An elevator door machine system anomaly detection and prediction method

Abstract

The application discloses an elevator door machine system anomaly detection and prediction method, and relates to the technical field of elevator door machine anomaly detection.The application comprises the following steps: S1, multi-dimensional data acquisition; S2, operation characteristic extraction; S3, dynamic benchmark establishment; S4, abnormal characteristic identification; and S5, output of fault type and confidence degree.The application trains an LSTM model by using historical operation data of a door machine system, so that the operation cycle of the elevator door machine system can be automatically divided, the defects in the prior art are overcome, a plurality of detection data in the operation cycle can be accurately segmented, and the value of the detection data for subsequent fault prediction of the door machine system is ensured.The application establishes a dynamic benchmark range curve associated with each characteristic parameter and the number of door machine operations, improves the accuracy of fault prediction of the door machine system, reduces the false alarm rate of the door machine system, and saves maintenance resources to a certain extent.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of elevator door machine anomaly detection, and in particular relates to an elevator door machine system anomaly detection and prediction method. BACKGROUND

[0002] In modern urban buildings, elevators have become the arteries connecting vertical space. The door machine system of the elevator is a complex system composed of mechanical hooks, light curtain sensors, door machine controllers and other components, and its anomaly detection is directly related to the safety of taking the elevator and the service life of the equipment. Therefore, it is extremely necessary to detect the anomaly of the elevator door machine system.

[0003] The prior art such as the industrial equipment intelligent operation and maintenance cloud platform disclosed in the patent application with the publication number CN118154174B significantly improves the accuracy and efficiency of industrial equipment fault diagnosis by integrating advanced data analysis and machine learning technology, optimizes maintenance decisions, reduces operation and maintenance costs, and improves the reliability and safety of equipment operation. The elevator door machine system anomaly detection and prediction method disclosed in the patent application with the publication number CN117446611A realizes comprehensive detection and prediction of elevator door machine system operation data, helping maintenance personnel to discover and solve problems in a timely manner.

[0004] In combination with the above-mentioned scheme, it can be found that in the prior art, the running period of the elevator door machine system is mostly determined by manual labeling or transmission signals of the door machine system, and there are few uses of LSTM models to cut the running period. Manual labeling is time-consuming and laborious, and the transmission signals of the door machine system have a delay, which leads to inaccurate running period determined according to the transmission signals, making it difficult to accurately segment the detection data within the running period, reducing the value of the fault prediction of the subsequent door machine system, and the attention to the dynamic reference range curve associated with the number of door machine operations is not high. With the increase of the number of door machine operations, the door machine gradually ages, and the division of the dynamic reference range curve is also different. The negligence of the prior art in this aspect increases the fault false alarm rate, which to some extent wastes maintenance resources. SUMMARY

[0005] The purpose of the present application is to provide an elevator door machine system anomaly detection and prediction method that solves the problems in the background art.

[0006] To solve the above technical problems, the present application adopts the following technical scheme: the present application provides an elevator door machine system anomaly detection and prediction method, comprising: S1, multi-dimensional data acquisition: arranging vibration sensors, current sensors, position encoders and temperature sensors in the elevator door machine system, and collecting vibration waveforms, electrode current curves, door position trajectories and temperature rise data in real time during door machine operation to form the original data of the door machine.

[0007] S2, running feature extraction: the original data of the door machine collected is divided according to the running period of the door machine, and a plurality of feature parameters corresponding to a plurality of feature groups are extracted for each running period.

[0008] S3, dynamic reference establishment: based on the factory parameters and historical normal running data of the door machine system, a dynamic reference range curve associated with the running number of each feature parameter is established, the dynamic reference range curve contains the allowed upper and lower limits of each feature parameter changing with the increase of the running number, and one running period corresponds to one running number.

[0009] S4, abnormal feature identification: based on a plurality of feature parameters corresponding to a plurality of feature groups extracted in real time, a plurality of abnormal features are screened, and the deviation direction and amplitude of the plurality of abnormal features are recorded.

[0010] S5, output fault type and confidence: a fault type and feature deviation mode library is established, and based on the deviation direction and amplitude of the plurality of abnormal features, the most possible fault type and confidence are output.

[0011] The beneficial effects of the present application are: (1) the present application trains the LSTM model through the historical running data of the door machine system, so as to achieve the purpose of automatically dividing the running period of the elevator door machine system, overcomes the defects in the prior art, accurately divides a plurality of detection data in the running period, and ensures the value of the subsequent door machine system fault prediction.

[0012] (2) the present application establishes a dynamic reference range curve associated with the running number of each feature parameter, improves the accuracy of door machine system fault prediction, reduces the false alarm rate of the door machine system, and saves maintenance resources to a certain extent. BRIEF DESCRIPTION OF DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiment or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0014] Figure 1 The method embodiment of the present application is shown in the flow chart. DETAILED DESCRIPTION

[0015] The technical solutions in the embodiments of the present application will be clearly and completely described with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort belong to the scope of protection of the present application.

[0016] Referring to Figure 1 As shown in the drawings, the present application provides an elevator door machine system anomaly detection and prediction method, comprising: S1, multi-dimensional data acquisition: arranging vibration sensors, current sensors, position encoders and temperature sensors in the elevator door machine system, collecting vibration waveform, electrode current curve, door position trajectory and temperature rise data in real time when the door machine is running, to form original data of the door machine.

[0017] S2, operation feature extraction: dividing the collected original data of the door machine according to the door machine operation cycle, and extracting a plurality of feature parameters corresponding to a plurality of feature groups for each operation cycle.

[0018] In the specific embodiments of the present application, the original data of the door machine collected is divided according to the door machine operation cycle, and the specific steps are: S2001: time synchronizing the vibration waveform, electrode current curve, door position trajectory and temperature rise data in the original data of the door machine collected.

[0019] S2002: slidingly extracting time domain features, frequency domain features and time-frequency features with a fixed window length, to obtain time sequence features after multi-source data fusion.

[0020] The time domain features are: vibration: peak factor, pulse factor, waveform factor.

[0021] Current: starting current peak value, ending current peak value, braking time length, current harmonic distortion.

[0022] Position: starting moving speed, ending moving speed, positioning accuracy, trajectory smoothness.

[0023] Temperature rise: temperature rise rate, steady-state temperature difference, over-temperature threshold.

[0024] The frequency domain features are: vibration: main frequency energy ratio, frequency spectrum centroid.

[0025] Current: fundamental amplitude, harmonic component.

[0026] The time-frequency features are: transient impact features of vibration and current.

[0027] It should be noted that, since the position trajectory is usually a low-frequency, discrete or segmented continuous signal, the frequency domain feature has little contribution to the period division, so the time-frequency analysis is limited, and the frequency domain feature is not extracted, and the temperature rise data presents a slow change trend, and the frequency domain feature is mainly concentrated in the very low frequency band, which is not sensitive to the transient process of period division, so the frequency domain feature is not extracted, and the time-frequency analysis is limited, and the frequency domain feature is not extracted.

[0028] S2003: The time sequence features after multi-source data fusion are taken as the input of the LSTM model.

[0029] S2004: Whether it is a period boundary is predicted for each time step, so that the original data of the gate machine is divided according to the running period of the gate machine.

[0030] It should be noted that, for each time step, whether it is a period boundary is predicted, and the specific steps are as follows:

[0031] The boundary probability of each time step output by the LSTM model is compared with the boundary probability threshold, and the points exceeding the boundary probability threshold are recorded as candidate boundaries.

[0032] Merge adjacent candidate boundaries to avoid repeated labeling, and keep the point with the highest probability during the merging process.

[0033] Add physical property constraints of the gate machine, constraint condition 1 is the minimum period length, and constraint condition 2 is the braking stage proportion.

[0034] In specific embodiments of the application, the LSTM model has the following specific training method: A1, collect the historical running data of the gate machine system, and synchronize the time.

[0035] A2, segment the historical running data according to the running period of the gate machine, independently standardize each segment, the standardization is Z-Score standardization, and the historical running data is preliminarily processed. (The preliminary processing includes detecting abnormal values by using IQR method, replacing with neighborhood interpolation, using wavelet threshold denoising for vibration waveform, and performing power frequency filtering for electrode current curve at specific frequency)

[0036] A3, use the labeling tool to mark the boundaries of the period of the historical running data, generate an initial label set, and extract the training time sequence features after multi-source data fusion according to the method of S2002.

[0037] Exemplarily, the specific structure of the LSTM model is as follows:

[0038] LSTM layer: stack 2-3 layers of LSTM, each layer contains 64-148 hidden units, attention mechanism: add a self-attention layer after LSTM, calculate the weight of each time step, strengthen key features (such as signals in the start / brake phase), fully connected layer: finally connect a Dense layer, output the boundary probability (sigmoid function) of each time step, the boundary probability ranges from 0 to 1.

[0039] A4, the training time sequence features of the labeled multi-source data fusion of several periods are divided into training set, validation set and test set according to the set proportion, the training set, the validation set and the test set contain complete periods, and the final LSTM model is obtained by training and optimization.

[0040] In specific embodiments of the application, the final LSTM model is obtained by training and optimization, and the specific steps are: A501, dividing the training set into multiple batches, each batch containing a fixed number of samples, inputting a batch of samples into the LSTM model, outputting the prediction result of the current batch, and calculating the value of the loss function according to the prediction result and the initial label set.

[0041] A502, calculate the gradient of the loss function to the LSTM model parameters, the gradient represents the parameter update direction and size, and the loss function is minimized, use the optimizer to update the LSTM model parameters according to the gradient, continue iteration until the loss function is minimized.

[0042] It should be noted that the gradient is a vector, which represents the rate of change of the function in each direction at a certain point. In deep learning, the gradient represents the rate of change of the loss function value with respect to the LSTM model parameters. Its physical meaning is: the direction in which the loss function grows fastest, and its size represents the speed of change. In training, the LSTM model parameters are usually updated in the opposite direction of the gradient, that is, the direction in which the loss function decreases fastest, in order to minimize the loss function.

[0043] It also needs to be explained that the gradient of the loss function to the LSTM model parameter is calculated, and the specific method is: the input data is calculated through each layer of the LSTM model to obtain the prediction result, and the loss function value is calculated, and the gradient of the loss function to each layer parameter is calculated from the output layer to the input layer direction by the back propagation algorithm (using the chain rule), and the gradient of the final loss function is decomposed into the product of the gradient of each layer parameter, in this embodiment, the LSTM model includes a weight matrix and a bias vector, the weight matrix is specifically the weight of the input gate, the forgetting gate and the output gate, the gradient of the loss function value to all parameters (weight matrix and bias vector) in the LSTM model is calculated by the back propagation algorithm, the calculated gradient is input to the optimizer, the optimizer calculates the update amount of the LSTM model parameter according to the preset learning rate, momentum and other hyperparameters, and the update amount of the LSTM model parameter is applied to the original parameter to obtain the updated parameter value.

[0044] The function of the optimizer is to update the parameters of the LSTM model according to the calculated gradient of the loss function to the LSTM model parameter, so as to minimize the loss function. Common optimizers include stochastic gradient descent (SGD) and Adam.

[0045] A503, evaluate the performance of the trained LSTM model on the validation set, analyze the reasons for the validation set loss, adjust the LSTM model according to the reasons, retrain the model, and evaluate it on the validation set, repeat the above steps until the validation set loss meets the requirements.

[0046] It needs to be explained that the reasons for the validation set loss are analyzed, and the LSTM model is adjusted according to the reasons, the specific process is that if the training set loss continues to decrease, but the validation set loss first decreases and then increases, it means that the model complexity is too high, then methods such as increasing regularization, reducing the number of model layers or hidden units can be used, if the training set loss and the validation set loss are both high, it means that the model complexity is insufficient, then methods such as increasing the number of model layers or hidden units, reducing the regularization strength can be used, if the validation set loss fluctuates greatly or the convergence speed is slow, it means that the hyperparameters are not reasonable, then the hyperparameters are adjusted.

[0047] A504, select the model with the lowest loss on the validation set as the best LSTM model, and evaluate the performance of the best LSTM model on the test set, calculate the performance evaluation index, if the performance evaluation index does not meet the requirements, adjust the LSTM model, evaluate it on the validation set, select the best model and then verify it on the test set, until the test set performance meets the requirements, and get the final LSTM model.

[0048] The performance evaluation index in the above includes accuracy, recall and F1 value, etc. Since the invention is to detect the period boundary, F1 value is selected in this embodiment.

[0049] The adjusting LSTM model is specifically adjusted according to overfitting and underfitting of the LSTM model.

[0050] In specific embodiments of the present application, the plurality of feature parameters corresponding to the plurality of feature groups specifically include: the vibration feature group includes a vibration peak value, a main frequency amplitude value and a high-frequency energy proportion.

[0051] The current feature group includes a starting current maximum value, a steady-state current average value and a shutdown current drop duration.

[0052] The motion feature group includes a total door opening duration, a total door closing duration and a middle section uniform speed zone speed fluctuation value.

[0053] The temperature feature group includes a temperature rise rate, a steady-state temperature difference and an over-temperature threshold value.

[0054] The present application trains an LSTM model through historical operation data of the door machine system to achieve the purpose of automatically dividing the operation cycle of the elevator door machine system, overcomes the defects in the prior art, accurately segments a plurality of detection data in the operation cycle, and ensures the value of subsequent fault prediction of the door machine system.

[0055] S3, dynamic reference establishment: based on factory parameters and historical normal operation data of the door machine system, a dynamic reference range curve associated with each feature parameter and the number of door machine operations is established, the dynamic reference range curve contains an allowed upper and lower limit of each feature parameter changing with the increase of the number of operations, and one operation cycle corresponds to one number of operations.

[0056] In specific embodiments of the present application, the dynamic reference range curve associated with each feature parameter and the number of door machine operations is specifically established by: performing correlation analysis on each feature parameter and the number of operations of the door machine, drawing a scatter plot and fitting a trend line, constructing a linear model if it is a linear trend, constructing a nonlinear model if it is a nonlinear trend, bringing each feature parameter and the number of operations of the door machine into the linear model or the nonlinear model, and solving to obtain linear model coefficients or nonlinear model coefficients, to obtain a fitting formula associated with each feature parameter and the number of operations of the door machine.

[0057] It should be noted that the linear model and the nonlinear model are relatively mature and simple in the prior art, and will not be described here.

[0058] According to the factory parameters of the door machine system, the life cycle stage of the door machine system is identified, the reference range adjustment parameters of the door machine system in each life cycle stage are obtained from the data warehouse, and the reference range adjustment parameters of the door machine system in the current life cycle are matched.

[0059] Due to the current life cycle of the door machine system is different, the benchmark range adjustment parameters are also different, for example, the life cycle stages of the door machine system are running-in period, stable period and aging period, the characteristic fluctuation of the running-in period is large, the benchmark range is wide, the characteristic of the stable period is stable, the benchmark range is narrowed, and the characteristic of the aging period starts to deteriorate, and the benchmark range is dynamically widened.

[0060] According to the fitting formula that each characteristic parameter of the door machine is associated with the running frequency, and in combination with the benchmark range adjustment parameters of the door machine system in the current life cycle, a dynamic benchmark range curve of each characteristic parameter associated with the running frequency of the door machine is established.

[0061] It should be noted that the dynamic benchmark range curve includes an upper limit curve and a lower limit curve.

[0062] The present application establishes a dynamic benchmark range curve of each characteristic parameter associated with the running frequency of the door machine, improves the accuracy of fault prediction of the door machine system, reduces the false positive rate of the door machine system, and to some extent, saves maintenance resources.

[0063] S4, Abnormal feature recognition: based on the several characteristic parameters corresponding to the several feature groups extracted in real time, several abnormal features are screened, and the deviation direction and amplitude of the several abnormal features are recorded.

[0064] In specific embodiments of the present application, the several abnormal features are screened, and the deviation direction and amplitude of the several abnormal features are recorded, and the specific method is: according to the dynamic benchmark range curve of each characteristic parameter associated with the running frequency of the door machine, in combination with the current running frequency of the door machine, the dynamic benchmark range of each characteristic parameter is circled.

[0065] Compare the several characteristic parameters corresponding to the several feature groups extracted in real time with the dynamic benchmark range under the corresponding running frequency, if a certain characteristic parameter of a certain feature group is greater than the upper limit of the corresponding dynamic benchmark range, the characteristic parameter is recorded as an abnormal feature, the deviation direction is recorded as positive, and the part exceeding the upper limit value of the dynamic benchmark range is recorded as the amplitude, if a certain characteristic parameter of a certain feature group is less than the lower limit of the corresponding dynamic benchmark range, the characteristic parameter is recorded as an abnormal feature, the deviation direction is recorded as negative, and the part exceeding the lower limit value of the dynamic benchmark range is recorded as the amplitude.

[0066] S5, output fault type and confidence: establish a fault type and feature deviation mode library, based on the deviation direction and amplitude of the several abnormal features, output the most possible fault type and confidence.

[0067] In specific embodiments of the present application, the fault type and feature deviation mode library specifically includes the weight factor of each feature group and the deviation direction and amplitude interval of several characteristic parameters corresponding to each fault type of the door machine system.

[0068] It should be noted that the fault type and feature deviation mode library is specifically constructed by analyzing data of the fault log of the door machine system and combing expert knowledge, which is clear and reasonable in structure, can be completely realized relying on existing data and tools, and can significantly improve the efficiency and reliability of the door machine system operation, and will not be described here.

[0069] The fault type is, for example, motor bearing wear, rail foreign matter blockage, door lock switch failure, etc.

[0070] In specific embodiments of the present application, the output of the most likely fault type and confidence is specifically as follows: based on the deviation direction and amplitude of a plurality of feature parameters, the fault type and feature deviation mode library is traversed to obtain the deviation direction similarity and amplitude interval similarity of the plurality of feature parameters of the door machine system and each feature group of each fault type corresponding to the feature parameters, the similarity including values of 0 and 1, and the deviation direction similarity and amplitude similarity of the plurality of feature parameters of the door machine system and each fault type in each feature group are summarized.

[0071] The feature parameters with a deviation direction similarity of 0 or an amplitude interval similarity of 0 are removed, the number of similar feature parameters of the door machine system and each fault type in each feature group is counted, the number of similar feature parameters of the door machine system and each fault type in each feature group is divided by the number of total feature parameters to obtain the similar proportion of the door machine system and each fault type in each feature group, and the similar proportions are weighted and summed to obtain the confidence of the door machine system and each fault type, the fault type corresponding to the maximum confidence is screened as the most likely fault type of the door machine system, and the maximum confidence is output, so as to obtain the most likely fault type and confidence.

[0072] It should be noted that the similar proportions of the door machine system and each fault type in each feature group are weighted and summed in the above, and the weight factor is specifically the weight factor of each feature group corresponding to each fault type.

[0073] The above content is only an example and description of the concept of the present application, and those skilled in the art can make various modifications or supplements or use similar ways to replace the described specific embodiments, as long as they do not deviate from the concept of the present application or exceed the scope defined by the present application, which should belong to the protection scope of the present application.

Claims

1. An elevator door machine system anomaly detection and prediction method, characterized by, The application relates to a method for identifying elevator door machine fault types, and belongs to the field of elevator door machine fault identification. The method comprises the following steps: S1, multidimensional data acquisition: arranging vibration sensors, current sensors, position encoders and temperature sensors in an elevator door machine system, collecting vibration waveforms, electrode current curves, door position trajectories and temperature rise data in real time during door machine operation, and constructing original data of the door machine; S2, operation feature extraction: dividing the collected original data of the door machine according to a door machine operation cycle, and extracting a plurality of feature parameters corresponding to a plurality of feature groups for each operation cycle; S3, dynamic benchmark establishment: establishing a dynamic benchmark range curve associated with the number of door machine operations for each feature parameter based on factory parameters and historical normal operation data of the door machine system, wherein the dynamic benchmark range curve comprises allowed upper and lower limits of each feature parameter changing with the increase of the number of operations, and one operation cycle corresponds to one number of operations; S4, abnormal feature identification: screening a plurality of abnormal features based on a plurality of feature parameters corresponding to a plurality of feature groups extracted in real time, and recording the deviation direction and amplitude of the plurality of abnormal features; 2. The elevator machine system abnormality detection and prediction method according to claim 1, characterized by, S5, output of fault type and confidence: establishing a fault type and feature deviation mode library, and outputting the most possible fault type and confidence based on the deviation direction and amplitude of the plurality of abnormal features. The collected original data of the door machine is divided according to the door machine operation cycle, and the specific steps are as follows: S2001: time synchronization of vibration waveforms, electrode current curves, door position trajectories and temperature rise data in the collected original data of the door machine; S2002: sliding extraction of time domain features, frequency domain features and time-frequency features with a fixed window length, to obtain time sequence features after multi-source data fusion; The time domain features are as follows: vibration: peak factor, pulse factor, waveform factor; current: starting current peak value, ending current peak value, braking time, current harmonic distortion; position: starting moving speed, ending moving speed, positioning accuracy, trajectory smoothness; temperature rise: temperature rise rate, steady-state temperature difference, over-temperature threshold; The frequency domain features are as follows: vibration: main frequency energy ratio, frequency spectrum centroid; current: fundamental wave amplitude, harmonic component; The time-frequency features are as follows: transient impact features of vibration and current; S2003: taking the time sequence features after multi-source data fusion as the input of an LSTM model; S2004: predicting whether it is a cycle boundary for each time step, so as to divide the original data of the door machine according to the door machine operation cycle. The specific training method of the LSTM model is as follows: A1, collecting historical operation data of the door machine system, and performing time synchronization on the historical operation data; A2, segmenting the historical operation data according to the door machine operation cycle, independently normalizing each segment, and performing preliminary processing on the historical operation data, wherein the normalization is Z-Score normalization; 3. The elevator machine system anomaly detection and prediction method of claim 2, wherein, A3, using a labeling tool to mark the boundaries of the cycles of the historical operation data, generating an initial label set, and extracting training time sequence features after multi-source data fusion according to the method of S2002; A4, dividing the labeled training time sequence features of a plurality of cycles according to a set proportion to obtain a training set, a verification set and a test set, wherein the training set, the verification set and the test set contain complete cycles, and the final LSTM model is obtained through training and optimization. ​ ​ ​ 4. The elevator machine system abnormality detection and prediction method according to claim 3, characterized by, The final LSTM model is obtained by training and optimizing, and the specific steps are as follows: A501, divide the training set into multiple batches, each batch containing a fixed number of samples, input a batch of samples into the LSTM model, output the prediction results of the current batch, and calculate the value of the loss function according to the prediction results and the initial label set; A502, calculate the gradient of the loss function with respect to the LSTM model parameters, which represents the parameter update direction and size, and minimize the loss function, update the LSTM model parameters according to the gradient using the optimizer, and continue iteration until the loss function is minimized; A503, evaluate the performance of the trained LSTM model on the validation set, analyze the reasons for the loss on the validation set, adjust the LSTM model according to the reasons, retrain the model, and evaluate it on the validation set, repeat the above steps until the loss on the validation set meets the requirements; A504, select the model with the lowest loss on the validation set as the best LSTM model, and evaluate the performance of the best LSTM model on the test set, calculate the performance evaluation index, if the performance evaluation index does not meet the requirements, adjust the LSTM model, evaluate it on the validation set, select the best model and then verify it on the test set, until the performance on the test set meets the requirements, and get the final LSTM model.

5. The elevator machine system anomaly detection and prediction method of claim 1, wherein, The plurality of feature groups correspond to a plurality of feature parameters, specifically including: The vibration feature group includes vibration peak value, main frequency amplitude, and high frequency energy ratio; The current feature group includes the maximum starting current, the average steady-state current, and the shutdown current drop duration; The motion feature group includes the total door opening time, the total door closing time, and the speed fluctuation value in the middle uniform speed area; The temperature feature group includes temperature rise rate, steady-state temperature difference, and over-temperature threshold.

6. The elevator machine system anomaly detection and prediction method of claim 1, wherein, The specific establishment method of the dynamic reference range curve associated with the number of door machine operations for each feature parameter is as follows: Correlation analysis is performed on each feature parameter of the door machine and the number of operations, a scatter plot is drawn and a trend line is fitted, if it is a linear trend, a linear model is constructed, if it is a nonlinear trend, a nonlinear model is constructed, each feature parameter of the door machine and the number of operations are brought into the linear model or the nonlinear model, and the linear model coefficient or the nonlinear model coefficient is solved to obtain a fitting formula of each feature parameter of the door machine and the number of operations; According to the door machine system factory parameters, the life cycle stage of the door machine system is identified, the reference range adjustment parameters of the door machine system in each life cycle stage are obtained from the data warehouse, and the reference range adjustment parameters of the door machine system in the current life cycle are matched; According to the fitting formula of each feature parameter of the door machine and the number of operations, and combined with the reference range adjustment parameters of the door machine system in the current life cycle, a dynamic reference range curve associated with the number of door machine operations for each feature parameter is established.

7. The elevator machine system anomaly detection and prediction method of claim 1, wherein, The specific method for screening a plurality of abnormal features and recording the deviation direction and amplitude of a plurality of abnormal features is as follows: According to the dynamic reference range curve associated with the number of door machine operations for each feature parameter, and combined with the current number of operations of the door machine, the dynamic reference range of each feature parameter is circled. The several feature parameters corresponding to the several feature groups extracted in real time are compared with the dynamic benchmark range under the corresponding operation times. If a feature parameter of a feature group is greater than the upper limit of the corresponding dynamic benchmark range, the feature parameter is recorded as an abnormal feature, the deviation direction is recorded as a positive direction, and the part exceeding the upper limit value of the dynamic benchmark range is recorded as an amplitude. If a feature parameter of a feature group is less than the lower limit of the corresponding dynamic benchmark range, the feature parameter is recorded as an abnormal feature, the deviation direction is recorded as a negative direction, and the part exceeding the lower limit value of the dynamic benchmark range is recorded as an amplitude.

8. The elevator machine system anomaly detection and prediction method of claim 1, wherein, The fault type and feature deviation mode library specifically includes the weight factor of each feature group corresponding to each fault type of the door machine system and the deviation direction and amplitude interval of the several feature parameters.

9. The elevator machine system anomaly detection and prediction method of claim 8, wherein, The output of the most possible fault type and the confidence degree specifically comprises the following steps: Based on the deviation direction and amplitude of the several feature parameters, the fault type and feature deviation mode library is traversed to obtain the deviation direction similarity and amplitude interval similarity of the several feature parameters of the door machine system and the corresponding feature parameters of each feature group of each fault type. The similarity includes the numerical values of 0 and 1, and the deviation direction similarity and amplitude similarity of the several feature parameters of the door machine system and each fault type in each feature group are obtained. The feature parameters with the deviation direction similarity of 0 or the amplitude interval similarity of 0 are removed, the number of similar feature parameters of the door machine system and each fault type in each feature group is counted, the number of similar feature parameters of the door machine system and each fault type in each feature group is divided by the number of total feature parameters to obtain the similar proportion of the door machine system and each fault type in each feature group, and the similar proportions are weighted and summed to obtain the confidence degree of the door machine system and each fault type. The fault type corresponding to the maximum confidence degree is screened as the most possible fault type of the door machine system, and the maximum confidence degree is output, so that the most possible fault type and the confidence degree are obtained.

Images