A method and system for field monitoring equipment maintenance based on multiple pressure indexes
By constructing multiple pressure indices and extracting dynamic temporal features, and combining physical association rules and gradient boosting decision tree models, the problem of high-precision fault early warning for field monitoring equipment was solved, achieving efficient and accurate fault detection and prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 中国地质环境监测院(自然资源部地质灾害技术指导中心)
- Filing Date
- 2026-03-17
- Publication Date
- 2026-06-19
Smart Images
Figure CN122241511A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of geological disaster monitoring and Internet of Things (IoT) equipment operation and maintenance technology, and in particular to a method and system for maintaining field monitoring equipment based on multiple pressure indices. Background Technology
[0002] Geological hazard monitoring is a crucial component of disaster prevention and mitigation systems. Currently, over 300,000 universal geological hazard monitoring devices (such as rain gauges, crack gauges, and GNSS receivers) have been deployed nationwide. These devices are typically deployed in remote mountainous areas, landslide-prone regions, and debris flow-prone areas to collect real-time data on critical parameters like rainfall and surface displacement. Due to their long-term exposure to the elements, these devices face significant survival challenges: extreme environmental conditions (high temperature, high humidity, heavy rainfall), reliance on unstable solar power, and weak and easily interrupted communication network coverage. Statistics show a high annual failure rate for field monitoring equipment, with most failures caused by performance degradation due to environmental factors. Once equipment failure leads to data interruption, it directly creates early warning blind spots, seriously threatening the lives and property of downstream residents.
[0003] To ensure the normal operation of equipment, the existing technologies mainly adopt the following maintenance strategies, but all of them have significant limitations: The first type is an alarm mechanism based on fixed thresholds. The monitoring platform sets fixed upper and lower limits for parameters such as battery voltage and signal strength (e.g., an alarm when the voltage is <3.5V). The drawback of this method is that it cannot comprehensively assess the impact of multiple factors and is usually a "hindsight" approach. When an alarm is triggered, the equipment is often already on the verge of failure or has already stopped, lacking sufficient lead time for maintenance personnel to respond, making it difficult to achieve true preventative maintenance.
[0004] The second method is statistical prediction based on equipment runtime (such as MTBF). This method calculates the mean time between failures (MTBF) based on a general reliability model to predict the probability of failure. However, the field environment has strong regional differences and dynamism. The same type of equipment deployed in humid and hot conditions and dry and cold conditions will have completely different lifespan degradation curves. Duration-based statistical models ignore the dynamic evolution of environmental pressures and cannot cope with the accelerated aging caused by sudden extreme weather.
[0005] The third approach is to draw inspiration from the HealthIndex used in the industrial sector. While the industry has mature equipment health management solutions, these are typically applied in controlled indoor environments (such as factory workshops) and rely on a large number of specialized, high-precision sensors (such as oil analysis and high-frequency vibration monitoring). For field monitoring equipment, directly adopting the industrial model is not suitable: on the one hand, field environmental parameters (temperature, humidity, light intensity) fluctuate dramatically, and existing models lack the ability to characterize the cumulative effects of environmental stress; on the other hand, the low-power design of field equipment (solar power supply) and the high cost of retrofitting existing equipment (high cost per unit) prevent the acquisition of more diagnostic data through simply "stacking in sensors."
[0006] The fourth type is fault prediction based on general machine learning. Some existing studies attempt to use neural networks to predict single parameters (such as voltage curves). However, these methods often lack engineering of physical mechanism characteristics for field conditions. Directly inputting raw data leads to poor model generalization ability, and the model output lacks interpretability. Maintenance personnel cannot know the specific cause of the fault (such as whether it is battery aging or solar panel obstruction), making it difficult to guide actual maintenance.
[0007] In summary, current technologies lack a predictive maintenance solution capable of quantitatively assessing the stress of complex field environments, capturing the dynamic evolution characteristics of faults, and requiring no additional hardware costs. How to leverage existing, limited equipment data to uncover the deep correlation between environmental factors and equipment health status, and achieve long-lead-time, high-precision fault warnings, is a pressing technical problem that needs to be solved. Summary of the Invention
[0008] The purpose of this invention is to provide a maintenance method and system for field monitoring equipment based on multiple pressure indices, and to solve the problem of how to use the limited data already available in the equipment to achieve high-precision fault early warning.
[0009] To address the aforementioned problems, a first aspect of the present invention provides a method and system for maintaining field monitoring equipment based on multiple pressure indices, comprising: The system acquires time-stamped operational status data of field monitoring equipment and environmental perception data that matches the environment in which the monitoring equipment is located. Based on the operational status data and the environmental perception data, it performs multi-dimensional quantitative calculations to obtain the corresponding pressure index and sorts the pressure index in time sequence based on the timestamp. The sorted pressure index is subjected to dynamic temporal feature extraction based on a rolling time window to obtain basic temporal features. Based on preset physical association rules, feature interaction and coupling calculations are performed on the basic temporal features. The calculation results are concatenated with the basic temporal features to obtain an enhanced feature vector. The basic time-series features and the temperature-pressure exponential enhanced feature vector are respectively subjected to feature selection and fusion processing to obtain the processing results; Based on the processing results, fault risk is detected and predicted to obtain fault risk prediction results.
[0010] Furthermore, in the above method, the step of performing feature selection and fusion processing on the basic temporal features and the enhanced feature vector to obtain the processing result includes: Based on a preset feature contribution evaluation strategy, feature filtering is performed on multiple feature dimensions contained in the basic time-series features and the enhanced feature vector, and features with a contribution value lower than a preset threshold are removed. Based on the gradient boosting decision tree model, the filtered feature vectors are mapped into a high-dimensional space and classified for inference. Based on the results of the feature selection and the classification reasoning, the processing result is obtained, which includes the failure risk probability and the feature influence weight of each feature dimension.
[0011] Furthermore, in the above method, the step of performing feature interaction and coupling calculations on the basic time-series features based on preset physical association rules includes: Based on the physical mechanism model, the pressure indices of different dimensions are cross-producted to obtain the environmental-energy coupled stress characteristics. Based on the exponentially weighted moving average algorithm, the pressure index is calculated over a long period to obtain the pressure cumulative fatigue characteristics. The time series of the pressure index is subjected to sample entropy calculation to obtain the wave morphology entropy characteristics; The step of concatenating the calculation results with the basic time series features includes: concatenating the environmental-energy coupled stress features, the pressure cumulative fatigue features, and the wave morphology entropy features with the basic time series features to construct the enhanced feature vector.
[0012] Furthermore, in the above method, obtaining the environmental-energy coupled stress characteristics includes at least one of the following steps: The temperature-pressure index and energy-pressure index are obtained from the pressure index. The product of the statistical value of the temperature-pressure index and the statistical value of the energy-pressure index is calculated to obtain the temperature-voltage coupling index, which is used to characterize the combined risk under high temperature and power shortage conditions. The humidity pressure index and communication pressure index are obtained from the pressure index. The product of the statistical value of the humidity pressure index and the statistical value of the communication pressure index is calculated to obtain the humidity-signal coupling index, which is used to characterize the combined risk of communication obstruction in high humidity environments.
[0013] Furthermore, in the above method, obtaining the corresponding pressure index includes: The operational status data and the environmental perception data are normalized to calculate the energy pressure index, temperature pressure index, humidity pressure index and communication pressure index, which have values ranging from [0, 1]. The formulas for calculating each pressure index are as follows: The formula for calculating the energy pressure index is as follows:
[0014] in, Energy stress index, This is the battery's rated voltage. This is the current battery voltage. This is the lowest voltage at which the device can operate. The formula for calculating the temperature-pressure index is as follows:
[0015] in, The current ambient temperature. and These are the lower and upper limits of the optimal operating temperature range for the equipment, respectively. Temperature tolerance range; The formula for calculating the humidity pressure index is:
[0016] in, Given the current ambient humidity, and These are the lower and upper limits of the optimal operating humidity range for the equipment, respectively. Humidity tolerance range; The formula for calculating the communication pressure index is as follows:
[0017] in, This is the current signal strength value. This is the minimum signal strength value at which the device can communicate. This is the preset optimal signal strength value.
[0018] Furthermore, in the above method, the step of extracting dynamic temporal features from the sorted pressure index based on a rolling time window includes: Set the length W of the scrolling time window and the sliding step S; The time point furthest from the current time point among the sorted pressure indices is determined as the sliding start point; The scrolling time window is controlled to start from the sliding start point and gradually slide towards the current time point according to the sliding step size S until it reaches the current time point; At each stop position of the scrolling time window, the pressure index sequence within the window is extracted, and the mean, standard deviation, extreme values, linear trend coefficient, and rate of change of the sequence are calculated as the basic time series characteristics.
[0019] Furthermore, in the above method, the window length W is adaptively adjusted based on the fluctuation intensity of the environmental perception data; Calculate the variance or standard deviation of the environmental perception data. When the calculation result exceeds the preset fluctuation threshold, it is determined that the current environment is in a high-fluctuation environment, and W is set to 3 to 5 days.
[0020] Furthermore, in the above method, the training label generation method of the gradient boosting decision tree model includes: For any point in time in the historical data, detect whether there is a data interruption event that lasts for a second preset time within a first preset time period in the future; If it exists, then the sample corresponding to that time point is marked as high-risk; otherwise, it is marked as low-risk. The first preset time is determined based on the maintenance response cycle of the monitoring equipment; the second preset time is determined based on the charging and discharging cycle or data transmission frequency of the monitoring equipment.
[0021] According to another aspect of the present invention, a predictive maintenance system for monitoring equipment is also provided, comprising: The data acquisition and preprocessing unit is used to acquire time-stamped operational status data of the field monitoring equipment and environmental perception data that matches the environment in which the monitoring equipment is located, and to perform multi-dimensional quantitative calculations based on the operational status data and the environmental perception data to obtain the corresponding pressure index, and to sort the pressure index in time sequence based on the timestamp. The feature extraction and enhancement unit is used to perform dynamic temporal feature extraction based on a rolling time window on the sorted pressure index to obtain basic temporal features, and to perform feature interaction and coupling calculation on the basic temporal features based on preset physical association rules. The calculation results are then concatenated with the basic temporal features to obtain an enhanced feature vector. The feature selection and fusion unit is used to perform feature selection and fusion processing on the basic temporal features and the enhanced feature vectors respectively to obtain the processing results; The fault prediction unit is used to detect and predict fault risks based on the processing results, and obtain fault risk prediction results.
[0022] This invention constructs an "environmental pressure index" to normalize operational status data such as voltage and current with environmental sensing data such as temperature and humidity, eliminating differences between different physical dimensions. It transforms raw numerical fluctuations into a standardized indicator measuring the "degree of environmental severity" experienced by the equipment, intuitively quantifying the comprehensive pressure exerted on the equipment by the external environment. This solves the problem that traditional methods struggle to adapt to complex and variable field environments using a single physical quantity. Building upon this, the solution introduces deep physical enhancement features to address pain points in field monitoring: nonlinear dynamics features can capture internal mechanism changes within the equipment through numerical data, identifying hidden faults where "numerical values are within limits but patterns are abnormal"; spatial neighborhood consistency features utilize group data to verify individual states, effectively distinguishing between severe environmental conditions and sensor drift, reducing false alarm rates; and physical time-delay coupling features eliminate time differences caused by thermal inertia or penetration lag, restoring the true physical causality of the environment's impact on the equipment. Furthermore, by introducing a feature selection mechanism based on contribution to eliminate redundant dimensions, the computational load is reduced to accommodate edge nodes in the field with limited computing power. On the other hand, noise interference is removed to prevent model overfitting. Thus, while significantly improving the accuracy of fault identification, the efficiency and robustness of the model in unknown environments are also taken into account. Attached Figure Description
[0023] Figure 1 This is a flowchart based on the first embodiment of the present invention; Detailed Implementation
[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments and the accompanying drawings. It should be understood that these descriptions are merely exemplary and not intended to limit the scope of the invention. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.
[0025] like Figure 1 As shown, this embodiment provides a method and system for maintaining field monitoring equipment based on multiple pressure indices, including the following steps: S1: Data Acquisition and Stress Index Construction Acquire timestamped operational status data of the field monitoring equipment and environmental perception data matching the environment in which the monitoring equipment is located. The operational status data includes at least battery voltage, charging current, and load current; the environmental perception data includes at least ambient temperature, ambient humidity, and received signal strength indication.
[0026] Considering the inconsistency of data dimensions across different dimensions, this embodiment performs multidimensional quantification and normalization on the above data to obtain a pressure index with a value range of [0, 1].
[0027] The pressure index is calculated based on operational status data and environmental perception data, respectively.
[0028] Specifically, the energy pressure index is calculated based on the battery voltage, using the following formula: (1) in, This is the battery's rated voltage. This is the current battery voltage. This is the lowest voltage at which the device can operate. The temperature-pressure index is calculated based on ambient temperature, specifically including: segmented normalization calculation based on the optimal operating temperature range and temperature tolerance range. The calculation formula is as follows: (2) in, The current ambient temperature. and These are the lower and upper limits of the optimal operating temperature range for the equipment, respectively. Temperature tolerance range; The humidity pressure index is calculated based on ambient humidity and is used to characterize the risk of short circuits caused by high humidity environments or the risk of electrostatic breakdown caused by dry environments. The calculation formula is as follows: (3) in, Given the current ambient humidity, and These are the lower and upper limits of the optimal operating humidity range for the equipment, respectively. Humidity tolerance range; Its calculation logic is similar to that of the temperature and pressure index, and it performs segmented normalization calculation based on the optimal working humidity range and humidity tolerance range: when the current ambient humidity is higher than the upper limit of the optimal humidity, the ratio of the excess part to the humidity tolerance range is calculated; when the current ambient humidity is lower than the lower limit of the optimal humidity, the ratio of the lower part to the humidity tolerance range is calculated; if it is within the optimal humidity range, the index is 0.
[0029] The communication stress index is calculated based on the received signal strength indicator. The calculation is normalized based on the optimal value of the received signal strength indicator and the drop-out threshold. The calculation formula is as follows: (4) in, This is the current signal strength value. This is the minimum signal strength value at which the device can communicate. This is the preset optimal signal strength value.
[0030] After obtaining the corresponding pressure index, all the calculated pressure indices are strictly sorted according to the timestamp to form a pressure index sequence arranged in chronological order.
[0031] S2: Dynamic Temporal Feature Extraction and Multidimensional Physical Augmentation Next, dynamic time-series features of the sorted pressure index are extracted based on a rolling time window. The length and sliding step of the rolling time window are preset, and the window is controlled to slide on the sorted sequence. At each stop position of the window, the data sequence within the window is extracted, and the mean, standard deviation, extreme values, and rate of change are calculated to obtain the basic time-series features.
[0032] In a preferred embodiment, an adaptive window adjustment strategy is adopted to calculate the variance of environmental perception data in real time. When the calculation result exceeds a preset fluctuation threshold, it indicates that a sudden change has occurred in the environment. Therefore, the window length is automatically shortened (e.g., set to 3 to 5 days) to capture transient features in a highly volatile environment.
[0033] Building upon this, in order to capture the qualitative changes in the internal nonlinear dynamics of the monitoring equipment under long-term environmental pressure (i.e., numerical values do not exceed limits but the evolutionary pattern is abnormal), this embodiment further extracts trajectory morphology features based on phase space reconstruction. The specific processing procedure is as follows: The pressure index sequence generated in step S1 and sorted by time is selected as the basic one-dimensional time series, denoted as... .
[0034] First, determine the key parameters for phase space reconstruction: Delay time The mutual information method is used to calculate the mutual information value of the sequence under different delays, and the delay time corresponding to the first minimum value of the mutual information value is selected as the time. ; Embedding dimension The pseudo-nearest neighbor method is used to calculate the proportion of pseudo-nearest neighbors in the phase space under different embedding dimensions. The smallest dimension at which this proportion converges to 0 or is lower than a preset threshold (e.g., 5%) is selected as the minimum dimension. Next, based on the determined and For this pressure index sequence Equal-interval delayed sampling is performed, and the samples are mapped to a high-dimensional phase space to construct a dynamic trajectory vector reflecting the evolution of environmental pressure. This vector is denoted as... .
[0035] A baseline attraction factor for the pressure index of the equipment under normal conditions is pre-built. During real-time monitoring, the maximum value of the minimum spatial distance from each point in the trajectory point cloud (i.e., the vector set composed of dynamic trajectory vectors) generated by the current pressure index to the baseline attraction factor point set is calculated, and this maximum value is defined as the maximum trajectory separation. This feature is used to quantify whether the impact mode of the current environmental pressure on the equipment has deviated from the normal fault baseline.
[0036] Meanwhile, in order to verify the authenticity of current field monitoring equipment by utilizing data from existing nearby field monitoring equipment, this embodiment further extracts spatial neighborhood consistency features. The specific processing procedure is as follows: Based on the principle of spatial variogram, the spatial correlation coefficient of the pressure index of each field monitoring equipment within the region is calculated. The calculation formula is: (5) in, and They represent the first The and the first The geographical coordinates of each field monitoring device; This refers to the spatial distance between the two field monitoring devices mentioned above; and These represent the pressure index values at these two locations, respectively; The spacing within the region is equal to The total number of field monitoring equipment pairs.
[0037] Traverse different intervals Select to make The distance at which the minimum value is reached or the steady-state region is entered is defined as the effective physical correlation radius. Subsequently, all neighboring field monitoring devices located within this correlation radius of the current field monitoring device are selected. The absolute value of the deviation between the pressure index of the current field monitoring device and the average pressure index of these neighboring field monitoring devices is calculated, and this deviation is defined as the spatial dispersion index. This feature is used to quantify the degree of deviation of the current field monitoring device's state from the consensus of the surrounding group, thereby assisting in determining whether there are any anomalies at the sensing end.
[0038] Furthermore, to eliminate the physical thermal inertia or penetration hysteresis inherent in the transmission of external environmental changes to the field monitoring equipment, this embodiment introduces an adaptive time-delay alignment mechanism to construct enhanced features. The specific processing procedure is as follows: The correlation value between the environmental perception data sequence and the operational status data sequence is calculated, and the optimal hysteresis time that maximizes the correlation value is determined. According to this The environmental perception data is shifted backward along the time axis to ensure strict physical causal alignment with the operational status data of field monitoring equipment. Subsequently, a cross-product operation is performed on the aligned data to extract physical coupling features.
[0039] The features calculated based on this strategy specifically include: Environmental-energy coupled stress characteristics after time delay correction: For example, multiplying the shifted environmental temperature data with the battery voltage data can be used to quantify the combined risk when environmental thermal stress actually acts on the battery; Pressure cumulative fatigue characteristics: long-term cumulative pressure values calculated using an exponentially weighted moving average algorithm; Wave morphology entropy features: Feature values calculated based on the sample entropy algorithm to quantify the complexity and irregularity of environmental stress sequences. Finally, the above trajectory morphology features, spatial neighborhood consistency features, physical coupling features, and basic temporal features are concatenated with each other to form a high-dimensional enhanced feature vector.
[0040] S3: Feature Selection and Model Inference Before model training or inference, to remove redundant information and improve model efficiency, feature filtering is first performed on the basic time-series features and multiple feature dimensions contained in the enhanced feature vectors generated in step S2, based on a preset feature contribution evaluation strategy. The contribution score of each feature is calculated, and features with a contribution score below a preset threshold are removed, retaining only key features. Subsequently, a fault prediction model is constructed, and the model is trained using the filtered feature vectors from historical data. Specifically, the filtered high-dimensional enhanced feature vectors are input into a gradient boosting decision tree model for training, establishing a mapping relationship between feature vectors and fault risks.
[0041] Furthermore, to address the problem of blurred model training boundaries caused by the scarcity of fault samples in field scenarios, this embodiment introduces a boundary simulation sample generation mechanism based on physical constraints before model training. The specific process is as follows: First, a multi-dimensional physical feature space based on the pressure index is constructed. For each normal sample point in the space, the direction of the maximum risk gradient pointing from the sample point to the critical value of equipment failure is calculated. Subsequently, a numerical perturbation conforming to physical laws is applied to the original normal sample along this gradient direction to generate a physical boundary simulation sample. If the physical state parameters of the simulation sample exceed the preset tolerance limit of the field monitoring equipment (such as the maximum tolerance temperature or minimum operating voltage specified in the design manual), it is forcibly marked as a "virtual fault sample" and added to the training set. This aims to force the model to learn hard fault criteria determined by physical boundaries, rather than relying solely on historical statistical patterns. After training, the model is used to infer the real-time input high-dimensional enhanced feature vector, outputting the fault risk prediction probability and feature influence weights.
[0042] S4: Fault Risk Detection, Prediction, and Attribution. Based on model inference results, the predicted probability of fault risk for the current field monitoring equipment is obtained. If the predicted probability exceeds a preset threshold, a fault warning signal is generated and an alarm is triggered.
[0043] Building upon this foundation, to accurately distinguish the source of the fault, this embodiment introduces a feature marginal contribution measurement mechanism based on full permutation combinations. The input features of the prediction model are considered as co-variables jointly influencing the prediction result. By calculating the average marginal contribution value of each feature across all possible feature combination subsets, the prediction probability deviation output by the model is decomposed into independent contribution components of each feature. This calculation process satisfies complete additivity, meaning the sum of the contribution values of all features equals the difference between the current prediction probability and the baseline probability. Based on the calculated contribution values, a fault attribution map is constructed, and corresponding strategies are executed according to the feature with the highest contribution value: If the contribution value of the spatial dispersion index is significantly the highest, the dominant fault factor is determined to be an anomaly at the sensing end, the data is marked as questionable, and a remote calibration work order is generated. If the contribution value of the maximum trajectory separation is the highest, it is determined to be a hidden fault in the equipment itself, and it is recommended to prioritize on-site replacement with spare parts. If the contribution value of the environmental-energy coupling stress feature after time-delay correction is the highest, it is determined to be an external environmental pressure fault, automatically triggering a degradation and keep-alive command, shutting down non-core loads and reducing the sampling frequency.
[0044] The present invention also provides a predictive maintenance system for monitoring equipment, comprising: The data acquisition and preprocessing unit is used to acquire time-stamped operational status data of the field monitoring equipment and environmental perception data that matches the environment in which the monitoring equipment is located, and to perform multi-dimensional quantitative calculations based on the operational status data and the environmental perception data to obtain the corresponding pressure index, and to sort the pressure index in time sequence based on the timestamp. The feature extraction and enhancement unit is used to perform dynamic temporal feature extraction based on a rolling time window on the sorted pressure index to obtain basic temporal features, and to perform feature interaction and coupling calculation on the basic temporal features based on preset physical association rules. The calculation results are then concatenated with the basic temporal features to obtain an enhanced feature vector. The feature selection and fusion unit is used to perform feature selection and fusion processing on the basic temporal features and the enhanced feature vectors respectively to obtain the processing results; The fault prediction unit is used to detect and predict fault risks based on the processing results, and obtain fault risk prediction results.
[0045] It should be understood that the specific embodiments described above are merely illustrative or explanatory of the principles of the invention and do not constitute a limitation thereof. Therefore, any modifications, equivalent substitutions, improvements, etc., made without departing from the spirit and scope of the invention should be included within the protection scope of the invention. Furthermore, the appended claims are intended to cover all variations and modifications falling within the scope and boundaries of the appended claims, or equivalent forms of such scope and boundaries.
Claims
1. A method and system for field monitoring equipment maintenance based on multi-pressure index, characterized in that, include: The system acquires time-stamped operational status data of field monitoring equipment and environmental perception data that matches the environment in which the monitoring equipment is located. Based on the operational status data and the environmental perception data, it performs multi-dimensional quantitative calculations to obtain the corresponding pressure index and sorts the pressure index in time sequence based on the timestamp. The sorted pressure index is subjected to dynamic temporal feature extraction based on a rolling time window to obtain basic temporal features. Based on preset physical association rules, feature interaction and coupling calculations are performed on the basic temporal features. The calculation results are concatenated with the basic temporal features to obtain an enhanced feature vector. The basic temporal features and the enhanced feature vectors are subjected to feature selection and fusion processing respectively to obtain the processing results; Based on the processing results, fault risk is detected and predicted to obtain fault risk prediction results.
2. The method according to claim 1, wherein, The process of performing feature selection and fusion on the basic temporal features and the enhanced feature vector to obtain the processing result includes: Based on a preset feature contribution evaluation strategy, feature filtering is performed on multiple feature dimensions contained in the basic time-series features and the enhanced feature vector, and features with a contribution value lower than a preset threshold are removed. Based on the gradient boosting decision tree model, the filtered feature vectors are mapped into a high-dimensional space and classified for inference. Based on the results of the feature selection and the classification reasoning, the processing result is obtained, which includes the failure risk probability and the feature influence weight of each feature dimension.
3. The method according to claim 2, wherein, The feature interaction and coupling calculation based on the preset physical association rules for the basic time-series features includes: Based on the physical mechanism model, the pressure indices of different dimensions are cross-producted to obtain the environmental-energy coupled stress characteristics. Based on the exponentially weighted moving average algorithm, the pressure index is calculated over a long period to obtain the pressure cumulative fatigue characteristics. The time series of the pressure index is subjected to sample entropy calculation to obtain the wave morphology entropy characteristics; The step of concatenating the calculation results with the basic time series features includes: concatenating the environmental-energy coupled stress features, the pressure cumulative fatigue features, and the wave morphology entropy features with the basic time series features to construct the enhanced feature vector.
4. The method according to claim 3, wherein, The process of obtaining environmental-energy coupled stress characteristics includes at least one of the following steps: The temperature-pressure index and energy-pressure index are obtained from the pressure index. The product of the statistical value of the temperature-pressure index and the statistical value of the energy-pressure index is calculated to obtain the temperature-voltage coupling index, which is used to characterize the combined risk under high temperature and power shortage conditions. The humidity pressure index and communication pressure index are obtained from the pressure index. The product of the statistical value of the humidity pressure index and the statistical value of the communication pressure index is calculated to obtain the humidity-signal coupling index, which is used to characterize the combined risk of communication obstruction in high humidity environments.
5. The method according to any one of claims 1-4, wherein, Obtaining the corresponding pressure index includes: The operational status data and the environmental perception data are normalized to calculate the energy pressure index, temperature pressure index, humidity pressure index and communication pressure index with values ranging from [0,1]. The formulas for calculating each pressure index are as follows: The formula for calculating the energy pressure index is as follows: , in, Energy stress index, This is the battery's rated voltage. This is the current battery voltage. This is the lowest voltage at which the device can operate. The formula for calculating the temperature-pressure index is as follows: , in, The current ambient temperature. and These are the lower and upper limits of the optimal operating temperature range for the equipment, respectively. Temperature tolerance range; The formula for calculating the humidity pressure index is: , in, Given the current ambient humidity, and These are the lower and upper limits of the optimal operating humidity range for the equipment, respectively. Humidity tolerance range; The formula for calculating the communication pressure index is as follows: , in, This is the current signal strength value. This is the minimum signal strength value at which the device can communicate. This is the preset optimal signal strength value.
6. The method according to any one of claims 1-5, wherein, The step of extracting dynamic temporal features from the sorted pressure indices based on a rolling time window includes: Set the length W of the scrolling time window and the sliding step S; The time point furthest from the current time point among the sorted pressure indices is determined as the sliding start point; The scrolling time window is controlled to start from the sliding start point and gradually slide towards the current time point according to the sliding step size S until it reaches the current time point; At each stop position of the scrolling time window, the pressure index sequence within the window is extracted, and the mean, standard deviation, extreme values, linear trend coefficient, and rate of change of the sequence are calculated as the basic time series characteristics.
7. The method according to claim 6, wherein, The window length W is adaptively adjusted based on the fluctuation intensity of the environmental perception data. Calculate the variance or standard deviation of the environmental perception data. When the calculation result exceeds the preset fluctuation threshold, it is determined that the current environment is in a high-fluctuation environment, and W is set to 3 to 5 days.
8. The method according to claim 2, wherein, The training label generation method for the gradient boosting decision tree model includes: For any point in time in the historical data, detect whether there is a data interruption event that lasts for a second preset time within a first preset time period in the future; If it exists, then the sample corresponding to that time point is marked as high-risk; otherwise, it is marked as low-risk. The first preset time is determined based on the maintenance response cycle of the monitoring equipment; the second preset time is determined based on the charging and discharging cycle or data transmission frequency of the monitoring equipment.
9. A predictive maintenance system for monitoring equipment, characterized in that, include: The data acquisition and preprocessing unit is used to acquire time-stamped operational status data of the field monitoring equipment and environmental perception data that matches the environment in which the monitoring equipment is located, and to perform multi-dimensional quantitative calculations based on the operational status data and the environmental perception data to obtain the corresponding pressure index, and to sort the pressure index in time sequence based on the timestamp. The feature extraction and enhancement unit is used to perform dynamic temporal feature extraction based on a rolling time window on the sorted pressure index to obtain basic temporal features, and to perform feature interaction and coupling calculation on the basic temporal features based on preset physical association rules. The calculation results are then concatenated with the basic temporal features to obtain an enhanced feature vector. The feature selection and fusion unit is used to perform feature selection and fusion processing on the basic temporal features and the enhanced feature vectors respectively to obtain the processing results; The fault prediction unit is used to detect and predict fault risks based on the processing results, and obtain fault risk prediction results.