, and the prediction probability refers to the probability that the target device is predicted to fail within the preset fixed time period in the prediction period. When the prediction signal is selected as the number 0, it means that the prediction period will not predict a failure within a preset fixed duration. When the prediction signal is 1, the fault prediction unit obtains the intervention index of the target device stored within it, and adds or subtracts the intervention index from the prediction probability contained in the prediction data. If the value after the addition or subtraction is greater than or equal to P5, it is determined that the fault prediction of the target device in the prediction period needs intervention. P5 is a preset intervention critical probability threshold. Intervention is performed on the fault prediction of the target device in the prediction period according to the preset intervention rules, which are as follows: An intervention instruction for the target device in the prediction period is generated and transmitted to the device management module. After receiving the intervention instruction, the device management module obtains the extraction interval duration and the static extraction duration and dynamic extraction duration of all monitoring parameters from the extraction strategy of the target device stored in its internal storage. For any monitoring parameter, the intervention start time is calculated by combining the time corresponding to the start point of the current prediction period with the dynamic extraction duration of the monitoring parameter. The calculation formula for the intervention start time is the time + the dynamic extraction duration. The intervention end time is calculated by combining the dynamic extraction duration and the static extraction duration. The calculation formula for the intervention end time is the intervention start time + the static extraction duration - the dynamic extraction duration. All monitoring values ​​of the monitoring parameter between the collected time and the intervention start time and the intervention end time are extracted to obtain the intervention data of the monitoring parameter in the prediction period. Based on the intervention data of all monitoring parameters obtained in the current prediction period, the input intervention data of the target device in the prediction period is generated, and the input intervention data is transmitted to the fault prediction unit. After receiving the input intervention data of the target device in the prediction period, the fault prediction unit inputs it together with the input data of the target device in the prediction period into the fault prediction model. The fault prediction model outputs the prediction data of the target device after intervention in the prediction period and transmits the prediction data to the electronic device of the manager. The electronic device includes, but is not limited to, mobile phones, laptops and desktop computers. In this application, the purpose of intervention is to avoid the following situation: In a scenario where the amount of data is not optimized, assuming that the prediction probability of the fault prediction model based on the complete monitoring data output is 81%, this value has reached the threshold at which managers need to pay close attention and initiate maintenance assessment. However, if the amount of data is blindly reduced in pursuit of transmission efficiency and prediction speed, the prediction probability output by the model may drop to 78%, which is just below the maintenance attention threshold. This would cause managers to miss the opportunity to intervene in potential high-risk faults in a timely manner, thereby leading to problems such as the expansion of equipment failure and increased operation and maintenance costs. If any value after addition or subtraction is less than P5, it is determined that the fault prediction of the target device in the prediction cycle does not require intervention, and the prediction data of the target device in the prediction cycle is directly transmitted to the electronic device of the management personnel. It should be noted here that any input data corresponds to the prediction data output by the fault prediction model after it is input into the fault prediction model. Example 2 is implemented based on Example 1; For any target device, the data analysis unit analyzes it after storing a preset fixed amount of input data and prediction data of the target device. The fixed amount ensures that the amount of stored input data and prediction data can fully cover the various operating states of the target device, providing a complete data foundation for subsequent analysis and a reliable basis for fault prediction and operation and maintenance decisions. The analysis steps are as follows: S11: Extract all fault types from all predicted data of the target device stored in the data analysis unit and remove duplicates. Label all remaining fault types after deduplication as A1, A2, ..., Aa, where a≥1; S12: From all the predicted data of the target device stored in the data analysis unit, the predicted data containing fault type A1 are marked as B1, B2, ..., Bb, b≥1 respectively, and all the monitoring parameters contained in the monitoring information of the target device are marked as C1, C2, ..., Cc, c≥1 respectively; S13: Based on the preset number of segments P1 of the monitoring parameter C1, create a numerical list F1 of the monitoring parameter C1 relative to the predicted data B1, and determine the optional subsets G1, G2, ..., Gg of the monitoring parameter C1 based on the numerical list F1; Among them, the elements in the digital list F1 are arranged in consecutive positive integer order from left to right, which are 1, 2, ..., P1 in sequence. In the standard expression form in the programming field, it is F1 = [1, 2, ..., P1]. The meaning of P1 is to evenly cut the acquisition moments of the monitoring values of the monitoring parameter C1 into P1 consecutive time periods; The optional subsets G1, G2, ..., Gg are {1}, {1, 2}, {1, 2, 3}, ..., {1, 2, ..., P2} respectively. P2 is the preset optional maximum threshold of the monitoring parameter C1. P2 < P1, and the value of P2 is selected by the management personnel based on historical data analysis, which is used to predict the shortest acquisition duration of the monitoring value of the monitoring parameter C1 required for the fault type A1; Here, it should be noted that the digital list F1 corresponds to the optional subsets G1, G2, ..., Gg respectively. Since the digital list F1 is the digital list of the monitoring parameter C1 relative to the prediction data B1, the optional subsets G1, G2, ..., Gg also correspond to the monitoring parameter C1; S14: Obtain the input data corresponding to the prediction data B1. First, extract the set of monitoring values of the monitoring parameter C1 from the input data, and then perform the binding operation between the elements in the digital list F1 and the set of monitoring values. The set of monitoring values contains all the monitoring values of the monitoring parameter C1 extracted from the input data; The content of the binding operation is as follows: Extract several monitoring values from the set of monitoring values in batches in the order of the acquisition moment from far to near to the current moment. The total duration of the acquisition moments of the monitoring values extracted in each batch accumulatively equals the cutting duration E1. A total of P1 batches of data can be extracted. Among them, the calculation formula of E1 is D1 / P1, and D1 is the dynamic extraction duration of the monitoring parameter C1 in the extraction strategy of the target device; Bind the P1 batches of data extracted in the order of extraction with the elements 1, 2, ..., P1 in the digital list F1 in sequence. At this time, the elements 1, 2, ..., P1 in the digital list F1 correspond to a batch of data respectively; S15: Create the digital lists F2, F3, ..., Fc of the monitoring parameters C2, C3, ..., Cc relative to the prediction data B1 respectively according to S14, and determine all the optional subsets of the monitoring parameters C2, C3, ..., Cc according to their correspondence, and perform the binding operation between the elements in the digital lists F2, F3, ..., Fc and the corresponding sets of monitoring values respectively; S16: Determine all the combination lists of the prediction data B1 according to the digital lists F1, F2, ..., Fc and all the optional subsets of the monitoring parameters C1, C2, ..., Cc. The determination content is as follows: S161: Calculate the number of optional subsets of monitoring parameters C1, C2, ..., Cc respectively, and take the monitoring parameter with the most optional subsets as the benchmark parameter of the predicted data B1. If there are multiple monitoring parameters with the most optional subsets, randomly select one of them. S162: For each optional subset of the reference parameters, perform the following steps on the c-1 lists of numbers: There are 1 ≤ f ≤ c-1 selection operations. Each selection operation involves selecting several elements from each of the c-1 number lists, summarizing all the selected elements with the selected elements in the optional set and the optional subset, and adding them to an empty list to obtain a combination list of the predicted data B1. Here, if represents the number of optional subsets corresponding to each of the c-1 number lists. It should be noted that the minimum number of optional subsets selected from all optional subsets corresponding to any number list is 0, and the maximum is the sum of all its optional subsets. The c-1 number lists refer to all number lists in the number lists F1, F2, ..., Fc except for the number list corresponding to the baseline parameter. It should be noted that the index of the reference parameter in any combination list increases sequentially from left to right; S17: For each combination list of the predicted data B1, remove the data bound to numbers in each optional subset of the combination list from the input data corresponding to the predicted data B1 to obtain the analysis data of the combination list, and calculate the total data capacity of the data bound to numbers in each optional subset of the combination list, and use the total as the optimized data volume of the analysis data; S18: Select several combination lists from all combination lists of predicted data B1 as standard combinations of predicted data B1 according to preset filtering rules, and obtain their optimized data volume and probability error volume. The filtering rules are as follows: For any combination list of predicted data B1, the analysis data of the combination list is input into the fault prediction model to obtain the predicted data of the corresponding combination list output by the fault prediction model. The fault type and prediction probability are then extracted from the predicted data. The extracted fault types are compared with the fault types in the predicted data B1. If the extracted fault types match the fault types in the predicted data B1, the difference between the extracted predicted probability and the predicted probability in the predicted data B1 is calculated. If the absolute value of the difference is less than or equal to P3, the combination list is selected as the standard combination of the predicted data B1, the optimized data volume of the analysis data in the combination list is used as the optimized data volume of the standard combination, and the difference is used as the probability error of the standard combination. If the extracted fault types do not match the fault types in the predicted data B1, no processing is performed. If the absolute value of the difference is greater than P3, no processing is performed. P3 is a preset acceptable error threshold, where the difference is the predicted probability in the predicted data B1 minus the extracted predicted probability. S19: Obtain all standard combinations of the prediction data B2, B3, ..., Bb, along with their optimized data volume and probability error, in sequence from S12 to S18; S110: Select several standard combinations that meet the first selection condition from all standard combinations of predicted data B1, B2, ..., Bb as all original combinations of fault type A1, and calculate their optimization amount and fault tolerance amount. The first selection criterion is as follows: Before any original combination is selected, it is one of all standard combinations of the prediction data B1, B2, ..., Bb; For any native combination, the optimization amount is calculated as follows: the optimization data amount is obtained when predicting data B1, B2, ..., Bb in the standard combination; the discrete point filtering algorithm is used to process all the obtained optimization data amounts; the average value of all remaining optimization data amounts after data processing is calculated; and the average value is defined as the optimization amount of the native combination. For any native combination, the fault tolerance is calculated as follows: obtain the probability error when predicting the standard combination of data B1, B2, ..., Bb; use a discrete point filtering algorithm to process all the obtained probability error quantities; calculate the average value of all remaining probability error quantities after data processing; and define the absolute value of the average value as the fault tolerance of the native combination. In this application, the discrete point filtering algorithm can be any one of the Z-score filtering algorithm, IQR filtering algorithm, and density filtering algorithm; S111: Obtain all original combinations of fault types A2, A3, ..., Aa, along with their optimization and fault tolerance quantities, according to S11 to S110 respectively; S112: Select several native combinations that meet the second selection condition from all native combinations of fault types A1, A2, ..., Aa as all configuration combinations, and calculate their optimization amount and fault tolerance amount; The second selection condition is as follows: any original combination is a combination of fault types A1, A2, ..., Aa before it is selected; For any given configuration combination, the optimized values ​​are calculated as follows: Obtain the optimization amount when the fault type is the original combination of A1, A2, ..., Aa. Use the discrete point filtering algorithm to process all the obtained optimization amounts and calculate the average value of all remaining optimization amounts after data processing. Define the average value as the optimization amount of the configuration combination. For any given configuration combination, the fault tolerance is calculated as follows: Obtain the fault tolerance amount when the fault types A1, A2, ..., Aa are natively combined. Use a discrete point filtering algorithm to process all the obtained fault tolerance amounts and calculate the average value of all remaining fault tolerance amounts after data processing. Define the average value as the fault tolerance amount of the configuration combination. S113: Label all configuration combinations as J1, J2, ..., Jj, j≥1. Calculate the analysis optimization index M1 of configuration combination J1 using the formula M1=K1×ɑ1+L1×ɑ2. In the formula, K1 and L1 are the optimization amount and fault tolerance amount of configuration combination J1 after normalization, respectively, and ɑ1 and ɑ2 are the preset weights of the optimization amount and fault tolerance amount. The normalization process is to remove the influence of dimensions from the optimization amount and fault tolerance amount during the calculation. It should be noted that the analysis optimization index M1 is artificially defined to measure the efficiency and error impact of fault prediction after data removal of the input data using the optional subset within the configuration combination. Similarly, the analysis and optimization indices of configuration combinations J2, J3, ..., Jj are calculated and obtained in sequence. Based on the configuration combination and its fault tolerance corresponding to the analysis and optimization indices of configuration combinations J1, J2, ..., Jj with the largest value, the update strategy of the target device is generated. The data analysis unit transmits the generated update strategy for the target device to the device management module and the fault prediction unit, respectively. After receiving the transmitted target device update policy, the device management module updates the target device extraction policy. The update steps are as follows: First, extract all optional subsets contained in the update strategy; Then, for each optional subset, the monitoring parameters corresponding to the optional subset are obtained, and the dynamic extraction duration Q1 of the monitoring parameters within the extraction strategy of the target device currently stored in the device management module is obtained; The new dynamic extraction duration T1 of the monitoring parameter is calculated using the formula T1=U1×P4, where U1 is the element with the largest value in the selectable set, and P4 is the preset number of segments of the monitoring parameter. As described above, the new dynamic extraction duration of the monitoring parameters corresponding to all optional subsets is calculated and obtained in sequence. Based on the new dynamic extraction duration of the monitoring parameters corresponding to all optional subsets, the dynamic extraction duration of all monitoring parameters in the extraction strategy of the target device stored in the device management module is replaced and updated accordingly. After receiving the updated strategy of the target device, the fault prediction unit uses the fault tolerance of the configuration combination contained therein as the new intervention index of the target device to replace and update the intervention index of the target device stored in the fault prediction unit.

[0016] Some of the data in the above formulas are numerical calculations with dimensions removed, and the contents not described in detail in this specification are all prior art known to those skilled in the art.

[0017] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A remote fault monitoring system for power equipment, characterized in that, include: The device management module is used to store the monitoring data received from the target device. The monitoring data contains the monitoring values ​​of several monitoring parameters. The device management module is also used to extract and generate the input data of the target device for the prediction period from the stored monitoring data of the target device according to the extraction strategy of the target device at each prediction period. The extraction strategy includes the extraction interval duration, the static extraction duration of several monitoring parameters, and the dynamic extraction duration. The remote fault prediction platform includes a fault prediction unit and a data analysis unit. The fault prediction unit receives input data from the target device for each prediction cycle and inputs it into the fault prediction model. The fault prediction model outputs the prediction data for the target device for the prediction cycle. The prediction data includes a prediction signal, which can only be selected from the numbers 1 and 0. When the prediction signal is selected as 1, it indicates that the target device is predicted to fail within a preset fixed time period in the prediction cycle. At this time, the prediction data also includes the fault type and prediction probability. When the prediction signal is 1, the fault prediction unit determines whether the fault prediction of the target device in the prediction cycle needs intervention based on the intervention index of the target device stored in it and the prediction probability contained in the prediction data. The data analysis unit is used to store the input data and prediction data received from the target device for each prediction cycle. The data analysis unit is also used to analyze the target device and generate an update strategy for the target device after the amount of input data and prediction data of the stored target device reaches a preset fixed amount. The device management module is also used to update the extraction policy of the target device after receiving the target device update policy; The fault prediction unit is also used to replace and update the intervention indicators of the target device stored in the fault prediction unit after receiving the updated strategy of the target device transmitted.

2. The remote fault monitoring system for power equipment according to claim 1, characterized in that, It also includes an equipment monitoring module, which is used to collect monitoring data of the target equipment in real time based on the preset monitoring information of the target equipment. The monitoring information contains several monitoring parameters.

3. The remote fault monitoring system for power equipment according to claim 1, characterized in that, The steps for analyzing and generating the update strategy for the target device are as follows: S11: Extract all fault types from all predicted data of the target device stored in the data analysis unit and remove duplicates. Label all remaining fault types after deduplication as A1, A2, ..., Aa, where a≥1; S12: From all the predicted data of the target device stored in the data analysis unit, the predicted data containing fault type A1 are marked as B1, B2, ..., Bb, b≥1 respectively, and all the monitoring parameters contained in the monitoring information of the target device are marked as C1, C2, ..., Cc, c≥1 respectively; S13: Based on the preset number of segments P1 of the monitoring parameter C1, create a numerical list F1 of the monitoring parameter C1 relative to the predicted data B1, and determine the optional subsets G1, G2, ..., Gg of the monitoring parameter C1 based on the numerical list F1; wherein, the elements in the numerical list F1 are arranged in consecutive positive integer order from left to right, namely 1, 2, ..., P1; The optional subsets G1, G2, ..., Gg are {1}, {1,2}, {1,2,3}, ..., {1,2,..., P2}, respectively, where P2 is the maximum optional threshold of the preset monitoring parameter C1, and P2 is the maximum optional threshold of the preset monitoring parameter C1. <P1; S14: Obtain the input data corresponding to the prediction data B1. First, extract the set of monitoring values ​​of monitoring parameter C1 from the input data. Then, perform the binding operation between the elements in the number list F1 and the set of monitoring values. The set of monitoring values ​​contains all the monitoring values ​​of monitoring parameter C1 extracted from the input data. S15: According to S14, create numerical lists F2, F3, ..., Fc of monitoring parameters C2, C3, ..., Cc relative to the predicted data B1, and determine all optional subsets of monitoring parameters C2, C3, ..., Cc based on them, and perform binding operations between the elements in the numerical lists F2, F3, ..., Fc and the corresponding sets of monitoring values. S16: Determine all possible combinations of the predicted data B1 based on the list of numbers F1, F2, ..., Fc and all possible subsets of the monitoring parameters C1, C2, ..., Cc; S17: For each combination list of the predicted data B1, remove the data bound to numbers in each optional subset of the combination list from the input data corresponding to the predicted data B1 to obtain the analysis data of the combination list, and calculate the total data capacity of the data bound to numbers in each optional subset of the combination list, and use the total as the optimized data volume of the analysis data; S18: Select several combination lists from all combination lists of predicted data B1 according to the preset filtering rules as standard combinations of predicted data B1, and obtain its optimized data volume and probability error volume at the same time. S19: Obtain all standard combinations of the prediction data B2, B3, ..., Bb, along with their optimized data volume and probability error, in sequence from S12 to S18; S110: Select several standard combinations that meet the first selection condition from all standard combinations of predicted data B1, B2, ..., Bb as all original combinations of fault type A1, and calculate their optimization amount and fault tolerance amount. The first selection criterion is as follows: Before any original combination is selected, it is one of all standard combinations of the prediction data B1, B2, ..., Bb; For any native combination, obtain its optimized data quantity when predicting the standard combination of data B1, B2, ..., Bb. Use a discrete point filtering algorithm to process all the obtained optimized data quantities and calculate the average value of all remaining optimized data quantities after data processing. Define the average value as the optimization quantity of the native combination. Similarly, the probability error quantity when predicting the standard combination of data B1, B2, ..., Bb can be obtained using a discrete point filtering algorithm. S111: Obtain all original combinations of fault types A2, A3, ..., Aa, along with their optimization and fault tolerance quantities, according to S11 to S110 respectively; S112: Select several native combinations that meet the second selection condition from all native combinations of fault types A1, A2, ..., Aa as all configuration combinations, and calculate their optimization amount and fault tolerance amount; The second selection condition is as follows: any original combination is a combination of fault types A1, A2, ..., Aa before it is selected; For any configuration combination, obtain its optimization amount when it is the original combination of fault types A1, A2, ..., Aa. Use a discrete point filtering algorithm to process all the obtained optimization amounts and calculate the average value of all remaining optimization amounts after data processing. Use the average value as the optimization amount of the configuration combination. Similarly, its fault tolerance amount when it is the original combination of fault types A1, A2, ..., Aa can be obtained by using a discrete point filtering algorithm. S113: Label all configuration combinations as J1, J2, ..., Jj, j≥1. Calculate the analysis and optimization index M1 of configuration combination J1 using the formula M1=K1×ɑ1+L1×ɑ2. In the formula, K1 and L1 are the optimization amount and fault tolerance amount of configuration combination J1 after normalization, respectively, and ɑ1 and ɑ2 are the preset weights of the optimization amount and fault tolerance amount. Similarly, the analysis and optimization indices of configuration combinations J2, J3, ..., Jj are calculated and obtained in sequence. Based on the configuration combination and its fault tolerance corresponding to the analysis and optimization indices with the largest values ​​among the configuration combinations J1, J2, ..., Jj, the update strategy for the target device is generated.

4. The remote fault monitoring system for power equipment according to claim 3, characterized in that, S14, the binding operation is performed as follows: Based on the extraction order from the furthest to the nearest time from the current time, several monitoring values ​​are extracted from the set of monitoring values ​​in batches. The total collection time of each batch of extracted monitoring values ​​is equal to the cutting time E1. A total of P1 batches of data can be extracted. The formula for E1 is D1 / P1, where D1 is the dynamic extraction time of monitoring parameter C1 in the extraction strategy of the target device. The extracted batch P1 data is bound to elements 1, 2, ..., P1 in the number list F1 in the order of extraction. At this time, elements 1, 2, ..., P1 in the number list F1 correspond to a batch of data.

5. A remote fault monitoring system for power equipment according to claim 3, characterized in that, S16, the following content is determined: S161: Calculate the number of optional subsets for monitoring parameters C1, C2, ..., Cc respectively, and take the monitoring parameter with the most optional subsets as the benchmark parameter for predicted data B1; S162: For each optional subset of the reference parameters, perform the following steps on the c-1 lists of numbers: There are 1≤f≤c-1 selection operations. Each selection operation is as follows: for each number list in the c-1 number list, select several elements from the number list, summarize all the selected elements with the selected elements in the optional set and the optional subset, and put them into an empty list to obtain a combination list of the predicted data B1. Here, if represents the number of optional subsets corresponding to each number list in the c-1 number lists.

6. A remote fault monitoring system for power equipment according to claim 5, characterized in that, S18, the filtering rules are as follows: For any combination list of predicted data B1, the analysis data of the combination list is input into the fault prediction model to obtain the predicted data of the corresponding combination list output by the fault prediction model. The fault type and prediction probability are then extracted from the predicted data. The extracted fault types are compared with the fault types in the predicted data B1. If the extracted fault types match the fault types in the predicted data B1, the difference between the extracted predicted probability and the predicted probability in the predicted data B1 is calculated. If the absolute value of the difference is less than or equal to P3, the combination list is selected as the standard combination for the predicted data B1. The optimized data volume of the analysis data in the combination list is used as the optimized data volume of the standard combination, and the difference is used as the probability error of the standard combination. If the extracted fault types do not match the fault types in the predicted data B1, no processing is performed. If the absolute value of the difference is greater than P3, no processing is performed. P3 is a preset acceptable error threshold, where the difference is the predicted probability in the predicted data B1 minus the extracted predicted probability.

7. A remote fault monitoring system for power equipment according to claim 1, characterized in that, The criteria for determining whether intervention is needed for fault prediction of the target equipment in the prediction cycle are as follows: The intervention index of the target device is obtained from the storage. The intervention index is added to and subtracted from the prediction probability in the prediction data. If the value after either the addition or subtraction operation is greater than or equal to P5, it is determined that the fault prediction of the target device in the prediction period needs to be intervened. P5 is a preset intervention critical probability threshold. The fault prediction of the target device in the prediction period is intervened according to the preset intervention rules. If the value after addition or subtraction is less than P5, it is determined that the fault prediction of the target device in the prediction cycle does not require intervention, and the prediction data of the target device in the prediction cycle is directly transmitted to the electronic equipment of the management personnel.

8. A remote fault monitoring system for power equipment according to claim 7, characterized in that, The intervention rules are as follows: An intervention instruction for the target device in the prediction period is generated and transmitted to the device management module. After receiving the intervention instruction, the device management module obtains the extraction interval duration and the static extraction duration and dynamic extraction duration of all monitoring parameters from the extraction strategy of the target device stored in its internal storage, and extracts the input intervention data of the target device in the prediction period from the stored monitoring data of the target device based on it, and transmits the input intervention data to the fault prediction unit. After receiving the input intervention data of the target device in the prediction period, the fault prediction unit inputs it together with the received input data of the target device in the prediction period into the fault prediction model. The fault prediction model outputs the prediction data of the target device after intervention in the prediction period and transmits the prediction data to the electronic device of the management personnel.

Images