An artificial intelligence-based power engineering terminal device operation and maintenance management system

Abstract

The application discloses an artificial intelligence-based power engineering terminal device operation and maintenance management system and relates to the technical field of power engineering operation and maintenance.The system comprises a data sensing and forecasting module, a task generation module, a resource scheduling module and an execution control module.The data sensing and forecasting module collects data from terminal devices and inputs the data into a time series forecasting model to generate a predicted failure probability sequence.The task generation module selects devices with a predicted failure probability exceeding a threshold value as target devices and generates initial operation and maintenance tasks containing device identification, failure type estimation and recommended maintenance duration.The resource scheduling module acquires operation and maintenance personnel information and spare parts inventory information, calculates task comprehensive priorities by using a dynamic priority allocation algorithm and sorts the tasks.The execution control module dispatches tasks to operation and maintenance personnel terminals in sequence and receives real-time execution state feedback.The system can actively predict failures and generate tasks in advance and accurately allocate operation and maintenance resources by combining dynamic priority scheduling.

Description

Technical Field

[0001] This invention relates to the field of power engineering operation and maintenance technology, specifically to an artificial intelligence-based power engineering terminal equipment operation and maintenance management system. Background Technology

[0002] The operation and maintenance management of power engineering terminal equipment relies on the monitoring and analysis of equipment status data. Existing technical solutions typically determine whether the equipment's operating status is abnormal based on fixed thresholds or simple rules. After collecting real-time equipment data, if the data exceeds a preset alarm threshold, an alarm is triggered and a maintenance work order is generated. Then, dispatchers allocate maintenance personnel and spare parts resources based on experience. This type of mode is essentially a passive response-based operation and maintenance, only initiating maintenance actions when equipment parameters have deviated significantly.

[0003] Existing technical solutions have several shortcomings. On the one hand, relying solely on threshold alarms cannot detect potential equipment degradation trends in advance; maintenance processes are only initiated after a failure occurs, leading to long unplanned downtimes and high repair costs. On the other hand, the manual dispatching method after maintenance tasks are generated struggles to comprehensively consider the skill matching of maintenance personnel, geographical distribution, spare parts inventory locations, and dynamic changes in the urgency of different tasks. This easily results in task backlog, idle personnel, and unreasonable spare parts requisition paths, leading to overall low maintenance efficiency. Furthermore, most existing scheduling strategies are fixed and cannot be self-corrected based on actual execution results, making it difficult to adapt to the increasing number of power grid terminal devices and the increasing complexity of maintenance scenarios.

[0004] The problem to be solved is how to predict equipment failures in advance and directly translate the prediction results into executable maintenance tasks. Another problem to be solved is how to perform dynamic priority quantification calculations for batch maintenance tasks based on multi-factor fusion, and on this basis, achieve joint optimized assignment of tasks, personnel, and spare parts. Summary of the Invention

[0005] This invention provides an artificial intelligence-based power engineering terminal equipment operation and maintenance management system, which aims to solve the problems that existing threshold alarm-based operation and maintenance cannot detect potential equipment faults in advance, and that manual dispatching methods are difficult to achieve dynamic optimization and scheduling of tasks and resources under multiple constraints.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an artificial intelligence-based power engineering terminal equipment operation and maintenance management system, comprising a data perception and prediction module, a task generation module, a resource scheduling module, and an execution control module.

[0007] The data perception and prediction module collects real-time operating status, historical fault records and environmental parameter data from multiple terminal devices, and inputs them into the time series prediction model to generate a predicted fault probability sequence for each device in the next time window, thereby identifying potential faults in advance.

[0008] The task generation module selects terminal devices whose predicted failure probability exceeds the fault threshold as target devices based on the predicted failure probability sequence and the preset fault threshold. It also generates an initial maintenance task for each target device, which includes device identification, fault type estimation and recommended maintenance duration, to ensure that high-risk devices are included in the maintenance plan in a timely manner.

[0009] The resource scheduling module obtains currently available maintenance personnel information and spare parts inventory information, and calculates the comprehensive priority of each initial maintenance task based on the initial maintenance task, the maintenance personnel information, and the spare parts inventory information using a dynamic priority allocation algorithm. The initial maintenance tasks are sorted from high to low according to the comprehensive priority to obtain a sorted list of tasks to be executed, thereby realizing the reasonable allocation of limited maintenance resources among multiple tasks.

[0010] The execution control module dispatches each task in the list of tasks to be executed to the corresponding maintenance personnel terminal in sorting order, and receives the task execution status feedback from the maintenance personnel terminal in real time, forming a closed-loop operation and maintenance management.

[0011] As a technical solution of the present invention, the process of generating the predicted fault probability sequence of each device in the next time window in the data perception and prediction module is specifically as follows:

[0012] Real-time operating status data, historical fault record data, and environmental parameter data are collected from multiple terminal devices at the power engineering site;

[0013] The real-time operating status data of each terminal device is divided into multiple state feature vectors for consecutive time segments by a sliding window.

[0014] The historical fault record data of each terminal device is timestamped, and the historical state feature sequence within a preset time before each historical fault occurs is extracted. The historical state feature sequence and the corresponding fault type label are used as training samples.

[0015] The environmental parameter data is concatenated with the real-time operating status data according to the timestamp to obtain the fused input features at each time point;

[0016] The fused input features are sequentially passed through the convolutional layer, recurrent layer, and fully connected layer in the time-series prediction model to output the probability value of each terminal device belonging to different fault types in each time unit within a future preset time window. The probability values ​​of all time units are arranged in chronological order to form the predicted fault probability sequence.

[0017] Preferably, the recurrent layer in the time-series prediction model uses a gated recurrent unit, and the feature map output by the convolutional layer is expanded by time step and then input into the gated recurrent unit, thereby accurately capturing the evolution of the running state over time.

[0018] Furthermore, the step in the task generation module to filter target devices based on the predicted fault probability sequence and the preset fault threshold is as follows:

[0019] Extract the maximum probability value for each terminal device and the fault type corresponding to the maximum probability value from the predicted fault probability sequence;

[0020] Compare the maximum probability value with the preset fault threshold. If the maximum probability value is greater than the fault threshold, mark the corresponding terminal device as the target device and use the fault type corresponding to the maximum probability value as the fault type prediction for the target device.

[0021] For each target device, query the historical average repair time corresponding to the fault type in the historical fault record data, and multiply the historical average repair time by a preset margin coefficient to obtain the recommended repair time.

[0022] The device identifier of the target device, the fault type prediction, and the recommended maintenance time are combined into a structured data object, which serves as the initial maintenance task.

[0023] The preset fault threshold is dynamically adjusted according to the importance level of each terminal device. The higher the importance level of the terminal device, the lower the corresponding fault threshold, so that critical equipment is included in the operation and maintenance scope when a low probability fault is predicted, thereby improving the power supply guarantee capability.

[0024] The resource scheduling module obtains currently available maintenance personnel information and spare parts inventory information in the following way:

[0025] Send an availability query request to the operations and maintenance personnel management server, and receive a list of operations and maintenance personnel who are currently idle or will complete the current task within a preset time, returned by the operations and maintenance personnel management server. Each operations and maintenance personnel record in the list includes personnel ID, skill tag set, current location coordinates and average maintenance efficiency.

[0026] Send an inventory query request to the spare parts warehouse management system and receive the spare parts identifier, spare parts type, storage location and inventory quantity of each spare part in the current inventory returned by the spare parts warehouse management system;

[0027] For each initial maintenance task, the fault type is estimated, and the spare parts type that matches the fault type is selected from the spare parts inventory information. The inventory quantity of the spare parts type is checked to see if it is greater than zero. If it is greater than zero, the corresponding spare parts identifier is associated with the initial maintenance task to avoid the task being unable to be executed due to lack of materials.

[0028] As a preferred embodiment of the present invention, the process by which the resource scheduling module calculates the comprehensive priority of each initial operation and maintenance task using a dynamic priority allocation algorithm is as follows:

[0029] Extract the device identifier from each initial maintenance task, and query the historical fault record data based on the device identifier to obtain the fault reproduction interval of the corresponding terminal device;

[0030] Extract the fault type prediction from each initial operation and maintenance task, and match the fault severity score corresponding to the fault type prediction from the preset fault level mapping table;

[0031] Based on the recommended maintenance duration and the average maintenance efficiency in the currently available maintenance personnel information, calculate the theoretical duration of each initial maintenance task;

[0032] Obtain the geographical location of the terminal device corresponding to each initial operation and maintenance task, and calculate the spatial distance between the geographical location and the current location of each operation and maintenance personnel in the currently available operation and maintenance personnel information;

[0033] The fault reproduction interval, the fault severity score, the theoretical occupancy time, and the spatial distance are input into a weighted summation function, which outputs a comprehensive priority value for each initial maintenance task. The initial maintenance tasks are then sorted from largest to smallest according to the comprehensive priority value.

[0034] The weighting coefficient for the fault reproduction interval in the weighted summation function is set to a negative value, so that the shorter the fault reproduction interval, the higher the overall priority value, highlighting the emergency repair needs for sudden and recurring faults.

[0035] The calculation of the theoretical time taken further includes: extracting the average maintenance efficiency of all maintenance personnel from the currently available maintenance personnel information, calculating the reciprocal of the average maintenance efficiency to obtain the standard working hours required for unit fault repair; multiplying the recommended maintenance time by the standard working hours to obtain the baseline time taken; obtaining the device complexity level of the terminal device corresponding to each initial maintenance task, querying the complexity adjustment coefficient corresponding to the device complexity level from a preset complexity coefficient mapping table; and multiplying the baseline time taken by the complexity adjustment coefficient to obtain the theoretical time taken for each initial maintenance task, so that the task time estimation is closer to the actual situation on site.

[0036] The calculation of the spatial distance further includes: querying the latitude and longitude coordinates of the installation location corresponding to the equipment identifier of each initial maintenance task from the equipment ledger database, as the first coordinate; extracting the latitude and longitude coordinates of the current location of each maintenance personnel from the currently available maintenance personnel information, as the second coordinate; for each combination of initial maintenance task and each maintenance personnel, converting the first coordinate and the second coordinate into projection plane coordinates under the same spatial reference system; calculating the Euclidean distance value between the projection plane coordinates, and multiplying the Euclidean distance value by a preset road network tortuosity coefficient to obtain the spatial distance corresponding to each combination, effectively simulating the impact of the actual path on scheduling.

[0037] The execution control module dispatches each task in the list of tasks to be executed to the corresponding maintenance personnel's terminal according to the sorting order as follows:

[0038] The current task with the highest overall priority is retrieved from the list of tasks to be executed. Based on the device identifier and fault type in the current task, the corresponding spare part storage location is extracted from the spare part identifier associated with the resource scheduling module.

[0039] Obtain information on all currently available maintenance personnel, calculate the first path distance from the current location of each maintenance personnel to the equipment installation location in the current task, and the second path distance from the current location of each maintenance personnel to the spare parts storage location, and use the sum of the first path distance and the second path distance as the total path distance for each maintenance personnel;

[0040] Select the maintenance personnel with the smallest total path distance as the assignment target of the current task, encapsulate the current task into a dispatch instruction, and send it to the terminal of the maintenance personnel corresponding to the assignment target.

[0041] Remove the currently assigned task from the list of tasks to be executed, update the status of the assigned object in the information of the currently available maintenance personnel to the occupied status, and then continue to assign the remaining tasks in the sorted order, thereby minimizing the overall travel time for maintenance personnel to go to the equipment and pick up spare parts.

[0042] The steps in the execution control module for receiving real-time feedback on the task execution status from the maintenance personnel's terminal are as follows:

[0043] Receive a status update message sent by the maintenance personnel's terminal, the status update message including personnel identifier, task identifier, current status code and status occurrence timestamp;

[0044] The task execution progress is determined based on the current status code, which includes: order accepted, dispatched, spare parts retrieved, arrived at site, under repair, and completed.

[0045] When a completed status code is received, the actual repair time and post-repair equipment operating parameters are obtained from the maintenance personnel's terminal. The actual repair time is compared with the recommended repair time in the corresponding initial maintenance task, and a repair efficiency record is generated and stored in the historical maintenance database.

[0046] The operating parameters of the repaired equipment are sent to the maintenance demand prediction module to update the online learning sample pool of the time series prediction model, so that the prediction model can continuously and adaptively optimize itself from the field data that has never been shut down.

[0047] As a further improvement of the present invention, the system further includes a strategy self-evolution module, which is used to adaptively adjust the weighted summation coefficient in the dynamic priority allocation algorithm according to the task execution status received by the execution control module and the actual occupancy time of the executed tasks in the resource scheduling module.

[0048] Specifically, the comprehensive priority value of completed tasks, actual dispatch waiting time, and actual repair completion time are obtained from the execution control module, and the sum of the actual dispatch waiting time and the actual repair completion time is taken as the actual total processing time.

[0049] According to the initial comprehensive priority of the completed tasks in descending order, calculate the Spearman correlation coefficient between the comprehensive priority ranking number of each completed task and the actual total processing time ranking number.

[0050] The Spearman correlation coefficient is compared with a preset target correlation coefficient range. If the Spearman correlation coefficient is lower than the lower limit of the target correlation coefficient range, the weight coefficient of the fault severity level score in the weighted summation function is increased by a preset step size, and the weight coefficient of spatial distance is decreased.

[0051] The weight coefficients in the adjusted weighted summation function are sent to the resource scheduling module to replace the original weighted summation coefficients, thereby ensuring that the task scheduling order is continuously aligned with the actual processing urgency, and realizing the self-evolution and optimal approximation of the scheduling strategy.

[0052] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0053] The data perception and prediction module integrates real-time operational status data, historical fault record data, and environmental parameter data. This data is then input into a time-series prediction model composed of convolutional layers, gated recurrent units, and fully connected layers. The model outputs the fault type probability for each terminal device unit over time, arranging them chronologically as a predicted fault probability sequence. Based on this, the task generation module automatically filters devices with predicted fault probabilities exceeding a preset fault threshold, marking them as target devices and generating initial maintenance tasks containing fault type estimates and recommended repair durations. This approach overcomes the limitation that threshold alarms can only trigger fault notifications after parameters exceed limits, shifting maintenance activities from passive response to proactive intervention based on equipment performance degradation trends. Task generation and preparation are completed before the actual fault occurs, thereby reducing fault duration and unplanned downtime losses.

[0054] After acquiring information on maintenance personnel skills, location, efficiency, and spare parts inventory, the resource scheduling module inputs a weighted summation function with the fault reproduction interval, fault severity score, theoretical occupancy time adjusted for personnel efficiency and equipment complexity, and spatial distance converted by the road network tortuosity coefficient. This calculates the comprehensive priority value for each initial maintenance task and sorts them from highest to lowest. Compared to fixed-rule or single-factor sorting methods, this comprehensive priority calculation incorporates dynamic information from multiple dimensions, including equipment fault urgency, maintenance resource consumption, and personnel arrival time, making the sorting results more closely reflect actual on-site constraints. During the task dispatch phase, the execution control module assigns tasks based on minimizing the total path distance from each maintenance personnel to the equipment installation location and spare parts storage location, combined with the comprehensive priority sorting. This achieves joint optimization between task urgency and personnel movement costs, reducing resource idleness and time delays caused by cross-regional spare parts requisition and unreasonable scheduling. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0056] Figure 1 This is a schematic diagram of the operation and maintenance management system for power engineering terminal equipment based on artificial intelligence.

[0057] Figure 2 This is a flowchart of power engineering terminal equipment fault prediction based on a time-series prediction model;

[0058] Figure 3 This is a flowchart of the operation and maintenance task generation process;

[0059] Figure 4 This is a flowchart of the strategy adaptive evolution process;

[0060] Figure 5 It is the probability density distribution of historical repair time and recommended repair time;

[0061] Figure 6 This is a comparison chart of the overall priority and fault severity levels of initial operation and maintenance tasks;

[0062] Figure 7 This is a statistical curve showing the total path distance of objects assigned by operations and maintenance personnel;

[0063] Figure 8 It is the curve showing the change of the Spearman correlation coefficient in the strategy self-evolution module with the cumulative number of completed tasks. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0065] See Figure 1This invention provides an artificial intelligence-based operation and maintenance management system for power engineering terminal equipment, including a data perception and prediction module, a task generation module, a resource scheduling module, and an execution control module. The data perception and prediction module collects real-time operating status, historical fault records, and environmental parameter data from multiple terminal devices, inputs them into a time-series prediction model, and generates a predicted fault probability sequence for each device within the next time window. The task generation module, based on the predicted fault probability sequence and a preset fault threshold, selects terminal devices with predicted fault probabilities exceeding the threshold as target devices, and generates an initial operation and maintenance task for each target device, including device identification, fault type estimation, and recommended maintenance duration. The resource scheduling module obtains currently available operation and maintenance personnel information and spare parts inventory information, and calculates the comprehensive priority of each initial operation and maintenance task using a dynamic priority allocation algorithm based on the initial operation and maintenance task, the operation and maintenance personnel information, and the spare parts inventory information. The initial operation and maintenance tasks are then sorted from high to low according to the comprehensive priority to obtain a sorted list of tasks to be executed. The execution control module dispatches each task in the list of tasks to be executed to the corresponding operation and maintenance personnel terminal according to the sorting order, and receives the task execution status feedback from the operation and maintenance personnel terminal in real time.

[0066] Example 1:

[0067] In specific implementation, please refer to Figure 2 The data sensing and prediction module collects real-time operating status data, historical fault record data, and environmental parameter data from multiple terminal devices at the power engineering site. The real-time operating status data includes multiple continuously monitored analog parameters such as current, voltage, active power, reactive power, and equipment temperature from the terminal devices. Each parameter is recorded at a fixed sampling frequency, forming a time-series data stream. The historical fault record data is stored in a fault management database, with each record containing a fault occurrence timestamp, fault device identifier, fault type code, and a description of the fault handling process. The environmental parameter data is collected by environmental sensors deployed at the power engineering site, including ambient temperature, ambient humidity, atmospheric pressure, and vibration intensity; this environmental parameter data also includes timestamp information.

[0068] The real-time operating status data of each terminal device is segmented using a sliding window to obtain multiple state feature vectors for consecutive time segments. In the specific implementation, the sliding window length parameter and the sliding step size parameter are preset. The sliding window length parameter is denoted as... , representing the number of consecutive sampling points contained in each time segment, with the sliding step parameter denoted as . This represents the number of sampling points between the start points of two adjacent time segments. Starting from the first sampling point of the real-time running status data, the number of sampling points is determined by the sliding step parameter. Move backward step by step, each time capturing a segment of the sliding window length. The continuous sampling point sequence, for each extracted time segment, expands the values ​​of all monitored parameters within the segment in a fixed order to form a one-dimensional state feature vector. When the total number of sampling points for real-time operating status data is insufficient to form a complete sliding window length parameter... When the time is reached, the sliding window segmentation operation is terminated. Through the above segmentation process, the dimension of each state feature vector obtained is equal to the number of monitored parameters multiplied by the sliding window length parameter. .

[0069] The historical fault record data of each terminal device is timestamped, and a historical state feature sequence within a preset time period before each historical fault occurs is extracted. This historical state feature sequence and its corresponding fault type label are then used as training samples. In specific implementation, each historical fault event is read sequentially from the historical fault record data to obtain the faulty device identifier and fault occurrence timestamp corresponding to the historical fault event. Based on the faulty device identifier, a preset time period parameter is extracted from the historical database, starting from the fault occurrence timestamp. Real-time operational status data and environmental parameter data within the specified range. Preset duration parameter. The value of is determined based on the operating cycle characteristics and fault development patterns of the terminal equipment, and is set to one-tenth of the average operating cycle of the terminal equipment. The captured historical real-time operating status data and historical environmental parameter data are processed using the same sliding window segmentation and environmental parameter concatenation method as the aforementioned real-time operating status data, generating a historical fused input feature sequence composed of multiple fused input feature vectors arranged in chronological order. This historical fused input feature sequence is associated with the fault type encoding corresponding to the current historical fault event, serving as a training sample. All historical fault events are traversed to construct a training sample set.

[0070] The environmental parameter data is concatenated with the real-time operating status data according to the timestamp to obtain the fused input feature for each time point. In specific implementation, for each sampling time point, the monitoring parameter value corresponding to that time point is extracted from the real-time operating status data to form a real-time state vector; the environmental parameter value corresponding to the same time point is extracted from the environmental parameter data to form an environmental state vector; the real-time state vector and the environmental state vector are concatenated sequentially to form the fused input feature vector for that time point. If the environmental parameter data for a certain time point is missing, it is filled using the environmental parameter data of the nearest previous time point. After performing the concatenation operation on all time points, a temporally continuous fused input feature sequence is obtained, which serves as the input to the time series prediction model.

[0071] The fused input features are sequentially passed through the convolutional layer, recurrent layer, and fully connected layer in the time-series prediction model to output the probability value of each terminal device belonging to different fault types in each time unit within a future preset time window. The probability values ​​of all time units are arranged in chronological order to form the predicted fault probability sequence.

[0072] In its implementation, the core architecture of the time series prediction model consists of convolutional layers, gated recurrent unit (ROU) layers, and fully connected layers connected sequentially. The model's input is a time series with a length equal to the preset number of input time steps. The vector at each time step in the sequence is the fused input feature vector, with dimension denoted as . Enter the number of time steps. Based on the time-dependent settings of terminal device status changes, it covers the typical cycle from normal state to fault occurrence.

[0073] The convolutional layer uses one-dimensional convolution operations, including There are 1 convolutional kernel, and the kernel size is 1. The convolutional stride is set to 1, and the activation function is a linear rectified function. The convolutional layer scans the input sequence along the time dimension, extracts feature patterns within each local time window, and outputs a feature map. The size of the feature map output by the convolutional layer is... That is, having There are time steps, each time step corresponding to a length of . eigenvectors.

[0074] The feature map output from the convolutional layer is unrolled along the time step and then input into the gated recurrent unit (ROU) recurrent layer. Unrolling along the time step means that each feature vector arranged along the time dimension in the feature map is sequentially used as the input to the GU recurrent layer at the corresponding time step. The hidden state dimension of the GU recurrent layer is set to... The internal calculations of the gated loop unit's loop layer include update gate and reset gate mechanisms, specifically implemented as follows: at time step... The gated loop unit receives the feature vector of the current time step. and the hidden state vector of the previous time step The hidden state vector at the current time step is calculated. The gated recurrent unit (ROU) recursively performs computations on all input time steps and outputs a sequence of hidden states consisting of the hidden state vectors at each time step.

[0075] A fully connected layer contains a fully connected weight matrix. and bias vector The fully connected layer maps the hidden state vector output by the gated recurrent unit layer at the last time step to a dimension equal to the total number of fault categories. The original score vector. Total number of fault categories. This includes all possible fault types and a category representing the normal state. Then, a softmax function is used to transform the original score vector into a probability distribution, calculated as follows:

[0076]

[0077] in, This indicates that, given the original score vector Predicting categories under the condition For the first The probability of a class; This is the original score vector output by the fully connected layer. The first in element Corresponding to the The original scores for each fault category; This is a category index, with values ​​ranging from 1 to the total number of fault categories. Integers; Represented by natural constant An exponential function with base 0; This is a temporary index variable used to iterate through all categories during the summation operation.

[0078] For each time unit within the preset future time window, a probability distribution vector is obtained using the above method. The dimension of this probability distribution vector is the total number of fault categories. Each element in the vector represents the probability value of the terminal device belonging to the corresponding fault category within that time unit. Arranging the probability distribution vectors of all time units within a future preset time window in chronological order yields the predicted fault probability sequence.

[0079] The training process of the time-series prediction model is as follows: Using the constructed training sample set, the historical fused input feature sequence in each training sample is used as the model input, and the corresponding fault category label is used as the supervision signal. For each training sample, the last input time step in the historical fused input feature sequence is taken. A fused input feature vector is input into the time series prediction model, and the model outputs the probability distribution of each time unit within a future preset time window. During the training phase, the target probability distribution for the time unit within the future preset time window that aligns with the historical fault occurrence timestamp is set as follows: the probability value for the corresponding fault category is 1, and the probability value for other categories is 0. For other time units not aligned with the fault occurrence timestamp, the target probability distribution is set as follows: the probability value for the normal state category is 1, and the probability value for other categories is 0. The cross-entropy loss between the model's output probability distribution and the target probability distribution is calculated, and the parameters of the convolutional layers, gated recurrent units, and fully connected layers in the time series prediction model are updated using an adaptive moment estimation optimizer. Multiple iterations are performed on all training samples until the cross-entropy loss value converges, and the model parameters obtained at convergence are saved.

[0080] In the actual prediction phase, for each terminal device, the length of the most recently collected data is taken as the input time step. The fused input feature sequence is used as the model input, and the trained time series prediction model is used to calculate the predicted failure probability sequence of the terminal device within a preset time window in the future.

[0081] Example 2:

[0082] In specific implementation, please refer to Figure 3 The task generation module obtains the predicted fault probability sequence for each terminal device from the data perception and prediction module. The data structure of the predicted fault probability sequence is a two-dimensional array. The first dimension of the array corresponds to each time unit within the preset future time window, and the second dimension corresponds to the total number of fault categories. Each element is a probability value.

[0083] The task generation module extracts the maximum probability value and the corresponding fault type for each terminal device from the predicted fault probability sequence. For a single terminal device, it iterates through the probability values ​​of all time units and all fault categories in the predicted fault probability sequence, determining the maximum probability value in the entire sequence through comparisons. Simultaneously, it records the fault category index that generates the maximum probability value and, based on a pre-built mapping table between fault category indices and fault type names, converts this index into a corresponding fault type name string, which serves as the fault type corresponding to the maximum probability value.

[0084] The task generation module compares the maximum probability value with a preset fault threshold. If the maximum probability value is greater than the fault threshold, the corresponding terminal device is marked as a target device, and the fault type corresponding to the maximum probability value is used as the fault type prediction for that target device. The comparison operation uses a direct numerical comparison method. When the maximum probability value is strictly greater than the fault threshold, the terminal device is added to the target device set, and the fault type name string obtained from the mapping table is assigned to the fault type prediction field of the target device. If the maximum probability value is less than or equal to the fault threshold, no task generation processing is performed on that terminal device.

[0085] For each target device, the task generation module queries the historical fault record data for the historical average repair time corresponding to the estimated fault type. Each fault record in the historical fault record data contains a fault type name string and a repair time value, which represents the number of hours elapsed from the start to the completion of the repair. The query process is as follows: using the estimated fault type name string as the query condition, all records in the historical fault record data are filtered to obtain all historical fault records with the same fault type name string; then, the repair time values ​​from all filtered records are extracted, and the arithmetic mean of these repair time values ​​is calculated to obtain the historical average repair time.

[0086] The task generation module multiplies the historical average repair time by a preset margin coefficient to obtain the recommended repair time. The calculation method is as follows:

[0087]

[0088] in, Indicates the recommended repair time; This represents the historical average maintenance time, obtained from the aforementioned query and calculation. This represents the preset margin factor. The value is set to 1.3. The basis for this value is: statistically analyzing the distribution of maintenance time values ​​in historical fault record data, calculating the coefficient of variation of maintenance time values ​​under each fault type, that is, the standard deviation of maintenance time values ​​divided by the corresponding historical average maintenance time, taking the 95th percentile of the coefficient of variation of all fault types, and then adding 1 to this percentile to get 1.3, so that the recommended maintenance time can cover the extended maintenance time caused by fluctuations in on-site conditions in most cases.

[0089] The task generation module combines the target device's device identifier, estimated fault type, and recommended repair time into a structured data object, which serves as the initial maintenance task. The device identifier is a unique string code retrieved from the terminal device management database, using the terminal device's asset number. Within the structured data object, the device identifier is stored as a string field, the estimated fault type as a string field, and the recommended repair time as a floating-point number in hours. These three fields are packaged into a single data record, and all initial maintenance tasks constitute an initial maintenance task list, which is then passed to the resource scheduling module.

[0090] In some embodiments, the preset fault threshold is dynamically adjusted based on the importance level of each terminal device; the higher the importance level of the terminal device, the lower the corresponding fault threshold. In specific implementations, the importance level of the terminal device is predefined in the device ledger database, with three levels: core device, important device, and general device, represented by values ​​3, 2, and 1 respectively. A higher value indicates a higher degree of importance. The dynamic adjustment of the preset fault threshold is achieved by setting a baseline fault threshold. Baseline fault threshold The value is set to 0.7 according to the operation and maintenance management strategy. Then, it is determined based on the importance level of the terminal device. Calculate the fault threshold corresponding to the terminal device. :

[0091]

[0092] in, This indicates the fault threshold corresponding to the terminal device. This represents the baseline fault threshold, with a value of 0.7. This indicates the importance level of the terminal device, with values ​​of 3, 2, and 1. When... hour, Approximately 0.233; when hour, It is 0.35; when hour, The threshold value is 0.7. The higher the importance level, the smaller the calculated fault threshold value. This makes core equipment more likely to be selected as the target equipment even if the predicted failure probability is low, thus prioritizing the operational status of highly important equipment. When comparing the maximum probability value with the fault threshold, the task generation module calculates the corresponding fault threshold for each terminal device based on its importance level value, and then uses this fault threshold for subsequent comparison operations.

[0093] See Figure 5In the graph, the horizontal axis represents maintenance time in hours, ranging from 0 to 40 hours, and the vertical axis represents probability density, indicating the frequency of maintenance tasks within the corresponding maintenance time interval. The solid curve represents the historical maintenance time distribution, and the dashed curve represents the recommended maintenance time distribution. Both curves exhibit a unimodal shape, with the peak values ​​concentrated in shorter maintenance time intervals.

[0094] Specifically, the peak of the historical maintenance time distribution is located at approximately 4 hours, with a peak probability density of about 0.148, indicating that most maintenance tasks in actual operations have a duration concentrated between 3 and 5 hours, showing a relatively concentrated maintenance duration characteristic. As the maintenance duration increases, the historical maintenance time probability density gradually decreases, and becomes significantly flat after 10 hours. The probability of long-duration maintenance tasks exceeding 30 hours is extremely low, approaching zero.

[0095] The peak of the recommended maintenance time distribution is located at approximately 5 hours, corresponding to a probability density of about 0.115, slightly shifted to the right compared to the historical maintenance time peak. This reflects that the recommended maintenance time was designed with an appropriate margin adjustment for the average maintenance time. The overall recommended maintenance time curve is smoother than the historical maintenance time curve, extending to a longer maintenance time range, indicating that the recommended maintenance time covers common maintenance times while also considering the possibility of extended maintenance times. Although the probability density for recommended maintenance times exceeding 20 hours is low, it is still higher than the historical maintenance time distribution, demonstrating the role of the margin coefficient in extending the maintenance time.

[0096] Overall, the recommended maintenance time distribution shows a reasonable offset and expansion compared to the historical maintenance time distribution, covering most of the historical maintenance time range. At the same time, it provides appropriate time margin for maintenance tasks and supports the technical solution in the task generation module to calculate the recommended maintenance time based on the historical average maintenance time multiplied by a margin coefficient of 1.3, effectively balancing maintenance efficiency and on-site uncertainties.

[0097] Example 3:

[0098] In practice, the resource scheduling module obtains information on currently available operations and maintenance (O&M) personnel. Internally, the resource scheduling module maintains a communication interface that exchanges data with the O&M personnel management server via encrypted Hypertext Transfer Protocol (HTTP). The resource scheduling module constructs an availability query request message, encoded in JSON format, containing a request timestamp and the resource scheduling module's identifier. The resource scheduling module then sends the availability query request message to the O&M personnel management server through the communication interface and awaits a response from the O&M personnel management server.

[0099] After receiving an availability query request, the operations and maintenance (O&M) personnel management server queries its internally stored O&M personnel status table, filtering out records of O&M personnel currently in an idle state, and records of O&M personnel currently in a busy state whose estimated remaining completion time for their tasks is less than or equal to a preset time threshold of 30 minutes. The O&M personnel management server assembles the filtered O&M personnel records into an O&M personnel list. Each O&M personnel record in the list contains four data items: personnel identifier, skill tag set, current location coordinates, and average maintenance efficiency. The personnel identifier uses the O&M personnel's employee number and is stored as a string. The skill tag set consists of multiple skill tags, each a string describing the maintenance skills possessed by the O&M personnel, such as relay protection maintenance, switchgear maintenance, transformer maintenance, secondary circuit maintenance, etc. The current location coordinates are represented as latitude and longitude coordinate pairs, with both longitude and latitude stored as floating-point numbers in degrees, with precision retained to six decimal places. Average maintenance efficiency is expressed as a dimensionless ratio, calculated as the average of the ratio of actual working hours to standard working hours for maintenance tasks historically completed by maintenance personnel. A value closer to 1 indicates maintenance efficiency closer to the standard level, while a value less than 1 indicates maintenance efficiency higher than the standard level. The maintenance personnel management server encapsulates the maintenance personnel list into a response message and returns it to the resource scheduling module via the communication interface. The resource scheduling module receives and parses the response message, storing the maintenance personnel list in the currently available maintenance personnel cache.

[0100] The resource scheduling module obtains currently available spare parts inventory information. It constructs an inventory query request message containing a request timestamp, a resource scheduling module identifier, and a query scope identifier set to all spare parts. The resource scheduling module sends this message to the spare parts warehouse management system. Upon receiving the message, the system queries the spare parts inventory database to retrieve inventory information for all spare parts currently in the warehouse. The system organizes this information into a list of spare parts inventory records. Each record contains a spare parts identifier, spare parts type, storage location, and inventory quantity. The spare parts identifier is a unique code for the spare parts, using an alphanumeric string. The spare parts type is a string describing the functional category of the spare parts, sharing the same type coding system as the fault type name string in fault type estimation. The storage location is represented by a string combining the warehouse number and shelf number, such as Warehouse A-03-12. The inventory quantity is an integer representing the actual number of spare parts currently stored in the warehouse. The spare parts warehouse management system encapsulates the list of spare parts inventory records into a response message and returns it to the resource scheduling module. The resource scheduling module receives the response message and stores it in the current spare parts inventory cache.

[0101] The resource scheduling module estimates the fault type in each initial maintenance task, filters the spare parts inventory information to find spare parts that match the estimated fault type, and checks if the inventory quantity of the spare parts is greater than zero. If it is, the corresponding spare parts identifier is associated with the initial maintenance task. In practice, the resource scheduling module iterates through the initial maintenance task list. For each initial maintenance task, it extracts the estimated fault type string and uses it as a query condition to search the current spare parts inventory cache for spare parts inventory records where the spare parts type string is exactly the same as the estimated fault type string. If a matching spare parts inventory record exists, it checks the value of the inventory quantity field in that record. If the inventory quantity is greater than zero, it retrieves the spare parts identifier field from that record and appends it to the structured data object of the initial maintenance task as a new spare parts identifier list field. If the inventory quantity is zero, or no matching spare parts inventory record is found, the spare parts identifier list field remains empty. After the traversal is completed, each initial maintenance task carries available spare parts identification information corresponding to the fault type prediction.

[0102] The resource scheduling module uses a dynamic priority allocation algorithm to calculate the overall priority of each initial maintenance task. In practice, the dynamic priority allocation algorithm calculates four input parameters in sequence: fault reproduction interval, fault severity score, theoretical occupancy time, and spatial distance. Then, it inputs these four input parameters into a weighted summation function and outputs the overall priority value of each initial maintenance task.

[0103] The resource scheduling module extracts the device identifier from each initial maintenance task and queries historical fault record data based on the device identifier to obtain the fault reproduction interval for the corresponding terminal device. The query process is as follows: using the device identifier as the query condition, the fault device identifier field of all fault records in the historical fault record data is matched to filter out all historical fault records for that terminal device. The filtered historical fault records are sorted in ascending order of fault occurrence timestamp, and the time interval between two adjacent fault occurrence timestamps is calculated in hours. The arithmetic mean of all calculated adjacent fault time intervals is obtained to obtain the fault reproduction interval. If the number of historical fault records for a terminal device is less than two, and the adjacent fault time interval cannot be calculated, the fault reproduction interval is set to a preset default value of 8760 hours, corresponding to the number of natural hours in a year.

[0104] The resource scheduling module extracts the fault type estimate from each initial maintenance task and matches it with the corresponding fault severity score from a pre-defined fault level mapping table. The pre-defined fault level mapping table is a key-value pair data structure, where the key is the fault type name string and the value is the fault severity score, which is an integer. The fault level mapping table is constructed based on the degree of threat the fault poses to the power system's operational safety and the potential load loss range, classifying fault types into three levels: Level 1 fault types that cause main equipment shutdown or protection malfunctions, with a corresponding fault severity score of 10; Level 2 fault types that cause equipment performance degradation but do not shut down, with a corresponding fault severity score of 6; and Level 3 fault types that only affect monitoring or auxiliary functions, with a corresponding fault severity score of 3. The matching process involves using the fault type estimate string as the key to perform an exact search in the fault level mapping table, retrieving the matched integer value as the fault severity score.

[0105] The resource scheduling module calculates the theoretical duration of each initial maintenance task based on the recommended maintenance duration and the average maintenance efficiency from the currently available maintenance personnel information. In practice, the resource scheduling module extracts the average maintenance efficiency values ​​of all available maintenance personnel from the cache, calculates the arithmetic mean of these average maintenance efficiency values, and obtains the mean average maintenance efficiency, denoted as . Then, the average maintenance efficiency was calculated. The reciprocal of the number of hours required for unit fault repair is denoted as . Standard working hours It indicates the time required to complete one unit of maintenance work at a standard efficiency level.

[0106] The resource scheduling module extracts the recommended maintenance duration from the structured data object of each initial maintenance task. Recommended repair time Multiply by standard working hours Obtain the baseline usage time Baseline usage time This indicates the estimated time to complete a maintenance task under standard efficiency levels and standard equipment complexity.

[0107] The resource scheduling module obtains the device complexity level of the terminal device corresponding to each initial maintenance task. Device complexity levels are pre-stored in the device ledger database and are divided into three levels: high complexity, medium complexity, and low complexity, represented by the values ​​3, 2, and 1, respectively. The resource scheduling module uses the device identifier as the query condition to read the device complexity level value from the device ledger database. Then, the resource scheduling module queries a preset complexity coefficient mapping table to find the complexity adjustment coefficient corresponding to the device complexity level. The preset complexity coefficient mapping table is a mapping relationship where the complexity adjustment coefficient for high complexity devices is set to 1.5, for medium complexity devices to 1.0, and for low complexity devices to 0.7. These coefficients are set based on the statistical multiple relationship of historical maintenance time for devices of different complexity levels: the median ratio of the historical maintenance time of all high complexity devices to that of medium complexity devices is calculated, resulting in 1.5; the median ratio of the historical maintenance time of all low complexity devices to that of medium complexity devices is calculated, resulting in 0.7. The resource scheduling module will use the baseline occupancy time. Multiply by the complexity adjustment factor to obtain the theoretical time taken. :

[0108]

[0109] in, This indicates the theoretical duration, in hours. The recommended repair duration is provided by the task generation module and is in hours. This represents the average maintenance efficiency, which is obtained by calculating the arithmetic mean of the average maintenance efficiency of all maintenance personnel in the currently available maintenance personnel information. It is a dimensionless ratio. This represents the complexity adjustment factor, which is obtained from a preset complexity factor mapping table based on the device complexity level of the terminal device. It is 1.5 for high complexity, 1.0 for medium complexity, and 0.7 for low complexity.

[0110] The resource scheduling module obtains the geographical location of the terminal device corresponding to each initial maintenance task and calculates the spatial distance between the geographical location and the current location of each maintenance personnel in the currently available maintenance personnel information. In specific implementation, the resource scheduling module queries the device ledger database for the latitude and longitude coordinates of the installation location corresponding to the device identifier of each initial maintenance task, using this as the first coordinate. The device ledger database stores the installation location information of each terminal device, including longitude and latitude values. The resource scheduling module uses the device identifier as the query condition to obtain the longitude and latitude values, forming the first coordinate.

[0111] The resource scheduling module extracts the current location latitude and longitude coordinates of each available operations and maintenance (O&M) personnel from the cache of currently available O&M personnel, using them as the second coordinate. The current location latitude and longitude coordinates are derived from the current location coordinates field in the O&M personnel record returned by the O&M personnel management server.

[0112] For each initial maintenance task and each maintenance personnel combination, the resource scheduling module converts the first and second coordinates into projected plane coordinates under the same spatial reference system. The spatial reference system adopts the universal transverse Mercator projection coordinate system. The conversion process includes: determining the universal transverse Mercator projection zone number based on the longitude values ​​of the first and second coordinates; then, using the universal transverse Mercator projection forward calculation formula, converting the latitude and longitude coordinates into east and north coordinate values ​​in a plane rectangular coordinate system, both in meters. The first coordinate transformation yields the first projected plane coordinates, and the second coordinate transformation yields the second projected plane coordinates.

[0113] The resource scheduling module calculates the Euclidean distance between the coordinates of the projection planes. Let the coordinates of the first projection plane be... The coordinates of the second projection plane are Euclidean distance value The calculation method is as follows: The results are in meters. Then, the resource scheduling module multiplies the Euclidean distance value by a preset road network tortuosity coefficient to obtain the spatial distance corresponding to each combination. The preset road network tortuosity coefficient is set to 1.4. This coefficient is set based on: sampling and calculating the actual road distance and straight-line Euclidean distance between major power plants at the power engineering site, collecting 50 sets of samples, calculating the ratio of the actual road distance to the straight-line Euclidean distance in each set of samples, and taking the arithmetic mean of all ratios, resulting in 1.4. Spatial distance The calculation method is as follows .

[0114] The resource scheduling module inputs the fault reproduction interval, fault severity score, theoretical occupancy time, and spatial distance into a weighted summation function, outputting a comprehensive priority value for each initial maintenance task. The weighted summation function is formed by multiplying each of the four input parameters by its corresponding weight coefficient and then summing the results. The weight coefficient for the fault reproduction interval is set to a negative value, ensuring that shorter intervals result in higher overall priority values. Let the fault reproduction interval be denoted as... The severity level of the fault is: The theoretical time required is The spatial distance is The corresponding weighting coefficients are denoted as follows: , , and Overall priority value The calculation formula is:

[0115]

[0116] in, This represents the overall priority value; a larger value indicates a higher priority. The weighting coefficient for the fault reproduction interval is set to -0.1. A negative value means that the shorter the fault reproduction interval, the larger the product term value, and thus the higher the overall priority value. The weighting coefficient for the severity level score of the fault is set to 2.0; The weighting coefficient for the theoretical duration is set to -0.5. A negative value means that the shorter the theoretical duration, the larger the product term value, and the higher the overall priority. The weighting coefficient for spatial distance is set to -0.01. A negative value means that the shorter the spatial distance, the larger the product term value, and the higher the overall priority. The numerical setting of the above weighting coefficient is determined based on the analytic hierarchy process (AHP): construct pairwise comparison judgment matrices of the relative importance of the four input parameters, calculate the eigenvectors of the judgment matrices, and use each component of the eigenvectors as the baseline value of the weighting coefficients. Then, based on the negative correlation between the three factors—fault recurrence interval, theoretical occupancy time, and spatial distance—and priority, reverse the sign of these three weighting coefficients to obtain the final weighting coefficient values.

[0117] The resource scheduling module sorts the initial maintenance tasks according to their overall priority values ​​from highest to lowest, resulting in a sorted list of tasks to be executed. The sorting method uses numerical comparison, with the initial maintenance task having the highest overall priority value placed at the top, and so on down, forming the list of tasks to be executed. Each element in the list still retains the complete structured data object of the initial maintenance task.

[0118] See Figure 6 The horizontal axis of the graph represents the initial sequence number of the maintenance tasks, totaling 500 tasks, with the sequence number increasing from 0 to 500; the vertical axis represents the priority value or normalized score, used to display the numerical values ​​based on different sorting criteria. The graph contains three curves, corresponding to "comprehensive priority sorting", "fault severity score sorting", and "theoretical occupancy time sorting", respectively.

[0119] The solid curve represents the comprehensive priority ranking calculated based on the dynamic priority allocation algorithm in the resource scheduling module. Its priority value shows a trend of gradually decreasing from about -20 to about -130, and the overall trend is smooth and declining. This indicates that the comprehensive priority value has a high degree of differentiation for tasks and can prioritize tasks relatively evenly, avoiding large-scale duplication of values. It reflects the comprehensive effect of the weighted sum of four factors: fault recurrence interval, fault severity level, theoretical occupancy time, and spatial distance.

[0120] The dashed curve represents the ranking based on the severity score, with the vertical axis representing the normalized result of the corresponding score. The curve exhibits a clear discrete distribution, with the values ​​concentrated in three discrete level intervals, reflecting the discrete characteristics of the three fixed scores (10, 6, 3) in the severity level mapping table. The curve shows a large number of jumps, indicating that when ranking solely based on the severity score, the task priority differentiation is low and cannot distinguish task priorities in detail.

[0121] The dotted-line curve corresponds to the normalized score of the theoretical time taken for ranking. Overall, it shows a small fluctuation, ranging from 10 to 30, and there is no obvious increasing or decreasing trend. This shows that the theoretical time taken for ranking contributes relatively evenly and stably to the ranking of tasks. However, it is difficult to form a strong distinguishing ranking by using this indicator alone.

[0122] Overall, the comparison of the three curves shown in the figure verifies that the dynamic priority allocation algorithm adopted by the resource scheduling module in this embodiment significantly improves the granularity and effectiveness of the initial operation and maintenance task priority ranking through multi-dimensional weighted comprehensive calculation. This is beneficial for the subsequent execution control module to reasonably dispatch tasks based on the ranking results, thereby achieving optimized allocation of operation and maintenance resources and improved scheduling efficiency.

[0123] Example 4:

[0124] In practice, the execution control module receives a sorted list of tasks to be executed from the resource scheduling module. This list contains multiple initial maintenance tasks, each carrying complete information such as device identifier, estimated fault type, recommended repair duration, and a list of associated spare parts. Internally, the execution control module maintains a dispatch loop that processes tasks in the list one by one according to their overall priority from highest to lowest, until all tasks in the list have been dispatched.

[0125] The execution control module retrieves the current task with the highest overall priority from the list of tasks to be executed. The retrieval process involves reading the first element of the list and designating it as the current task, while maintaining the original order of the remaining elements. Based on the device identifier and fault type prediction within the current task, the execution control module extracts the corresponding spare part storage location from the spare part identifier associated with the resource scheduling module. Specifically, the execution control module retrieves the spare part identifier list field from the structured data object of the current task. If the spare part identifier list field is not empty, it retrieves the first spare part identifier from the list. Using the retrieved spare part identifier as a query condition, the execution control module sends a single spare part query request to the spare part warehouse management system. The spare part warehouse management system returns the string representing the spare part storage location corresponding to that identifier. The execution control module stores the returned spare part storage location string in a temporary variable. If the spare part identifier list field is empty, the execution control module sets the spare part storage location string to an empty value.

[0126] The execution control module obtains information on all currently available maintenance personnel. This information comes from the maintenance personnel list transmitted in real-time by the resource scheduling module at the time of dispatch. Each maintenance personnel record in the list includes a personnel identifier, a set of skill tags, current location coordinates, and average maintenance efficiency. The execution control module iterates through each maintenance personnel in the list, calculates the first path distance from each maintenance personnel's current location to the equipment installation location in the current task, and the second path distance from each maintenance personnel's current location to the spare parts storage location. The sum of the first and second path distances is taken as the total path distance for each maintenance personnel.

[0127] The process of calculating the first path distance is as follows: The execution control module retrieves the device identifier from the structured data object of the current task. Using the device identifier as a query condition, it queries the latitude and longitude coordinates of the device installation location in the current task from the device ledger database, recording them as the device longitude and latitude. The execution control module retrieves the current location coordinates from the currently traversed maintenance personnel records, recording them as the maintenance personnel longitude and latitude. The execution control module converts the device latitude and longitude coordinates and the maintenance personnel latitude and longitude coordinates into projection plane coordinates under the same spatial reference system. The spatial reference system adopts the universal transverse Mercator projection coordinate system. The conversion process is as follows: the universal transverse Mercator projection zone number is determined according to the device longitude value and the maintenance personnel longitude value. If the two longitude values ​​are in the same projection zone, the same central meridian parameter is used for projection conversion; if the two longitude values ​​are in different projection zones, the projection zone corresponding to the maintenance personnel longitude value is uniformly converted to the projection zone corresponding to the device longitude value. The general transverse Mercator projection formula is used to convert the equipment's latitude and longitude coordinates into east-facing and north-facing coordinates on the equipment plane, and the maintenance personnel's latitude and longitude coordinates into east-facing and north-facing coordinates on the maintenance personnel plane. The east-facing and north-facing coordinates of the equipment plane constitute the equipment's projected plane coordinates, and the east-facing and north-facing coordinates of the maintenance personnel plane constitute the maintenance personnel's projected plane coordinates. The execution control module calculates the Euclidean distance between the equipment's projected plane coordinates and the maintenance personnel's projected plane coordinates, denoted as [Equation missing]. The coordinates of the projected plane for maintenance personnel are Euclidean distance value The calculation method is as follows:

[0128]

[0129] in, This represents the straight-line Euclidean distance between the equipment installation location and the current location of the maintenance personnel, in meters. This represents the eastward coordinate value of the device's projected plane coordinates, in meters; This represents the north coordinate value of the device's projected plane coordinates, in meters; This represents the eastward coordinate value of the projected plane coordinates of the maintenance personnel, in meters; This represents the north coordinate value of the projected plane coordinates of the maintenance personnel, in meters. The execution control module will then convert the Euclidean distance value... Multiply by the preset road network tortuosity coefficient to obtain the first path distance. The preset road network tortuosity coefficient is set to 1.4. The basis for this coefficient setting is as follows: sampling the actual road length and corresponding straight-line Euclidean distance between the main power plants and substations at the power engineering site, collecting 50 pairs of data, calculating the ratio of the actual road length to the straight-line Euclidean distance in each pair of data, and taking the arithmetic mean of the 50 ratios to obtain 1.4.

[0130] The process of calculating the second path distance is as follows: If the spare parts storage location string is empty, the second path distance is directly set to zero. If the spare parts storage location string is not empty, the execution control module uses the spare parts storage location string as the query condition to query the latitude and longitude coordinates corresponding to the spare parts storage location from the spare parts warehouse management system, which are recorded as the spare parts longitude and latitude. The execution control module converts the spare parts latitude and longitude coordinates into the spare parts plane east-west coordinates and spare parts plane north-north coordinates in the universal transverse Mercator projection coordinate system, using the same projection zone parameters as the equipment projection plane coordinates. The execution control module calculates the Euclidean distance between the maintenance personnel projection plane coordinates and the spare parts projection plane coordinates, and multiplies this Euclidean distance by a preset road network tortuosity coefficient of 1.4 to obtain the second path distance.

[0131] The execution control module adds the first path distance to the second path distance to obtain the total path distance for each operations and maintenance personnel, with the unit being meters. After completing the calculation for all operations and maintenance personnel, the execution control module obtains a set of total path distances, where each element contains the personnel identifier and the corresponding total path distance value.

[0132] The execution control module selects the maintenance personnel with the smallest total path distance as the assignment target for the current task. In practice, the execution control module iterates through the set of total path distances, finding the element with the smallest total path distance by comparing each element, extracting the personnel identifier from that element, and identifying the maintenance personnel corresponding to that identifier as the assignment target. If multiple maintenance personnel have the same minimum total path distance, the maintenance personnel with the average maintenance efficiency value closest to 1 is selected as the assignment target.

[0133] The execution control module encapsulates the current task into a dispatch instruction and sends it to the terminal of the corresponding maintenance personnel. The dispatch instruction is encoded in JSON format and includes the following fields: task identifier, device identifier, device installation location, estimated fault type, recommended repair duration, spare parts identifier list, and spare parts storage location. The task identifier is automatically generated by the execution control module during dispatch. It is generated by concatenating the current system timestamp with the device identifier, performing a hash operation, and using the first twelve characters of the hash value as the task identifier. The execution control module pushes the dispatch instruction to the maintenance personnel's terminal via a mobile communication network. The maintenance personnel's terminal has a maintenance task receiving application installed. After receiving the dispatch instruction, the application parses the JSON content and displays the task details on the interface, while simultaneously returning a confirmation message to the execution control module.

[0134] The execution control module removes the currently assigned task from the list of tasks to be executed. The removal operation is performed by deleting the element with index zero in the list of tasks to be executed, and shifting the index of all remaining elements one position forward.

[0135] The execution control module updates the status of assigned objects in the currently available operations and maintenance (O&M) personnel information to an occupied status. The update method is as follows: In the list of currently available O&M personnel, find the O&M personnel record whose personnel ID matches the personnel ID of the assigned object. Change the status field of that O&M personnel record from idle or soon to idle to occupied, and record the occupied task ID. The updated currently available O&M personnel information will be used for the calculation of the next round of task dispatch.

[0136] The execution control module checks whether the list of tasks to be executed is empty. If it is not empty, it continues to retrieve the current task with the highest overall priority and repeats the above dispatch process. If it is empty, it ends the task dispatch loop for this batch.

[0137] In some embodiments, the execution control module receives task execution status feedback from the operation and maintenance personnel's terminals in real time. Specifically, after receiving a dispatch instruction, the operation and maintenance personnel trigger a status change through an application, and the terminal generates a status update message in real time and sends it to the execution control module. The execution control module internally runs a status monitoring service that continuously listens for status update messages from each operation and maintenance personnel's terminal.

[0138] The execution control module receives status update messages sent from the maintenance personnel's terminal. These messages are encoded in JSON format and contain four fields: personnel identifier, task identifier, current status code, and status occurrence timestamp. The personnel identifier field contains the employee ID of the maintenance personnel performing the operation. The task identifier field contains the task identifier generated by the execution control module when the task is dispatched. The current status code field is a one-digit integer representing the task execution progress. The predefined mapping is: 0 indicates order accepted, 1 indicates dispatched, 2 indicates spare parts retrieved, 3 indicates arrived at the site, 4 indicates under repair, and 5 indicates completed. The status occurrence timestamp field contains the local time, accurate to seconds, when the status change was triggered by the maintenance personnel's terminal.

[0139] The execution control module determines the task execution progress based on the current status code. It parses the value of the current status code field in the status update message, matches the value with a predefined encoding mapping relationship, and obtains the current task execution progress stage. The execution control module stores all information from the status update message in the task execution status log database for subsequent querying and statistical analysis.

[0140] When the execution control module receives a completion status code (current status code 5), it retrieves the actual repair time and post-repair equipment operating parameters from the maintenance personnel's terminal. In practice, after the maintenance personnel's terminal sends a status update message with a current status code of 5, the application on the terminal displays a repair data entry interface, requiring the maintenance personnel to input the actual repair time and post-repair equipment operating parameters. The actual repair time is the actual time taken by the maintenance personnel from the start of the repair operation to its completion, in hours, rounded to one decimal place. The post-repair equipment operating parameters are a set of real-time operating status data collected during the first power-on of the terminal equipment after repair, including current, voltage, active power, reactive power, and equipment temperature. The maintenance personnel's terminal encapsulates the actual repair time and post-repair equipment operating parameters into a repair completion data message and sends it to the execution control module. The repair completion data message also includes the task identifier and personnel identifier for the corresponding task.

[0141] The execution control module compares the actual repair time with the recommended repair time in the corresponding initial maintenance task, generating a repair efficiency record and storing it in the historical maintenance database. The comparison method is as follows: read the recommended repair time from the initial maintenance task, divide the actual repair time by the recommended repair time, and obtain a ratio of actual time to recommended time. The execution control module combines this ratio, task identifier, personnel identifier, and comparison timestamp into a repair efficiency record and writes it to the historical maintenance database. The historical maintenance database stores the repair efficiency records for each repair task, used for subsequent evaluation of maintenance personnel's repair efficiency and calibration of recommended repair time calculation parameters.

[0142] The execution control module sends the post-maintenance equipment operating parameters to the data perception and prediction module to update the online learning sample pool of the time-series prediction model. The online learning sample pool is a circular buffer that stores the sequence of operating parameters for the terminal equipment that recently completed maintenance. Upon receiving the post-maintenance equipment operating parameters, the data perception and prediction module concatenates them with the operating parameters continuously collected after the maintenance completion time, forming a new sample of the equipment's full lifecycle state sequence. When the number of samples in the online learning sample pool reaches a preset batch update threshold, the data perception and prediction module uses these samples to incrementally train the time-series prediction model, updating the convolutional kernel weights of the convolutional layers, the gating matrix parameters of the gated recurrent unit layers, and the connection weights of the fully connected layers, enabling the time-series prediction model to continuously adapt to the changing trends of the equipment's operating state.

[0143] See Figure 7In the graph, the horizontal axis represents the task number, and the vertical axis represents the total path distance, in meters. The graph contains three curves: the total path distance of the assigned object (solid line), the average total path distance (dashed line), and the maximum total path distance (dotted line). This graph reflects the spatial distance distribution between maintenance personnel and the locations of equipment and spare parts when the execution control module calculates and dispatches tasks based on the list of tasks to be executed after comprehensive priority sorting in Example 4.

[0144] The solid line represents the total path distance for each assigned maintenance personnel, which is the sum of the path distances from the maintenance personnel's current location to the equipment installation location and spare parts storage location. This curve fluctuates significantly throughout the task sequence, but overall shows a trend that is significantly lower than the dashed and dotted lines. This indicates that the execution control module prioritizes assigning tasks to maintenance personnel with shorter total path distances, thereby optimizing the dispatch path and reducing the travel burden and response time for maintenance personnel.

[0145] The dashed line represents the average total path distance for all currently available maintenance personnel, ranging from approximately 1000 to 2200 meters. This reflects the average spatial distribution distance from maintenance personnel to the locations of various task devices and spare parts. The dashed line exhibits greater fluctuations than the solid line, and its values ​​are generally higher, indicating that the path distance of assigned objects is typically significantly lower than the average level. This demonstrates the effective consideration of spatial distance factors by the execution control module during scheduling.

[0146] The dotted line represents the maximum total path distance, with values ​​mostly concentrated in the 2500-3000 meter range. This shows the group of path distances furthest from the equipment and spare parts locations among all available maintenance personnel. The dotted line curve fluctuates dramatically, with values ​​much higher than the solid and dashed lines, highlighting that the system avoids selecting these remote maintenance personnel during task dispatch to ensure the timeliness and efficiency of task execution.

[0147] Overall, as the task number increases, all three curves show a relatively stable fluctuation trend, without significant unilateral increases or decreases. This indicates that the system's task scheduling maintains a balanced distribution of path distances among different tasks, avoiding significant path extremes. The execution control module effectively reduces the waste of operational resources and improves task response speed by calculating the total path distance in real time and assigning tasks to the maintenance personnel with the shortest path distance. This meets the technical requirements of the execution control module in Example 4 for calculating and optimizing task dispatch path distances.

[0148] Example 5:

[0149] In specific implementation, please refer to Figure 4The strategy self-evolution module interacts with the execution control module and resource scheduling module as an independent module. The trigger condition for the strategy self-evolution module is set to automatically initiate a strategy evolution process when the cumulative number of completed tasks in the system reaches a preset evaluation period threshold. The preset evaluation period threshold is set to complete twenty tasks. When the count of completed tasks reaches twenty, the strategy self-evolution module starts, and after completing the strategy evolution, it resets the count of completed tasks to zero and restarts the next round of accumulation.

[0150] The strategy self-evolution module obtains the overall priority value, actual dispatch waiting time, and actual repair completion time of completed tasks from the execution control module. In specific implementation, after receiving a completed status code from the maintenance personnel's terminal and generating a repair efficiency record, the execution control module writes the corresponding task's overall priority value, actual dispatch waiting time, and actual repair completion time as a completed task record in the completed task record table. The overall priority value is calculated by the resource scheduling module before task dispatch and is transmitted to the execution control module along with the dispatch instruction. The execution control module binds and stores this overall priority value with the task identifier. The actual dispatch waiting time is defined as the time interval from when the task is generated and sorted by the resource scheduling module to when the execution control module successfully dispatches the task to the assigned maintenance personnel's terminal, in minutes. The actual repair completion time is defined as the time interval from when the maintenance personnel's terminal sends an accepted status code to when the maintenance personnel's terminal sends a completed status code, in minutes. The strategy self-evolution module sends a request for completed task data to the execution control module. The execution control module reads the most recent twenty completed task records from the completed task record table. Each record contains a comprehensive priority value, actual dispatch waiting time, and actual maintenance completion time. These records are then encapsulated into response data and returned to the strategy self-evolution module.

[0151] The strategy self-evolution module uses the sum of the actual dispatch waiting time and the actual repair completion time as the actual total processing time. For each completed task record, the strategy self-evolution module extracts the actual dispatch waiting time and the actual repair completion time, adds the two values ​​together to obtain the actual total processing time in minutes, and adds the calculated actual total processing time to the corresponding completed task record.

[0152] The strategy self-evolution module calculates the Spearman correlation coefficient between the overall priority ranking number of each completed task and its actual total processing time ranking number, based on the initial overall priority of the completed tasks from highest to lowest. In practice, the module first sorts the twenty completed task records according to their overall priority values ​​from highest to lowest. Each completed task is then assigned an overall priority ranking number, starting from 1 and increasing sequentially. The completed task with the highest overall priority value is assigned ranking number 1, the second highest is assigned ranking number 2, and so on. If two or more completed tasks have the same overall priority value, their ranking numbers are assigned the average of their respective position indices in the sorted sequence.

[0153] The strategy self-evolution module sorts the twenty completed task records in ascending order of their actual total processing time. The completed task with the shortest actual total processing time is assigned a sorting number of 1, the next shortest is assigned a sorting number of 2, and so on. If two or more completed tasks have the same actual total processing time, then the sorting numbers of these completed tasks are assigned the average of their respective position indices in the sorted sequence.

[0154] The strategy self-evolution module calculates the Spearman correlation coefficient between the overall priority ranking number and the actual total processing time ranking number. Let the total number of completed tasks be... In this embodiment The value is 20. For the... For each completed task, its overall priority ranking number is _____. The actual total processing time is sorted by the number as follows: Calculate the first The difference between the two sort numbers of a completed task The calculation formula is: Spearman correlation coefficient The calculation formula is:

[0155]

[0156] in, This represents the Spearman correlation coefficient, with values ​​ranging from -1 to 1. The closer the result is to 1, the stronger the positive correlation between the overall priority ranking and the actual total processing time ranking. In other words, the priority ranking can more accurately reflect the urgency and efficiency of task processing. This indicates the number of completed tasks included in the calculation, with a value of 20. The index variable is for completed tasks, ranging from 1 to... ; Indicates the first The difference between the overall priority sort number of each completed task and the sort number of the actual total processing time. This represents the sum of the squared differences in the sorting indices of all completed tasks.

[0157] The strategy self-evolution module compares the Spearman correlation coefficient with a preset target correlation coefficient range. This target range is a closed interval with a lower limit of 0.80 and an upper limit of 1.00. The basis for this target correlation coefficient range is as follows: a Spearman correlation coefficient above 0.80 indicates a strong positive correlation between the overall priority ranking and the actual processing time ranking, meaning the weighted summation coefficient configuration of the dynamic priority allocation algorithm is sufficient to prioritize high-priority tasks and shorten their total processing time. If the Spearman correlation coefficient is below 0.80, it indicates insufficient correlation between the overall priority ranking and the actual processing time ranking, meaning high-priority tasks are not being processed quickly enough, and the weighted summation coefficient needs adjustment to enhance the influence of fault severity factors in priority calculation.

[0158] If the Spearman correlation coefficient is below the lower limit of the target correlation coefficient range (0.80), the strategy self-evolution module increases the weight coefficient of the fault severity level score in the weighted summation function by a preset step size, and decreases the weight coefficient of spatial distance. In specific implementation, the weight coefficients in the weighted summation function include the weight coefficient of the fault recurrence interval. Weighting coefficients for severity level scores of faults Weighting coefficient for theoretical time commitment Weighting coefficients for spatial distance The preset weight adjustment step size is used as the weighting coefficient for the fault severity level score. Set the increment to 0.1 for the weighting coefficient of spatial distance. Set the reduction to 0.005. This step size is based on the weighting coefficient of the fault severity score. The baseline value is 2.0, and a single adjustment of 0.1 is equivalent to five percent of the baseline value, enabling fine-grained and gradual adjustments and preventing drastic fluctuations in priority ranking caused by excessively large single adjustments; spatial distance weighting coefficient. The baseline value is -0.01, and the adjustment is in units of 0.005. The adjustments are matched in magnitude to maintain a smooth evolution of the overall weighting system. During the adjustment operation, the strategy self-evolution module reads the current weighted summation coefficients currently used by the resource scheduling module. Value, will Increase by 0.1 to obtain the new fault severity level score weighting coefficient; read the current... Value, will The new spatial distance weighting coefficient is obtained by reducing it by 0.005. The upper limit of the adjusted fault severity score weighting coefficient is set to 5.0; if it increases to exceed 5.0, it will remain at 5.0 and will not increase further. The lower limit of the adjusted spatial distance weighting coefficient is set to -0.05; if it decreases to below -0.05, it will remain at -0.05 and will not decrease further. The weighting coefficient for fault recurrence interval... Weighting coefficients for theoretical time occupancy This will remain unchanged in this adjustment.

[0159] If the Spearman correlation coefficient is greater than or equal to the lower limit of the target correlation coefficient range of 0.80, the strategy self-evolution module will not make any adjustments to the weighted summation coefficient and will directly end the current strategy evolution process.

[0160] The strategy self-evolution module distributes the adjusted weight coefficients of the weighted summation function to the resource scheduling module to replace the original weighted summation coefficients. In specific implementation, the strategy self-evolution module will include... , , and The parameter set of the four weight coefficients is encapsulated into a weight update message, encoded in JSON format, containing key-value pairs of weight coefficient names and corresponding values. The policy self-evolution module sends the weight update message to the resource scheduling module through the inter-module communication interface. Upon receiving the weight update message, the resource scheduling module parses the JSON content, extracts the values ​​of the four weight coefficients, and replaces the corresponding parameters in the weighted summation function of the dynamic priority allocation algorithm with the newly received weight coefficient values. The resource scheduling module uses the updated weight coefficients in the comprehensive priority calculation of all subsequent initial operation and maintenance tasks until it receives the next weight update message from the policy self-evolution module. Through this method, the weighted summation coefficients in the dynamic priority allocation algorithm can be adaptively adjusted in a closed loop based on feedback from the actual task execution, continuously optimizing the match between the comprehensive priority ranking and the urgency of actual task processing.

[0161] See Figure 8In the figure, the horizontal axis represents the "cumulative number of completed tasks," indicating the cumulative count of completed task records recorded by the strategy self-evolution module in Example 5, ranging from 0 to 1000; the vertical axis represents the "Spearman correlation coefficient," reflecting the correlation between the overall priority ranking of completed tasks and the ranking of actual total processing time, with a value ranging from 0.6 to 1.0. The solid curve represents the trend of the Spearman correlation coefficient calculated by the system as the cumulative number of completed tasks increases. The dashed line and dotted line in the figure represent the upper limit (1.0) and lower limit (0.8) of the preset target correlation coefficient range, respectively, and the interval formed by the two reference lines is the ideal correlation coefficient target interval set by the strategy self-evolution module.

[0162] The Spearman correlation coefficient fluctuated between 0.7 and 0.97 throughout the curve, and remained above the lower limit of the target correlation coefficient range of 0.8 for most of the time. This indicates a strong positive correlation between the overall priority ranking and the ranking of actual total processing time, suggesting that the weighted summation coefficient configuration of the current dynamic priority allocation algorithm effectively reflects the urgency and efficiency of task processing. Multiple peaks in the curve correspond to correlation coefficients close to or exceeding 0.95, indicating that the strategy self-evolution module effectively improved the accuracy of priority ranking by periodically adjusting the weight coefficients. The lowest trough in the curve is close to 0.7, slightly below the lower limit of the target range, suggesting that the strategy self-evolution module may have adjusted the weights during the accumulation phase of some tasks to optimize the adaptability of the priority system.

[0163] The periodic fluctuations in the Spearman correlation coefficient in the figure reflect the triggering mechanism of the strategy self-evolution module in Example 5, which updates weights every twenty completed tasks. The trend of the curve reflects the immediate impact of weight adjustment on the correlation coefficient and its subsequent stationary process. In summary, this figure vividly illustrates how the strategy self-evolution module, under a closed-loop feedback mechanism, analyzes the execution data of completed tasks and dynamically adjusts the weights of each input parameter in the weighted summation function to ensure a strong correlation between the comprehensive priority ranking and the actual task processing order, thereby continuously optimizing the task dispatching effect of the resource scheduling module.

[0164] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. An artificial intelligence-based operation and maintenance management system for power engineering terminal equipment, characterized in that, include: The data perception and prediction module collects real-time operating status, historical fault records and environmental parameter data from multiple terminal devices, and inputs them into the time series prediction model to generate a predicted fault probability sequence for each device in the next time window. The task generation module, based on the predicted fault probability sequence and the preset fault threshold, selects terminal devices whose predicted fault probability exceeds the fault threshold as target devices, and generates an initial maintenance task for each target device, including device identifier, fault type estimation and recommended maintenance time. The resource scheduling module obtains currently available operation and maintenance personnel information and spare parts inventory information, and calculates the comprehensive priority of each initial operation and maintenance task based on the initial operation and maintenance task, the operation and maintenance personnel information and the spare parts inventory information using a dynamic priority allocation algorithm. The initial operation and maintenance tasks are sorted from high to low according to the comprehensive priority to obtain a sorted list of tasks to be executed. The execution control module dispatches each task in the list of tasks to be executed to the corresponding maintenance personnel terminal according to the sorting order, and receives the task execution status feedback from the maintenance personnel terminal in real time.

2. The power engineering terminal equipment operation and maintenance management system based on artificial intelligence according to claim 1, characterized in that, The data perception and prediction module generates a predicted failure probability sequence for each device within the next time window, specifically including: Real-time operating status data, historical fault record data, and environmental parameter data are collected from multiple terminal devices at the power engineering site; The real-time operating status data of each terminal device is divided into multiple state feature vectors for consecutive time segments by a sliding window. The historical fault record data of each terminal device is timestamped, and the historical state feature sequence within a preset time before each historical fault occurs is extracted. The historical state feature sequence and the corresponding fault type label are used as training samples. The environmental parameter data is concatenated with the real-time operating status data according to the timestamp to obtain the fused input features at each time point; The fused input features are sequentially passed through the convolutional layer, recurrent layer, and fully connected layer in the time-series prediction model to output the probability value of each terminal device belonging to different fault types in each time unit within a future preset time window. The probability values ​​of all time units are arranged in chronological order to form the predicted fault probability sequence.

3. The power engineering terminal equipment operation and maintenance management system based on artificial intelligence according to claim 2, characterized in that, The recurrent layer in the time-series prediction model uses a gated recurrent unit, and the feature map output by the convolutional layer is expanded by time step and then input into the gated recurrent unit.

4. The power engineering terminal equipment operation and maintenance management system based on artificial intelligence according to claim 2, characterized in that, The step in the task generation module of selecting terminal devices whose predicted failure probability exceeds the fault threshold as target devices based on the predicted failure probability sequence and the preset fault threshold specifically includes: Extract the maximum probability value for each terminal device and the fault type corresponding to the maximum probability value from the predicted fault probability sequence; Compare the maximum probability value with the preset fault threshold. If the maximum probability value is greater than the fault threshold, mark the corresponding terminal device as the target device and use the fault type corresponding to the maximum probability value as the fault type prediction for the target device. For each target device, query the historical average repair time corresponding to the fault type in the historical fault record data, and multiply the historical average repair time by a preset margin coefficient to obtain the recommended repair time. The device identifier of the target device, the fault type prediction, and the recommended maintenance time are combined into a structured data object, which serves as the initial maintenance task.

5. The power engineering terminal equipment operation and maintenance management system based on artificial intelligence according to claim 4, characterized in that, The preset fault threshold is dynamically adjusted according to the importance level of each terminal device; the higher the importance level of the terminal device, the lower the corresponding fault threshold.

6. The power engineering terminal equipment operation and maintenance management system based on artificial intelligence according to claim 4, characterized in that, The steps in the resource scheduling module to obtain currently available maintenance personnel information and spare parts inventory information specifically include: Send an availability query request to the operations and maintenance personnel management server, and receive a list of operations and maintenance personnel who are currently idle or will complete the current task within a preset time, returned by the operations and maintenance personnel management server. Each operations and maintenance personnel record in the list includes personnel ID, skill tag set, current location coordinates and average maintenance efficiency. Send an inventory query request to the spare parts warehouse management system and receive the spare parts identifier, spare parts type, storage location and inventory quantity of each spare part in the current inventory returned by the spare parts warehouse management system; For each initial maintenance task, the fault type is estimated, and the spare parts type that matches the fault type is selected from the spare parts inventory information. The inventory quantity of the spare parts type is checked to see if it is greater than zero. If it is greater than zero, the corresponding spare parts identifier is associated with the initial maintenance task.

7. The power engineering terminal equipment operation and maintenance management system based on artificial intelligence according to claim 6, characterized in that, The steps in the resource scheduling module that use a dynamic priority allocation algorithm to calculate the overall priority of each initial operation and maintenance task specifically include: Extract the device identifier from each initial maintenance task, and query the historical fault record data based on the device identifier to obtain the fault reproduction interval of the corresponding terminal device; Extract the fault type prediction from each initial operation and maintenance task, and match the fault severity score corresponding to the fault type prediction from the preset fault level mapping table; Based on the recommended maintenance duration and the average maintenance efficiency in the currently available maintenance personnel information, calculate the theoretical duration of each initial maintenance task; Obtain the geographical location of the terminal device corresponding to each initial operation and maintenance task, and calculate the spatial distance between the geographical location and the current location of each operation and maintenance personnel in the currently available operation and maintenance personnel information; The fault reproduction interval, the fault severity score, the theoretical occupancy time, and the spatial distance are input into a weighted summation function, which outputs a comprehensive priority value for each initial maintenance task. The initial maintenance tasks are then sorted from largest to smallest according to the comprehensive priority value.

8. The power engineering terminal equipment operation and maintenance management system based on artificial intelligence according to claim 7, characterized in that, The weighting coefficient for the fault reproduction interval in the weighted summation function is set to a negative value, so that the shorter the fault reproduction interval, the higher the overall priority value.

9. The power engineering terminal equipment operation and maintenance management system based on artificial intelligence according to claim 7, characterized in that, The step in the resource scheduling module to calculate the theoretical duration of each initial maintenance task based on the recommended maintenance time and the average maintenance efficiency in the currently available maintenance personnel information specifically includes: Extract the average maintenance efficiency of all maintenance personnel from the currently available maintenance personnel information, and calculate the reciprocal of the average maintenance efficiency to obtain the standard man-hours required to repair a unit fault; Multiply the recommended maintenance time by the standard working hours to obtain the baseline occupancy time; Obtain the device complexity level of the terminal device corresponding to each initial operation and maintenance task, and query the complexity adjustment coefficient corresponding to the device complexity level from the preset complexity coefficient mapping table; Multiply the baseline duration by the complexity adjustment factor to obtain the theoretical duration of each initial operation and maintenance task.

10. The power engineering terminal equipment operation and maintenance management system based on artificial intelligence according to claim 7, characterized in that, The steps in the resource scheduling module to obtain the geographical location of the terminal device corresponding to each initial maintenance task and to calculate the spatial distance between the geographical location and the current location of each maintenance worker in the currently available maintenance personnel information specifically include: Query the latitude and longitude coordinates of the installation location corresponding to the equipment identifier of each initial maintenance task from the equipment ledger database, and use them as the first coordinates; Extract the latitude and longitude coordinates of each maintenance personnel from the currently available maintenance personnel information, and use them as the second coordinates; For each combination of initial maintenance task and each maintenance personnel, the first coordinate and the second coordinate are respectively converted into projection plane coordinates under the same spatial reference system; Calculate the Euclidean distance between the coordinates of the projection plane, and multiply the Euclidean distance by a preset road network tortuosity coefficient to obtain the spatial distance corresponding to each combination.

Images