Intelligent elevator monitoring method and platform

Abstract

The invention relates to the technical field of video monitoring, in particular to an intelligent elevator monitoring method and platform. The intelligent elevator monitoring method comprises the steps that a monitoring video of a target elevator and monitoring data monitored by all sensors are preprocessed through edge nodes of the target elevator; obtaining a video analysis result of the monitoring video and feature data of each piece of monitoring data, wherein the video analysis result comprises illegal behaviors and data generating the illegal behaviors; the feature data are fused through a prediction model, and the equipment health index of the target elevator is output; the abnormal state of the target elevator is determined according to the video clips and data of the illegal behaviors and the feature data of the monitoring data, and early warning is conducted; and an operation and maintenance work order is generated according to the abnormal state of the target elevator and the equipment health index. In this way, the monitoring efficiency, the monitoring reliability, the real-time performance and the safety of elevator monitoring are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of elevator intelligent monitoring, and particularly relates to an elevator intelligent monitoring method and platform. Background Art

[0002] An elevator is a mechanical device used for vertical transportation of people or goods inside a building. It is driven by an electric motor and uses a pulley system, steel cables, and other mechanical components to lift and lower the car, thereby realizing the transportation of people and goods between different floors. In order to ensure the safe operation of the elevator, it is necessary to monitor the elevator.

[0003] In some scenarios, video monitoring is often used to monitor elevators. The video monitoring method mainly records the images inside the elevator through a camera, and then sends the monitoring video to the background, where operation and maintenance personnel analyze the monitoring video to determine whether there are potential problems with the elevator. However, this passive monitoring solution has obvious limitations. Since the video monitoring method for elevators is single, the monitoring reliability is low. In addition, the monitored elevator video needs to be analyzed by operation and maintenance personnel later, and it is impossible to identify illegal behaviors and abnormal states in the elevator in real time, which has a certain lag. This lag not only reduces the monitoring efficiency but also increases potential safety hazards. In some other scenarios, some monitoring data of the elevator often rely on the cloud for analysis and processing. Over-reliance on the cloud for analysis and processing of monitoring data results in high bandwidth occupancy and large latency, leading to low real-time performance, safety, and reliability of elevator monitoring.

[0004] Therefore, the method of using video monitoring to monitor elevators not only has low monitoring efficiency but also low monitoring reliability, real-time performance, and safety. Summary of the Invention

[0005] In order to solve the technical problems of low monitoring efficiency and low monitoring reliability, real-time performance, and safety, the purpose of the present invention is to provide an elevator intelligent monitoring method, and the specific technical solution adopted is as follows: In a first aspect, an embodiment of the present invention provides an elevator intelligent monitoring method, including: preprocessing the monitoring video of the target elevator and the monitoring data monitored by each sensor through the edge node of the target elevator to obtain the video analysis result of the monitoring video and the characteristic data of each monitoring data, where the video analysis result includes illegal behaviors and the data generating the illegal behaviors; performing fusion processing on each characteristic data through a prediction model to output the equipment health index of the target elevator; determining the abnormal state of the target elevator based on the video segments, data of the illegal behaviors, and the characteristic data of each monitoring data and giving an early warning; generating an operation and maintenance work order according to the abnormal state and the equipment health index of the target elevator.

[0006] Optionally, after the feature data is fused through the prediction model and the equipment health index of the target elevator is output, the method further includes: inputting the monitoring data monitored by each sensor into the prediction model, and calculating the over-limit value of each sensor through the prediction model. The over-limit value is determined by the prediction model based on the initial over-limit value of the sensor, the mean value, and the standard deviation of the monitoring data.

[0007] Optionally, the method for determining the over-limit value of the sensor includes: calculating the product between the initial over-limit value and the standard deviation; determining the sum value between the mean value and the product as the over-limit value of the sensor.

[0008] Optionally, preprocessing the monitoring video of the target elevator and the monitoring data monitored by each sensor through the edge node of the target elevator to obtain the video analysis result of the monitoring video and the feature data of each monitoring data includes: analyzing the first data of the personnel, the second data of the items, and the elevator door area in each video frame of the monitoring video through the lightweight YOLOv5s model in the edge node to obtain the violation behaviors in the monitoring video and the data generating the violation behaviors; extracting the feature data of the monitoring data monitored by each sensor through the edge node, where the monitoring data includes vibration data, current data, and temperature data, and the feature data includes the vibration spectrum of the vibration data, the current trend of the current data, and the temperature gradient of the temperature data.

[0009] Optionally, analyzing the first data of the personnel, the second data of the items, and the elevator door area in each video frame of the monitoring video through the lightweight YOLOv5s model in the edge node to obtain the violation behaviors in the monitoring video and the data generating the violation behaviors includes: determining the first number of personnel and the second number of items in the video frame through the lightweight YOLOv5s model, and determining the load weight in the target elevator according to the preset personnel weight, item weight, the first number of personnel, and the second number of items through the lightweight YOLOv5s model. When the load weight exceeds the overload threshold, it is determined that the target elevator has an overload behavior, and the data generating the overload behavior includes the overweight exceeding the overload threshold; detecting the door-prying action of the personnel in the elevator door area in the video frame through the lightweight YOLOv5s model, and estimating the door-prying force of the personnel based on the action amplitude of the door-prying action of the personnel in the video frame. When the door-prying force is greater than the door-prying force threshold, it is determined that the target elevator has a door-prying behavior, and the data generating the door-prying behavior includes the door-prying force.

[0010] Optionally, the lightweight YOLOv5s model in the edge node is used to analyze the first data of the personnel, the second data of the items, and the elevator door area in each video frame of the surveillance video, and the illegal behaviors in the surveillance video and the data generating the illegal behaviors are obtained, including: detecting whether there are personnel or items in the closing area in the elevator door area in the video frame through the lightweight YOLOv5s model. If there are personnel or items in the elevator door area in the closing area, it is determined that the target elevator generates a behavior of blocking the door closing, and the data generating the behavior of blocking the door closing includes the type of the blocking object; detecting the actions of the personnel in the video frame through the lightweight YOLOv5s model. If the action of the personnel is a jumping action, it is determined that the target elevator generates a jumping behavior, and the data generating the jumping behavior includes the detection result of the jumping action.

[0011] Optionally, before the prediction model performs fusion processing on the feature data and outputs the equipment health index of the target elevator, the method further includes: collecting fault samples from the historical operation and maintenance records of the elevator, where the fault samples include historical fault types and historical feature data of sensors; performing data cleaning on the fault samples to obtain cleaned samples; inputting the cleaned samples into a pre-trained LSTM model for iterative training to obtain a prediction model, where the training mechanism of the pre-trained LSTM model includes an early stopping mechanism. When the loss of the pre-trained LSTM model remains unchanged for N consecutive times, the training of the LSTM model is stopped in advance, and N is a natural number greater than or equal to 6.

[0012] Optionally, determining the abnormal state of the target elevator and giving an early warning according to the video clip, data of the illegal behavior, and the feature data of each monitoring data includes: when the data of the illegal behavior in the video clip of the illegal behavior is greater than the alarm threshold, determining that the abnormal state of the target elevator is that the target elevator violates the regulations and triggering a voice alarm; when the mean value of the vibration data in the monitoring data exceeds the corresponding first fault threshold and the mean value of the current data in the monitoring data exceeds the second fault threshold, determining that the abnormal state of the target elevator is a bearing fault and giving a voice alarm.

[0013] In a second aspect, an embodiment of the present invention provides an elevator intelligent monitoring platform, including: an edge node and a cloud. The edge node is communicatively connected to the cloud. The edge node is configured to preprocess the monitoring video of the target elevator and the monitoring data monitored by each sensor to obtain the video analysis result of the monitoring video and the feature data of each monitoring data. The video analysis result includes the violation behavior and the data generating the violation behavior, and send the feature data, the video segment of the violation behavior, and the data generating the violation behavior to the cloud. The cloud is deployed with a prediction model. The cloud is configured to perform fusion processing on each feature data through the prediction model, output the equipment health index of the target elevator, determine the abnormal state of the target elevator based on the video segment, data of the violation behavior, and the feature data of each monitoring data and give an early warning, and generate an operation and maintenance work order according to the abnormal state and the equipment health index of the target elevator.

[0014] In a third aspect, an embodiment of the present invention provides an elevator intelligent monitoring platform, including: a processor and a memory. Wherein, the memory is used to store a computer program that can run on the processor. The processor is configured to execute the program stored on the memory to implement the steps of the elevator intelligent monitoring method mentioned in the first aspect.

[0015] The present invention has the following beneficial effects: In the embodiment of the present invention, first, the edge node of the target elevator preprocesses the monitoring video of the target elevator and the monitoring data monitored by each sensor to obtain the video analysis result of the monitoring video and the feature data of each monitoring data. The video analysis result includes the violation behavior and the data generating the violation behavior. In this way, the monitoring video and the monitoring data monitored by each sensor can be processed in real time at the edge node, thereby reducing the bandwidth occupation and latency. This can not only improve the real-time performance, security and reliability of the monitoring, but also reduce the dependence on the cloud. Secondly, the prediction model performs fusion processing on each feature data and outputs the equipment health index of the target elevator; and determines the abnormal state of the target elevator based on the video segment, data of the violation behavior, and the feature data of each monitoring data and gives an early warning. In this way, the embodiment of the present invention performs fusion processing on the multi-modal data of the elevator. Through the linkage between various types of data, it can not only determine the violation behavior occurring in the elevator in real time with low hysteresis, but also determine the equipment health degree and abnormal state of the elevator. The data types are diverse and can process various monitoring data of the elevator in real time, improving the monitoring efficiency, monitoring reliability, real-time performance and security of the elevator monitoring. Description of the Drawings

[0016] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0017] Figure 1 It is a schematic structural diagram of an elevator intelligent monitoring platform provided by an embodiment of the present invention.

[0018] Figure 2 It is a schematic flowchart of an elevator intelligent monitoring method provided by an embodiment of the present invention.

[0019] Figure 3 It is a schematic flowchart of another elevator intelligent monitoring method provided by an embodiment of the present invention.

[0020] Figure 4 It is a schematic structural diagram of another elevator intelligent monitoring platform provided by an embodiment of the present invention. Detailed implementation manners

[0021] In order to further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following, in combination with the accompanying drawings and preferred embodiments, details the specific implementation manners, structures, features and effects of an elevator intelligent monitoring method proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0022] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0023] In some scenarios, the elevator is monitored by means of video monitoring. The video monitoring method mainly records the images inside the elevator through a camera, and then sends the monitoring video to the background, where the operation and maintenance personnel analyze the monitoring video to determine whether there are potential problems with the elevator. However, this passive monitoring solution has obvious limitations. Since the video monitoring method for monitoring the elevator is single, the monitoring reliability is low. In addition, the monitored elevator monitoring video needs to be analyzed by the operation and maintenance personnel later, and it is impossible to identify illegal behaviors and abnormal states in the elevator in real time, which has a certain lag. This lag not only reduces the monitoring efficiency but also increases the safety hazards. Therefore, the method of monitoring the elevator by means of video monitoring not only has low monitoring efficiency but also low monitoring reliability and safety.

[0024] In some other scenarios, vibration sensors or temperature sensors are deployed to monitor the operating status of the elevator. These sensors can real-time monitor vibration data, temperature data, etc. during the elevator operation, providing certain data support for the elevator. However, generally, the data monitored by these sensors are isolated and often lack the ability of multi-dimensional data fusion. For example, a certain sensor can only monitor the temperature of the motor and cannot associate vibration data to judge the wear degree of the bearing. This information island phenomenon limits the accuracy and reliability of elevator monitoring.

[0025] In some other scenarios, in the traditional elevator maintenance mode, maintenance personnel need to regularly check the elevator status and rely on personal experience to judge the health of the elevator. However, this manual inspection method is easily affected by human factors, such as missed inspections and misjudgments often occur, resulting in low maintenance efficiency and accuracy of the elevator.

[0026] In some other scenarios, some monitoring data of the elevator often rely on the cloud for analysis and processing. Over-reliance on the cloud for analyzing and processing monitoring data results in high bandwidth occupancy and large latency, leading to low real-time performance, security and reliability of elevator monitoring.

[0027] Based on the above analysis, there are obvious limitations in the existing elevator monitoring technology in terms of data integration, monitoring integrity, reliability, security, real-time performance, and cloud dependence, which affect the stability and reliability of the elevator.

[0028] Next, the specific solutions of an elevator intelligent monitoring method and platform provided by the present invention will be specifically described with reference to the accompanying drawings.

[0029] Please refer to Figures 1 to 4 , Figure 1 which is a schematic structural diagram of an elevator intelligent monitoring platform provided by an embodiment of the present invention. Figure 2 which is a schematic flow diagram of an elevator intelligent monitoring method provided by an embodiment of the present invention. Figure 3 which is a schematic flow diagram of another elevator intelligent monitoring method provided by an embodiment of the present invention. Figure 4 which is a schematic structural diagram of another elevator intelligent monitoring platform provided by an embodiment of the present invention.

[0030] As Figure 1As shown in the figure, the elevator intelligent monitoring platform 100 includes: an edge node 101 and a cloud 102, and the edge node 101 is communicatively connected to the cloud 102; the edge node 101 is used to preprocess the monitoring video of the target elevator and the monitoring data monitored by each sensor to obtain the video analysis result of the monitoring video and the characteristic data of each monitoring data. The video analysis result includes violation behaviors and the data generating the violation behaviors, and sends the characteristic data, the video segments of the violation behaviors, and the data generating the violation behaviors to the cloud; a prediction model is deployed in the cloud 102, and the cloud 102 is used to perform fusion processing on each characteristic data through the prediction model, output the equipment health index of the target elevator, determine the abnormal state of the target elevator based on the video segments, data of the violation behaviors, and the characteristic data of each monitoring data and give an early warning, and generate an operation and maintenance work order according to the abnormal state and the equipment health index of the target elevator.

[0031] Specifically, in the embodiment of the present invention, an embedded device is deployed in the elevator machine room as an edge node, such as NVIDIA Jetson Orin Nano. The video stream analysis of the monitoring video and the preprocessing of the monitoring data monitored by the sensor are completed locally. The cloud can be the Alibaba Cloud IoT platform, and a prediction model is deployed in the cloud. Through this prediction model, the equipment health index of the elevator can be generated, etc. The prediction model can be a Long Short-Term Memory networks (LSTM).

[0032] Further, an infrared camera is installed on the top of the target elevator in the embodiment of the present invention for monitoring the target elevator and generating a monitoring video. A three-axis vibration sensor is installed on the top or bottom of the target elevator, a non-contact current sensor is installed at the outlet end of the frequency converter in the control cabinet of the target elevator, and a temperature sensor is installed on the inner side wall of the target elevator, etc. The three-axis vibration sensor is used to monitor the vibration data of the target elevator, the non-contact current sensor is used to monitor the current data of the target elevator, and the temperature sensor is used to monitor the temperature data of the target elevator.

[0033] Further, a lightweight YOLOv5s model is deployed in the edge node. The lightweight YOLOv5s model is used to analyze the monitoring video. As an optional embodiment of the present invention, the lightweight YOLOv5s model in the edge node analyzes the first data of the personnel, the second data of the items, and the elevator door area in each video frame of the monitoring video to obtain the violation behaviors in the monitoring video and the data generating the violation behaviors; the edge node extracts the characteristic data of the monitoring data monitored by each sensor. The monitoring data includes vibration data, current data, and temperature data, and the characteristic data includes the vibration spectrum of the vibration data, the current trend of the current data, and the temperature gradient of the temperature data.

[0034] Specifically, the first data includes, but is not limited to, the number of people, the weight of people, and the actions of people, etc. The second data of the item includes, but is not limited to, weight, quantity, type, and volume, etc. The elevator door area refers to the location area where the elevator door is located. The violation behaviors include, but are not limited to, overloading, prying the door, blocking the door from closing, jumping, and jamming the door, etc. The data generating the violation behaviors includes, but is not limited to, the overloading rate, the overweight exceeding the overloading threshold when overloading, the prying force of the door, the type of blocking the elevator door, the jumping action of people, etc.

[0035] Furthermore, in the embodiment of the present invention, YOLOv5s is first selected as the lightweight model, and then it is deployed on the edge node. By model quantization and pruning, the size of the YOLOv5s model is compressed and the inference speed is improved. Specifically, the FP32 precision of the YOLOv5s model is compressed to INT8 to achieve model quantization, and redundant neurons are removed to achieve pruning. In this way, the size of the YOLOv5s model is compressed from 27MB to 7MB, and the inference speed is increased by 2.3 times. It not only saves the storage resources in the edge node, optimizes the resource utilization rate, but also improves the processing speed of the model and the real-time performance of elevator monitoring.

[0036] Furthermore, before applying the YOLOv5s model, the embodiment of the present invention iteratively trains the YOLOv5s model to obtain a trained YOLOv5s model. Specifically: First, determine the data source. The embodiment of the present invention collects 1000 hours of video clips from the elevator monitoring videos of the power plant and labels 5 types of violation behaviors, including overloading, prying the door, blocking the door from closing, jumping, and jamming the door. Then, data augmentation is performed on the video clips, specifically, enhancement operations such as rotation, scaling, and brightness adjustment are performed on the video clips to improve the generalization ability of the model. In the embodiment of the present invention, the LabelImg tool is used as the annotation tool to annotate the targets (such as the number of overloaded people, the prying action) in the video frames of the above video clips. Then, model training is carried out. Among them, the training environment: 4 NVIDIA Tesla V100 GPUs are used for training for 100 epochs. Loss function: CIoU Loss (Complete Intersection over Union) is adopted to optimize the positioning accuracy of the targets in the video frames. The prepared data is sent into the pre-trained lightweight YOLOv5s model for iterative training until the loss function converges to obtain a trained lightweight YOLOv5s model. In this way, the trained lightweight YOLOv5s model is used to analyze the first data of people, the second data of items, and the elevator door area in each video frame of the monitoring video to obtain the violation behaviors in the monitoring video and the data generating the violation behaviors.

[0037] Further, as an alternative embodiment of the present invention, the lightweight YOLOv5s model in the edge node analyzes the first data of the personnel, the second data of the items, and the elevator door area in each video frame of the surveillance video, and the obtained illegal behaviors and the data generating the illegal behaviors in the surveillance video include: determining the first number of personnel and the second number of items in the video frame through the lightweight YOLOv5s model, determining the load weight in the target elevator according to the preset weight of personnel, the weight of items, the first number of personnel, and the second number of items through the lightweight YOLOv5s model, and determining that the target elevator generates an overloading behavior when the load weight exceeds the overload threshold, and the data generating the overloading behavior includes the overweight exceeding the overload threshold; detecting the door-prying action of the personnel in the elevator door area in the video frame through the lightweight YOLOv5s model, and estimating the door-prying force of the personnel based on the amplitude of the door-prying action of the personnel in the video frame, and determining that the target elevator generates a door-prying behavior when the door-prying force is greater than the door-prying force threshold, and the data generating the door-prying behavior includes the door-prying force.

[0038] Specifically, the preset weight of personnel, the weight of items, and the overload threshold in the embodiments of the present invention can be custom-set according to actual experience. For example, the preset weight of personnel is 70 kg, the weight of items is 10 kg, and the overload threshold is 1000 kg. Among them, the data generating the overloading behavior further includes the overload probability, and the overload probability refers to that the total weight in the current target elevator exceeds 90% of the overload threshold. When the overload probability exceeds 90% of the overload threshold, it is determined that the elevator generates an overloading behavior.

[0039] Exemplarily, the present invention provides a specific implementation manner for determining that the target elevator generates an overloading behavior through the lightweight YOLOv5s model: Input: Personnel in the video frame and the target detection box.

[0040] def detect_overload(frame): # Use YOLOv5s to detect personnel and items results = model(frame) # Calculate the load probability load_prob = len(results['person']) * 70 # Estimate 70 kg per person load_prob += len(results['object']) * 10 # Estimate 10 kg per item # Determine whether it is overloaded if load_prob > 1000: # Overload threshold 1000 kg return True else: return False Output: Overload probability (an alarm is triggered if it is greater than 90% of the overload threshold) or the overweight when the total weight in the target elevator exceeds the overload threshold.

[0041] Furthermore, for the door prying action, the door prying force is estimated based on the amplitude of the door prying action of the person in the video frame, such as the arm stretching situation, the displacement distance and angle between the hand and the elevator door frame, and the change in the hand speed of the person in the surveillance video, etc. to estimate the door prying force. In addition, a safety force sensor can be directly installed at the elevator door of the target elevator to directly measure the door prying force. The door prying force threshold can be determined according to actual experience, and in the embodiments of the present invention, it is set to 50 Newtons.

[0042] Exemplarily, the present invention provides a specific implementation method for determining the door prying behavior of the target elevator through the lightweight YOLOv5s model: Input: The elevator door area in the video frame.

[0043] def detect_door_force(frame): # Use YOLOv5s to detect the door prying action results = model(frame) # Calculate the door prying force force = results['door_force'] # Estimated based on the amplitude of the door prying action # Determine whether it exceeds the door prying force threshold if force > 50N: # The door prying force threshold is 50 Newtons return True else: return False Output: Door prying force (an alarm is triggered if it is greater than 50N).

[0044] Further, as an optional embodiment of the present invention, the lightweight YOLOv5s model in the edge node analyzes the first data of the personnel, the second data of the items, and the elevator door area in each video frame of the surveillance video, and the obtained illegal behaviors in the surveillance video and the data generating the illegal behaviors include: detecting whether there are personnel or items in the closing area in the elevator door area in the video frame through the lightweight YOLOv5s model. If there are personnel or items in the closing area in the elevator door area, it is determined that the target elevator generates a behavior of blocking the door closing, and the data generating the behavior of blocking the door closing includes the type of the blocking object; detecting the actions of the personnel in the video frame through the lightweight YOLOv5s model. If the action of the personnel is a jumping action, it is determined that the target elevator generates a jumping behavior, and the data generating the jumping behavior includes the detection result of the jumping action.

[0045] Specifically, when the elevator door is closing in the embodiment of the present invention, there may be personnel or items in the elevator door area. At this time, it is necessary to identify whether there are objects in the elevator door area to avoid directly closing the elevator door and causing safety problems.

[0046] Exemplarily, the present invention provides a specific implementation manner for determining that the target elevator generates a behavior of blocking the door closing through the lightweight YOLOv5s model: Input: The elevator door area in the video frame.

[0047] def detect_blocking(frame): # Use YOLOv5s to detect blocking objects results = model(frame) # Determine whether there are objects in the door gap if results['blocking_object']: return True else: return False Output: The detection result of the blocking object (trigger an alarm if it exists).

[0048] Exemplarily, the present invention provides a specific implementation manner for determining that the target elevator generates a jumping behavior through the lightweight YOLOv5s model: Input: The actions of the personnel in the video frame.

[0049] def detect_jumping(frame): # Use YOLOv5s to detect jumping actions results = model(frame) # Determine if there is a person jumping if results['jumping']: return True else: return False Output: Detection result of jumping action (trigger an alarm if it exists).

[0050] Furthermore, the edge node in the embodiment of the present invention extracts the characteristic data of the monitoring data monitored by each sensor. Specifically, first, the monitoring data monitored by each sensor is filtered, noise-reduced, etc., and then the vibration spectrum of the vibration data, the current trend of the current data, the temperature gradient of the temperature data, and other characteristic data are extracted from the processed monitoring data.

[0051] Furthermore, in order to avoid excessive bandwidth occupation and reduce latency, after the edge node preprocesses the monitoring video and monitoring data in the embodiment of the present invention, only the characteristic data, the video segment of the violation behavior, and the data generating the violation behavior are transmitted to the cloud. In this way, the amount of data transmitted to the cloud is significantly reduced, avoiding excessive bandwidth occupation, thereby reducing the latency and improving the real-time performance of elevator monitoring. In addition, in order to avoid the leakage of personal information of users in the video, the embodiment of the present invention performs video desensitization processing when transmitting the video segment to the cloud. Specifically, the face information of the person can be automatically hidden through a face blurring algorithm to avoid the leakage of personal information of the person and improve the security of personal information.

[0052] Furthermore, before the prediction model in the cloud performs fusion processing on each characteristic data and outputs the equipment health index of the target elevator, the prediction model can be trained first. As an optional embodiment of the present invention, fault samples are collected from the historical operation and maintenance records of the elevator. The fault samples include historical fault types and historical characteristic data of the sensors; the fault samples are subjected to data cleaning to obtain cleaned samples; the cleaned samples are input into a pre-trained LSTM model for iterative training to obtain a prediction model. The training mechanism of the pre-trained LSTM model includes an early stopping mechanism. When the loss of the pre-trained LSTM model remains unchanged for N consecutive times, the training of the LSTM model is stopped in advance, and N is a natural number greater than or equal to 6.

[0053] Specifically, in the embodiment of the present invention, 1000 groups of fault samples are collected from the historical operation and maintenance records of the power plant elevator. The fault samples include historical fault types and historical characteristic data of the sensors. Exemplarily, the historical fault types include but are not limited to the following types: Fault type Sample quantity Description Bearing wear 300 Vibration abnormality caused by wear of the traction machine bearing Motor overload 250 The motor load is too high, causing abnormal current Guide rail offset 200 Abnormal vibration caused by guide rail installation deviation Electrical overheat 150 The temperature inside the control cabinet is too high Other faults 100 Including door system fault, sensor failure, etc. Exemplarily, the historical feature data provided by the embodiment of the present invention includes but is not limited to the following types: Feature Name Description Vibration frequency distribution after FFT transformation of vibration spectrum Current trend Current mean and fluctuation over the past 72 hours The temperature gradient controls the rate of temperature change inside the cabinet.

[0054] Furthermore, the fault labels of the pre-trained LSTM model are fault types, such as bearing wear and motor overload.

[0055] Furthermore, after obtaining the fault samples, data cleaning is performed to remove noise and outliers in the data. Specifically, first remove missing values: use linear interpolation to fill in the missing vibration data, current data, and temperature data in the historical feature data. Then remove outliers: remove abnormal data points based on the 3σ principle (three times the standard deviation). Finally, standardize the features: normalize the vibration spectrum, current trend, and temperature gradient so that their value range is between [0, 1]. After the data cleaning is completed, the 1000 groups of cleaned samples are divided into training sets (800 groups) and test sets (200 groups) in a ratio of 8:2.

[0056] Further, the model structure of the pre-trained LSTM model in the embodiment of the present invention includes an input layer, an LSTM layer, a Dropout layer, a fully connected layer, and an output layer. Among them, the input layer is used to accept vibration spectrum, current trend, and temperature gradient as input (time step = 72). The LSTM layer contains 64 LSTM units for capturing long-term dependencies of time series data. The dropout rate of the Dropout layer is 0.2 to prevent overfitting. The fully connected layer contains 32 neurons for connecting the input layer and the output layer. The output layer contains 1 neuron, which outputs the equipment health index EHI value (range 0-100). The larger the value of the equipment health index EHI value, the higher the health of the target elevator. In addition, the loss function of the pre-trained LSTM model adopts the mean square error loss function for regression tasks. The optimizer adopts Adam, and the learning rate = 0.001. The evaluation index is the mean absolute error. In addition, the number of training rounds of the pre-trained LSTM model is 100 epochs. The batch size is 32.

[0057] Furthermore, an early stopping mechanism is also set for the pre-trained LSTM model. When the loss of the pre-trained LSTM model remains unchanged for N consecutive times, the training of the LSTM model is stopped in advance, where N can be set according to the actual situation and is set to 10 in the embodiments of the present invention. In this way, by setting an early stopping mechanism for the pre-trained LSTM model, when the training accuracy of the pre-trained LSTM model reaches the predetermined accuracy in advance, the training can be ended in advance, improving the training efficiency of the pre-trained LSTM model.

[0058] Furthermore, after the pre-trained LSTM model is trained, the performance of the prediction model is evaluated on the test set. If the performance of the prediction model meets the requirements, the prediction model can be applied to fuse the feature data and output the equipment health index of the target elevator.

[0059] Furthermore, the prediction model in the embodiments of the present invention can also dynamically predict the over-limit values of each sensor, where the over-limit value refers to the maximum value within the normal range allowed by each sensor. As an optional embodiment of the present invention, the monitoring data monitored by each sensor is input into the prediction model, and the over-limit value of each sensor is calculated by the prediction model. The over-limit value is determined by the prediction model based on the initial over-limit value of the sensor, the mean value and the standard deviation of the monitoring data.

[0060] Specifically, in the embodiments of the present invention, the sensor collects vibration data, current data, and temperature data every 24 hours as the input of the prediction model, and updates the over-limit value of each sensor every predetermined number of days. The predetermined number of days can be determined according to the actual situation and is set to 7 days in the embodiments of the present invention. The prediction model updates the over-limit value of the sensor according to the monitoring data of each sensor within 7 days. The mean value and the standard deviation of the monitoring data of each sensor refer to the mean value and the standard deviation of the monitoring data of each sensor from the moment when the target elevator is newly installed to the latest acquisition moment. For example, the prediction model determines a new over-limit based on the mean value and the standard deviation of the monitoring data of a certain sensor within the time period from the moment when the target elevator is newly installed to the latest moment. The initial over-limit value of the sensor can be custom-set according to the actual scenario, and is not limited in the embodiments of the present invention.

[0061] Furthermore, as this optional embodiment of the present invention, the determination method of the over-limit value of the sensor includes: calculating the product between the initial over-limit value and the standard deviation; determining the sum value between the mean value and the product as the over-limit value of the sensor.

[0062] Exemplarily, in the embodiments of the present invention, taking the sensor as a vibration sensor as an example, the process of updating the over-limit value is described: First, in the embodiments of the present invention, based on the historical data when the target elevator is newly installed, the initial vibration over-limit value of the vibration sensor is statistically obtained and initially set to 2.0 mm / s².

[0063] During the new installation stage of the target elevator: The mean value of the vibration data: 1.5 mm / s², the standard deviation: 0.2 mm / s².

[0064] The initial vibration overrun value: 2.0 mm / s².

[0065] One year after the target elevator operates: The mean value of the vibration data rises to 1.9 mm / s², and the standard deviation is 0.3 mm / s².

[0066] The overrun value of the vibration sensor is updated to: 1.9 + 2×0.3 = 2.5 mm / s².

[0067] In this way, the embodiments of this aspect can use the prediction model to dynamically adjust the overrun values of each sensor according to the operating time of the target elevator, so that the overrun values of each sensor are more adapted to the elevator in the current state, reducing the false judgment rate of elevator faults and improving the accuracy of elevator fault identification and the safety and reliability of elevator monitoring.

[0068] Further, in the cloud, determining the abnormal state of the target elevator and giving an early warning based on the video clips, data of the violation behaviors, and the characteristic data of each monitoring data includes: when the data of the violation behavior in the video clip of the violation behavior is greater than the alarm threshold, determining that the abnormal state of the target elevator is that the target elevator violates the regulations and triggering a voice alarm; when the mean value of the vibration data in the monitoring data exceeds the corresponding first fault threshold and the mean value of the current data in the monitoring data exceeds the second fault threshold, determining that the abnormal state of the target elevator is a bearing fault and giving a voice alarm.

[0069] Specifically, the alarm threshold can be determined according to the actual situation, and the embodiments of the present invention do not limit this here. For example, when the violation behavior is an overloading behavior and the data of the violation behavior is the overloading probability, the alarm threshold can be set to 90%. When the violation behavior is a door prying behavior and the data of the violation behavior is the door prying force, the alarm threshold can be set to 50 Newtons. The first fault threshold and the second fault threshold can also be determined according to the actual situation, and the embodiments of the present invention do not limit this here. For example, in the embodiments of the present invention, the first fault threshold is taken as 2.5 mm / s² and the second fault threshold is taken as 15 A. By the interconnection of the monitoring data of different types of sensors, the specific fault type in the target elevator is determined, thereby improving the accuracy and reliability of the fault identification of the target elevator.

[0070] Exemplarily, the embodiments of the present invention provide the following specific implementation manners for fusing and processing the video clips, data of the violation behaviors, and the characteristic data of each monitoring data to determine the abnormal state of the target elevator and give an early warning: def fusion(video_result, sensor_data): Video analysis results: overload probability, door-prying force marking overload_prob = video_result['overload'] Obtain the overload probability from the video analysis results door_force = video_result['door_force'] Obtain the door-prying force from the video analysis results Sensor data: vibration data (vibration RMS value), average current vibration_rms = sensor_data['vibration'] Obtain the vibration RMS value from the sensor data current_avg = sensor_data['current'] Obtain the average current Comprehensive judgment rule if overload_prob > 0.9 or door_force > 50N: If the overload probability is greater than 90% or the door-prying force exceeds 50 Newtons trigger_voice_alert() Trigger a voice alarm if vibration_rms > 2.5mm / s² and current_avg > 15A: If the vibration RMS value is greater than 2.5 mm / s² and the average current exceeds 15 Amperes predict_bearing_failure() Predict bearing failure.

[0071] It should be noted that according to the actual operating conditions of the target elevator, other types of monitoring data can also be used for mutual linkage to determine the fault conditions of the target elevator.

[0072] Further, after the cloud obtains the abnormal status and the equipment health index of the target elevator, when the equipment health index < health threshold, it is determined that the target elevator is in a high-risk state, and a maintenance work order needs to be generated immediately, thereby improving the maintenance efficiency and safety of the target elevator. The health threshold can be determined according to the actual situation, and the value in the embodiment of the present invention is 60. When the target elevator has an abnormal status, a maintenance work order also needs to be generated immediately. For example, when the target elevator has a bearing failure or a motor overload failure, spare parts are accurately recommended to improve the maintenance efficiency and safety of the target elevator. Among them, the embodiment of the present invention can be docked with the power plant ERP system to realize the linkage of automatic work order dispatching and spare parts inventory.

[0073] Further, the embodiment of the present invention can also develop a mobile APP. The mobile APP can receive real-time alarm information (such as overloading, door prying) from the cloud or edge nodes, and provide the equipment health status in combination with the equipment health index EHI value of the LSTM model. In addition, multi-modal data such as vibration spectrum, current trend, and temperature gradient can also be displayed on the mobile APP to help maintenance personnel quickly locate faults.

[0074] Further, in order to ensure the data security when the edge node sends the video analysis result and feature data to the cloud, the embodiment of the present invention uses the AES-256 algorithm to encrypt the video analysis result and feature data before transmission, improving the data security. In addition, a secure transmission protocol (such as HTTPS), a firewall and an intrusion detection system can be deployed on the edge node, and regular updates and patch management can be carried out to further improve the data security.

[0075] Based on the same inventive concept of the elevator intelligent monitoring platform provided in the above embodiments, the present invention also discloses an elevator intelligent monitoring method, as Figure 2 shown, Figure 2 is a schematic flowchart of an elevator intelligent monitoring method provided by an embodiment of the present invention. The disclosed elevator intelligent monitoring method includes: Step S201, preprocess the monitoring video of the target elevator and the monitoring data monitored by each sensor through the edge node of the target elevator to obtain the video analysis result of the monitoring video and the feature data of each monitoring data. The video analysis result includes violation behaviors and the data generating the violation behaviors.

[0076] Among them, as an optional embodiment of the present invention, the edge node of the target elevator preprocesses the monitoring video of the target elevator and the monitoring data monitored by each sensor to obtain the video analysis result of the monitoring video and the characteristic data of each monitoring data, including: analyzing the first data of the personnel, the second data of the items, and the elevator door area in each video frame of the monitoring video through the lightweight YOLOv5s model in the edge node to obtain the illegal behaviors in the monitoring video and the data generating the illegal behaviors; extracting the characteristic data of the monitoring data monitored by each sensor through the edge node, the monitoring data includes vibration data, current data, and temperature data, and the characteristic data includes the vibration spectrum of the vibration data, the current trend of the current data, and the temperature gradient of the temperature data.

[0077] Among them, as an optional embodiment of the present invention, analyzing the first data of the personnel, the second data of the items, and the elevator door area in each video frame of the monitoring video through the lightweight YOLOv5s model in the edge node to obtain the illegal behaviors in the monitoring video and the data generating the illegal behaviors includes: determining the first number of personnel and the second number of items in the video frame through the lightweight YOLOv5s model, and determining the load weight in the target elevator according to the preset personnel weight, item weight, the first number of personnel, and the second number of items through the lightweight YOLOv5s model. When the load weight exceeds the overload threshold, it is determined that the target elevator has an overload behavior, and the data generating the overload behavior includes the overweight exceeding the overload threshold; detecting the door-prying action of the personnel in the elevator door area in the video frame through the lightweight YOLOv5s model, and estimating the door-prying force of the personnel based on the action amplitude of the door-prying action of the personnel in the video frame. When the door-prying force is greater than the door-prying force threshold, it is determined that the target elevator has a door-prying behavior, and the data generating the door-prying behavior includes the door-prying force.

[0078] Among them, as an optional embodiment of the present invention, analyzing the first data of the personnel, the second data of the items, and the elevator door area in each video frame of the monitoring video through the lightweight YOLOv5s model in the edge node to obtain the illegal behaviors in the monitoring video and the data generating the illegal behaviors includes: Detecting whether there are personnel or items in the elevator door area in the closing area in the video frame through the lightweight YOLOv5s model. If there are personnel or items in the elevator door area in the closing area, it is determined that the target elevator has a behavior of blocking the door closing, and the data generating the behavior of blocking the door closing includes the type of the blocking object; detecting the action of the personnel in the video frame through the lightweight YOLOv5s model. If the action of the personnel is a jumping action, it is determined that the target elevator has a jumping behavior, and the data generating the jumping behavior includes the jumping action detection result.

[0079] Step S203: Perform fusion processing on each piece of feature data through a prediction model, and output the equipment health index of the target elevator.

[0080] Step S205: Determine the abnormal state of the target elevator based on the video clips, data of the violation behavior, and the feature data of each monitoring data, and give an alarm.

[0081] Among them, as an optional embodiment of the present invention, determining the abnormal state of the target elevator based on the video clips, data of the violation behavior, and the feature data of each monitoring data, and giving an alarm includes: when the data of the violation behavior in the video clip of the violation behavior is greater than the alarm threshold, determining that the abnormal state of the target elevator is that the target elevator violates the regulations and triggering a voice alarm; when the mean value of the vibration data in the monitoring data exceeds the corresponding first fault threshold and the mean value of the current data in the monitoring data exceeds the second fault threshold, determining that the abnormal state of the target elevator is a bearing fault and giving a voice alarm.

[0082] Step S207: Generate an operation and maintenance work order according to the abnormal state and the equipment health index of the target elevator.

[0083] In the embodiment of the present invention, first, the edge node of the target elevator preprocesses the monitoring video of the target elevator and the monitoring data monitored by each sensor to obtain the video analysis result of the monitoring video and the feature data of each monitoring data. The video analysis result includes the violation behavior and the data generating the violation behavior. In this way, the monitoring video and the monitoring data monitored by each sensor can be processed at the edge node, thereby reducing the bandwidth occupation and latency. This can not only improve the monitoring real-time performance, security and reliability, but also reduce the dependence on the cloud. Secondly, perform fusion processing on each piece of feature data through a prediction model, and output the equipment health index of the target elevator; and determine the abnormal state of the target elevator based on the video clips, data of the violation behavior, and the feature data of each monitoring data, and give an alarm. In this way, the embodiment of the present invention performs fusion processing on the multi-modal data of the elevator. Through the linkage between various types of data, not only can the violation behavior occurring in the elevator be determined, but also the equipment health degree and abnormal state of the elevator can be determined. The data types are diverse and can process various monitoring data of the elevator in real time, improving the monitoring efficiency, monitoring reliability, real-time performance and security of the elevator monitoring.

[0084] Further, as an optional embodiment of the present invention, before the fusion processing of each feature data by the prediction model and the output of the equipment health index of the target elevator, the method further includes: collecting fault samples from the historical operation and maintenance records of the elevator, where the fault samples include historical fault types and historical feature data of sensors; performing data cleaning on the fault samples to obtain cleaned samples; inputting the cleaned samples into a pre-trained LSTM model for iterative training to obtain a prediction model, where the training mechanism of the pre-trained LSTM model includes an early stopping mechanism. When the loss of the pre-trained LSTM model remains unchanged for N consecutive times, the training of the LSTM model is stopped in advance, and N is a natural number greater than or equal to 6.

[0085] It should be noted that the elevator intelligent monitoring method provided in the embodiment of the present invention and the elevator intelligent monitoring platform provided in the above embodiment belong to the same inventive concept. Their similarities, similarities and beneficial effects can be referred to each other. For the sake of simplicity of the text, the embodiments of the present invention will not be described in detail here.

[0086] Based on the same inventive concept as the elevator intelligent monitoring platform provided in the above embodiment, the present invention also discloses an elevator intelligent monitoring method, as Figure 3 shown in Figure 3 is a schematic flow chart of another elevator intelligent monitoring method provided by an embodiment of the present invention. The disclosed elevator intelligent monitoring method includes: Step S301, preprocess the monitoring video of the target elevator and the monitoring data monitored by each sensor through the edge node of the target elevator to obtain the video analysis result of the monitoring video and the feature data of each monitoring data. The video analysis result includes violation behaviors and data generating violation behaviors.

[0087] Step S303, perform fusion processing on each feature data through the prediction model, and output the equipment health index of the target elevator.

[0088] Step S304, input the monitoring data monitored by each sensor into the prediction model, and calculate the overrun value of each sensor through the prediction model. The overrun value is determined by the prediction model through the initial overrun value of the sensor, the mean value and the standard deviation of the monitoring data.

[0089] Among them, as an optional embodiment of the present invention, the determination method of the overrun value of the sensor includes: calculating the product between the initial overrun value and the standard deviation; determining the sum value between the mean value and the product as the overrun value of the sensor.

[0090] Step S305, determine the abnormal state of the target elevator based on the video segment, data of the violation behavior, and the feature data of each monitoring data, and give an early warning.

[0091] Step S307: Generate an operation and maintenance work order based on the abnormal status of the target elevator and the equipment health index.

[0092] It should be noted that the elevator intelligent monitoring method provided in the embodiments of the present invention and the elevator intelligent monitoring platform provided in the above embodiments belong to the same inventive concept. Their similarities, beneficial effects can be referred to each other. For the sake of simplicity of the text, the embodiments of the present invention will not be elaborated here.

[0093] Based on the same inventive concept as the elevator intelligent monitoring platform provided in the above embodiments, the present invention also discloses an elevator intelligent monitoring platform, as Figure 4 shown in Figure 4 FIG. 10 is a schematic structural diagram of another elevator intelligent monitoring platform provided by an embodiment of the present invention. The disclosed elevator intelligent monitoring platform 400 includes: a processor 401 and a memory 402. The memory 402 is used to store a computer program that can run on the processor 401. The processor 401 is configured to execute the program stored in the memory 402 to implement each step in the above Figure 2 or Figure 3 method embodiments. Among them, the memory 402 can be a transient storage or a persistent storage. The application program stored in the memory 402 may include one or more modules (not shown in the figure), and each module may include a series of computer-executable instructions for the elevator intelligent monitoring method.

[0094] Furthermore, the processor 401 can be set to communicate with the memory 402 and execute a series of computer-executable instructions in the memory 402 on the elevator intelligent monitoring platform. The elevator intelligent monitoring platform may further include one or more power supplies 403, one or more wired or wireless network interfaces 404, one or more input / output interfaces 405, and one or more keyboards 406.

[0095] Specifically, in this embodiment, the elevator intelligent monitoring platform includes a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete communication with each other through the bus. The memory is used to store a computer program. The processor is configured to execute the program stored in the memory to implement each step in the above Figure 2 or Figure 3 method embodiments, and has the beneficial effects of the above method embodiments. To avoid repetition, the embodiments of the present invention will not be elaborated here.

[0096] It should be noted that the elevator intelligent monitoring platform provided in the embodiments of the present invention and the elevator intelligent monitoring method provided in the embodiments of the present invention are based on the same application concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the foregoing elevator intelligent monitoring method, and has the same or similar beneficial effects. The repeated parts will not be elaborated.

[0097] It should be noted that the elevator intelligent monitoring platform and the elevator intelligent monitoring method provided in the above embodiments are all based on the same inventive concept. For the same or similar technical solutions and beneficial effects, reference can be made to each other. The embodiments of the present invention will not repeat the same or similar parts.

[0098] It should be noted that the above sequence of the embodiments of the present invention is only for description and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0099] Each embodiment in this specification is described in a progressive manner. For the same or similar parts between the embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments.

Claims

1. An intelligent elevator monitoring method, characterized in that: include: Preprocessing the monitoring video of the target elevator and the monitoring data monitored by each sensor through the edge node of the target elevator to obtain the video analysis result of the monitoring video and the feature data of each monitoring data, wherein the video analysis result includes the violation behavior and the data generating the violation behavior; The characteristic data are fused and processed by a prediction model to output the equipment health index of the target elevator; Determine the abnormal state of the target elevator and issue an early warning based on the video clip of the illegal behavior, the data and the characteristic data of each monitoring data; An operation and maintenance work order is generated according to the abnormal state of the target elevator and the equipment health index.

2. The intelligent elevator monitoring method according to claim 1, characterized in that: After fusing the characteristic data through the prediction model and outputting the equipment health index of the target elevator, the method further includes: The monitoring data monitored by each sensor is input into the prediction model, and the over-limit value of each sensor is calculated by the prediction model. The over-limit value is determined by the prediction model through the initial over-limit value of the sensor, the mean value and the standard deviation of the monitoring data.

3. The intelligent elevator monitoring method according to claim 2, characterized in that: The method for determining the over-limit value of the sensor includes: Calculating the product of the initial over-limit value and the standard deviation; A sum of the mean value and the product is determined as an out-of-limit value of the sensor.

4. The intelligent elevator monitoring method according to claim 1, characterized in that: The monitoring video of the target elevator and the monitoring data monitored by each sensor are preprocessed by the edge node of the target elevator to obtain the video analysis result of the monitoring video and the feature data of each monitoring data, including: Analyze the first data of the person, the second data of the object, and the elevator door area in each video frame in the surveillance video through the lightweight YOLOv5s model in the edge node to obtain the violation in the surveillance video and the data generating the violation; The edge node extracts characteristic data of the monitoring data monitored by each sensor, wherein the monitoring data includes vibration data, current data and temperature data, and the characteristic data includes the vibration spectrum of the vibration data, the current trend of the current data and the temperature gradient of the temperature data.

5. The intelligent elevator monitoring method according to claim 4, characterized in that: The lightweight YOLOv5s model in the edge node is used to analyze the first data of the person, the second data of the object, and the elevator door area in each video frame in the surveillance video to obtain the violation in the surveillance video and the data generating the violation, including: Determine a first number of people and a second number of objects in the video frame by using the lightweight YOLOv5s model, determine a load in the target elevator by using the lightweight YOLOv5s model according to preset weights of people, weights of objects, the first number of people, and the second number of objects, and determine that the target elevator generates an overload behavior when the load exceeds an overload threshold, wherein data generating the overload behavior includes an overweight that exceeds the overload threshold; The lightweight YOLOv5s model is used to detect the door prying action of the person in the elevator door area in the video frame, and the door prying force of the person is estimated based on the action amplitude of the door prying action of the person in the video frame. When the door prying force is greater than the door prying force threshold, it is determined that the target elevator has generated the door prying behavior, and the data generating the door prying behavior includes the door prying force.

6. The intelligent elevator monitoring method according to claim 4, characterized in that: The lightweight YOLOv5s model in the edge node is used to analyze the first data of the person, the second data of the object, and the elevator door area in each video frame in the surveillance video to obtain the violation in the surveillance video and the data generating the violation, including: Using the lightweight YOLOv5s model, it is detected whether there is a person or an object in the door closing area in the elevator door area in the video frame. If there is a person or an object in the door closing area in the elevator door area, it is determined that the target elevator generates a door-blocking behavior, and the data generating the door-blocking behavior includes the type of the blocking object. The action of the person in the video frame is detected by the lightweight YOLOv5s model. If the action of the person is a jumping action, it is determined that the target elevator generates a jumping behavior, and the data generating the jumping behavior includes a jumping action detection result.

7. The intelligent elevator monitoring method according to claim 1, characterized in that: Before fusing the characteristic data through the prediction model and outputting the equipment health index of the target elevator, the method further includes: Collecting fault samples from the historical operation and maintenance records of the elevator, wherein the fault samples include historical fault types and historical characteristic data of sensors; Performing data cleaning on the fault sample to obtain a cleaned sample; The cleaned samples are input into a pre-trained LSTM model for iterative training to obtain the prediction model, wherein the training mechanism of the pre-trained LSTM model includes an early stopping mechanism. When the loss of the pre-trained LSTM model remains unchanged for N consecutive times, the training of the LSTM model is stopped in advance, and N is a natural number greater than or equal to 6.

8. The intelligent elevator monitoring method according to claim 1, characterized in that: The determining the abnormal state of the target elevator and issuing an early warning based on the video clip of the illegal behavior, the data and the characteristic data of each monitoring data includes: When the data of the violation in the video clip of the violation is greater than the alarm threshold, determining the abnormal state of the target elevator as a violation of the target elevator and triggering a voice alarm; When the mean value of the vibration data in the monitoring data exceeds the corresponding first fault threshold, and the mean value of the current data in the monitoring data exceeds the second fault threshold, the abnormal state of the target elevator is determined to be a bearing fault and a voice alarm is issued.

9. An elevator intelligent monitoring platform, characterized in that: include: An edge node and a cloud, wherein the edge node is communicatively connected with the cloud; The edge node is used to pre-process the monitoring video of the target elevator and the monitoring data monitored by each sensor, obtain the video analysis result of the monitoring video and the feature data of each monitoring data, the video analysis result includes the violation and the data that generates the violation, and send the feature data, the video clip of the violation and the data that generates the violation to the cloud; A prediction model is deployed on the cloud, and the cloud is used to fuse the characteristic data through the prediction model, output the equipment health index of the target elevator, determine the abnormal state of the target elevator and issue an early warning based on the video clips of the violation, the data and the characteristic data of each monitoring data, and generate an operation and maintenance work order based on the abnormal state of the target elevator and the equipment health index.

10. An intelligent elevator monitoring platform, characterized in that: include: A processor and a memory; wherein the memory is used to store a computer program that can be run on the processor; The processor is used to execute the program stored in the memory to implement the steps of the elevator intelligent monitoring method as described in any one of claims 1 to 8.

Images