An intelligent elevator abnormal state detection method

Abstract

The application belongs to the field of elevator detection, and particularly discloses an intelligent elevator abnormal state detection method, which comprises the following steps: multi-source sensing data acquisition, data preprocessing, multi-source signal time-frequency domain feature extraction, mixed classification model construction, and health state classification. The application achieves all-around real-time monitoring of the running state of the elevator permanent magnet synchronous motor drive system by deploying a multi-source sensor array, extracts rich time-frequency domain features from multi-dimensional signals through discrete wavelet transform, comprehensively reflects the electrical, mechanical and thermal states of the motor, constructs a mixed classification model combining Rhizostoma optimization algorithm and random forest, and automatically optimizes the hyperparameters of the random forest through the Rhizostoma optimization algorithm, so as to realize fine grading of the motor health state.

Description

Technical Field

[0001] This invention relates to the field of elevator inspection, specifically to an intelligent method for detecting abnormal elevator conditions. Background Technology

[0002] As an indispensable vertical transportation tool in modern high-rise buildings, the safe and stable operation of elevators is directly related to the convenience of daily travel and the safety of people's lives. Due to its high efficiency, low noise and good low-speed high torque characteristics, permanent magnet synchronous motors have become the mainstream drive solution for modern elevator traction systems. If faults are not detected in time during long-term continuous operation of permanent magnet synchronous motors, it will lead to abnormal elevator operation and even safety accidents.

[0003] Traditional fault detection methods rely heavily on manual periodic inspections or post-incident maintenance, lacking the ability to monitor the elevator's operating status in real time and continuously. Manual inspection is not only inefficient and costly, but also struggles to detect subtle early signs of faults, often only being discovered when the fault has progressed to a more serious stage. Existing monitoring schemes based on single sensors are easily affected by noise in complex operating conditions, making it difficult to comprehensively reflect the multi-dimensional operating status of the elevator drive system, resulting in insufficient reliability of the detection results. Among existing fault diagnosis methods based on machine learning, traditional models such as support vector machines have limited processing capabilities for large-scale data and are sensitive to parameter selection, while deep learning methods, although possessing high feature extraction capabilities, have high computational complexity and long training times, making it difficult to meet the low-latency requirements of real-time elevator monitoring. Summary of the Invention

[0004] To address the above issues and overcome the shortcomings of existing technologies, this invention provides an intelligent elevator abnormality detection method. Addressing the technical problems of existing elevator fault detection methods, such as difficulty in achieving real-time continuous monitoring and delayed fault detection, this invention achieves comprehensive real-time monitoring of the elevator permanent magnet synchronous motor drive system's operating status by deploying a multi-source sensor array. Combined with efficient signal processing and machine learning classification models, it can detect abnormalities and issue early warnings before the motor's health deteriorates to a critical level, realizing a shift from reactive maintenance to predictive maintenance. Furthermore, addressing the technical problems of limited information sources and insufficient detection reliability in existing single-sensor-based monitoring schemes, this invention integrates multi-source sensor data and utilizes discrete wavelet... The transformation extracts rich time-frequency domain features from multi-dimensional signals, comprehensively reflecting the electrical, mechanical, and thermal states of the motor, significantly improving the reliability and robustness of fault detection. Addressing the technical problems of existing machine learning models, such as limited ability to process large-scale data, coarse classification granularity, and reliance on manual experience in setting hyperparameters, this invention constructs a hybrid classification model combining the Rhizostoma optimization algorithm and random forest. The Rhizostoma optimization algorithm automatically optimizes the hyperparameters of the random forest globally, overcoming the limitations of manual parameter tuning. The random forest integrates the voting results of multiple decision trees, exhibiting strong anti-overfitting ability and robustness to noisy data, enabling refined classification of the motor's health status.

[0005] The technical solution adopted by this invention is as follows: This invention provides an intelligent method for detecting abnormal elevator conditions, which specifically includes the following steps:

[0006] Step S1: Multi-source sensor data acquisition. A multi-source sensor array is deployed on the elevator permanent magnet synchronous drive motor to collect multi-dimensional status data during motor operation in real time. The multi-source sensor array includes: current sensor, vibration sensor, temperature sensor, and Hall effect sensor. Each sensor collects data synchronously at a preset sampling frequency. One multi-dimensional status data is used as a sample, and one sensor data is used as a feature variable. The actual health status of the motor is marked as a label.

[0007] Step S2: Data preprocessing, preprocessing the multidimensional state data, including missing value handling, outlier removal, data normalization and feature selection, to obtain input features;

[0008] Step S3: Time-frequency domain feature extraction of multi-source signals. Discrete wavelet transform is used to extract time-frequency domain features of the input features to obtain the time-frequency domain features of the signal in different frequency bands.

[0009] Step S4: Construct a hybrid classification model. Based on the Rhizostoma optimization algorithm and random forest, construct the RO-RF hybrid classification model to classify and identify the health status of the elevator permanent magnet synchronous drive motor. The health status includes three categories: normal state, edge state and critical state.

[0010] Step S5: Health status classification. The time-frequency domain feature vectors extracted in step S3 are divided into training set and test set according to a preset ratio, and the hybrid classification model constructed in step S4 is used for classification.

[0011] Further, step S2, data preprocessing, specifically includes the following sub-steps:

[0012] Step S21: Clean the multidimensional state data, identify and process missing values ​​in the data. For samples with a missing value ratio lower than a preset threshold, use linear interpolation to fill in the missing values. For samples with a missing value ratio higher than the preset threshold, delete them.

[0013] Step S22: Use the interquartile range (IQR) method to remove outliers. Specifically, calculate the first and third quartiles of each feature variable. The IQR is defined as the difference between the third and first quartiles. Data that are less than Q1−1.5×IQR or greater than Q3+1.5×IQR are identified as outliers and removed. Q1 is the first quartile, Q3 is the third quartile, and IQR is the IQR.

[0014] Step S23: For consecutive missing data segments, a data repair method based on similar sample matching is adopted to find the complete sample with the most similar features to the missing sample in all multidimensional state data, and the corresponding feature value of the complete sample is used as the repair value of the missing sample.

[0015] Step S24: The multidimensional state data is normalized using the min-max normalization method, mapping the numerical range of all feature variables to the interval [0, 1]. The formula used is as follows: ;

[0016] In the formula, These are the original eigenvalues. These are the normalized eigenvalues. It is the minimum value of this feature variable among all samples. It is the maximum value of this feature variable across all samples;

[0017] Step S25: The Pearson correlation coefficient method is used to measure the correlation between each feature variable and the motor health status label, and the feature variable with the strongest correlation to the abnormal state of the motor is selected as the input feature of the subsequent prediction model. The formula for calculating the Pearson correlation coefficient is as follows: ;

[0018] In the formula, It is a characteristic variable and tags The Pearson correlation coefficient between them It is the total number of samples. It is a sample index. It is the first The values ​​of feature variable XX in each sample. It is the first The values ​​of label YY in each sample;

[0019] Based on the calculation results of the Pearson correlation coefficient, feature variables with an absolute value of correlation coefficient greater than a preset threshold with the motor health status label are selected as input features for the prediction model.

[0020] Further, in step S3, the time-frequency domain feature extraction of the multi-source signal is performed by using discrete wavelet transform to decompose the input features into low-frequency approximate components and high-frequency detail components through multi-scale decomposition. The decomposition process of the discrete wavelet transform is as follows:

[0021] The discrete signal in the input features is decomposed into multiple scales, and the low-frequency approximation coefficients and high-frequency detail coefficients of each decomposition level are calculated using the following recursive formulas: ; ;

[0022] In the formula, These are low-frequency approximation coefficients. It is a high-frequency detail factor. It is a decomposition level index. It is the coefficient position index. It is the index of the summation variable. These are the coefficients of the low-pass filter. These are the coefficients of the high-pass filter. These are the low-frequency approximation coefficients of the previous level of decomposition;

[0023] The discrete signal is fully reconstructed using the detail coefficients of each layer and the final layer approximation coefficients to obtain the reconstructed discrete signal. The reconstruction formula used is as follows: ;

[0024] In the formula, It is to reconstruct discrete signals. It is the maximum number of decomposition levels. It is the first Approximation coefficients of the layer It is the first The detail factor of the layer, It is a scaling function. It is a wavelet function;

[0025] By using discrete wavelet transform, the following time-frequency domain features are extracted from the detail coefficients of each layer: energy value, energy ratio, entropy value, mean, standard deviation, peak factor and kurtosis of each layer detail coefficient, which constitute the original feature vector.

[0026] Further, step S4, constructing a hybrid classification model, specifically includes the following sub-steps:

[0027] Step S41: Construct a random forest base classifier. The random forest consists of N decision trees. Each decision tree is trained on a subset of the training data obtained by sampling from the bootstrap dataset. For a new input sample, calculate the output of the random forest using the following formula: ;

[0028] In the formula, It is the final prediction category of the random forest. These are category labels: 0 represents a normal state, 1 represents an edge state, and 2 represents a critical state. It is the total number of decision trees. It is the index of the decision tree. It is the first Decision trees for input samples Prediction categories, It is an indicator function that takes the value 1 when the condition inside the parentheses is true, and takes the value 0 otherwise.

[0029] When splitting a node, each decision tree randomly selects a subset of features from all time-frequency domain features, and chooses the best splitting feature from it. The Gini impurity is used as the splitting criterion for node splitting. The formula for calculating the Gini impurity for any node is as follows: ;

[0030] In the formula, It is a node The impurity of the gin. It is a node Belongs to the category The proportion of samples, the feature and splitting threshold that minimize the weighted sum of Gini impurities of child nodes after splitting when splitting nodes;

[0031] Step S42: Use the Rhizostoma optimization algorithm to globally optimize the hyperparameters of the random forest. The hyperparameters to be optimized include the number of decision trees, the maximum depth of the decision trees, the minimum number of samples required for node splitting, the minimum number of samples for leaf nodes, and the maximum number of features.

[0032] The Rhizostoma optimization algorithm simulates the foraging behavior of jellyfish in the ocean and achieves global optimization through the following steps:

[0033] Step S421: Population initialization. R candidate solutions are randomly generated. The position vector of the j-th candidate solution represents the hyperparameter to be optimized. The initial values ​​of the candidate solutions in each dimension are randomly generated in the corresponding search space, using the following formula: ;

[0034] In the formula, It is the first The candidate solution at the th... Initial values ​​in each dimension and They are the first The lower and upper bounds of the search in each dimension. It is a random number uniformly distributed within the interval (0, 1);

[0035] Step S422: Boundary constraint handling. When a candidate solution exceeds the search space boundary during the search process, it is remapped to the inside of the boundary. The formula used is as follows: ;

[0036] In the formula, These are the position values ​​after boundary constraint processing. It is the current position value;

[0037] Step S423: Position Update. The position update of the candidate solution is achieved through Levy flight. The position update formula is as follows: ;

[0038] In the formula, This is the updated location. It is the step size, which follows a Levy distribution. It is a uniformly distributed random number within the interval (0, 1), with a step size of Sampling is performed from the Cauchy distribution using the following formula: ;

[0039] In the formula, It is the probability density function of the Cauchy distribution. It is a temperature parameter that controls the amplitude of step size fluctuation;

[0040] Step S424: Acceptance probability. The simulated annealing mechanism is used to determine whether to accept the updated position. The formula used is as follows: ;

[0041] In the formula, It is the probability of acceptance. It is the difference between the objective function values ​​before and after the update, that is, the difference in the classification error rate of the random forest on the validation set. This is the current temperature parameter;

[0042] Step S425: Motion control factor. This step introduces the exploration and development phase of the dynamic adjustment algorithm for the motion control factor. The formula used is as follows: ;

[0043] In the formula, It is the first The motion control factor for the next iteration. It is the current iteration number. It is the maximum number of iterations;

[0044] When the motion control factor is less than the threshold, the algorithm enters the development phase and performs a fine search near the current optimal solution; otherwise, it enters the exploration phase and searches for new potential optimal regions over a wide range.

[0045] Step S426: Iterative optimization. Repeat the above position update, boundary constraint processing and acceptance probability judgment until the maximum number of iterations is reached, and output the optimal hyperparameter combination that minimizes the classification error rate of the random forest.

[0046] Step S43: Retrain the random forest model using the optimal hyperparameter combination.

[0047] Further, step S5, health status classification, specifically includes the following sub-steps:

[0048] Step S51: Divide the preprocessed dataset into training and test sets according to the proportions, and use a stratified sampling strategy to make the proportions of each category of samples in the training and test sets consistent with the original dataset.

[0049] Step S52: Using the optimal hyperparameter combination obtained from the Rhizostoma optimization algorithm in step S42, train the random forest model on the training set. During training, each decision tree grows independently based on the sub-training set sampled by Bootstrap. When a node splits, select the best splitting feature and splitting threshold from a randomly selected feature subset.

[0050] Step S53: Evaluate the classification performance of the trained hybrid classification model on the test set, using precision, recall, F1 score and accuracy as evaluation metrics.

[0051] Step S54: Deploy the trained hybrid classification model to the elevator real-time monitoring system to perform online abnormal state detection on the real-time collected multi-source sensor data. The hybrid model outputs the current motor health status category and triggers the corresponding early warning mechanism according to the output category. When the detection result is an edge state, a maintenance reminder is issued, and when the detection result is a critical state, a fault alarm is issued.

[0052] The beneficial effects achieved by the present invention using the above solution are as follows:

[0053] (1) In view of the technical problems that existing elevator fault detection methods are difficult to achieve real-time continuous monitoring and fault detection is delayed, this invention realizes all-round real-time monitoring of the operating status of elevator permanent magnet synchronous motor drive system by deploying a multi-source sensor array. Combined with efficient signal processing and machine learning classification model, it can detect abnormalities and issue warnings in time before the motor health status deteriorates to the critical point, realizing the transformation from post-maintenance to predictive maintenance.

[0054] (2) In view of the technical problems of limited information sources and insufficient detection reliability of existing monitoring schemes based on single sensors, this invention integrates multi-source sensor data and extracts rich time-frequency domain features from multi-dimensional signals through discrete wavelet transform, comprehensively reflecting the electrical, mechanical and thermal state of the motor, and significantly improving the reliability and robustness of fault detection.

[0055] (3) In view of the technical problems of existing machine learning models having limited ability to process large-scale data, coarse classification granularity, and hyperparameters relying on manual experience setting, this invention constructs a hybrid classification model that combines the Rhizostoma optimization algorithm with random forest. The Rhizostoma optimization algorithm is used to automatically optimize the hyperparameters of the random forest globally, overcoming the limitations of manual parameter tuning. The random forest integrates the voting results of multiple decision trees, which has strong anti-overfitting ability and robustness to noisy data, and realizes the fine classification of motor health status. Attached Figure Description

[0056] Figure 1 The present invention provides a flowchart of the steps of an intelligent elevator abnormality detection method.

[0057] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0058] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0059] Example 1, see Figure 1 This invention provides an intelligent method for detecting abnormal elevator conditions, which specifically includes the following steps:

[0060] Step S1: Multi-source sensor data acquisition. A multi-source sensor array is deployed on the elevator permanent magnet synchronous drive motor to collect multi-dimensional status data during motor operation in real time. The multi-source sensor array includes: a current sensor for acquiring the three-phase stator current signal of the motor; a vibration sensor for acquiring the vibration acceleration signal of the motor in the X, Y, and Z axes; a temperature sensor for acquiring the temperature signal of the stator winding of the motor; and a Hall effect sensor for acquiring the rotor position signal of the motor. Each sensor collects data synchronously at a preset sampling frequency. One multi-dimensional status data is used as a sample, and one sensor data is used as a feature variable to label the actual health status of the motor.

[0061] Step S2: Data preprocessing, preprocessing the multidimensional state data, including missing value handling, outlier removal, data normalization and feature selection, to obtain input features;

[0062] Step S3: Time-frequency domain feature extraction of multi-source signals. Discrete wavelet transform is used to extract time-frequency domain features of the input features to obtain the time-frequency domain features of the signal in different frequency bands.

[0063] Step S4: Construct a hybrid classification model. Based on the Rhizostoma optimization algorithm and random forest, construct the RO-RF hybrid classification model to classify and identify the health status of the elevator permanent magnet synchronous drive motor. The health status includes three categories: normal state, edge state and critical state.

[0064] Step S5: Health status classification. The time-frequency domain feature vectors extracted in step S3 are divided into training set and test set according to a preset ratio, and the hybrid classification model constructed in step S4 is used for classification.

[0065] Example 2, see Figure 1 This embodiment is based on the above embodiment. Step S2, data preprocessing, specifically includes the following sub-steps:

[0066] Step S21: Clean the multidimensional state data, identify and process missing values ​​in the data. For samples with a missing value ratio lower than a preset threshold, use linear interpolation to fill in the missing values. For samples with a missing value ratio higher than the preset threshold, delete them.

[0067] Step S22: Use the interquartile range (IQR) method to remove outliers. Specifically, calculate the first and third quartiles of each feature variable. The IQR is defined as the difference between the third and first quartiles. Data that are less than Q1−1.5×IQR or greater than Q3+1.5×IQR are identified as outliers and removed. Q1 is the first quartile, Q3 is the third quartile, and IQR is the IQR.

[0068] Step S23: For consecutive missing data segments, a data repair method based on similar sample matching is adopted to find the complete sample with the most similar features to the missing sample in all multidimensional state data, and the corresponding feature value of the complete sample is used as the repair value of the missing sample.

[0069] Step S24: The multidimensional state data is normalized using the min-max normalization method, mapping the numerical range of all feature variables to the interval [0, 1]. The formula used is as follows: ;

[0070] In the formula, These are the original eigenvalues. These are the normalized eigenvalues. It is the minimum value of this feature variable among all samples. It is the maximum value of this feature variable across all samples;

[0071] Step S25: The Pearson correlation coefficient method is used to measure the correlation between each feature variable and the motor health status label, and the feature variable with the strongest correlation to the abnormal state of the motor is selected as the input feature of the subsequent prediction model. The formula for calculating the Pearson correlation coefficient is as follows: ;

[0072] In the formula, It is a characteristic variable and tags The Pearson correlation coefficient between them It is the total number of samples. It is a sample index. It is the first The values ​​of feature variable XX in each sample. It is the first The values ​​of label YY in each sample;

[0073] Based on the calculation results of the Pearson correlation coefficient, feature variables with an absolute value of correlation coefficient with the motor health status label greater than the preset threshold of 0.3 are selected as input features for the prediction model.

[0074] Example 3, see Figure 1 This embodiment is based on the above embodiment. In step S3, multi-source signal time-frequency domain feature extraction is performed by using discrete wavelet transform to decompose the input features into low-frequency approximate components and high-frequency detail components through multi-scale decomposition. The decomposition process of the discrete wavelet transform is as follows:

[0075] The discrete signal in the input features is decomposed into multiple scales, and the low-frequency approximation coefficients and high-frequency detail coefficients of each decomposition level are calculated using the following recursive formulas: ; ;

[0076] In the formula, These are low-frequency approximation coefficients. It is a high-frequency detail factor. It is a decomposition level index. It is the coefficient position index. It is the index of the summation variable. These are the coefficients of the low-pass filter. These are the coefficients of the high-pass filter. These are the low-frequency approximation coefficients of the previous level of decomposition;

[0077] The discrete signal is fully reconstructed using the detail coefficients of each layer and the final layer approximation coefficients to obtain the reconstructed discrete signal. The reconstruction formula used is as follows: ;

[0078] In the formula, It is to reconstruct discrete signals. It is the maximum number of decomposition levels. It is the first Approximation coefficients of the layer It is the first The detail factor of the layer, It is a scaling function. It is a wavelet function;

[0079] By using discrete wavelet transform, the following time-frequency domain features are extracted from the detail coefficients of each layer: energy value, energy ratio, entropy value, mean, standard deviation, peak factor and kurtosis of each layer detail coefficient, which constitute the original feature vector.

[0080] Example 4, see Figure 1 This embodiment is based on the above embodiment. Step S4, constructing a hybrid classification model, specifically includes the following sub-steps:

[0081] Step S41: Construct a random forest base classifier. The random forest consists of N decision trees. Each decision tree is trained on a subset of the training data obtained by sampling from the bootstrap dataset. For a new input sample, calculate the output of the random forest using the following formula: ;

[0082] In the formula, It is the final prediction category of the random forest. These are category labels: 0 represents a normal state, 1 represents an edge state, and 2 represents a critical state. It is the total number of decision trees. It is the index of the decision tree. It is the first Decision trees for input samples Prediction categories, It is an indicator function that takes the value 1 when the condition inside the parentheses is true, and takes the value 0 otherwise.

[0083] When splitting a node, each decision tree randomly selects a subset of features from all time-frequency domain features, and chooses the best splitting feature from it. The Gini impurity is used as the splitting criterion for node splitting. The formula for calculating the Gini impurity for any node is as follows: ;

[0084] In the formula, It is a node The impurity of the gin. It is a node Belongs to the category The proportion of samples, the feature and splitting threshold that minimize the weighted sum of Gini impurities of child nodes after splitting when splitting nodes;

[0085] Step S42: Use the Rhizostoma optimization algorithm to globally optimize the hyperparameters of the random forest. The hyperparameters to be optimized include the number of decision trees, the maximum depth of the decision trees, the minimum number of samples required for node splitting, the minimum number of samples for leaf nodes, and the maximum number of features.

[0086] The Rhizostoma optimization algorithm simulates the foraging behavior of jellyfish in the ocean and achieves global optimization through the following steps:

[0087] Step S421: Population initialization. R candidate solutions are randomly generated. The position vector of the j-th candidate solution represents the hyperparameter to be optimized. The initial values ​​of the candidate solutions in each dimension are randomly generated in the corresponding search space, using the following formula: ;

[0088] In the formula, It is the first The candidate solution at the th... Initial values ​​in each dimension and They are the first The lower and upper bounds of the search in each dimension. It is a random number uniformly distributed within the interval (0, 1);

[0089] Step S422: Boundary constraint handling. When a candidate solution exceeds the search space boundary during the search process, it is remapped to the inside of the boundary. The formula used is as follows: ;

[0090] In the formula, These are the position values ​​after boundary constraint processing. It is the current position value;

[0091] Step S423: Position Update. The position update of the candidate solution is achieved through Levy flight. The position update formula is as follows: ;

[0092] In the formula, This is the updated location. It is the step size, which follows a Levy distribution. It is a uniformly distributed random number within the interval (0, 1), with a step size of Sampling is performed from the Cauchy distribution using the following formula: ;

[0093] In the formula, It is the probability density function of the Cauchy distribution. It is a temperature parameter that controls the amplitude of step size fluctuation;

[0094] Step S424: Acceptance probability. The simulated annealing mechanism is used to determine whether to accept the updated position. The formula used is as follows: ;

[0095] In the formula, It is the probability of acceptance. It is the difference between the objective function values ​​before and after the update, that is, the difference in the classification error rate of the random forest on the validation set. This is the current temperature parameter;

[0096] Step S425: Motion control factor. This step introduces the exploration and development phase of the dynamic adjustment algorithm for the motion control factor. The formula used is as follows: ;

[0097] In the formula, It is the first The motion control factor for the next iteration. It is the current iteration number. It is the maximum number of iterations;

[0098] When the motion control factor is less than the threshold of 0.5, the algorithm enters the development stage and performs a fine search near the current optimal solution; otherwise, it enters the exploration stage and searches for new potential optimal regions over a wide range.

[0099] Step S426: Iterative optimization. Repeat the above position update, boundary constraint processing and acceptance probability judgment until the maximum number of iterations is reached, and output the optimal hyperparameter combination that minimizes the classification error rate of the random forest.

[0100] Step S43: Retrain the random forest model using the optimal hyperparameter combination.

[0101] Example 5, see Figure 1 This embodiment is based on the above embodiment. Step S5, health status classification, specifically includes the following sub-steps:

[0102] Step S51: Divide the preprocessed dataset into a training set and a test set in a ratio of 8:2, and use a stratified sampling strategy to make the proportion of each category of samples in the training set and the test set consistent with the original dataset.

[0103] Step S52: Using the optimal hyperparameter combination obtained from the Rhizostoma optimization algorithm in step S42, train the random forest model on the training set. During training, each decision tree grows independently based on the sub-training set sampled by Bootstrap. When a node splits, select the best splitting feature and splitting threshold from a randomly selected feature subset.

[0104] Step S53: Evaluate the classification performance of the trained hybrid classification model on the test set, using precision, recall, F1 score and accuracy as evaluation metrics.

[0105] Step S54: Deploy the trained hybrid classification model to the elevator real-time monitoring system to perform online abnormal state detection on the real-time collected multi-source sensor data. The hybrid model outputs the current motor health status category and triggers the corresponding early warning mechanism according to the output category. When the detection result is an edge state, a maintenance reminder is issued, and when the detection result is a critical state, a fault alarm is issued.

[0106] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0107] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0108] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. An intelligent method for detecting abnormal elevator conditions, characterized in that, Specifically, the following steps are included: Step S1: Multi-source sensor data acquisition. A multi-source sensor array is deployed on the elevator permanent magnet synchronous drive motor to collect multi-dimensional status data during motor operation in real time. The multi-source sensor array includes: current sensor, vibration sensor, temperature sensor, and Hall effect sensor. Each sensor collects data synchronously at a preset sampling frequency. One multi-dimensional status data is used as a sample, and one sensor data is used as a feature variable. The actual health status of the motor is marked as a label. Step S2: Data preprocessing, preprocessing the multidimensional state data, including missing value handling, outlier removal, data normalization and feature selection, to obtain input features; Step S3: Time-frequency domain feature extraction of multi-source signals. Discrete wavelet transform is used to extract time-frequency domain features of the input features to obtain the time-frequency domain features of the signal in different frequency bands. Step S4: Construct a hybrid classification model. Based on the Rhizostoma optimization algorithm and random forest, construct the RO-RF hybrid classification model to classify and identify the health status of the elevator permanent magnet synchronous drive motor. The health status includes three categories: normal state, edge state and critical state. Step S5: Health status classification. The time-frequency domain feature vectors extracted in step S3 are divided into training set and test set according to a preset ratio, and the hybrid classification model constructed in step S4 is used for classification.

2. The intelligent elevator abnormality detection method according to claim 1, characterized in that, Step S2, data preprocessing, specifically includes the following sub-steps: Step S21: Clean the multidimensional state data, identify and process missing values ​​in the data. For samples with a missing value ratio lower than a preset threshold, use linear interpolation to fill in the missing values. For samples with a missing value ratio higher than the preset threshold, delete them. Step S22: Use the interquartile range (IQR) method to remove outliers. Specifically, calculate the first and third quartiles of each feature variable. The IQR is defined as the difference between the third and first quartiles. Data that are less than Q1−1.5×IQR or greater than Q3+1.5×IQR are identified as outliers and removed. Q1 is the first quartile, Q3 is the third quartile, and IQR is the IQR. Step S23: For consecutive missing data segments, a data repair method based on similar sample matching is adopted to find the complete sample with the most similar features to the missing sample in all multidimensional state data, and the corresponding feature value of the complete sample is used as the repair value of the missing sample. Step S24: Use the min-max normalization method to normalize the multidimensional state data, mapping the numerical range of all feature variables to the interval [0, 1]. Step S25: Use the Pearson correlation coefficient method to measure the correlation between each feature variable and the motor health status label, and select the feature variable with the strongest correlation with the abnormal state of the motor as the input feature of the subsequent prediction model.

3. The intelligent elevator abnormality detection method according to claim 1, characterized in that, Step S3, multi-source signal time-frequency domain feature extraction, involves using discrete wavelet transform to decompose the input features into low-frequency approximate components and high-frequency detail components through multi-scale decomposition. The decomposition process of the discrete wavelet transform is as follows: The discrete signal in the input features is decomposed into multiple scales, and the low-frequency approximation coefficients and high-frequency detail coefficients of each decomposition level are calculated. The discrete signal is completely reconstructed using the detail coefficients of each layer and the final layer approximation coefficients to obtain the reconstructed discrete signal. The following time-frequency domain features are extracted from the detail coefficients of each layer using discrete wavelet transform: The energy value, energy percentage, entropy value, mean, standard deviation, peak factor, and kurtosis of each layer's detail coefficients constitute the original feature vector.

4. The intelligent elevator abnormality detection method according to claim 3, characterized in that, The low-frequency approximation coefficients and high-frequency detail coefficients for each decomposition layer are calculated using the following recursive formulas: ； ； In the formula, These are low-frequency approximation coefficients. It is a high-frequency detail factor. It is a decomposition level index. It is the coefficient position index. It is the index of the summation variable. These are the coefficients of the low-pass filter. These are the coefficients of the high-pass filter. These are the low-frequency approximation coefficients of the previous level of decomposition.

5. The intelligent elevator abnormality detection method according to claim 3, characterized in that, The reconstruction formula used is as follows: ； In the formula, It is to reconstruct discrete signals. It is the maximum number of decomposition levels. It is the first Approximation coefficients of the layer It is the first The detail factor of the layer, It is a scaling function. It is a wavelet function.

6. The intelligent elevator abnormality detection method according to claim 5, characterized in that, Step S4, constructing a hybrid classification model, specifically includes the following sub-steps: Step S41: Construct a random forest base classifier. The random forest consists of N decision trees. Each decision tree is trained on a subset of the training data obtained by sampling from the bootstrap dataset. For a new input sample, calculate the output of the random forest using the following formula: ； In the formula, It is the final prediction category of the random forest. These are category labels: 0 represents a normal state, 1 represents an edge state, and 2 represents a critical state. It is the total number of decision trees. It is the index of the decision tree. It is the first Decision trees for input samples Prediction categories, It is an indicator function that takes the value 1 when the condition inside the parentheses is true, and takes the value 0 otherwise. When splitting a node, each decision tree randomly selects a subset of features from all time-frequency domain features and chooses the best splitting feature from them. The Gini impurity is used as the splitting criterion for node splitting. Step S42: Use the Rhizostoma optimization algorithm to globally optimize the hyperparameters of the random forest. The hyperparameters to be optimized include the number of decision trees, the maximum depth of the decision trees, the minimum number of samples required for node splitting, the minimum number of samples for leaf nodes, and the maximum number of features. Step S43: Retrain the random forest model using the optimal hyperparameter combination.

7. The intelligent elevator abnormality detection method according to claim 6, characterized in that, For any node, the formula for calculating Gini impurity is as follows: ； In the formula, It is a node The impurity of the gin. It is a node Belongs to the category The proportion of samples, the feature and splitting threshold that minimize the weighted sum of Gini impurities of child nodes after splitting are selected when nodes split.

8. The intelligent elevator abnormality detection method according to claim 6, characterized in that, The Rhizostoma optimization algorithm simulates the foraging behavior of jellyfish in the ocean and achieves global optimization through the following steps: Step S421: Population initialization. R candidate solutions are randomly generated. The position vector of the j-th candidate solution represents the hyperparameter to be optimized. The initial values ​​of the candidate solutions in each dimension are randomly generated in the corresponding search space, using the following formula: ； In the formula, It is the first The candidate solution at the th... Initial values ​​in each dimension and They are the first The lower and upper bounds of the search in each dimension. It is a random number uniformly distributed within the interval (0, 1); Step S422: Boundary constraint handling. When a candidate solution exceeds the search space boundary during the search process, it is remapped to the inside of the boundary. The formula used is as follows: ； In the formula, These are the position values ​​after boundary constraint processing. It is the current position value; Step S423: Position Update. The position update of the candidate solution is achieved through Levy flight. The position update formula is as follows: ； In the formula, This is the updated location. It is the step size, which follows a Levy distribution. It is a random number uniformly distributed within the interval (0, 1); Step S424: Acceptance probability. The simulated annealing mechanism is used to determine whether to accept the updated position. The formula used is as follows: ； In the formula, It is the probability of acceptance. It is the difference between the objective function values ​​before and after the update, that is, the difference in the classification error rate of the random forest on the validation set. This is the current temperature parameter; Step S425: Motion control factor. This step introduces the exploration and development phase of the dynamic adjustment algorithm for the motion control factor. The formula used is as follows: ； In the formula, It is the first The motion control factor for the next iteration. It is the current iteration number. It is the maximum number of iterations; When the motion control factor is less than the threshold, the algorithm enters the development phase and performs a fine search near the current optimal solution; otherwise, it enters the exploration phase and searches for new potential optimal regions over a wide range. Step S426: Iterative optimization. Repeat the above position update, boundary constraint processing and acceptance probability judgment until the maximum number of iterations is reached, and output the optimal hyperparameter combination that minimizes the classification error rate of the random forest.

9. The intelligent elevator abnormality detection method according to claim 8, characterized in that, Step length Sampling is performed from the Cauchy distribution using the following formula: ； In the formula, It is the probability density function of the Cauchy distribution. It is a temperature parameter that controls the amplitude of step size fluctuations.

10. The intelligent elevator abnormality detection method according to claim 9, characterized in that, Step S5, health status classification, specifically includes the following sub-steps: Step S51: Divide the preprocessed dataset into a training set and a test set in a ratio of 8:2, and use a stratified sampling strategy to make the proportion of each category of samples in the training set and the test set consistent with the original dataset. Step S52: Using the optimal hyperparameter combination obtained from the Rhizostoma optimization algorithm in step S42, train the random forest model on the training set. During training, each decision tree grows independently based on the sub-training set sampled by Bootstrap. When a node splits, select the best splitting feature and splitting threshold from a randomly selected feature subset. Step S53: Evaluate the classification performance of the trained hybrid classification model on the test set, using precision, recall, F1 score and accuracy as evaluation metrics. Step S54: Deploy the trained hybrid classification model to the elevator real-time monitoring system to perform online abnormal state detection on the real-time collected multi-source sensor data. The hybrid model outputs the current motor health status category and triggers the corresponding early warning mechanism according to the output category. When the detection result is an edge state, a maintenance reminder is issued, and when the detection result is a critical state, a fault alarm is issued.

Images