Lifting system fault diagnosis method based on convex hull and CNN-LSTM

By applying the fault diagnosis method of convex hull and CNN-LSTM in the mine lifting system, combining data preprocessing and multivariate analysis, the problem of inaccurate fault prediction in the existing technology is solved, and efficient fault detection and development of the lifting system is achieved. diagnosis.

CN118981952APending Publication Date: 2024-11-19YANKUANG ENERGY GRP CO LTD +2
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411052655.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-01
Publication Date
2024-11-19

AI Technical Summary

Technical Problem

The existing technology cannot accurately predict faults of mine lifting systems, and there are problems of reaction lag, false alarms and missed alarms.

Method used

Using a fault diagnosis method based on convex hull and CNN-LSTM, the fault diagnosis method of the improvement system is achieved through data preprocessing, Pearson correlation coefficient analysis, convex hull model and CNN-LSTM neural network, combined with multivariate process alarm monitoring and trend prediction. Test and diagnosis.

Benefits of technology

Improves the accuracy and reliability of fault detection, can effectively identify abnormal states and predict potential faults, and improves the accuracy of data analysis and modeling.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118981952A_ABST
    Figure CN118981952A_ABST
Patent Text Reader

Abstract

The invention relates to an elevator multivariate data collaborative diagnosis method based on a convex hull and a CNN-LSTM neural network. Aiming at anomaly detection and diagnosis of an elevator system, the method comprises the following steps: firstly, carrying out Pearson correlation analysis on parameters of the system by utilizing a Pearson correlation analysis technology, and then screening out parameter combinations with high Pearson correlation to carry out multi-dimensional collaborative analysis so as to realize effective detection and analysis of an abnormal state; secondly, for non-continuous parameters such as torque and rotating speed, a convex hull algorithm is adopted to monitor and diagnose an abnormal state in real time; for continuous parameters such as stator winding temperature, convolutional neural network-long and short term memory is used as a prediction model for predicting potential abnormal conditions.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of hoisting system fault diagnosis, and in particular to a hoisting system fault diagnosis method based on convex hull and CNN-LSTM. Background Art

[0002] Mine hoisting system is a special equipment used to hoist personnel, ore, waste rock, materials and equipment in coal, metal and non-metal mines. Due to the complexity of mine hoisting system and the functional limitations of each module, there are problems such as delayed response to some special situations, even false alarms, missed alarms and blind spots in actual use of the system.

[0003] Therefore, there is a need for a lifting system fault diagnosis method based on convex hull and CNN-LSTM that can accurately predict the faults of the lifting system. Summary of the invention

[0004] The main purpose of the present invention is to provide a lifting system fault diagnosis method based on convex hull and CNN-LSTM to solve the problem that the prior art cannot predict faults of the lifting system.

[0005] To achieve the above object, the present invention provides a lifting system fault diagnosis method based on convex hull and CNN-LSTM, which specifically includes the following steps:

[0006] S1, the original lifting system data is preprocessed by normalization method;

[0007] S2, first observe whether the data points are stable within the time interval. If the data fluctuates less within the time interval, it means that the temporal continuity of the data is strong, otherwise it means that the temporal continuity of the data is weak. Then use the Pearson correlation coefficient method to determine the Pearson correlation between variables with strong continuity and the Pearson correlation between variables with weak continuity. If the absolute value of the Pearson coefficient tends to 1, it means that the Pearson correlation is high. If the absolute value of the Pearson coefficient tends to 0, it means that the Pearson correlation is low. Variables with strong continuity and high Pearson correlation coefficients are combined, and variables with weak continuity and high Pearson correlation coefficients are combined.

[0008] S3, combining the stator current and rotor current with weak continuity and high Pearson correlation coefficient, and analyzing the algorithm of multivariate process alarm monitoring based on the normal working area of ​​convex hull;

[0009] S4, the torque, speed and power combinations with weak continuity and high Pearson correlation coefficient are analyzed by the algorithm of multivariate process alarm monitoring based on the normal working area of ​​convex hull;

[0010] S5, multiple stator winding temperature combinations with strong continuity and high Pearson correlation coefficient are input into the CNN-LSTM neural network to predict the trends of multiple stator winding temperatures respectively.

[0011] Furthermore, step S1 specifically uses formula (1) to perform data preprocessing on the system variables of the lifting system:

[0012]

[0013] In the formula, x max and x min They represent the maximum and minimum values ​​of the features, respectively, and x and x represent the features before and after normalization, respectively.

[0014] Furthermore, the trend graphs of the parameters are used to determine whether the continuity is strong or weak. Figures 1 to 11 It is a trend diagram of each parameter. By observing the trend diagram, it can be seen that the parameters with weak continuity in step S2 include the stator current, rotor current, power, torque and speed of the lifting system, and the parameters with strong continuity include multiple stator winding temperatures of the lifting system:

[0015] Furthermore, step S3 specifically includes the following steps:

[0016] S2.1, the torque, speed and power of the lifting system are represented by three simulated process variables x1, x2, x3 respectively. The three simulated process variables constitute a multivariable process X. The historical data points under normal conditions are represented as: {X(t)|t∈[1,N]}, where X(t): = [x1(t), x2(t), x3(t)]; where t and N are the sampling time index and the number of data points respectively.

[0017] S2.2, the convex hull model composed of the torque, speed and power of the lifting system is a three-dimensional convex hull model, and the equation of each convex hull surface is expressed by formula (2):

[0018] a j x+b j y+c j z+d j =0 (2);

[0019] Among them, a j , b j , c j , d j They correspond to the coefficients of the jth convex hull plane in the three-dimensional convex hull, and x, y, and z correspond to the three dimensions x1, x2, and x3 respectively.

[0020] S2.3, t i The data point at time [x1(t i ), x2(ti ), x3(t i )] into the equation of the convex hull. If the calculated results of all convex hulls are less than or equal to 0, it means that the point is inside the convex hull.

[0021] Furthermore, step S4 specifically includes the following steps:

[0022] S3.1, the stator current and rotor current of the lifting system are represented by two simulated process variables x1 and x2. The multivariable process X composed of the two simulated process variables, the historical data points under normal conditions are represented as {X(t)|t∈[1,N]}, where X(t): = [x1(t), x2(t)].

[0023] S3.2, the convex hull model composed of the stator current and rotor current of the lifting system is a two-dimensional convex hull model. The data points constituting the convex hull are arranged in order. If t i The improvement system data point [x1(t i ), x2(t i )] satisfies formula (3), then the data point is considered to be outside the convex hull or on the boundary of the convex hull:

[0024] (y j+1 -y j )*(x1(t i )-x j+1 )-(x j+1 -x j )*(x2(t i )-x j+1 )≤0 (3);

[0025] Among them, x j and x j+1 Respectively represent the values ​​of the jth and j+1th convex hull points in the x1 dimension, y j and j+1 Represent the values ​​of the j-th and j+1-th convex hull points in the x2 dimension respectively.

[0026] S3.3, if t i If the lifting system data point at time is outside or on the boundary of all the boundaries of the two-dimensional convex hull model, then it is considered that t i The data points of the lifting system at the moment are within the convex hull, that is, the stator current and rotor current of the lifting system at this moment are normal.

[0027] Furthermore, in step S5, the temperatures of six stator windings of the lifting system are combined and input into the CNN-LSTM network model, and the CNN-LSTM neural network model includes: an input layer, a CNN layer, an LSTM layer and a linear output layer connected in sequence.

[0028] The present invention has the following beneficial effects:

[0029] Based on the operation mechanism of the lifting system, the present invention uses data-based modeling and analysis methods, and combines them with abnormal warning and fault diagnosis technology to perform multi-parameter analysis on the operation status of the main equipment of the lifting system. Through the Pearson correlation analysis of the variables, the variable change area is obtained for the parameters with weak continuity, and for the parameters with strong continuity, an abnormal alarm can be given according to the change trend of the variables. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the specific implementation of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the specific implementation or the prior art description. Obviously, the drawings described below are some implementations of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0031] Figure 1 shows the trend diagram of the stator current;

[0032] Figure 2 shows the trend graph of the rotor current;

[0033] Figure 3 A trend graph of power is shown;

[0034] Figure 4 A trend diagram of stator winding temperature 1 is shown;

[0035] Figure 5 shows a trend diagram of stator winding temperature 2;

[0036] Figure 6 shows a trend diagram of the stator winding temperature 3;

[0037] Figure 7 shows a trend diagram of the stator winding temperature 4;

[0038] Figure 8 shows a trend diagram of the stator winding temperature 5;

[0039] Fig. 9 shows a trend diagram of the stator winding temperature 6;

[0040] Fig.10 A trend graph of torque is shown;

[0041] Fig.11 A trend graph of the rotation speed is shown;

[0042] Fig.12 Pearson correlation coefficient plots for the continuous strong parameters are shown;

[0043] Fig.13 Pearson correlation coefficient plots for continuous weak parameters are shown;

[0044] Fig.14 shows the relationship between the stator current and the rotor current under normal operating conditions;

[0045] Fig.15 A diagram showing the relationship between stator current and rotor current under abnormal operating conditions;

[0046] Fig.16 A graph showing the relationship between torque, speed and power under normal operating conditions is shown;

[0047] Fig.17 A graph showing the relationship between torque, speed and power under abnormal operating conditions;

[0048] Fig.18 The structural diagram of CNN-LSTM is shown;

[0049] Fig.19 The structural diagram of LSTM is shown;

[0050] Fig. 20 The figure shows the prediction effect of stator winding temperature 1 based on the CNN-LSTM neural network model;

[0051] Fig.21 The figure shows the prediction effect of stator winding temperature 2 based on the CNN-LSTM neural network model; DETAILED DESCRIPTION

[0052] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] A lifting system fault diagnosis method based on convex hull and CNN-LSTM specifically includes the following steps:

[0054] S1, the original lifting system data is preprocessed by normalization method, aiming to eliminate the scale differences between different features, thereby improving the convergence speed of model training and enhancing the stability of the model. This preprocessing method helps to establish a data feature set with a unified scale, effectively eliminating the impact of the dimensional differences between different features on model training and prediction, and providing a more robust and reliable foundation for subsequent data analysis and modeling processes;

[0055] S2, first observe whether the data points are stable within the time interval. If the data fluctuates less within the time interval, it means that the temporal continuity of the data is strong, otherwise it means that the temporal continuity of the data is weak. Then use the Pearson correlation coefficient method to determine the Pearson correlation between variables with strong continuity and the Pearson correlation between variables with weak continuity. If the absolute value of the Pearson coefficient tends to 1, it means that the Pearson correlation is high. If the absolute value of the Pearson coefficient tends to 0, it means that the Pearson correlation is low. Variables with strong continuity and high Pearson correlation coefficients are combined, and variables with weak continuity and high Pearson correlation coefficients are combined.

[0056] Pearson correlation coefficient analysis is a statistical tool designed to evaluate the linear correlation and direction between two variables. By calculating the Pearson correlation coefficient, the strength of this association can be measured. Its value ranges from -1 to 1. The value of the Pearson correlation coefficient can reveal the degree of association between the variables, including positive correlation, negative correlation or no correlation.

[0057] In terms of constructing indicators to improve the correlation of system parameters, the Pearson correlation coefficient is calculated to explore the correlation between variables, and then determine whether there is a positive correlation, negative correlation or no correlation. By examining multiple highly correlated variables, outliers that may have a negative impact on the model can be more accurately identified, and outliers can be effectively diagnosed and corrected. This method helps to improve the reliability and accuracy of outlier diagnosis, thereby improving the accuracy of data analysis and modeling.

[0058] S3, combining the stator current and rotor current with weak continuity and high Pearson correlation coefficient, and analyzing them by using a multivariate process alarm monitoring algorithm based on the normal working area of ​​the convex hull. For variables with weak continuity, the present invention adopts a multivariate process alarm monitoring algorithm based on the normal working area of ​​the convex hull to perform multi-parameter data collaborative analysis. The convex hull algorithm fits the convex hull of the data in a multidimensional space, thereby comprehensively considering the overall structure and relationship of the data and effectively detecting outliers. Especially in high-dimensional data sets, the convex hull algorithm can more comprehensively capture the impact of outliers and improve the accuracy and reliability of anomaly detection.

[0059] Based on the convex hull theory, the normal operating state of the multivariable process is described by establishing the convex hull of the normal operating area. When multiple parameters are abnormal at the same time, the convex hull algorithm can monitor the position of these parameters in the convex hull for collaborative analysis, thereby improving the efficiency and accuracy of fault detection. Specifically, the convex hull algorithm first samples and processes multiple parameters, and then uses the convex hull model to describe the relationship between these parameters. Once a parameter or multiple parameters in the system exceed the normal operating area represented by the convex hull, an alarm will be triggered.

[0060] S4, the torque, speed and power combinations with weak continuity and high Pearson correlation coefficient are analyzed by the algorithm of multivariate process alarm monitoring based on the normal working area of ​​convex hull;

[0061] S5, multiple stator winding temperature combinations with strong continuity and high Pearson correlation coefficient are input into the CNN-LSTM neural network, and the trends of multiple stator winding temperatures are predicted respectively. When predicting the trend of variables with strong continuity, the present invention adopts convolutional neural network-long short-term memory (CNN-LSTM) as a model to evaluate whether the equipment may be abnormal through trend prediction, wherein CNN processes sequence data by applying a sliding convolution filter to the input, and CNN can learn features from spatial and temporal dimensions. At the same time, LSTM is a recurrent neural network suitable for sequence data, and LSTM can achieve end-to-end modeling and prediction of time series data. The excellent sequence modeling ability of this model enables CNN-LSTM to more comprehensively analyze the characteristics of the entire time series in outlier diagnosis, rather than just being limited to the abnormality of a single data point.

[0062] The structure diagram of CNN-LSTM is shown in the figure attached to the specification. Fig.18 As shown, the structure diagram of LSTM is shown in the figure of the specification. Fig.19 As shown in the figure, CNN treats time series data as one-dimensional data. Through the convolutional layer, CNN can learn important features in time series data. LSTM is a special recurrent neural network architecture that mainly controls the flow of information and the update of memory by introducing three gating units: input gate, forget gate, and output gate, thereby effectively capturing long-term dependencies. Through the three key gating units, LSTM can accurately control the update of cell states and the output of hidden states. These gating units use the sigmoid activation function to generate values ​​between 0 and 1, which respectively represent the degree of acceptance of new information, the degree of forgetting of old information, and the degree of cell state output, thereby achieving long-term storage of information and capturing long-term dependencies.

[0063] The present invention detects and diagnoses the abnormality of the lifting system. The present invention first proposes to use the Pearson correlation analysis technology to carry out Pearson correlation analysis on various parameters of the lifting system, and then screen out parameter combinations with high Pearson correlation for multi-dimensional collaborative analysis to achieve effective detection and analysis of abnormal conditions; secondly, for parameters with weaker continuity (i.e., non-continuous parameters) such as torque and speed, the convex hull algorithm is used to monitor and diagnose the abnormal condition in real time; for continuous parameters (i.e., parameters with stronger continuity) such as stator winding temperature, the convolutional CNN-LSTM neural network is used as a prediction model to predict potential abnormal conditions.

[0064] The present invention firstly performs data preprocessing on the raw data of the lifting system by using the normalization method, thereby compressing the data range to between 0 and 1. Fig.12 It can be seen that the Pearson correlation of the 6 stator winding temperatures in the lifting system is relatively high, and the stator winding temperature is a variable with strong continuity. Therefore, the CNN-LSTM neural network prediction is performed after the 6 stator winding temperatures are combined. Fig.13 It can be seen that the Pearson correlation of the stator current and the rotor current in the lifting system is relatively high, and the Pearson correlation of the torque, speed and power is relatively high. The stator current, rotor current, torque, speed and power are variables with strong continuity. Therefore, the stator current and the rotor current are analyzed using the multivariable process alarm monitoring algorithm based on the normal working area of ​​the convex hull, and the torque, speed and power are analyzed using the multivariable process alarm monitoring algorithm based on the normal working area of ​​the convex hull.

[0065] Specifically, in order to eliminate the influence of the large dimensional differences between different features of the lifting system, the minimum-maximum normalization process is used to scale the lifting system variables to between 0 and 1. Step S1 specifically uses formula (1) to preprocess the system variables of the lifting system:

[0066]

[0067] In the formula, x max and x min They represent the maximum and minimum values ​​of the features, respectively, and x and x represent the features before and after normalization, respectively.

[0068] Specifically, the trend graph of each parameter is used to determine whether the continuity is strong or weak. Figures 1 to 11 It is a trend diagram of each parameter. By observing the trend diagram, it can be seen that the parameters with weak continuity in step S2 include the stator current, rotor current, power, torque and speed of the lifting system, and the parameters with strong continuity include multiple stator winding temperatures of the lifting system:

[0069] Specifically, there are multiple parameters with weak continuity in the lifting system, and its normal state can be represented by the normal operating zone (NOZ) determined by multiple related process variables. The convex hull description of the NOZ is based on the adaptability between the measured data points and the historical normal data points, and is modeled by a formulated adaptability index. According to this model, if all process variables are within the NOZ, there is no need to issue an alarm; otherwise, an alarm notification will be triggered. The dynamic alarm limit of each process variable is associated with other related process variables and has the same physical meaning.

[0070] Step S3 specifically includes the following steps:

[0071] S2.1, the torque, speed and power of the lifting system are represented by three simulated process variables x1, x2, x3 respectively. The three simulated process variables constitute a multivariable process X. The historical data points under normal conditions are represented as: {X(t)|t∈[1, N]}, where X(t): = [x1(t), x2(t), x3(t)]; where t and N are the sampling time index and the number of data points, respectively. X(t) under normal conditions should be within the NOZ, and X(t) under abnormal conditions should be outside the NOZ. The historical normal data points are located in the normal operating zone NOZ, so X(t) under normal conditions should be within the NOZ, and X(t) under abnormal conditions should be outside the NOZ.

[0072] S2.2, the convex hull model composed of the torque, speed and power of the lifting system is a three-dimensional convex hull model, and the equation of each convex hull surface is expressed by formula (2):

[0073] a j x+b j y+c j z+d j =0 (2);

[0074] Among them, a j , b j , c j , d j They correspond to the coefficients of the jth convex hull plane in the three-dimensional convex hull, and x, y, and z correspond to the three dimensions x1, x2, and x3 respectively.

[0075] S2.3, in order to determine the t i Whether the data points consisting of the torque, speed and power of the lifting system are inside the convex hull at the moment t i The data point at time [x1(t i ), x2(t i ), x3(t i )] into the equation of the convex hull. If the calculated results of all convex hulls are less than or equal to 0, it means that the point is inside the convex hull.

[0076] Specifically, step S4 includes the following steps:

[0077] S3.1, the stator current and rotor current of the lifting system are represented by two simulated process variables x1 and x2. The multivariable process X composed of the two simulated process variables, the historical data points under normal conditions are represented as {X(t)|t∈[1,N]}, where X(t): = [x1(t), x2(t)].

[0078] S3.2, the convex hull model composed of the stator current and rotor current of the lifting system is a two-dimensional convex hull model. Therefore, when judging whether the data point composed of the stator current and rotor current of the lifting system at a certain moment is inside the convex hull, the property of the convex hull is used, that is, any point on the convex hull to the boundary of the convex hull is outside the boundary or on the boundary. Therefore, each edge of the convex hull is traversed to check whether the data point at that moment is outside the boundary. If there is an edge that makes the data point at that moment inside the boundary, the data point at that moment is not in the convex hull.

[0079] Arrange the data points that make up the convex hull in order. If t i The improvement system data point [x1(t i ), x2(t i )] satisfies formula (3), then the data point is considered to be outside the convex hull or on the boundary of the convex hull:

[0080] (y j+1 -y j )*(x1(t i )-x j+1 )-(x j+1 -x j )*(x2(t i )-x j+1 )≤0 (3);

[0081] Among them, x j and x j+1 Respectively represent the values ​​of the jth and j+1th convex hull points in the x1 dimension, y j and j+1 Represent the values ​​of the j-th and j+1-th convex hull points in the x2 dimension respectively.

[0082] S3.3, if t i If the lifting system data point at time is outside or on the boundary of all the boundaries of the two-dimensional convex hull model, then it is considered that t i The data points of the lifting system at the moment are within the convex hull, that is, the stator current and rotor current of the lifting system at this moment are normal.

[0083] Specifically, in step S5, the temperatures of the six stator windings of the lifting system are combined and input into the CNN-LSTM network model. The CNN-LSTM neural network model includes: an input layer, a CNN layer, an LSTM layer and a linear output layer connected in sequence.

[0084] The six stator winding temperatures of the lifting system are combined into the CNN-LSTM network model, and six CNN-LSTM neural network models are set to predict the six stator winding temperatures respectively. The length of the input layer size of each CNN-LSTM neural network model is 6, the hidden layer size is 20, and there are 1 CNN layer, 1 LSTM layer and 1 linear layer output. In the forward propagation process, the input data passes through the CNN layer and LSTM layer to obtain the output and hidden state, and then the hidden state passes through the linear layer to obtain the final output. In the model training process, the Adam optimizer is used for back propagation. Adam is adaptive when calculating the mean and variance of the gradient, and can automatically adjust the learning rate. It is suitable for different scales and sparse gradients of different parameters.

[0085] The CNN-LSTM neural network model is used to predict the trend of variables with strong continuity, which includes the following steps:

[0086] Build a neural network model that includes CNN layers and LSTM layers to capture long-term dependencies in the data. The model should also include fully connected layers to improve the accuracy and generalization ability of the model.

[0087] Use the historical data of the boost system to train the CNN-LSTM model, and continuously adjust the model parameters through the back-propagation algorithm to minimize the loss function between the predicted value and the actual value;

[0088] After the training is completed, the trained CNN-LSTM model is used to monitor the status of the data with strong continuity of the lifting system. The CNN-LSTM model is deployed in practical applications, and the data of the current time period is input. The historical laws and trends learned through the network structure are used to determine whether the status of the lifting system in the current time period will be abnormal.

[0089] The formula for the one-dimensional convolution kernel in the CNN layer in the CNN-LSTM structure is as follows:

[0090]

[0091] Among them, y i represents the i-th element in the output sequence, x i represents the fth element in the input sequence, w j represents the,th element in the convolution kernel, and k represents the length of the convolution kernel.

[0092] In a one-dimensional convolution operation, the convolution kernel moves on the input sequence in a sliding window manner, calculates the sum of the products at the corresponding positions, and obtains the corresponding elements in the output sequence. In this way, through the convolution operation, the feature information in the input sequence can be extracted, and the result processed by the convolution kernel can be output.

[0093] Cell states that store long-term information t The formula for modulation combined with a series of gates is shown below:

[0094] Input Gate i t :

[0095] Output gate o t :

[0096] Forget Gate t :

[0097] Among them, W i1 refers to the weight of the hidden state in the input gate, W i2 refers to the weight of the input data in the input gate, W o1 refers to the weight of the hidden state in the output gate, W o2 Refers to the weight of the input data in the output gate, W f1 refers to the weight of the hidden state in the forget gate, W f2 refers to the weight of the input data in the forget gate, b i is the bias of the input gate, b o is the bias of the output gate, b f is the bias of the forget gate, refers to the input data of LSTM, h t-1 is the hidden state of LSTM, and σ(.) is the sigmoid activation function.

[0098] In the above formula, the input gate is responsible for determining which new information should be added to the memory cell, the output gate determines which information should be stored in the hidden state at the current moment, and the forget gate determines which past memories should be retained at the current moment.

[0099] The hidden and cell states of the LSTM are as follows:

[0100] Hidden state h t :h t =o t ⊙tanh(c t );

[0101] Cell state c t :

[0102] Among them, t is the output gate, f t For the forget gate, i t is the input gate, ⊙ is the element-wise product, tanh(.) is the tanh activation function, W c1 refers to the weight that controls the hidden state in the cell state, Wc2 refers to the weight of the input data in the control cell state, b t is the bias that controls the cell state, Refers to the input data of LSTM.

[0103] In the above formula, the hidden state is the state value output by the LSTM model at each time step, which represents the information and features of the sequence learned by the model at that moment. The cell state is a state variable unique to the LSTM network, which is used to store long-term memory and information.

[0104] The key design of LSTM is the introduction of the gating mechanism of input gate, forget gate and output gate. The forget gate allows the network to selectively remember or forget past information, the input gate is responsible for determining the importance of the current input and integrating it into the new state, and the output gate determines which information in the cell state should be output. This design enables LSTM to effectively handle long-term dependencies and capture long-distance dependencies in sequence data without losing gradients or causing gradient explosions.

[0105] The present invention firstly performs data preprocessing on the original lifting system data, and uses deviation standardization to process the original lifting system data. Since the dimensional differences between different characteristics of the lifting system are large, and the dimensional differences will have a great impact on the training and prediction of the model, the data of the lifting machine system are processed by minimum-maximum normalization, and the lifting machine system variables are scaled to between 0 and 1. The continuity of each parameter is determined by observing the trend graph of each parameter, and then the Pearson correlation coefficient method is used to respectively determine the Pearson correlation between variables with strong continuity and the Pearson correlation between variables with weak continuity, and the variables with high Pearson coefficients are combined. For the variables with weak continuity, a multivariate process alarm monitoring algorithm based on the normal working area of ​​the convex hull is used to perform multi-parameter data collaborative analysis, and a CNN-LSTM neural network is used to perform trend prediction for the variables with strong continuity.

[0106] In order to verify the performance of the method provided by the present invention, the following simulation experiments were performed:

[0107] This study conducted a simulation experiment based on 8 days of data from a coal mine. The results showed that the method provided by this paper showed good performance in abnormal data detection and also showed high accuracy in predicting temperature trends.

[0108] The parameters selected by the present invention with weak continuity include the stator current, rotor current, power, torque and speed of the lifting system, and the parameters with strong continuity include the 6 stator winding temperatures of the lifting system. The thermal diagram is plotted under the Pearson correlation coefficient with strong continuity and the Pearson correlation coefficient with weak continuity, such as Fig.12 and Fig.13As shown. Fig.12 It can be seen that among the Pearson correlation coefficients with strong continuity, the Pearson coefficients between the six stator winding temperatures are relatively high. Fig.13 It can be seen that among the Pearson correlation coefficients with weak continuity, the Pearson coefficient between the stator current and the rotor current is relatively high, and the Pearson coefficients of torque, speed, and power are relatively high;

[0109] according to Fig.14 and Fig.15 The results shown in the figure show that under the multivariable process alarm monitoring algorithm based on the normal working area of ​​the convex hull, the present invention performs a two-dimensional convex hull analysis on the stator current and rotor current of the lifting system. Fig.14 The relationship between the stator current and the rotor current under normal working conditions is shown, where the blue represents the historical normal data points, the green represents the detected normal data points, and Fig.15 This shows the situation when the data is abnormal (rotor current is too large). Since there is no abnormal data in the data set, the rotor current is manually increased, and the effect is as follows: Fig.15 As shown in the figure, blue represents historical normal data points, and red represents detected abnormal data points. In addition, Fig.16 and Fig.17 The three-dimensional convex hull of the torque, speed and power of the lifting system is displayed to reflect the algorithm's monitoring of the normal working area. Fig.16 The relationship between torque, speed and power under normal working conditions is depicted, where blue represents historical normal data points, green represents detected normal data points, and Fig.17 This represents the situation when the data is abnormal (i.e. the power is too large). Since there is no abnormal data in the data set, the power is manually increased, and the effect is as follows: Fig.17 As shown, the blue in the figure represents historical normal data points, and the red represents detected abnormal data points. In summary, according to the above results, the algorithm performs well and successfully identifies abnormal data;

[0110] The structure diagram of CNN-LSTM is shown in the figure attached to the specification. Fig.18 As shown, the structure diagram of LSTM is shown in the figure of the specification. Fig.19 As shown in the figure, the CNN-LSTM network structure helps to capture the long-term dependencies in the time series, so that it can consider more historical information in the prediction process. At the same time, the CNN-LSTM network can effectively process time series data at different time scales, thereby adapting to various complex time change patterns. Through the design of the gate unit, the CNN-LSTM network can selectively remember or forget past information, thereby better adapting to the characteristics of time series data. Fig. 20 and Fig.21As shown, the red line segment represents the actual value of the temperature, and the green line segment represents the predicted value of the temperature. Fig. 20 The figure shows the prediction effect of stator winding temperature 1 based on the CNN-LSTM neural network model. Fig.21 The following figure shows the prediction effect of stator winding temperature 2 based on the CNN-LSTM neural network model. The prediction effects of the other four stator winding temperatures are similar to those of stator winding temperature 1 and stator winding temperature 2. Fig. 20 and Fig.21 The results show that when the CNN-LSTM network model is used, the prediction effect of the six stator winding temperatures shows significant superiority.

[0111] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the essential scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A lifting system fault diagnosis method based on convex hull and CNN-LSTM, characterized in that: The specific steps include: S1, the original lifting system data is preprocessed by normalization method; S2, first observe whether the data points are stable within the time interval. If the data fluctuates less within the time interval, it means that the temporal continuity of the data is strong, otherwise it means that the temporal continuity of the data is weak. Then use the Pearson correlation coefficient method to determine the Pearson correlation between variables with strong continuity and the Pearson correlation between variables with weak continuity. If the absolute value of the Pearson coefficient tends to 1, it means that the Pearson correlation is high. If the absolute value of the Pearson coefficient tends to 0, it means that the Pearson correlation is low. Variables with strong continuity and high Pearson correlation coefficients are combined, and variables with weak continuity and high Pearson correlation coefficients are combined. S3, combining the stator current and rotor current with weak continuity and high Pearson correlation coefficient, and analyzing the algorithm of multivariate process alarm monitoring based on the normal working area of ​​convex hull; S4, the torque, speed and power combinations with weak continuity and high Pearson correlation coefficient are analyzed by the algorithm of multivariate process alarm monitoring based on the normal working area of ​​convex hull; S5, multiple stator winding temperature combinations with strong continuity and high Pearson correlation coefficient are input into the CNN-LSTM neural network to predict the trends of multiple stator winding temperatures respectively.

2. A lifting system fault diagnosis method based on convex hull and CNN-LSTM according to claim 1, characterized in that: Step S1 specifically uses formula (1) to preprocess the system variables of the lifting system: In the formula, x max and x min represent the maximum and minimum values ​​of the features, respectively, and x and x represent the features before and after normalization, respectively.

3. A lifting system fault diagnosis method based on convex hull and CNN-LSTM according to claim 1, characterized in that: The trend graph of each parameter is used to determine whether it is a parameter with strong or weak continuity. By observing the trend graph, it can be seen that the parameters with weak continuity in step S2 include the stator current, rotor current, power, torque and speed of the lifting system, and the parameters with strong continuity include multiple stator winding temperatures of the lifting system.

4. The method for diagnosing a lifting system fault based on convex hull and CNN-LSTM according to claim 1, characterized in that: Step S3 specifically includes the following steps: S2.1, the torque, speed and power of the lifting system are represented by three simulated process variables x1, x2, x3 respectively, and the three simulated process variables constitute a multivariable process X. The historical data points under normal conditions are represented as: {X(t)|t∈[1,N]}, where X(t):=[x1(t), x2(t), x3(t)]; where t and N are the sampling time index and the number of data points respectively; S2.2, the convex hull model composed of the torque, speed and power of the lifting system is a three-dimensional convex hull model, and the equation of each convex hull surface is expressed by formula (2): a j x+b j y+c j z+d j =0 (2); Among them, a j , b j , c j , d j They correspond to the coefficients of the jth convex hull plane in the three-dimensional convex hull, and x, y, and z correspond to the three dimensions x1, x2, and x3 respectively; S2.3, t i The data point at time [x1(t i ), x2(t i ), x3(t i )] into the equation of the convex hull. If the calculated results of all convex hulls are less than or equal to 0, it means that the point is inside the convex hull.

5. The method for fault diagnosis of a lifting system based on convex hull and CNN-LSTM according to claim 1, characterized in that: Step S4 specifically includes the following steps: S3.1, the stator current and rotor current of the lifting system are represented by two simulated process variables x1, x2, and the multivariable process X composed of the two simulated process variables, the historical data points under normal conditions are represented as {X(t)|t∈[1,N]}, where X(t):=[x1(t),x2(t)]; S3.2, the convex hull model composed of the stator current and rotor current of the lifting system is a two-dimensional convex hull model. The data points constituting the convex hull are arranged in order. If t i The improvement system data point [x1(t i ), x2(t i )] satisfies formula (3), then the data point is considered to be outside the convex hull or on the boundary of the convex hull: (y j+1 -y j )*(x1(t i )-x j+1 )-(x j+1 -x j )*(x2(t i )-x j+1 )≤0 (3)? Among them, x j and x j+1 Respectively represent the values ​​of the jth and j+1th convex hull points in the x1 dimension, y j and j+1 Respectively represent the values ​​of the j-th and j+1-th convex hull points in the x2 dimension; S3.3, if t i If the lifting system data point at time is outside or on the boundary of all the boundaries of the two-dimensional convex hull model, then it is considered that t i The data points of the lifting system at the moment are within the convex hull, that is, the stator current and rotor current of the lifting system at this moment are normal.

6. A lifting system fault diagnosis method based on convex hull and CNN-LSTM according to claim 1, characterized in that: In step S5, the six stator winding temperatures of the lifting system are combined and input into the CNN-LSTM network model. The CNN-LSTM neural network model includes: an input layer, a CNN layer, an LSTM layer and a linear output layer connected in sequence.