Key component thermal anomaly detection method based on improved isolated forest algorithm

Abstract

The invention belongs to the field of machine learning fault prediction, and particularly relates to a key component thermal anomaly detection method based on an improved isolated forest algorithm. Firstly, a temperature sensor and a thermal imager are arranged on the surfaces of key components (such as a traction machine and a braking system) of the elevator to collect data; and then realizing data fusion by combining an improved dynamic convolution kernel with reinforcement learning, determining convolution kernel parameters by constructing an environment and designing a reward function, and integrating features by using an attention-based weighted fusion mode. And then modeling training is carried out by using an improved isolated forest algorithm, and an adaptive segmentation strategy and multi-granularity path length aggregation evaluation are improved. After modeling, heat dissipation modeling and dynamic threshold calibration are carried out, and an adversarial sample with heat dissipation noise is injected; and finally, anomaly is detected by using the trained model. According to the method, multi-source data can be fully utilized, features are accurately extracted, the method adapts to elevator running state changes, the accuracy and timeliness of thermal anomaly detection are improved, and safe running of the elevator is guaranteed.

Description

Technical Field

[0001] The present invention belongs to the field of machine learning fault prediction, and in particular relates to a key component thermal anomaly detection method based on an improved isolation forest algorithm. Background Art

[0002] Elevators are important vertical transportation tools, but key elevator components are prone to heat up after long-term operation, causing safety accidents, affecting passenger safety and elevator operation efficiency. Traditional detection methods have obvious shortcomings: the data fusion method is simple, often using direct splicing or fixed convolution kernel fusion, without considering the dynamic changes of data, making it difficult to mine the potential relationship of data, and the fused features have weak ability to characterize thermal anomalies, resulting in poor detection accuracy and timeliness. When the isolation forest algorithm is used for anomaly detection, the traditional construction method randomly selects features and segmentation points, does not consider data distribution characteristics and feature associations, and only relies on simple path length to evaluate anomalies, which is easy to miss potential anomalies, and the detection accuracy and reliability are not high. In addition, the heat dissipation of the elevator shaft is affected by many factors. The traditional model uses a fixed threshold, which cannot adapt to heat dissipation fluctuations, is prone to misjudgment, and cannot effectively distinguish between real faults and environmental heat dissipation problems. Summary of the invention

[0003] In view of the technical problems existing in the above-mentioned background technology, the present invention proposes a key component thermal anomaly detection method based on an improved isolation forest algorithm.

[0004] In order to achieve the above object, the technical solution adopted by the present invention comprises the following steps:

[0005] S1. First, deploy temperature sensors and thermal imagers on the surfaces of key elevator components;

[0006] S2. Use improved dynamic convolution kernel to achieve data fusion

[0007] S3, using the improved isolation forest algorithm to build a model, and using the original data to train the model, the improvements of the isolation forest are respectively to improve the adaptive segmentation strategy and to evaluate the multi-granularity path length aggregation;

[0008] S4. Finally, use the trained model to detect anomalies in the temperature data of key elevator components;

[0009] The specific implementation method of step S2 using the improved dynamic convolution kernel to achieve data fusion is:

[0010] S21, sort the collected temperature data in chronological order T=[T1,T2,...T n ], the thermal image collected by the thermal imager is recorded as I = [I1, I2, ... I m ], where n is the length of the time series, m is the number of frames of the thermal image, and the thermal image is normalized;

[0011] S22. Next, construct the reinforcement learning environment for the dynamic convolution kernel, and use the currently processed temperature data segment and thermal image frame as the state S t , and the action a t represents the parameters of the dynamic convolution kernel;

[0012] S23. Design the reward function, which is defined as:

[0013] where α1, α2, α3, α4 are weight coefficients, FD(F), CP(a t ), CA(V t ,L t ,F) are the accuracy rate, feature diversity metric, complexity of the convolution kernel corresponding to the action, and operating state of the key components of the elevator respectively;

[0014] S24. Then perform reinforcement learning training to learn to select the optimal dynamic convolution kernel parameters in different states;

[0015] S25. At each time step t, determine the convolution kernel parameters for feature extraction of temperature data and thermal image data according to the learned action;

[0016] S26. Finally, fuse the extracted data features.

[0017] Preferably, the key parts in step S1 include the tractor, braking system, control cabinet, cable and guide rail.

[0018] Preferably, the dynamic convolution kernel and parameters in step S22 include the size, stride and padding of the convolution kernel.

[0019] Preferably, CA(V t ,L t ,F) in step S23 represents the fitness of feature F to the current running speed V t and load condition L t , and the specific calculation method is: where, Correlation(F,V t ,L t ) is a measure of the correlation between feature F and running speed V t and load L t , and is obtained through the Pearson correlation coefficient.

[0020] Preferably, the specific fusion method in step S26 is an attention-based weighted fusion method, and the attention weights ω T of the thermal image data features and temperature data are calculated respectively through the Softmax function, ωI , and then perform weighted fusion to obtain the final feature where represents element-wise multiplication.

[0021] Preferably, the implementation of the improved isolation forest in step S3 is as follows:

[0022] S31. First, initialize the isolation forest model;

[0023] S32. Then, for each feature dimension in the sample subset, calculate the data density by adapting kernel density estimation: where h is the bandwidth and H is the Epanechnikov kernel function;

[0024] S33. On the selected feature dimension, find the splitting point s that maximizes the density difference of the subsets after splitting * , that is, where s - , s + represent the left and right neighborhoods of the splitting point, respectively;

[0025] S34. For the left and right child nodes obtained by splitting, repeat the splitting step until the depth of the tree reaches the preset maximum depth;

[0026] S35. According to the set number of trees, construct multiple isolation trees to form a complete isolation forest model;

[0027] S36. Finally, improve the anomaly score formula: where L(x) is the path length, c(n) is the normalization factor, is the depth weighting function, where d0 is the preset depth.

[0028] Preferably, after using the improved isolation forest algorithm for modeling and before anomaly detection, there are also steps of heat dissipation modeling and dynamic threshold calibration:

[0029] Construct a heat dissipation model for the elevator shaft, and define the heat dissipation coefficient h vent as: where A open is the area of the shaft ventilation opening, V shaft is the volume of the shaft, ΔP airflow is the intensity of natural convection driven by the pressure difference, and k1, k2 are experimentally calibrated coefficients;

[0030] And introduce a time integral term to correct the temperature threshold T thres , and the calculation formula is: where P heat (τ) is the heat generation power of the component.

[0031] Preferably, adversarial samples with heat dissipation noise are injected into the improved isolation forest model, and the calculation formula is as follows: where T real is the true temperature, ε is the perturbation intensity, is the sign of the gradient of the model loss function with respect to the temperature data T.

[0032] Compared with the prior art, the advantages and positive effects of the present invention are as follows. Detailed implementation manners

[0033] In order to more clearly understand the above objects, features, and advantages of the present invention, the present invention will be further described below with reference to the embodiments. It should be noted that, without conflict, the embodiments of the present application and the features in the embodiments may be combined with each other.

[0034] In the following description, many specific details are set forth in order to fully understand the present invention. However, the present invention may be implemented in other ways different from those described herein. Therefore, the present invention is not limited by the limitations of the specific embodiments disclosed in the following specification.

[0035] Embodiment: In today's cities with high-rise buildings everywhere, elevators, as an important tool for vertical transportation, their safety is of crucial importance. Key components of elevators, such as traction machines, braking systems, control cabinets, cables, and guide rails, are extremely prone to cause safety accidents due to heat generation during long-term operation. According to incomplete statistics, faults caused by thermal anomalies of elevator components account for a considerable proportion of the total faults, seriously affecting the safety of passengers' lives and the normal operation efficiency of elevators. In order to effectively solve the defects existing in the traditional elevator thermal anomaly detection method, the present invention proposes a thermal anomaly detection method for key components based on an improved isolation forest algorithm.

[0036] First, high-precision temperature sensors and high-resolution thermal imagers are deployed on the surfaces of the traction machine, braking system, control cabinet, cables, and guide rails of the elevator. These devices are reasonably distributed in the heat-prone areas of the key components to ensure accurate collection of temperature data and thermal image data.

[0037] Considering that the traditional data fusion method is relatively simple, usually directly splicing temperature data and thermal image data, or performing feature extraction using a fixed convolution kernel and then fusing. The parameters of this fixed convolution kernel are preset and will not be adjusted according to the dynamic changes of the data. This connection method does not consider the dynamic characteristics of the data, which will lead to incomplete feature extraction. In addition, simple data splicing or fixed convolution kernel fusion methods do not fully explore the potential relationships between data and cannot effectively integrate the complementary information of temperature data and thermal image data, resulting in insufficient characterization ability of the fused data features for thermal anomalies, thereby affecting the accuracy and timeliness of thermal anomaly detection. And the present invention adopts the method of combining an improved dynamic convolution kernel with reinforcement learning. Construct a reinforcement learning environment, take temperature data segments and thermal image frames as states, and take the parameters of the dynamic convolution kernel as actions. By designing a reward function that includes factors such as accuracy, feature diversity metric, convolution kernel complexity, and the operating state of key elevator components, guide the algorithm to learn and select the optimal dynamic convolution kernel parameters in different states. Sort the collected temperature data in chronological order T = [T1, T2,... T n , and the thermal images collected by the thermal imager are denoted as I = [I1, I2,... I m , where n is the length of the time series and m is the number of thermal image frames, and normalize the thermal images; then construct the reinforcement learning environment of the dynamic convolution kernel, and take the currently processed temperature data segment and thermal image frame as state S t , and action a t represents the parameters of the dynamic convolution kernel, including the size, stride, and padding of the convolution kernel. Then design the reward function, and the reward function is defined as: where α1, α2, α3, α4 are weight coefficients, FD(F), CP(a t ), CA(V t , L t , F) are the accuracy, feature diversity metric, complexity of the convolution kernel corresponding to the action, and the operating state of the key elevator components respectively, where the feature F's fitness for the current running speed V t and load condition L t is calculated as follows: where, Correlation(F, V t , L t ) is a measure of the correlation between feature F and running speed V t and load L tThe correlation is obtained through the Pearson correlation coefficient. Then, reinforcement learning training is carried out to learn to select the optimal dynamic convolution kernel parameters in different states. At each time step t, the convolution kernel parameters are determined according to the learned actions for feature extraction of temperature data and thermal image data. Finally, the extracted data features are weighted and fused based on attention, and the attention weights ω T , ω I are calculated through the Softmax function, and then weighted fusion is carried out to obtain the final features where represents element-wise multiplication. Through this improvement, it can adapt to the changes in the elevator operation state, dynamically adjust the convolution kernel parameters, and accurately extract the temperature and thermal image data features under different working conditions. For example, when the elevator is lightly loaded and heavily loaded, the most suitable convolution kernel parameters can be found to comprehensively capture the changes in the thermal state of components. In addition, the dynamic convolution kernel obtained based on reinforcement learning training can deeply explore the potential relationships between data, improve the characterization ability of the fused data features for thermal anomalies, and enhance the accuracy and timeliness of thermal anomaly detection.

[0038] Then, for traditional anomaly detection in a similar scenario, if the Isolation Forest algorithm is used, its implementation principle is to construct isolation trees based on randomly selected features and split points. During the construction process, samples are randomly selected from the dataset to build trees. For each node, a feature dimension is randomly selected, and then a split point is randomly selected within the value range of this dimension to divide the dataset into two subsets, the left and the right. This process continues until the preset tree depth is reached. Finally, the anomaly degree of the sample is evaluated by calculating the path length of the sample in the tree. The shorter the path length, the more likely the sample is an outlier. However, this traditional method does not consider the distribution characteristics of the data and the correlation between different features. On the other hand, only evaluating anomalies based on the simple path length does not fully utilize multi-granularity information and is prone to missing some potential anomaly situations, greatly reducing the accuracy and reliability of anomaly detection. The improvement of the present invention lies in the adaptive segmentation strategy. By calculating the data density through kernel density estimation, the split point that maximizes the density difference of the subsets after segmentation is found on the selected feature dimension. Compared with the traditional random selection of split points, this method can better segment according to the distribution characteristics of the data itself and accurately capture the key features. In terms of multi-granularity path length aggregation evaluation, the anomaly score formula is improved, and a depth weighting function is introduced to comprehensively consider multiple factors such as path length and normalization factor, and evaluate the data from multiple granularities to comprehensively capture the anomaly patterns of the data. First, initialize the Isolation Forest model; then, for each feature dimension in the sample subset, calculate the data density by adapting kernel density estimation: where h is the bandwidth and H is the Epanechnikov kernel function; on the selected feature dimension, find the split point s that maximizes the density difference of the subsets after segmentation * , that is, where s - and s + represent the left and right domains of the segmentation point respectively; for the left and right child nodes obtained by segmentation, repeat the segmentation step until the depth of the tree reaches the preset maximum depth; according to the set number of trees, construct multiple isolated trees to form a complete isolated forest model; finally, improve the anomaly score formula: where L(x) is the path length, c(n) is the normalization factor, is the depth-weighted function, where d0 is the preset depth. This improvement method enhances the stability and generalization ability of the model, reducing the cases of misjudgment and missed judgment. Data fusion provides feature data that is optimized and highly representative of thermal anomalies for the isolated forest algorithm, making the algorithm training more targeted and improving the recognition ability of abnormal data; the isolated forest algorithm then effectively analyzes the fused data to accurately judge the thermal anomaly situation.

[0039] In addition, considering that the elevator shaft is a semi-closed space, the heat dissipation efficiency is easily affected by factors such as environmental temperature and blockage of ventilation openings. The traditional model is prone to misjudgment due to heat dissipation fluctuations under a fixed threshold. Therefore, after modeling, steps of heat dissipation modeling and dynamic threshold calibration are also carried out: construct an elevator shaft heat dissipation model, and define the heat dissipation coefficient h vent as: where A open is the area of the shaft ventilation opening, V shaft is the volume of the shaft, ΔP airflow is the natural convection intensity driven by the pressure difference, and k1, k2 are experimentally calibrated coefficients; and introduce a time integral term to correct the temperature threshold T thres , and the calculation formula is: where P heat (τ) is the heat generation power of the component. And inject adversarial samples with heat dissipation noise into the improved isolated forest model, and the calculation formula is: where T real is the true temperature, ε is the perturbation intensity, It is the sign of the gradient of the model loss function with respect to the temperature data T. By simulating heat dissipation anomalies (such as high summer temperatures and dust accumulation at ventilation openings) with adversarial samples, the model learns to distinguish real component failures from environmental heat dissipation failures during the training phase. For example, when the ventilation efficiency in the hoistway decreases, the adversarial samples generate simulated temperature rise data, forcing the model to dynamically adjust the threshold to avoid false alarms about the temperature rise of normal components. In an embodiment, under high summer temperatures, the model accurately distinguishes overheating of cable joints (real anomaly) from environmental temperature rise (non-anomaly) through adversarial training. Introducing a time integration term combined with the heat generation power of the component to correct the temperature threshold in this step can adapt to heat generation changes in real time and avoid misjudgment caused by heat dissipation fluctuations. Injecting heat dissipation noise adversarial samples into the model to simulate heat dissipation anomaly situations enables the model to learn to distinguish real failures from environmental heat dissipation problems during training. For example, it can accurately determine real anomalies such as overheating of cable joints in high summer temperatures.

[0040] Finally, the trained model is used to detect anomalies in the temperature data of elevator key components, providing a more reliable anomaly detection guarantee for the safe operation of the elevator and enhancing the efficiency of the entire thermal anomaly detection.

[0041] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any person skilled in the art may use the disclosed technical content to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change, and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A key component thermal anomaly detection method based on an improved isolation forest algorithm, characterized in that: The following steps are involved: S1. First, deploy temperature sensors and thermal imagers on the surfaces of key elevator components; S2, using improved dynamic convolution kernel to achieve data fusion; S3, using the improved isolation forest algorithm to build a model, and using the original data to train the model, the improvements of the isolation forest are respectively to improve the adaptive segmentation strategy and to evaluate the multi-granularity path length aggregation; S4. Finally, use the trained model to detect anomalies in the temperature data of key elevator components; The specific implementation method of step S2 using the improved dynamic convolution kernel to achieve data fusion is: S21, sort the collected temperature data in chronological order T=[T1,T2,...T n ], the thermal image collected by the thermal imager is recorded as I = [I1, I2, ... I m ], where n is the length of the time series, m is the number of frames of the thermal image, and the thermal image is normalized; S22, then build a reinforcement learning environment for the dynamic convolution kernel, and use the currently processed temperature data fragment and thermal image frame as state S t , action a t Represents the parameters of the dynamic convolution kernel; S23. Design a reward function. The reward function is defined as: Among them, α1, α2, α3, α4 are weight coefficients, CP(a t ), CA(V t ,L t ,F) are respectively the accuracy, feature diversity metric, the complexity of the convolution kernel corresponding to the action, and the operating status of the key components of the elevator; S24, then conduct reinforcement learning training to learn to select the optimal dynamic convolution kernel parameters under different states; S25, at each time step t, determining convolution kernel parameters for temperature data and thermal image data feature extraction according to the learned actions; S26. Finally, the extracted data features are fused.

2. According to claim 1, a key component thermal anomaly detection method based on improved isolation forest algorithm is characterized in that: The key parts in step S1 include the traction machine, the braking system, the control cabinet, the cables and the guide rails.

3. According to claim 1, a key component thermal anomaly detection method based on improved isolation forest algorithm is characterized in that: The dynamic convolution kernel and parameters in step S22 include the size, step size and padding of the convolution kernel.

4. The key component thermal anomaly detection method based on improved isolation forest algorithm according to claim 1 is characterized in that: In step S23, CA(V t ,L t ,F) represents the effect of feature F on the current running speed V t and load condition L t The fitness of is calculated as follows: Among them, Correlation(F,Vt,Lt) is a measure of the relationship between feature F and running speed V t and load L t The correlation was obtained by Pearson's correlation coefficient.

5. The key component thermal anomaly detection method based on improved isolation forest algorithm according to claim 1 is characterized in that: The specific fusion method in step S26 is a weighted fusion method based on attention, and the attention weights ω of the thermal image data features and the temperature data are calculated respectively by the Softmax function. T ,ω I , and then perform weighted fusion to obtain the final feature in Stands for element-wise multiplication.

6. The key component thermal anomaly detection method based on improved isolation forest algorithm according to claim 1 is characterized in that: The implementation of improving the isolation forest in step S3 is as follows: S31, first initialize the isolation forest model; S32. Then, for each feature dimension in the sample subset, kernel density estimation is used to calculate the data density: Where h is the bandwidth and H is the Epanechnikov kernel function; S33. On the selected feature dimension, find the segmentation point s that maximizes the density difference of the segmented subsets. * ,Right now, where s - ,s + Respectively represent the left and right areas of the split point; S34, repeating the segmentation step for the left and right child nodes obtained by segmentation until the depth of the tree reaches a preset maximum depth; S35. construct multiple isolated trees according to the set number of trees to form a complete isolation forest model; S36. Finally, improve the anomaly score formula: Where L(x) is the path length, c(n) is the normalization factor, is the depth weighted function, where d0 is the preset depth.

7. According to the key component thermal anomaly detection method based on the improved isolation forest algorithm of claim 1, after using the improved isolation forest algorithm for modeling and before anomaly detection, there are also steps of heat dissipation modeling and dynamic threshold calibration: Construct the heat dissipation model of the elevator shaft and define the heat dissipation coefficient h vent for: Among them A open is the shaft ventilation opening area, V shaft is the well volume, ΔP airflow is the natural convection intensity driven by air pressure difference, k1 and k2 are experimental calibration coefficients; And introduce the time integral term to correct the temperature threshold T thres , the calculation formula is: Where P heat (τ) is the heat generation power of the component.

8. According to the key component thermal anomaly detection method based on the improved isolation forest algorithm of claim 7, the adversarial sample with heat dissipation noise is injected into the improved isolation forest model, and the calculation formula is: Where T real is the true temperature, ε is the disturbance intensity, is the gradient sign of the model loss function with respect to the temperature data T.

Images