Elevator door system fault detection method and system based on deep learning

Abstract

The present invention relates to the technical field of elevator door system data processing, and specifically discloses an elevator door system fault detection method and system based on deep learning. The present invention obtains an original data set of elevator doors through data collection; adopts a data preprocessing method of data cleaning, data encoding and data normalization; adopts a semi-supervised learning method of kernel K-means clustering combined with Gaussian process regression, which can extract valuable information from a large amount of unlabeled data, and effectively screen out data related to fault modes through clustering and regression; adopts a method for elevator door system fault detection using a bidirectional long short-term memory network model combined with an optimization algorithm, which can fully explore the complex dynamic behavior and timing dependency in the elevator door system, and has strong accuracy and adaptability.

Description

Technical Field

[0001] The present invention relates to the technical field of elevator door system data processing, and specifically to an elevator door system fault detection method and system based on deep learning. Background Art

[0002] Elevator door system fault detection uses sensors, data analysis, artificial intelligence and other technical means to monitor and diagnose the operating status of elevator doors in real time, promptly detect potential faults and take maintenance measures; it can effectively prevent elevators from being unable to operate normally, passengers from being trapped or accidents caused by door system failures, thereby improving the safety and reliability of elevators.

[0003] However, traditional elevator door system fault detection methods rely on a large amount of labeled data, making it difficult to handle large-scale, dynamically changing, and scarcely labeled elevator door system data. In addition, they have poor processing effects on noise and outliers, resulting in the loss of important feature information. Traditional elevator door system fault detection methods have the technical problem of being difficult to capture long-term and short-term dependencies and nonlinear changes when faced with a highly dynamic and complex environment for elevator door systems, and being unable to provide sufficient accuracy. Summary of the invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an elevator door system fault detection method and system based on deep learning. The traditional elevator door system fault detection method relies on a large amount of labeled data, is difficult to process large-scale, dynamically changing and scarce labeled data elevator door system data, and has poor processing effect on noise and outliers, resulting in the loss of important feature information. The present invention creatively adopts a semi-supervised learning method of kernel K-means clustering combined with Gaussian process regression, which can extract valuable information from a large amount of unlabeled data, and effectively screen out data related to the fault mode through clustering and regression, thereby improving the adaptability of the model in the case of scarce labeled data; the traditional elevator door system fault detection method has the technical problem that it is difficult to capture long-term and short-term dependencies and nonlinear changes when facing a highly dynamic and complex environment of the elevator door system, and cannot provide sufficient accuracy. The present invention creatively adopts a bidirectional long short-term memory network model combined with an optimization algorithm for elevator door system fault detection, which can fully explore the complex dynamic behavior and timing dependency in the elevator door system, and has strong accuracy and adaptability.

[0005] The technical solution adopted by the present invention is as follows: The elevator door system fault detection method based on deep learning provided by the present invention comprises the following steps:

[0006] Step S1: data collection;

[0007] Step S2: data preprocessing;

[0008] Step S3: elevator door data labeling;

[0009] Step S4: fault detection model construction;

[0010] Step S5: Elevator door system fault detection.

[0011] Further, in step S1, the data collection is used to collect data required for detecting the occurrence of elevator door system failures, specifically, obtaining an original data set of elevator doors through data collection from an elevator management system and an elevator sensor system;

[0012] The elevator door raw data set specifically includes a historical raw data set and a current raw data set. Both the historical raw data set and the current raw data set include elevator door switch status data, elevator sensor data, elevator door control signal data and environmental data. The historical raw data set also includes historical fault log data. The elevator door switch status data specifically includes an elevator door switch status label and an elevator door switch cycle. The elevator sensor data specifically includes door position sensor data, vibration sensor data, pressure sensor data, current sensor data and voltage sensor data. The environmental data specifically includes temperature and humidity data and elevator floor load data.

[0013] Furthermore, in step S2, the data preprocessing is used to preprocess the collected raw data, and specifically includes the following steps:

[0014] Step S21: data cleaning, which is used to clean the original data, specifically to remove missing values ​​and duplicate values ​​in the historical original data set and the current original data set to obtain a historical preliminary data set and a current preliminary data set;

[0015] Step S22: data encoding, which is used to encode the preliminary data, specifically, using a one-hot encoding method to encode the historical preliminary data set and the current preliminary data set to obtain a historical encoded data set and a current encoded data set;

[0016] Step S23: data normalization, which is used to normalize the coded data, specifically, using the minimum-maximum method to normalize the historical coded data set and the current coded data set to obtain a historical elevator door data set and a current elevator door data set;

[0017] Step S24: performing preprocessing, specifically preprocessing the historical original data set and the current original data set through the data cleaning, the data encoding and the data normalization to obtain a historical elevator door data set and a current elevator door data set.

[0018] Further, in step S3, the elevator door data labeling is used to label the unlabeled data, and the historical elevator door data set specifically includes unlabeled data and labeled data. Based on the labeled data, a kernel K-means clustering method is used to label the unlabeled data;

[0019] The elevator door data labeling specifically includes the following steps:

[0020] Step S31: constructing a clustering model for constructing a kernel K-means clustering model, the steps include:

[0021] Step S311: cluster model initialization, specifically initializing cluster centers and determining iteration termination conditions. The initialization of cluster centers specifically involves randomly selecting K sample data as initial cluster centers. The iteration termination conditions specifically include that the change in cluster centers is less than a threshold and that the maximum number of iterations is reached.

[0022] Step S312: assigning clusters, the steps include:

[0023] Step S3121: Calculate the similarity between the sample data and the cluster center. The formula used is as follows:

[0024] ;

[0025] In the formula, represents the Gaussian radial basis kernel function, represents the mth sample data, represents the nth sample data, represents the scale parameter, Indicates the calculation of Euclidean distance, represents the similarity between the mth sample data and the kth cluster center, N represents the total number of sample data, represents the kth cluster center;

[0026] Step S3122: Assign data to clusters using the following formula:

[0027] ;

[0028] In the formula, represents the cluster label of the mth sample data, Indicates the k value that maximizes the similarity between the mth sample data and the kth cluster center;

[0029] Step S3123: Update the cluster center. The formula used is as follows:

[0030] ;

[0031] In the formula, represents the updated cluster center;

[0032] Step S3124: reallocating clusters, specifically reallocating clusters based on updated cluster centers;

[0033] Step S313: iterative updating, specifically continuously iteratively updating the clustering model until the iteration termination condition is reached;

[0034] Step S314: clustering model construction, specifically, constructing a kernel K-means clustering model through the clustering model initialization, the clustering assignment and the iterative update to obtain a kernel K-means clustering model;

[0035] Step S32: data clustering processing, specifically, using the kernel K-means clustering model to perform clustering processing on the unlabeled data and the labeled data in the historical elevator door data set, respectively, to obtain clustered unlabeled data and clustered labeled data;

[0036] Step S33: data extraction, specifically, extracting data from the unlabeled data by calculating the distance between the clustered unlabeled data and the clustered labeled data, the steps include:

[0037] Step S331: Calculate the distance between the unlabeled data cluster center and the labeled data cluster center, using the following formula:

[0038] ;

[0039] In the formula, Indicates The cluster centers of the unlabeled data are The distance between the cluster centers of labeled data, Indicates Unlabeled data cluster centers, Indicates The cluster centers of labeled data;

[0040] Step S332: Calculate the average distance between the unlabeled data cluster center and the labeled data cluster center, using the following formula:

[0041] ;

[0042] In the formula, Indicates The average distance between the unlabeled data cluster centers and the labeled data cluster centers;

[0043] Step S333: extracting unlabeled data, specifically, sorting the unlabeled data cluster centers from large to small according to the average distance between the unlabeled data cluster centers and the labeled data cluster centers, selecting the clusters where the first num unlabeled data cluster centers are located, and randomly extracting unlabeled data from each selected cluster to obtain a preliminary unlabeled data set;

[0044] Step S34: data screening, specifically using Gaussian process regression to screen the preliminary unlabeled data set, the steps include:

[0045] Step S341: Design a function, the formula used is as follows:

[0046] ;

[0047] In the formula, represents the output generating function, represents a normal distribution, represents the mean function, represents the covariance function, x and represents the output generating function argument, represents the ath input data, represents the bth input data, represents the signal variance, represents the length scale;

[0048] Step S342: training a Gaussian process regression model, specifically training a Gaussian process regression model based on the labeled data in the historical elevator door data set to obtain a training data covariance matrix, and the formula used is as follows:

[0049] ;

[0050] In the formula, Km represents the size The covariance matrix of Represents the number of labeled data in the historical elevator door dataset;

[0051] Step S343: predict the mean and variance, using the following formula:

[0052] ;

[0053] In the formula, represents the covariance vector of the data to be predicted, Represents the data to be predicted, represents the predicted mean, T represents the transposition operation, represents the output value of the training data, represents the prediction variance;

[0054] Step S344: designing a threshold and screening, specifically, taking the data in the preliminary unlabeled data set as the data to be predicted, processing it with the trained Gaussian process regression model, and screening the data in the preliminary unlabeled data set by designing a threshold based on the prediction variance. The formula used for the threshold design is as follows:

[0055] ;

[0056] In the formula, represents the threshold control coefficient, It represents the mean of the predicted variance of the data to be predicted. Represents a threshold, filters out data with a prediction variance greater than the threshold, and merges it with the labeled data in the historical elevator door dataset into one dataset to obtain a historical fusion dataset;

[0057] Step S35: performing data labeling, specifically using the kernel K-means clustering model to process the historical fusion data set, adding labels to the data in the historical fusion data set, and obtaining a fault detection data set.

[0058] Further, in step S4, the fault detection model is constructed to construct a model required for detecting the occurrence of elevator door system faults, specifically, a bidirectional long short-term memory network model combined with an optimization algorithm is constructed and used as a fault detection model;

[0059] The fault detection model construction specifically includes the following steps:

[0060] Step S41: splitting the data set, specifically splitting the fault detection data set into a fault detection training set and a fault detection test set;

[0061] Step S42: constructing a bidirectional long short-term memory network model, the steps include:

[0062] Step S421: construct a forward module, the formula used is as follows:

[0063] ;

[0064] In the formula, represents the output of the forget gate at time t, represents the weight of the forward forget gate, represents the forward hidden state at time t-1, represents the model input at time t, represents the forward forget gate bias term, represents the output of the positive input gate at time t, represents the weight of the forward input gate, represents the forward input gate bias term, represents the candidate cell state at time t, represents the hyperbolic tangent function, represents the weight used to calculate the positive candidate cell state, represents the bias term used to calculate the positive candidate cell state, represents the cell state at the positive time t, represents the cell state at time t-1, Indicates the output of the positive output gate at time t, represents the weight of the forward output gate, represents the forward output gate bias term, represents the hidden state at the forward time t, Represents the sigmoid function;

[0065] Step S422: Construct a reverse module, and the formula used is as follows:

[0066] ;

[0067] In the formula, Represents the output of the forget gate at the reverse time t, represents the weight of the reverse forget gate, represents the reverse hidden state at time t-1, represents the reverse forget gate bias term, Represents the output of the input gate at time t, represents the reverse input gate weight, represents the reverse input gate bias term, represents the candidate cell state at the reverse time t, represents the weight used to calculate the reverse candidate cell state, represents the bias term used to calculate the reverse candidate cell state, represents the cell state at the reverse time t, represents the cell state at the reverse time t-1, Represents the output gate output at time t in the reverse direction, represents the reverse output gate weight, represents the reverse output gate bias term, Represents the hidden state at the reverse time t;

[0068] Step S423: construct an output module, the formula used is as follows:

[0069] ;

[0070] In the formula, represents the output of the model at time t, represents the softmax function, represents the model output weight, Represents the model output bias term;

[0071] Step S424: constructing a model and training it, specifically, constructing a bidirectional long short-term memory network model by constructing the forward module, the reverse module and the output module, training the model based on the fault detection training set, and verifying the model performance based on the fault detection test set, and the model loss function adopts a cross entropy loss function;

[0072] Step S43: hyperparameter optimization, specifically optimizing the bidirectional long short-term memory network model hyperparameters based on the optimization algorithm, the model hyperparameters specifically including the number of long short-term memory layers, the number of long short-term memory units, the model learning rate and the batch size;

[0073] The hyperparameter optimization steps include:

[0074] Step S431: initializing the algorithm, specifically initializing the search space and constructing an individual unit set, wherein the individual unit is used to represent a hyperparameter combination of a bidirectional long short-term memory network model, and the fitness function of the individual unit is a loss function of the bidirectional long short-term memory network model;

[0075] Step S432: Calculate the balance factor and mutation factor using the following formula:

[0076] ;

[0077] In the formula, Bf represents the balance factor, Mf represents the mutation factor, Indicates that a random number in the range (0,1) will be regenerated in each iteration, dt indicates the current number of iterations, Indicates the maximum number of iterations;

[0078] Step S433: global search, the formula used is as follows:

[0079] ;

[0080] In the formula, represents the position of the jth dimension of the ith individual unit at the dt+1th iteration, represents the position of the random dimension of the i-th individual unit at the dt-th iteration, Indicates the position of the random dimension of the random individual unit at the dt-th iteration, which is the same dimension as the position of the random dimension of the i-th individual unit at the dt-th iteration. Represents a random number in the range (0,1). Represents a random number in the range (0,1), even represents an even number, and odd represents an odd number;

[0081] Step S434: local search, the formula used is as follows:

[0082] ;

[0083] In the formula, represents the flight scale parameter, represents the gamma function, represents the flight index parameter, Lf represents the flight factor, u represents a random number that obeys a normal distribution, v represents a random number that obeys a normal distribution, represents the position of the i-th individual unit at the dt+1th iteration, represents the global optimal position of the current iteration, represents the position of the i-th individual unit at the dt-th iteration, represents the position of the random individual unit at the dtth iteration, Represents a random number in the range (0,1). Represents a random number in the range (0,1);

[0084] Step S435: position mutation, the formula used is as follows:

[0085] ;

[0086] In the formula, Represents a random number in the range (0,1). Represents a random number in the range (0,1). Represents a random number in the range (0,1). represents the mutation step size, ub represents the upper limit of the variable to be optimized, and lb represents the lower limit of the variable to be optimized. represents the total number of individual units;

[0087] Step S436: Determine the global optimal solution, specifically by continuously iteratively updating the optimization algorithm until the optimization algorithm iteration termination condition is reached, and take the position of the individual unit with the lowest fitness as the global optimal solution, the global optimal solution is specifically the optimal bidirectional long short-term memory network model hyperparameter combination, and the optimization algorithm iteration termination condition specifically includes that the individual unit fitness is less than a set threshold and reaches the maximum number of iterations;

[0088] Step S437: Optimizing model hyperparameters, specifically optimizing the hyperparameters of the bidirectional long short-term memory network model through the algorithm initialization, the calculation of balance factors and mutation factors, the global search, the local search, the position mutation and the determination of the global optimal solution, and obtaining a bidirectional long short-term memory network model combined with the optimization algorithm as a fault detection model.

[0089] Furthermore, in step S5, the elevator door system fault detection specifically uses the current elevator door data set as the input of the fault detection model to obtain fault detection reference data, and comprehensively analyzes the occurrence of elevator door system faults based on the fault detection reference data.

[0090] The elevator door system fault detection system based on deep learning provided by the present invention comprises a data acquisition module, a data preprocessing module, an elevator door data annotation module, a fault detection model construction module and an elevator door system fault detection module;

[0091] The data acquisition module is used for data acquisition, and obtains the original data set of the elevator door through data acquisition, and sends the original data set of the elevator door to the data preprocessing module;

[0092] The data preprocessing module is used for data preprocessing. Through data preprocessing, a historical elevator door data set and a current elevator door data set are obtained, and the historical elevator door data set is sent to the elevator door data annotation module, and the current elevator door data set is sent to the elevator door system fault detection module;

[0093] The elevator door data labeling module is used for elevator door data labeling, and performs data labeling by constructing a kernel K-means clustering model to obtain a fault detection data set, and sends the fault detection data set to the fault detection model construction module;

[0094] The fault detection model building module is used to build a fault detection model, obtain the fault detection model by building a bidirectional long short-term memory network model combined with an optimization algorithm, and send the fault detection model to the elevator door system fault detection module;

[0095] The elevator door system fault detection module is used for elevator door system fault detection. The fault detection model is used to perform elevator door system fault detection to obtain fault detection reference data.

[0096] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0097] (1) In view of the technical problems that traditional elevator door system fault detection methods rely on a large amount of labeled data and are difficult to process large-scale, dynamically changing, and scarce labeled data, and that the processing effect of noise and outliers is poor, resulting in the loss of important feature information, this solution creatively adopts a semi-supervised learning method combining kernel K-means clustering with Gaussian process regression, which can extract valuable information from a large amount of unlabeled data and effectively screen out data related to fault modes through clustering and regression, thereby improving the adaptability of the model when labeled data is scarce.

[0098] (2) In view of the technical problem that traditional elevator door system fault detection methods have difficulty in capturing long-term and short-term dependencies and nonlinear changes when faced with a highly dynamic and complex environment, and cannot provide sufficient accuracy, this solution creatively uses a bidirectional long short-term memory network model combined with an optimization algorithm to perform elevator door system fault detection. This method can fully explore the complex dynamic behavior and timing dependencies in the elevator door system, and has strong accuracy and adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0099] Figure 1 A schematic diagram of a flow chart of an elevator door system fault detection method based on deep learning provided by the present invention;

[0100] Figure 2 A module schematic diagram of an elevator door system fault detection system based on deep learning provided by the present invention;

[0101] Figure 3 This is a flow chart of elevator door data labeling in step S3;

[0102] Figure 4 A schematic diagram of the process of constructing a bidirectional long short-term memory network model in step S42;

[0103] Figure 5 This is a schematic diagram of the process of hyperparameter optimization in step S43.

[0104] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0105] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0106] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0107] Example 1, see Figure 1The technical solution adopted by the present invention is as follows: The present invention provides an elevator door system fault detection method based on deep learning, the method comprising the following steps:

[0108] Step S1: data collection;

[0109] Step S2: data preprocessing;

[0110] Step S3: elevator door data labeling;

[0111] Step S4: fault detection model construction;

[0112] Step S5: Elevator door system fault detection.

[0113] Example 2, see Figure 1 and Figure 2 In step S1, the data collection is used to collect data required for detecting the occurrence of elevator door system failures, specifically, obtaining the original data set of the elevator door from the elevator management system and the elevator sensor system through data collection;

[0114] The elevator door raw data set specifically includes a historical raw data set and a current raw data set. Both the historical raw data set and the current raw data set include elevator door switch status data, elevator sensor data, elevator door control signal data and environmental data. The historical raw data set also includes historical fault log data. The elevator door switch status data specifically includes an elevator door switch status label and an elevator door switch cycle. The elevator sensor data specifically includes door position sensor data, vibration sensor data, pressure sensor data, current sensor data and voltage sensor data. The environmental data specifically includes temperature and humidity data and elevator floor load data.

[0115] Example 3, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S2, the data preprocessing is used to preprocess the collected raw data, and specifically includes the following steps:

[0116] Step S21: data cleaning, which is used to clean the original data, specifically to remove missing values ​​and duplicate values ​​in the historical original data set and the current original data set to obtain a historical preliminary data set and a current preliminary data set;

[0117] Step S22: data encoding, which is used to encode the preliminary data, specifically, using a one-hot encoding method to encode the historical preliminary data set and the current preliminary data set to obtain a historical encoded data set and a current encoded data set;

[0118] Step S23: data normalization, which is used to normalize the coded data, specifically, using the minimum-maximum method to normalize the historical coded data set and the current coded data set to obtain a historical elevator door data set and a current elevator door data set;

[0119] Step S24: performing preprocessing, specifically preprocessing the historical original data set and the current original data set through the data cleaning, the data encoding and the data normalization to obtain a historical elevator door data set and a current elevator door data set.

[0120] Example 4, see Figure 1 , Figure 2 and Figure 3 , This embodiment is based on the above embodiment. In step S3, the elevator door data labeling is used to label the unlabeled data. The historical elevator door data set specifically includes unlabeled data and labeled data. Based on the labeled data, the kernel K-means clustering method is used to label the unlabeled data.

[0121] The elevator door data labeling specifically includes the following steps:

[0122] Step S31: constructing a clustering model for constructing a kernel K-means clustering model, the steps include:

[0123] Step S311: cluster model initialization, specifically initializing cluster centers and determining iteration termination conditions. The initialization of cluster centers specifically involves randomly selecting K sample data as initial cluster centers. The iteration termination conditions specifically include that the change in cluster centers is less than a threshold and that the maximum number of iterations is reached.

[0124] Step S312: assigning clusters, the steps include:

[0125] Step S3121: Calculate the similarity between the sample data and the cluster center. The formula used is as follows:

[0126] ;

[0127] In the formula, represents the Gaussian radial basis kernel function, represents the mth sample data, represents the nth sample data, represents the scale parameter, Indicates the calculation of Euclidean distance, represents the similarity between the mth sample data and the kth cluster center, N represents the total number of sample data, represents the kth cluster center;

[0128] Step S3122: Assign data to clusters using the following formula:

[0129] ;

[0130] In the formula, represents the cluster label of the mth sample data, Indicates the k value that maximizes the similarity between the mth sample data and the kth cluster center;

[0131] Step S3123: Update the cluster center. The formula used is as follows:

[0132] ;

[0133] In the formula, represents the updated cluster center;

[0134] Step S3124: reallocating clusters, specifically reallocating clusters based on updated cluster centers;

[0135] Step S313: iterative updating, specifically continuously iteratively updating the clustering model until the iteration termination condition is reached;

[0136] Step S314: clustering model construction, specifically, constructing a kernel K-means clustering model through the clustering model initialization, the clustering assignment and the iterative update to obtain a kernel K-means clustering model;

[0137] Step S32: data clustering processing, specifically, using the kernel K-means clustering model to perform clustering processing on the unlabeled data and the labeled data in the historical elevator door data set, respectively, to obtain clustered unlabeled data and clustered labeled data;

[0138] Step S33: data extraction, specifically, extracting data from the unlabeled data by calculating the distance between the clustered unlabeled data and the clustered labeled data, the steps include:

[0139] Step S331: Calculate the distance between the unlabeled data cluster center and the labeled data cluster center, using the following formula:

[0140] ;

[0141] In the formula, Indicates The cluster centers of the unlabeled data are The distance between the cluster centers of labeled data, Indicates Unlabeled data cluster centers, Indicates The cluster centers of labeled data;

[0142] Step S332: Calculate the average distance between the unlabeled data cluster center and the labeled data cluster center, using the following formula:

[0143] ;

[0144] In the formula, Indicates The average distance between the unlabeled data cluster centers and the labeled data cluster centers;

[0145] Step S333: extracting unlabeled data, specifically, sorting the unlabeled data cluster centers from large to small according to the average distance between the unlabeled data cluster centers and the labeled data cluster centers, selecting the clusters where the first num unlabeled data cluster centers are located, and randomly extracting unlabeled data from each selected cluster to obtain a preliminary unlabeled data set;

[0146] Step S34: data screening, specifically using Gaussian process regression to screen the preliminary unlabeled data set, the steps include:

[0147] Step S341: Design a function, the formula used is as follows:

[0148] ;

[0149] In the formula, represents the output generating function, represents a normal distribution, represents the mean function, represents the covariance function, x and represents the output generating function argument, represents the ath input data, represents the bth input data, represents the signal variance, represents the length scale;

[0150] Step S342: training a Gaussian process regression model, specifically training a Gaussian process regression model based on the labeled data in the historical elevator door data set to obtain a training data covariance matrix, and the formula used is as follows:

[0151] ;

[0152] In the formula, Km represents the size The covariance matrix of Represents the number of labeled data in the historical elevator door dataset;

[0153] Step S343: predict the mean and variance, using the following formula:

[0154] ;

[0155] In the formula, represents the covariance vector of the data to be predicted, Represents the data to be predicted, represents the predicted mean, T represents the transposition operation, represents the output value of the training data, represents the prediction variance;

[0156] Step S344: designing a threshold and screening, specifically, taking the data in the preliminary unlabeled data set as the data to be predicted, processing it with the trained Gaussian process regression model, and screening the data in the preliminary unlabeled data set by designing a threshold based on the prediction variance. The formula used for the threshold design is as follows:

[0157] ;

[0158] In the formula, represents the threshold control coefficient, It represents the mean of the prediction variance of the data to be predicted. Represents a threshold, filters out data with a prediction variance greater than the threshold, and merges it with the labeled data in the historical elevator door dataset into one dataset to obtain a historical fusion dataset;

[0159] Step S35: performing data labeling, specifically using the kernel K-means clustering model to process the historical fusion data set, adding labels to the data in the historical fusion data set, and obtaining a fault detection data set.

[0160] By performing the above operations, the traditional elevator door system fault detection method relies on a large amount of labeled data, is difficult to process large-scale, dynamically changing and scarce labeled data elevator door system data, and has poor processing effect on noise and outliers, resulting in the loss of important feature information. This solution creatively adopts a semi-supervised learning method of kernel K-means clustering combined with Gaussian process regression, which can extract valuable information from a large amount of unlabeled data, and effectively screen out data related to fault modes through clustering and regression, thereby improving the adaptability of the model when labeled data is scarce.

[0161] Example 5, see Figure 1 , Figure 2 , Figure 4 and Figure 5 , this embodiment is based on the above embodiment. In step S4, the fault detection model is constructed to construct a model required for detecting the occurrence of elevator door system faults, specifically, a bidirectional long short-term memory network model combined with an optimization algorithm is constructed and used as a fault detection model;

[0162] The fault detection model construction specifically includes the following steps:

[0163] Step S41: splitting the data set, specifically splitting the fault detection data set into a fault detection training set and a fault detection test set;

[0164] Step S42: constructing a bidirectional long short-term memory network model, the steps include:

[0165] Step S421: construct a forward module, the formula used is as follows:

[0166] ;

[0167] In the formula, represents the output of the forget gate at time t, represents the weight of the forward forget gate, represents the forward hidden state at time t-1, represents the model input at time t, represents the forward forget gate bias term, represents the output of the positive input gate at time t, represents the weight of the forward input gate, represents the forward input gate bias term, represents the candidate cell state at time t, represents the hyperbolic tangent function, represents the weight used to calculate the positive candidate cell state, represents the bias term used to calculate the positive candidate cell state, represents the cell state at the positive time t, represents the cell state at time t-1, Indicates the output of the positive output gate at time t, represents the weight of the forward output gate, represents the forward output gate bias term, represents the hidden state at the forward time t, Represents the sigmoid function;

[0168] Step S422: Construct a reverse module, and the formula used is as follows:

[0169] ;

[0170] In the formula, Represents the output of the forget gate at the reverse time t, represents the weight of the reverse forget gate, represents the reverse hidden state at time t-1, represents the reverse forget gate bias term, Represents the output of the input gate at time t, represents the reverse input gate weight, represents the reverse input gate bias term, represents the candidate cell state at the reverse time t, represents the weight used to calculate the reverse candidate cell state, represents the bias term used to calculate the reverse candidate cell state, represents the cell state at the reverse time t, represents the cell state at the reverse time t-1, Represents the output gate output at time t in the reverse direction, represents the reverse output gate weight, represents the reverse output gate bias term, Represents the hidden state at the reverse time t;

[0171] Step S423: construct an output module, the formula used is as follows:

[0172] ;

[0173] In the formula, represents the output of the model at time t, represents the softmax function, represents the model output weight, Represents the model output bias term;

[0174] Step S424: constructing a model and training it, specifically, constructing a bidirectional long short-term memory network model by constructing the forward module, the reverse module and the output module, training the model based on the fault detection training set, and verifying the model performance based on the fault detection test set, and the model loss function adopts a cross entropy loss function;

[0175] Step S43: hyperparameter optimization, specifically optimizing the bidirectional long short-term memory network model hyperparameters based on the optimization algorithm, the model hyperparameters specifically including the number of long short-term memory layers, the number of long short-term memory units, the model learning rate and the batch size;

[0176] The hyperparameter optimization steps include:

[0177] Step S431: initializing the algorithm, specifically initializing the search space and constructing an individual unit set, wherein the individual unit is used to represent a hyperparameter combination of a bidirectional long short-term memory network model, and the fitness function of the individual unit is a loss function of the bidirectional long short-term memory network model;

[0178] Step S432: Calculate the balance factor and mutation factor using the following formula:

[0179] ;

[0180] In the formula, Bf represents the balance factor, Mf represents the mutation factor, Indicates that a random number in the range (0,1) will be regenerated in each iteration, dt indicates the current number of iterations, Indicates the maximum number of iterations;

[0181] Step S433: global search, the formula used is as follows:

[0182] ;

[0183] In the formula, represents the position of the jth dimension of the ith individual unit at the dt+1th iteration, represents the position of the random dimension of the i-th individual unit at the dt-th iteration, Indicates the position of the random dimension of the random individual unit at the dt-th iteration, which is the same dimension as the position of the random dimension of the i-th individual unit at the dt-th iteration. Represents a random number in the range (0,1). Represents a random number in the range (0,1), even represents an even number, and odd represents an odd number;

[0184] Step S434: local search, the formula used is as follows:

[0185] ;

[0186] In the formula, represents the flight scale parameter, represents the gamma function, represents the flight index parameter, Lf represents the flight factor, u represents a random number that obeys a normal distribution, v represents a random number that obeys a normal distribution, represents the position of the i-th individual unit at the dt+1th iteration, represents the global optimal position of the current iteration, represents the position of the i-th individual unit at the dt-th iteration, represents the position of the random individual unit at the dtth iteration, Represents a random number in the range (0,1). Represents a random number in the range (0,1);

[0187] Step S435: position mutation, the formula used is as follows:

[0188] ;

[0189] In the formula, Represents a random number in the range (0,1). Represents a random number in the range (0,1). Represents a random number in the range (0,1). represents the mutation step size, ub represents the upper limit of the variable to be optimized, and lb represents the lower limit of the variable to be optimized. represents the total number of individual units;

[0190] Step S436: Determine the global optimal solution, specifically by continuously iteratively updating the optimization algorithm until the optimization algorithm iteration termination condition is reached, and take the position of the individual unit with the lowest fitness as the global optimal solution, the global optimal solution is specifically the optimal bidirectional long short-term memory network model hyperparameter combination, and the optimization algorithm iteration termination condition specifically includes that the individual unit fitness is less than a set threshold and reaches the maximum number of iterations;

[0191] Step S437: Optimizing model hyperparameters, specifically optimizing the hyperparameters of the bidirectional long short-term memory network model through the algorithm initialization, the calculation of balance factors and mutation factors, the global search, the local search, the position mutation and the determination of the global optimal solution, and obtaining a bidirectional long short-term memory network model combined with the optimization algorithm as a fault detection model.

[0192] By performing the above operations, in view of the technical problem that the traditional elevator door system fault detection method has difficulty in capturing long-term and short-term dependencies and nonlinear changes and cannot provide sufficient accuracy when facing the highly dynamic and complex environment of the elevator door system, this solution creatively adopts a bidirectional long short-term memory network model combined with an optimization algorithm for elevator door system fault detection, which can fully explore the complex dynamic behavior and timing dependency in the elevator door system and has strong accuracy and adaptability.

[0193] Example 6, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S5, the elevator door system fault detection specifically uses the current elevator door data set as the input of the fault detection model to obtain fault detection reference data, and comprehensively analyzes the occurrence of elevator door system faults based on the fault detection reference data.

[0194] Embodiment 7, see Figure 1 and Figure 2 , this embodiment is based on the above embodiment. The elevator door system fault detection system based on deep learning provided by the present invention includes a data acquisition module, a data preprocessing module, an elevator door data annotation module, a fault detection model construction module and an elevator door system fault detection module;

[0195] The data acquisition module is used for data acquisition, and obtains the original data set of the elevator door through data acquisition, and sends the original data set of the elevator door to the data preprocessing module;

[0196] The data preprocessing module is used for data preprocessing. Through data preprocessing, a historical elevator door data set and a current elevator door data set are obtained, and the historical elevator door data set is sent to the elevator door data annotation module, and the current elevator door data set is sent to the elevator door system fault detection module;

[0197] The elevator door data labeling module is used for elevator door data labeling, and performs data labeling by constructing a kernel K-means clustering model to obtain a fault detection data set, and sends the fault detection data set to the fault detection model construction module;

[0198] The fault detection model building module is used to build a fault detection model, obtain the fault detection model by building a bidirectional long short-term memory network model combined with an optimization algorithm, and send the fault detection model to the elevator door system fault detection module;

[0199] The elevator door system fault detection module is used for elevator door system fault detection. The fault detection model is used to perform elevator door system fault detection to obtain fault detection reference data.

[0200] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0201] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0202] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A method for detecting elevator door system faults based on deep learning, characterized in that: The method comprises the following steps: Step S1: data collection, through which the original data set of the elevator door is obtained, specifically including the historical original data set and the current original data set; Step S2: Data preprocessing, preprocessing the collected raw data to obtain a historical elevator door data set and a current elevator door data set; Step S3: elevator door data labeling, which is used for elevator door data labeling, specifically building a kernel K-means clustering model for data labeling to obtain a fault detection data set. The specific steps include building a clustering model, data clustering processing, data extraction, data screening and data labeling; The data extraction is specifically to extract data from the unlabeled data by calculating the distance between the clustered unlabeled data and the clustered labeled data to obtain a preliminary unlabeled data set; The data screening is specifically to use Gaussian process regression to screen the preliminary unlabeled data set, and the steps include: Step S341: Design a function, the formula used is as follows: ； In the formula, represents the output generating function, represents a normal distribution, represents the mean function, represents the covariance function, x and represents the output generating function argument, represents the ath input data, represents the bth input data, represents the signal variance, represents the length scale; Step S342: training a Gaussian process regression model, specifically training a Gaussian process regression model based on the labeled data in the historical elevator door data set to obtain a training data covariance matrix, and the formula used is as follows: ； In the formula, Km represents the size The covariance matrix of Represents the number of labeled data in the historical elevator door dataset; Step S343: predict the mean and variance, using the following formula: ； In the formula, represents the covariance vector of the data to be predicted, Represents the data to be predicted, represents the predicted mean, T represents the transposition operation, represents the output value of the training data, represents the prediction variance; Step S344: designing a threshold and screening, specifically, taking the data in the preliminary unlabeled data set as the data to be predicted, processing it with the trained Gaussian process regression model, and screening the data in the preliminary unlabeled data set by designing a threshold based on the prediction variance. The formula used for the threshold design is as follows: ； In the formula, represents the threshold control coefficient, It represents the mean of the predicted variance of the data to be predicted. Represents a threshold, filters out data with a prediction variance greater than the threshold, and merges it with the labeled data in the historical elevator door dataset into one dataset to obtain a historical fusion dataset; Step S4: Fault detection model construction, which is used to construct a model required for detecting the occurrence of elevator door system faults, specifically, to construct a bidirectional long short-term memory network model combined with an optimization algorithm and use it as a fault detection model. The specific steps include segmenting the data set, constructing a bidirectional long short-term memory network model, and optimizing hyperparameters. The hyperparameter optimization steps include algorithm initialization, calculation of balance factor and mutation factor, global search, local search, position mutation, determination of global optimal solution and optimization of model hyperparameters; Step S5: elevator door system fault detection, specifically, performing elevator door system fault detection through the fault detection model to obtain fault detection reference data.

2. The elevator door system fault detection method based on deep learning according to claim 1, characterized in that: In step S3, the elevator door data labeling is used to label the unlabeled data. The historical elevator door data set specifically includes unlabeled data and labeled data. Based on the labeled data, the kernel K-means clustering method is used to label the unlabeled data. The elevator door data labeling specifically includes the following steps: Step S31: constructing a clustering model for constructing a kernel K-means clustering model, the steps include: Step S311: cluster model initialization, specifically initializing cluster centers and determining iteration termination conditions. The initialization of cluster centers specifically involves randomly selecting K sample data as initial cluster centers. The iteration termination conditions specifically include that the change in cluster centers is less than a threshold and that the maximum number of iterations is reached. Step S312: assigning clusters, the steps include: Step S3121: Calculate the similarity between the sample data and the cluster center. The formula used is as follows: ； In the formula, represents the Gaussian radial basis kernel function, represents the mth sample data, represents the nth sample data, represents the scale parameter, Indicates the calculation of Euclidean distance, represents the similarity between the mth sample data and the kth cluster center, N represents the total number of sample data, represents the kth cluster center; Step S3122: Assign data to clusters using the following formula: ； In the formula, represents the cluster label of the mth sample data, Indicates the k value that maximizes the similarity between the mth sample data and the kth cluster center; Step S3123: Update the cluster center. The formula used is as follows: ； In the formula, represents the updated cluster center; Step S3124: reallocating clusters, specifically reallocating clusters based on updated cluster centers; Step S313: iterative updating, specifically continuously iteratively updating the clustering model until the iteration termination condition is reached; Step S314: clustering model construction, specifically, constructing a kernel K-means clustering model through the clustering model initialization, the clustering assignment and the iterative update to obtain a kernel K-means clustering model; Step S32: data clustering processing, specifically, using the kernel K-means clustering model to perform clustering processing on the unlabeled data and the labeled data in the historical elevator door data set, respectively, to obtain clustered unlabeled data and clustered labeled data; Step S33: data extraction, the steps include: Step S331: Calculate the distance between the unlabeled data cluster center and the labeled data cluster center, using the following formula: ； In the formula, Indicates The cluster centers of the unlabeled data are The distance between the cluster centers of labeled data, Indicates Unlabeled data cluster centers, Indicates The cluster centers of labeled data; Step S332: Calculate the average distance between the unlabeled data cluster center and the labeled data cluster center, using the following formula: ； In the formula, Indicates The average distance between the unlabeled data cluster centers and the labeled data cluster centers; Step S333: extracting unlabeled data, specifically, sorting the unlabeled data cluster centers from large to small according to the average distance between the unlabeled data cluster centers and the labeled data cluster centers, selecting the clusters where the first num unlabeled data cluster centers are located, and randomly extracting unlabeled data from each selected cluster to obtain a preliminary unlabeled data set; Step S34: data screening; Step S35: performing data labeling, specifically using the kernel K-means clustering model to process the historical fusion data set, adding labels to the data in the historical fusion data set, and obtaining a fault detection data set.

3. The elevator door system fault detection method based on deep learning according to claim 1, characterized in that: In step S4, the fault detection model is constructed to construct a model required for detecting the occurrence of elevator door system faults, specifically, a bidirectional long short-term memory network model combined with an optimization algorithm is constructed as a fault detection model; The fault detection model construction specifically includes the following steps: Step S41: splitting the data set, specifically splitting the fault detection data set into a fault detection training set and a fault detection test set; Step S42: constructing a bidirectional long short-term memory network model, the steps include: Step S421: construct a forward module, the formula used is as follows: ； In the formula, represents the output of the forget gate at time t, represents the weight of the forward forget gate, represents the forward hidden state at time t-1, represents the model input at time t, represents the forward forget gate bias term, represents the output of the positive input gate at time t, represents the weight of the forward input gate, represents the forward input gate bias term, represents the candidate cell state at time t, represents the hyperbolic tangent function, represents the weight used to calculate the positive candidate cell state, represents the bias term used to calculate the positive candidate cell state, represents the cell state at the positive time t, represents the cell state at time t-1, It indicates the output of the positive output gate at time t. represents the forward output gate weight, represents the forward output gate bias term, represents the hidden state at the forward time t, Represents the sigmoid function; Step S422: Construct a reverse module, and the formula used is as follows: ； In the formula, Represents the output of the forget gate at the reverse time t, represents the weight of the reverse forget gate, represents the reverse hidden state at time t-1, represents the reverse forget gate bias term, Represents the output of the input gate at the reverse time t, represents the reverse input gate weight, represents the reverse input gate bias term, represents the candidate cell state at the reverse time t, represents the weight used to calculate the reverse candidate cell state, represents the bias term used to calculate the reverse candidate cell state, represents the cell state at the reverse time t, represents the cell state at the reverse time t-1, Represents the output of the output gate at time t in the reverse direction, represents the reverse output gate weight, represents the reverse output gate bias term, Represents the hidden state at the reverse time t; Step S423: construct an output module, the formula used is as follows: ； In the formula, represents the output of the model at time t, represents the softmax function, represents the model output weight, Represents the model output bias term; Step S424: constructing a model and training it, specifically, constructing a bidirectional long short-term memory network model by constructing the forward module, the reverse module and the output module, training the model based on the fault detection training set, and verifying the model performance based on the fault detection test set, and the model loss function adopts a cross entropy loss function; Step S43: hyperparameter optimization, specifically optimizing the bidirectional long short-term memory network model hyperparameters based on the optimization algorithm, the model hyperparameters specifically including the number of long short-term memory layers, the number of long short-term memory units, the model learning rate and the batch size; The hyperparameter optimization steps include: Step S431: initializing the algorithm, specifically initializing the search space and constructing an individual unit set, wherein the individual unit is used to represent a hyperparameter combination of a bidirectional long short-term memory network model, and the fitness function of the individual unit is a loss function of the bidirectional long short-term memory network model; Step S432: Calculate the balance factor and mutation factor using the following formula: ； In the formula, Bf represents the balance factor, Mf represents the mutation factor, Indicates that a random number in the range (0,1) will be regenerated in each iteration, dt indicates the current number of iterations, Indicates the maximum number of iterations; Step S433: global search, the formula used is as follows: ； In the formula, represents the position of the jth dimension of the ith individual unit at the dt+1th iteration, represents the position of the random dimension of the i-th individual unit at the dt-th iteration, Indicates the position of the random dimension of the random individual unit at the dt-th iteration, which is the same dimension as the position of the random dimension of the i-th individual unit at the dt-th iteration. Represents a random number in the range (0,1). Represents a random number in the range (0,1), even represents an even number, and odd represents an odd number; Step S434: local search, the formula used is as follows: ； In the formula, represents the flight scale parameter, represents the gamma function, represents the flight index parameter, Lf represents the flight factor, u represents a random number that obeys a normal distribution, v represents a random number that obeys a normal distribution, represents the position of the i-th individual unit at the dt+1th iteration, represents the global optimal position of the current iteration, represents the position of the i-th individual unit at the dt-th iteration, represents the position of the random individual unit at the dtth iteration, Represents a random number in the range (0,1). Represents a random number in the range (0,1); Step S435: position mutation, the formula used is as follows: ； In the formula, Represents a random number in the range (0,1). Represents a random number in the range (0,1). Represents a random number in the range (0,1). represents the mutation step size, ub represents the upper limit of the variable to be optimized, and lb represents the lower limit of the variable to be optimized. represents the total number of individual units; Step S436: Determine the global optimal solution, specifically by continuously iteratively updating the optimization algorithm until the optimization algorithm iteration termination condition is reached, and take the position of the individual unit with the lowest fitness as the global optimal solution, the global optimal solution is specifically the optimal bidirectional long short-term memory network model hyperparameter combination, and the optimization algorithm iteration termination condition specifically includes that the individual unit fitness is less than a set threshold and reaches the maximum number of iterations; Step S437: Optimizing model hyperparameters, specifically optimizing the hyperparameters of the bidirectional long short-term memory network model through the algorithm initialization, the calculation of balance factors and mutation factors, the global search, the local search, the position mutation and the determination of the global optimal solution, and obtaining a bidirectional long short-term memory network model combined with the optimization algorithm as a fault detection model.

4. The elevator door system fault detection method based on deep learning according to claim 1, characterized in that: In step S1, the data collection is used to collect data required for detecting the occurrence of elevator door system failures, specifically, obtaining the original data set of the elevator door from the elevator management system and the elevator sensor system through data collection; The elevator door raw data set specifically includes a historical raw data set and a current raw data set. Both the historical raw data set and the current raw data set include elevator door switch status data, elevator sensor data, elevator door control signal data and environmental data. The historical raw data set also includes historical fault log data. The elevator door switch status data specifically includes an elevator door switch status label and an elevator door switch cycle. The elevator sensor data specifically includes door position sensor data, vibration sensor data, pressure sensor data, current sensor data and voltage sensor data. The environmental data specifically includes temperature and humidity data and elevator floor load data.

5. The elevator door system fault detection method based on deep learning according to claim 1, characterized in that: In step S2, the data preprocessing is used to preprocess the collected raw data, and specifically includes the following steps: Step S21: data cleaning, which is used to clean the original data, specifically to remove missing values ​​and duplicate values ​​in the historical original data set and the current original data set to obtain a historical preliminary data set and a current preliminary data set; Step S22: data encoding, which is used to encode the preliminary data, specifically, using a one-hot encoding method to encode the historical preliminary data set and the current preliminary data set to obtain a historical encoded data set and a current encoded data set; Step S23: data normalization, which is used to normalize the coded data, specifically, using the minimum-maximum method to normalize the historical coded data set and the current coded data set to obtain a historical elevator door data set and a current elevator door data set; Step S24: performing preprocessing, specifically preprocessing the historical original data set and the current original data set through the data cleaning, the data encoding and the data normalization to obtain a historical elevator door data set and a current elevator door data set.

6. The elevator door system fault detection method based on deep learning according to claim 1, characterized in that: In step S5, the elevator door system fault detection specifically includes taking the current elevator door data set as the input of the fault detection model to obtain fault detection reference data, and comprehensively analyzing the occurrence of elevator door system faults based on the fault detection reference data.

7. An elevator door system fault detection system based on deep learning, used to implement the elevator door system fault detection method based on deep learning as described in any one of claims 1 to 6, characterized in that: It includes a data acquisition module, a data preprocessing module, an elevator door data labeling module, a fault detection model building module and an elevator door system fault detection module.

8. The elevator door system fault detection system based on deep learning according to claim 7, characterized in that: The data acquisition module is used for data acquisition, and obtains the original data set of the elevator door through data acquisition, and sends the original data set of the elevator door to the data preprocessing module; The data preprocessing module is used for data preprocessing. Through data preprocessing, a historical elevator door data set and a current elevator door data set are obtained, and the historical elevator door data set is sent to the elevator door data annotation module, and the current elevator door data set is sent to the elevator door system fault detection module; The elevator door data labeling module is used for elevator door data labeling, and performs data labeling by constructing a kernel K-means clustering model to obtain a fault detection data set, and sends the fault detection data set to the fault detection model construction module; The fault detection model building module is used to build a fault detection model, obtain the fault detection model by building a bidirectional long short-term memory network model combined with an optimization algorithm, and send the fault detection model to the elevator door system fault detection module; The elevator door system fault detection module is used for elevator door system fault detection. The fault detection model is used to perform elevator door system fault detection to obtain fault detection reference data.

Images