Intelligent analysis and early warning method and system for elevator emergency rescue data

Abstract

The present invention provides an elevator emergency rescue data intelligent analysis and early warning method and system, which relates to the technical field of data analysis, including: collecting operating parameter data, performing preprocessing, performing local calculation through an elevator safety assessment model, outputting a safety assessment score and determining a safety level, judging whether to trigger an early warning mechanism, sending early warning information if triggered, classifying and summarizing the early warning information, obtaining a historical early warning data sequence, performing feature learning, generating deep features and predicting fault probability distribution, identifying key precursor factors and generating an elevator fault early warning report and a preventive maintenance strategy; judging whether to trigger an alarm mechanism, sending alarm information if triggered, performing association matching, representation learning, extracting semantic label information, modeling as a graph-based Markov decision process and solving to obtain a fault diagnosis report, selecting a rescue action combination to obtain an initial rescue plan, optimizing to obtain an optimal rescue plan and sending it to a safety management department.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and in particular to an intelligent analysis and early warning method and system for elevator emergency rescue data. Background Art

[0002] With the increasing number of high-rise buildings in cities, elevators have become an essential means of vertical transportation for people's daily travel. During operation, elevators may cause safety accidents due to aging of components, failures or interference from external factors, which seriously threaten the safety of life and property of passengers. Traditional elevator safety management mainly relies on regular inspections and passive responses, which have problems such as untimely discovery of safety hazards, inaccurate fault diagnosis, and inefficient emergency rescue.

[0003] The development of emerging information technologies such as the Internet of Things, big data, and artificial intelligence has provided new ideas and means to solve the above problems. However, most of the existing research focuses on a specific link in elevator safety management and lacks a systematic solution for the entire process. There are still problems such as inaccurate safety assessment and low level of intelligence.

[0004] Therefore, a solution is urgently needed to solve the problems existing in the prior art. Summary of the invention

[0005] The present invention provides an elevator emergency rescue data intelligent analysis and early warning method and system, which can at least solve some problems existing in the prior art.

[0006] A first aspect of the present invention provides an elevator emergency rescue data intelligent analysis and early warning method, comprising:

[0007] The operating parameter data of the elevator during operation is collected in real time by multiple sensors installed in the elevator car, and uploaded to the edge computing gateway through the industrial Ethernet bus for preprocessing. For the preprocessed operating parameter data, the elevator safety assessment model based on the adaptive reinforcement learning algorithm pre-set in the edge computing gateway is used for local calculation, and the nonlinear mapping relationship between the elevator safety state and the operating parameter data is learned in combination with the interaction process between the intelligent agent and the environment. The safety assessment score is output and the safety level is determined based on the pre-set safety level threshold, and a reward signal is generated and the hyperparameters of the elevator safety assessment model are dynamically adjusted;

[0008] Based on the safety level, it is determined whether to trigger the early warning mechanism. If the early warning mechanism is triggered, the early warning information is sent through the edge computing gateway. The cloud server receives multiple early warning information and classifies and summarizes the early warning information according to the elevator number to obtain a historical early warning data sequence. The historical early warning data sequence corresponding to each elevator is subjected to feature learning through a multivariate time series analysis method to generate deep features and construct an elevator fault early warning model. The fault probability distribution corresponding to the current elevator is predicted by combining Monte Carlo simulation and confidence interval estimation. Based on the fault probability distribution, key predictive factors are identified and an elevator fault early warning report and preventive maintenance strategy are generated.

[0009] Whether to trigger the alarm mechanism is determined based on the safety level. If the alarm mechanism is triggered, the alarm information is sent through the edge computing gateway, the operating parameter data in each alarm information is extracted and associated with the pre-set elevator equipment knowledge graph, and representation learning is performed through the fault diagnosis model deployed in the cloud to extract the semantic label information corresponding to the elevator fault. The fault diagnosis task is modeled as a graph-based Markov decision process and solved to obtain a fault diagnosis report. Based on the fault diagnosis report, a search and match is performed in the historical elevator fault case library through an intelligent optimization algorithm based on case reasoning. The historical elevator fault cases are sorted in descending order according to the similarity and the rescue action combinations corresponding to the first three historical elevator fault cases are selected to obtain an initial rescue plan. The initial rescue plan is optimized through a genetic algorithm to obtain the optimal rescue plan and send it to the safety management department.

[0010] Preferably,

[0011] The operating parameter data of the elevator during operation is collected in real time by multiple sensors set in the elevator car, uploaded to the edge computing gateway through the industrial Ethernet bus and preprocessed. For the preprocessed operating parameter data, the elevator safety assessment model based on the adaptive reinforcement learning algorithm pre-set in the edge computing gateway is used for local calculation, and the nonlinear mapping relationship between the elevator safety state and the operating parameter data is learned in combination with the interaction process between the intelligent agent and the environment. The safety assessment score is output and the safety level is determined based on the pre-set safety level threshold. The reward signal is generated and the hyperparameters of the elevator safety assessment model are dynamically adjusted, including:

[0012] Multiple sensors are set up in the elevator car to collect the acceleration, speed, temperature and vibration frequency of the current elevator as operating parameter data in real time, and a bus pre-set in the industrial Ethernet is connected to each sensor for data transmission, and the operating parameter data is uploaded to the edge computing gateway;

[0013] For the operating parameter data uploaded to the edge computing gateway, the neighbor data points of each data point are determined by a density-based local outlier factor algorithm, the first local density corresponding to the current data point and the second local density corresponding to the neighbor data point are calculated, the local outlier value is calculated based on the first local density and the second local density, the abnormal points are identified and removed based on the pre-set outlier threshold and the local outlier value, the maximum and minimum normalization method is used to perform normalization processing and the principal component analysis algorithm is combined to remove redundant information in the operating parameter data, and the preprocessed operating parameter data is obtained;

[0014] An elevator safety assessment model is constructed in the edge computing gateway based on an adaptive reinforcement learning algorithm, and the preprocessed operating parameter data is evaluated by the asynchronous dominant actor critic algorithm in the elevator safety assessment model to generate a safety assessment score and add random noise for exploration, and the safety assessment score is compared with the valuation and immediate reward of the critic network in the elevator safety assessment model, and the parameters of the critic network are updated in combination with an adaptive optimization algorithm, and the update is repeated until convergence, and the preset safety level threshold is updated according to the updated critic network and the elevator safety state is determined, and the operating parameter data is used as the agent action, and the elevator safety state is used as the environment state, and the nonlinear mapping relationship between the elevator safety state and the operating parameter data is determined by interacting with the environment, and the safety level corresponding to the current safety assessment score is determined based on the nonlinear mapping relationship and the updated safety level threshold;

[0015] Based on the safety level, a reward signal is generated and an action-value function is approximated by a neural network to dynamically adjust the hyperparameters of the elevator safety assessment model.

[0016] Preferably,

[0017] The local outlier factor is shown in the following formula:

[0018] ;

[0019] in, LOF k ( p ) represents a data point p The local outlier factor of N k ( p ) is the data point p of k -Neighborhood, representing the data point p The nearest k A set of neighbor points, w ( p , r ) represents a data point pand neighbors r The distance weight between w ( o , q ) indicates neighbor points o and neighbors q The distance weight between N ( o ) indicates the neighboring point o Adjacent k A set of neighbor points, Represents data points p Point with neighbors r The distance between Neighbor point o and neighbors q The distance between o Represents data points p Neighborhood point.

[0020] Preferably,

[0021] Based on the safety level, it is determined whether to trigger the early warning mechanism. If the early warning mechanism is triggered, the early warning information is sent through the edge computing gateway. The cloud server receives multiple early warning information and classifies and summarizes the early warning information according to the elevator number to obtain a historical early warning data sequence. The historical early warning data sequence corresponding to each elevator is subjected to feature learning through a multivariate time series analysis method to generate deep features and construct an elevator fault early warning model. The fault probability distribution corresponding to the current elevator is predicted by combining Monte Carlo simulation and confidence interval estimation. Based on the fault probability distribution, key predictive factors are identified and an elevator fault early warning report and preventive maintenance strategies are generated, including:

[0022] Based on the pre-acquired safety level corresponding to the current elevator, determine whether the current safety level is greater than the pre-set warning threshold. If so, trigger the warning mechanism; otherwise, do not trigger the warning mechanism;

[0023] If the warning mechanism is triggered, the warning information is sent to the cloud server through the edge computing gateway, wherein the warning information includes the current elevator number, the current safety level and the warning timestamp. The cloud server summarizes the received warning information and constructs the warning information corresponding to each elevator into a time series. The cloud server groups the warning information according to the station number to obtain the historical warning data sequence corresponding to each elevator.

[0024] For each elevator's historical warning data sequence, forward propagate the historical warning data sequence in multiple causal convolution blocks through a multivariate time series analysis method, extract the output of the last causal convolution block as a deep feature, determine the corresponding fault label for each deep feature, classify in combination with a support vector machine and set the objective function corresponding to the support vector machine, solve the objective function to obtain the decision function corresponding to the support vector machine, add the deep feature to the decision function, perform multiple Monte Carlo simulations to add random disturbances to the decision function, calculate the fault probability in each simulation, repeat the calculation and average the fault probabilities obtained from multiple Monte Carlo simulations to obtain the fault probability estimate corresponding to the current elevator, calculate the confidence interval corresponding to the fault probability estimate, and obtain the fault probability distribution corresponding to the current elevator;

[0025] Based on the fault probability distribution, the contribution of each feature in the warning information to the fault probability is calculated by the Shapley value method, the features in the warning information are sorted in descending order based on the contribution, the top 3 data are selected as the key predictive factors, and the elevator fault warning report and the preventive maintenance strategy are generated.

[0026] Preferably,

[0027] The contribution of each feature in the warning information to the fault probability is calculated by the Shapley value method as shown in the following formula:

[0028] ;

[0029] in, Represents the Shapley value, used to quantify features w ij Contribution to the probability of elevator failure, S Representation feature set F i A subset of w ij represents the first j Warning information features, | S | indicates a subset S The number of features in | F i | represents feature set F i The number of features in v i ( S ) is the failure probability estimation function, which means that when the feature set F i Restricted to a subset S At that time, i The probability of failure of an elevator, is the marginal contribution, indicating the S Based on this, we add the features w ij The change in the probability of failure after Indicates in the collection F i In the w ij All subsets after S .

[0030] Preferably,

[0031] Based on the safety level, it is determined whether to trigger the alarm mechanism. If the alarm mechanism is triggered, the alarm information is sent through the edge computing gateway, the operating parameter data in each alarm information is extracted and associated with the preset elevator equipment knowledge graph, and the representation learning is performed through the fault diagnosis model deployed in the cloud, and the semantic label information corresponding to the elevator fault is extracted. The fault diagnosis task is modeled as a graph-based Markov decision process and solved to obtain a fault diagnosis report. Based on the fault diagnosis report, the historical elevator fault case library is searched and matched through the case-based reasoning intelligent optimization algorithm, the historical elevator fault cases are sorted in descending order according to the similarity, and the rescue action combination corresponding to the first three historical elevator fault cases is selected to obtain the initial rescue plan, and the initial rescue plan is optimized by the genetic algorithm to obtain the optimal rescue plan and send it to the safety management department, including:

[0032] Based on the pre-acquired safety level corresponding to the current elevator, determine whether the current safety level is greater than the preset alarm threshold. If so, trigger the alarm mechanism; otherwise, do not trigger the alarm mechanism;

[0033] If the alarm mechanism is triggered, an alarm message is sent to the cloud server through the edge computing gateway, wherein the alarm message includes the operating parameter data corresponding to the elevator that triggered the alarm mechanism, and the operating parameter data includes speed, acceleration and temperature. The operating parameter data is associated and matched with a pre-set elevator equipment knowledge graph, and the operating parameter data is mapped to the embedding space corresponding to the elevator equipment knowledge graph through the knowledge graph embedding technology to obtain a node vector corresponding to the operating parameter data. The node vector is embedded and learned through a fault diagnosis model pre-set in the cloud server to obtain an alarm embedding vector corresponding to the node vector. For each node in the elevator equipment knowledge graph, the semantic similarity between the alarm embedding vector and each node is calculated, and the three nodes with the highest semantic similarity are selected to extract the corresponding semantic label information and use it as the input feature of the fault diagnosis model;

[0034] Based on the extracted semantic label information and the input features, the fault diagnosis model models the fault diagnosis task as a graph-based Markov decision process, defines the state space and action space based on the elevator equipment knowledge graph, sets a reward function and combines the discount factor and the transition probability, solves the fault diagnosis task according to a reinforcement learning algorithm based on a graph neural network, repeatedly solves and updates the parameters in the policy network and the value function network in the fault diagnosis model according to minimizing the temporal difference loss, and obtains the fault diagnosis report;

[0035] Based on the fault diagnosis report, a search is performed in a preset historical elevator fault case library through an intelligent optimization algorithm based on case reasoning, and the case similarity between each historical elevator fault case and the fault diagnosis report is calculated. The historical elevator fault cases are sorted in descending order according to the case similarity, and the rescue actions corresponding to the top three historical elevator fault cases are selected to obtain an initial rescue plan by combination. For the initial rescue plan, the initial rescue plan is encoded into a chromosome through a genetic algorithm, and a fitness function is defined. For each chromosome, a child chromosome is generated through a crossover operation and a mutation operation and the fitness value corresponding to the child chromosome is calculated. The child chromosomes are sorted in descending order based on the fitness value, and the child chromosomes in the top 10% are selected and the crossover operation and mutation operation are performed again. The iteration is repeated until the preset maximum number of iterations is reached to obtain the optimal rescue plan.

[0036] The optimal rescue plan is backed up and sent to the security management department.

[0037] Preferably,

[0038] The parameters in the strategy network and the value function network in the fault diagnosis model are updated according to minimizing the temporal difference loss as shown in the following formula:

[0039] ;

[0040] in, L ( θ , φ ) represents the time difference loss value, is the expected value operation, which represents the experience replay buffer The samples in the statistical average are s t Represents the time step t The state of a t Represents the time step t The action of time, r t Represents the time step t Instant rewards when st+1 Represents the time step t+ The state at 1 o'clock, γ represents the discount factor, Indicates the next state s t+1 Next, according to the current strategy π θ After selecting the action Q Value estimation, For the current Q Value estimation, indicating that in state s t Next action a t Expected cumulative reward.

[0041] A second aspect of the present invention provides an elevator emergency rescue data intelligent analysis and early warning system, comprising:

[0042] The first unit is used to collect the operating parameter data of the elevator in real time during operation through multiple sensors set in the elevator car, upload it to the edge computing gateway through the industrial Ethernet bus and pre-process it, and perform local calculation on the pre-processed operating parameter data through the elevator safety assessment model based on the adaptive reinforcement learning algorithm pre-set in the edge computing gateway, learn the nonlinear mapping relationship between the elevator safety state and the operating parameter data in combination with the interaction process between the intelligent agent and the environment, output the safety assessment score and determine the safety level based on the pre-set safety level threshold, generate a reward signal and dynamically adjust the hyperparameters of the elevator safety assessment model;

[0043] The second unit is used to determine whether to trigger the early warning mechanism based on the safety level. If the early warning mechanism is triggered, the early warning information is sent through the edge computing gateway. The cloud server receives multiple early warning information and classifies and summarizes the early warning information according to the elevator number to obtain a historical early warning data sequence. The historical early warning data sequence corresponding to each elevator is subjected to feature learning through a multivariate time series analysis method to generate deep features and construct an elevator fault early warning model. The fault probability distribution corresponding to the current elevator is predicted by combining Monte Carlo simulation and confidence interval estimation. Based on the fault probability distribution, key predictive factors are identified and an elevator fault early warning report and preventive maintenance strategy are generated.

[0044] The third unit is used to determine whether to trigger the alarm mechanism based on the safety level. If the alarm mechanism is triggered, the alarm information is sent through the edge computing gateway, the operating parameter data in each alarm information is extracted and associated with the pre-set elevator equipment knowledge graph, and representation learning is performed through the fault diagnosis model deployed on the cloud. The semantic label information corresponding to the elevator fault is extracted, and the fault diagnosis task is modeled as a graph-based Markov decision process and solved to obtain a fault diagnosis report. Based on the fault diagnosis report, the historical elevator fault case library is searched and matched through the case-based reasoning intelligent optimization algorithm, the historical elevator fault cases are sorted in descending order according to the similarity, and the rescue action combination corresponding to the first three historical elevator fault cases is selected to obtain an initial rescue plan, and the initial rescue plan is optimized by a genetic algorithm to obtain the optimal rescue plan and send it to the safety management department.

[0045] The third aspect of the present invention,

[0046] An electronic device is provided, comprising:

[0047] processor;

[0048] a memory for storing processor-executable instructions;

[0049] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0050] A fourth aspect of the present invention,

[0051] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0052] In the present invention, the real-time online monitoring method can timely discover abnormal situations in the operation of the elevator, and provide a reliable data basis for subsequent safety assessments, fault warnings, etc. The adaptive online learning mechanism enables the safety assessment model to respond quickly to changes in the operating conditions of the elevator, thereby improving the accuracy and real-time performance of the assessment. The intelligent early warning model based on big data analysis can detect signs of faults in advance and provide maintenance personnel with targeted preventive maintenance strategies, thereby reducing the frequency of faults, extending the service life of the elevator, and saving maintenance costs. In summary, the present invention realizes closed-loop management of the entire life cycle from data collection, safety assessment, early warning prediction to fault diagnosis and emergency rescue, which can minimize the probability of elevator safety accidents, control accident risks and losses within an acceptable range, and effectively protect the safety of passengers' lives and property, providing strong technical support for the high-quality development of smart cities. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1Schematic diagram of the process of the intelligent analysis and early warning method of elevator emergency rescue data according to an embodiment of the present invention;

[0054] Figure 2 The figure is a schematic diagram of the structure of the intelligent analysis and early warning system for elevator emergency rescue data according to an embodiment of the present invention. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0057] Figure 1 FIG. 1 is a flow chart of an intelligent analysis and early warning method for elevator emergency rescue data according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0058] S1. The operating parameter data of the elevator during operation is collected in real time by multiple sensors installed in the elevator car, and uploaded to the edge computing gateway through the industrial Ethernet bus for preprocessing. For the preprocessed operating parameter data, the elevator safety assessment model based on the adaptive reinforcement learning algorithm pre-set in the edge computing gateway is used for local calculation, and the nonlinear mapping relationship between the elevator safety state and the operating parameter data is learned in combination with the interaction process between the intelligent agent and the environment. The safety assessment score is output and the safety level is determined based on the pre-set safety level threshold, and a reward signal is generated and the hyperparameters of the elevator safety assessment model are dynamically adjusted;

[0059] The operating parameter data refers to various parameters generated when a device or system operates under specific conditions in an industrial system, such as temperature, pressure, flow, and current, which are used to monitor device performance, fault detection, and status evaluation. The industrial Ethernet bus is a communication network used for industrial automation, which uses Ethernet technology to achieve high-speed data exchange between devices. It has high reliability and real-time performance and is suitable for connecting devices such as sensors, controllers, and actuators. The edge computing gateway is a device that transfers data processing and analysis from the cloud to a location close to the data source (such as sensors and devices). The adaptive reinforcement learning algorithm is a reinforcement learning algorithm that can automatically adjust learning strategies and parameters according to environmental changes to improve the learning efficiency and performance of intelligent agents in complex and dynamic environments. The safety level threshold refers to the boundary set in system monitoring and fault diagnosis, which is used to determine whether the system is in a safe state.

[0060] In an optional embodiment,

[0061] The operating parameter data of the elevator during operation is collected in real time by multiple sensors set in the elevator car, uploaded to the edge computing gateway through the industrial Ethernet bus and preprocessed. For the preprocessed operating parameter data, the elevator safety assessment model based on the adaptive reinforcement learning algorithm pre-set in the edge computing gateway is used for local calculation, and the nonlinear mapping relationship between the elevator safety state and the operating parameter data is learned in combination with the interaction process between the intelligent agent and the environment. The safety assessment score is output and the safety level is determined based on the pre-set safety level threshold. The reward signal is generated and the hyperparameters of the elevator safety assessment model are dynamically adjusted, including:

[0062] Multiple sensors are set up in the elevator car to collect the acceleration, speed, temperature and vibration frequency of the current elevator as operating parameter data in real time, and a bus pre-set in the industrial Ethernet is connected to each sensor for data transmission, and the operating parameter data is uploaded to the edge computing gateway;

[0063] For the operating parameter data uploaded to the edge computing gateway, the neighbor data points of each data point are determined by a density-based local outlier factor algorithm, the first local density corresponding to the current data point and the second local density corresponding to the neighbor data point are calculated, the local outlier value is calculated based on the first local density and the second local density, the abnormal points are identified and removed based on the pre-set outlier threshold and the local outlier value, the maximum and minimum normalization method is used to perform normalization processing and the principal component analysis algorithm is combined to remove redundant information in the operating parameter data, and the preprocessed operating parameter data is obtained;

[0064] An elevator safety assessment model is constructed in the edge computing gateway based on an adaptive reinforcement learning algorithm, and the preprocessed operating parameter data is evaluated by the asynchronous dominant actor critic algorithm in the elevator safety assessment model to generate a safety assessment score and add random noise for exploration, and the safety assessment score is compared with the valuation and immediate reward of the critic network in the elevator safety assessment model, and the parameters of the critic network are updated in combination with an adaptive optimization algorithm, and the update is repeated until convergence, and the preset safety level threshold is updated according to the updated critic network and the elevator safety state is determined, and the operating parameter data is used as the agent action, and the elevator safety state is used as the environment state, and the nonlinear mapping relationship between the elevator safety state and the operating parameter data is determined by interacting with the environment, and the safety level corresponding to the current safety assessment score is determined based on the nonlinear mapping relationship and the updated safety level threshold;

[0065] Based on the safety level, a reward signal is generated and an action-value function is approximated by a neural network to dynamically adjust the hyperparameters of the elevator safety assessment model.

[0066] The density-based local outlier factor algorithm is an algorithm for anomaly detection. It identifies outliers by measuring the density of data points in their neighborhoods. It can effectively process high-dimensional data and adapt to the local characteristics of the data. The maximum and minimum normalization is a data preprocessing technology that scales the data to the interval [0, 1] to eliminate the impact of different dimensions on data analysis and improve the stability and convergence speed of the model. The asynchronous advantage actor-critic algorithm is a reinforcement learning algorithm that combines the actor-critic architecture and the asynchronous update mechanism. The actor is responsible for selecting actions and the critic is responsible for evaluating the value of actions. The two jointly optimize the strategy and value function. The neural network approximates the action value function through a neural network to approximate the action value function. It is used to evaluate the expected cumulative reward for performing a specific action in a given state, thereby improving the performance of the reinforcement learning algorithm.

[0067] Install various sensors such as acceleration sensors, speed sensors, temperature sensors and vibration sensors in the elevator car to collect parameter data such as acceleration, speed, temperature and vibration frequency during the operation of the elevator. Connect them to the edge computing gateway through the pre-set industrial Ethernet bus to achieve real-time data transmission.

[0068] When the elevator starts running, each sensor collects data at a frequency of 100 Hz and transmits it to the edge computing gateway through the industrial Ethernet bus. For example, the data collected at a certain moment is: acceleration 0.5 m / s 2, speed 1.2m / s, temperature 25℃, vibration frequency 10Hz. After the raw data is uploaded to the edge computing gateway, it needs to be preprocessed first. The density-based local outlier factor algorithm is used to calculate the local outlier value for each data point. First, the k nearest neighbors (k=5) of each data point are determined, and the average distance between the point and its k nearest neighbors is calculated as the first local density. Then, the average distance of the k nearest neighbors of the k nearest neighbor points is calculated as the second local density. The first local density is divided by the average value of the second local density to obtain the local outlier factor. If the local outlier factor of a data point is greater than the preset threshold 2, it is determined as an abnormal point and removed;

[0069] The maximum and minimum value normalization method is used to map each parameter to the [0, 1] interval. For example, the maximum value of acceleration is 2m / s 2 , the minimum value is 0m / s 2 , then 0.5m / s 2 The normalized value is 0.25. The principal component analysis (PCA) algorithm is used for dimensionality reduction to remove redundant information in the data, retain the principal components with a cumulative contribution rate of 95%, and reduce the original 4-dimensional data to 2-3 dimensions to obtain the reduced-dimensional and normalized operating parameter data as the input for subsequent safety assessment;

[0070] The edge computing gateway pre-builds an elevator safety assessment model based on adaptive reinforcement learning. The model uses the asynchronous advantage actor-critic (A3C) algorithm, which includes an actor network and a critic network. The actor network inputs the preprocessed operating parameter data and outputs the safety assessment score. The critic network estimates the state value function. During the assessment process, the actor network first evaluates the input data to obtain a preliminary safety assessment score. For example, if the initial assessment score of a set of data is 85 points, in order to increase the exploratory nature, a Gaussian noise with a mean of 0 and a standard deviation of 5 will be added to this score to obtain the final safety assessment score, such as 87 points.

[0071] The noisy evaluation score is compared with the state value estimated by the critic network, and the advantage function value is calculated. At the same time, an immediate reward signal is generated based on the current state and the evaluation score. For example, if the safety evaluation score is between 80-90 points, a reward of +1 is given; otherwise, a reward of -1 is given. Based on the calculated advantage function value and the immediate reward, the parameters of the critic network are updated using the adaptive moment estimation optimization algorithm. Specifically, the loss function of the critic network is calculated, and then the gradient of the loss function with respect to the network parameters is calculated, and the parameters are adjusted according to the update rule of the Adam algorithm. This is repeated many times until convergence, so that the critic network can accurately estimate the state value. After the critic network converges, the threshold of the safety level is readjusted. For example, a state value greater than 0.8 is judged as safe, between 0.5-0.8 is a warning, and less than 0.5 is dangerous. Based on the updated threshold, the safety level corresponding to the current safety evaluation score of 87 points is determined to be "safe".

[0072] The preprocessed operating parameter data is regarded as the action of the intelligent agent, and the determined safety level is regarded as the environmental state. Through multiple interactions between the intelligent agent and the environment, the nonlinear mapping relationship between the operating parameters and the safety state is gradually learned. For example, it may be learned that when the speed and acceleration are both large, the safety level decreases. The corresponding reward signal is generated according to the determined safety level. For example, a reward of +2 is given to the safe state, a reward of 0 is given to the warning state, and a reward of -2 is given to the dangerous state. The reward signal is used to approximate the action value function through a three-layer fully connected neural network. The input layer of the network is the preprocessed operating parameter data, the hidden layer uses the ReLU activation function, and the output layer is the value estimate of each action. Based on the approximated action value function, the hyperparameters of the elevator safety assessment model are dynamically adjusted. The main adjusted hyperparameters include: learning rate, discount factor, exploration rate, etc. For example, when the cumulative reward is on an upward trend, the learning rate can be appropriately reduced to stabilize the model. When the cumulative reward tends to be stable, the exploration rate can be increased to find potential better strategies.

[0073] In this embodiment, real-time collection, preprocessing, safety assessment and adaptive optimization of the model of elevator operating parameters are realized, which provides a strong guarantee for the safe operation of the elevator. It has strong real-time and adaptability, can dynamically adjust the assessment strategy according to the actual operating conditions of the elevator, and improves the accuracy and reliability of the safety assessment.

[0074] In an optional embodiment,

[0075] The local outlier factor is shown in the following formula:

[0076] ;

[0077] in, LOFk ( p ) represents a data point p The local outlier factor of N k ( p ) is the data point p of k -Neighborhood, representing the data point p The nearest k A set of neighbor points, w ( p , r ) represents a data point p and neighbors r The distance weight between w ( o , q ) indicates neighbor points o and neighbors q The distance weight between N ( o ) indicates the neighboring point o Adjacent k A set of neighbor points, Represents data points p Point with neighbors r The distance between Neighbor point o and neighbors q The distance between o Represents data points p Neighborhood point.

[0078] In this embodiment, when calculating the local outlier factor, not only the achievable distance between the data point and the neighboring point is considered, but also a distance weight factor is introduced. By introducing the distance weight, the local density relationship between the data points can be more finely characterized, and the accuracy of outlier detection can be improved. The negative exponential form can better distinguish the difference in local density, enhance the contrast between outliers and normal points, and help improve the sensitivity of outlier detection. Normalization processing can eliminate the influence of density differences in different local areas, making the value of the outlier factor more stable and comparable. In summary, this embodiment can more accurately characterize the local density relationship of data points, improve the accuracy and sensitivity of outlier detection, and reduce the computational complexity, and has a good technical effect.

[0079] S2. Determine whether to trigger the early warning mechanism based on the safety level. If the early warning mechanism is triggered, send the early warning information through the edge computing gateway. The cloud server receives multiple early warning information and classifies and summarizes the early warning information according to the elevator number to obtain the historical early warning data sequence. The historical early warning data sequence corresponding to each elevator is subjected to feature learning through the multivariate time series analysis method to generate deep features and construct an elevator fault early warning model. The fault probability distribution corresponding to the current elevator is predicted by combining Monte Carlo simulation and confidence interval estimation. Based on the fault probability distribution, the key predictive factors are identified and the elevator fault early warning report and preventive maintenance strategy are generated.

[0080] The historical warning data sequence refers to alarm data collected within a certain time range and used for fault prediction and anomaly detection, which is usually used for model training and evaluation. The multivariate time series analysis method is an analysis method that aims to process multiple time series data, identify the relationship and dynamic changes between different variables, and is used to predict future trends and anomaly detection. The Monte Carlo simulation is a numerical calculation method based on random sampling, which is used to estimate the behavior and performance of complex systems by simulating a large number of random samples, and is often used in risk assessment and decision analysis. The confidence interval estimation is a statistical method used to estimate the uncertainty of a parameter (such as the mean) and provide the possible value range of the parameter at a certain confidence level. The key predictive factors refer to those characteristics or indicators that can significantly affect the probability of failure before the failure occurs, and are often used in fault prediction and warning systems.

[0081] In an optional embodiment,

[0082] Based on the safety level, it is determined whether to trigger the early warning mechanism. If the early warning mechanism is triggered, the early warning information is sent through the edge computing gateway. The cloud server receives multiple early warning information and classifies and summarizes the early warning information according to the elevator number to obtain a historical early warning data sequence. The historical early warning data sequence corresponding to each elevator is subjected to feature learning through a multivariate time series analysis method to generate deep features and construct an elevator fault early warning model. The fault probability distribution corresponding to the current elevator is predicted by combining Monte Carlo simulation and confidence interval estimation. Based on the fault probability distribution, key predictive factors are identified and an elevator fault early warning report and preventive maintenance strategies are generated, including:

[0083] Based on the pre-acquired safety level corresponding to the current elevator, determine whether the current safety level is greater than the pre-set warning threshold. If so, trigger the warning mechanism; otherwise, do not trigger the warning mechanism;

[0084] If the warning mechanism is triggered, the warning information is sent to the cloud server through the edge computing gateway, wherein the warning information includes the current elevator number, the current safety level and the warning timestamp. The cloud server summarizes the received warning information and constructs the warning information corresponding to each elevator into a time series. The cloud server groups the warning information according to the station number to obtain the historical warning data sequence corresponding to each elevator.

[0085] For each elevator's historical warning data sequence, forward propagate the historical warning data sequence in multiple causal convolution blocks through a multivariate time series analysis method, extract the output of the last causal convolution block as a deep feature, determine the corresponding fault label for each deep feature, classify in combination with a support vector machine and set the objective function corresponding to the support vector machine, solve the objective function to obtain the decision function corresponding to the support vector machine, add the deep feature to the decision function, perform multiple Monte Carlo simulations to add random disturbances to the decision function, calculate the fault probability in each simulation, repeat the calculation and average the fault probabilities obtained from multiple Monte Carlo simulations to obtain the fault probability estimate corresponding to the current elevator, calculate the confidence interval corresponding to the fault probability estimate, and obtain the fault probability distribution corresponding to the current elevator;

[0086] Based on the fault probability distribution, the contribution of each feature in the warning information to the fault probability is calculated by the Shapley value method, the features in the warning information are sorted in descending order based on the contribution, the top 3 data are selected as the key predictive factors, and the elevator fault warning report and the preventive maintenance strategy are generated.

[0087] The causal convolution block is a convolutional neural network structure for time series modeling, which ensures that when processing sequence data, the current output depends only on the current and past inputs to maintain causality. The fault label refers to the identifier assigned to a specific fault type in a fault diagnosis or prediction task, which is usually used to train supervised learning models. The decision function is a function used to select an action under a given state, which is usually evaluated in combination with environmental feedback and a policy network to optimize the decision-making process of the intelligent agent. The Shapley value method is a distribution theory that aims to calculate the marginal contribution of each participant to the total revenue by fairly distributing the revenue in a cooperative game, and is often used for feature importance evaluation.

[0088] Get the current elevator safety level and compare the current safety level with the preset warning threshold. The warning threshold is generally set to 7 or 8 and can be adjusted according to actual conditions. If the current safety level is greater than the warning threshold, the warning mechanism is triggered; otherwise, the warning mechanism is not triggered and the elevator status continues to be monitored;

[0089] When the warning mechanism is triggered, the edge computing gateway generates a warning message, including the current elevator number (such as ELE001), the current safety level (such as 8) and the warning timestamp (such as 2023-04-01 10:30:15). The edge computing gateway transmits the warning message to the cloud server through the network. After receiving the warning message, the cloud server first groups the warning message according to the elevator number. For example, all warning messages of ELE001 are grouped into one group, and all warning messages of ELE002 are grouped into another group, and so on. For each elevator, the cloud server arranges its warning messages in chronological order and constructs a time series to form a historical warning data sequence for the elevator.

[0090] The historical warning data sequence of each elevator is processed by multivariate time series analysis method, and the feature extraction of the sequence data is performed using a multi-layer causal convolutional network. Each causal convolutional block contains a one-dimensional convolutional layer, a batch normalization layer, and an activation function layer. The sequence data passes through multiple causal convolutional blocks in sequence, and the output of the last convolutional block is used as the extracted deep feature. For example, for elevator ELE001, its historical warning data sequence is processed by 4 causal convolutional blocks to obtain a 128-dimensional feature vector as a deep feature. According to the actual fault record of the elevator, the corresponding fault label is determined for each deep feature (0 indicates no fault, 1 indicates fault);

[0091] Use support vector machine to classify the extracted deep features, set the objective function of support vector machine, including maximizing classification interval and minimizing classification error, and obtain the decision function of support vector machine by solving the objective function. Substitute the extracted deep features into the decision function to obtain the preliminary fault classification result. Use Monte Carlo simulation method to perform multiple random perturbations on the decision function, add random noise that obeys normal distribution to the decision function, repeat the calculation 1000 times, and obtain a fault probability each time. Take the average value of the fault probability obtained by these 1000 simulations as the estimated value of the current elevator's fault probability. For example, 1000 Monte Carlo simulations are performed on elevator ELE001, and the average fault probability estimate is 0.78. The 95% confidence interval of the estimate is further calculated, and the fault probability distribution is [0.75, 0.81], indicating that the probability of failure of the elevator in the future is between 75% and 81%, with a confidence level of 95%;

[0092] Based on the obtained fault probability distribution, it is necessary to further identify the key predictive factors that lead to faults. The Shapley value method based on game theory is used to calculate the contribution of each feature in the warning information to the fault probability. These features are arranged in descending order of contribution, and the top three features are selected as key predictive factors, such as door lock abnormality, motor overheating and brake wear. Based on these key predictive factors, an elevator fault warning report is generated, including information such as fault probability, key predictive factors and their importance. According to the identified key predictive factors, targeted preventive maintenance strategies are formulated. For example, for door lock abnormality, it is recommended to increase the inspection frequency of the door lock system and replace aging parts in time. For motor overheating, it is recommended to check the motor cooling system and clean the heat sink. For brake wear, it is recommended to measure the brake clearance and replace the brake lining when necessary.

[0093] In this embodiment, the advantages of edge computing and cloud computing are combined, and accurate and reliable fault prediction and key factor identification are achieved through multivariate time series analysis, support vector machine classification, Monte Carlo simulation and other technologies, which effectively reduces the risk of elevator failures, improves the safety of elevator operation, and provides strong support for elevator maintenance and management.

[0094] In an optional embodiment,

[0095] The contribution of each feature in the warning information to the fault probability is calculated by the Shapley value method as shown in the following formula:

[0096] ;

[0097] in, Represents the Shapley value, used to quantify features w ij Contribution to the probability of elevator failure, S Representation feature set F i A subset of w ij represents the first j Warning information features, | S | indicates a subset S The number of features in | F i | represents feature set F i The number of features in v i ( S ) is the failure probability estimation function, which means that when the feature set F i Restricted to a subset S At that time, i The probability of failure of an elevator, is the marginal contribution, indicating the S Based on this, we add the features w ij The change in the probability of failure after Indicates in the collection F i In the w ij All subsets after S .

[0098] In this embodiment, by calculating the marginal contribution of features in different subsets, the Shapley value can capture the nonlinear relationship between features, which helps to understand the feature interaction in complex systems. By calculating the Shapley value of each feature, a ranking of feature importance can be obtained, which helps to optimize the collection and processing of warning information and concentrate resources on the most important features. In summary, this embodiment helps to improve the performance of the fault warning model, improve the reliability and interpretability of the system, and has important technical effects.

[0099] S3. Determine whether to trigger the alarm mechanism based on the safety level. If the alarm mechanism is triggered, send the alarm information through the edge computing gateway, extract the operating parameter data in each alarm information and associate and match it with the pre-set elevator equipment knowledge graph, perform representation learning through the fault diagnosis model deployed in the cloud, extract the semantic label information corresponding to the elevator fault, model the fault diagnosis task as a graph-based Markov decision process and solve it to obtain a fault diagnosis report. Based on the fault diagnosis report, search and match in the historical elevator fault case library through the case-based reasoning intelligent optimization algorithm, sort the historical elevator fault cases in descending order according to the similarity, and select the rescue action combination corresponding to the first 3 historical elevator fault cases to obtain the initial rescue plan, optimize the initial rescue plan through the genetic algorithm, obtain the optimal rescue plan and send it to the safety management department.

[0100] The elevator equipment knowledge graph is a structured knowledge representation method that contains relevant information, attributes and relationships of elevator equipment, and is used to support fault diagnosis and intelligent decision-making. The semantic label information refers to meta-information that annotates data or knowledge, which helps the system understand the meaning and context of the data and improves the accuracy of data processing and analysis. The graph-based Markov decision process is a Markov decision process that uses a graph structure to represent the relationship between states and actions, and can effectively solve optimization and decision-making problems in the graph structure. The case-based reasoning intelligent optimization algorithm is an algorithm that solves new problems by drawing on historical cases, combining a case library and a reasoning mechanism to improve decision-making efficiency. The rescue action refers to emergency measures or operations taken after a fault occurs, aimed at restoring the normal operation of the system or avoiding greater losses.

[0101] In an optional embodiment,

[0102] Based on the safety level, it is determined whether to trigger the alarm mechanism. If the alarm mechanism is triggered, the alarm information is sent through the edge computing gateway, the operating parameter data in each alarm information is extracted and associated with the preset elevator equipment knowledge graph, and the representation learning is performed through the fault diagnosis model deployed in the cloud, and the semantic label information corresponding to the elevator fault is extracted. The fault diagnosis task is modeled as a graph-based Markov decision process and solved to obtain a fault diagnosis report. Based on the fault diagnosis report, the historical elevator fault case library is searched and matched through the case-based reasoning intelligent optimization algorithm, the historical elevator fault cases are sorted in descending order according to the similarity, and the rescue action combination corresponding to the first three historical elevator fault cases is selected to obtain the initial rescue plan, and the initial rescue plan is optimized by the genetic algorithm to obtain the optimal rescue plan and send it to the safety management department, including:

[0103] Based on the pre-acquired safety level corresponding to the current elevator, determine whether the current safety level is greater than the preset alarm threshold. If so, trigger the alarm mechanism; otherwise, do not trigger the alarm mechanism;

[0104] If the alarm mechanism is triggered, an alarm message is sent to the cloud server through the edge computing gateway, wherein the alarm message includes the operating parameter data corresponding to the elevator that triggered the alarm mechanism, and the operating parameter data includes speed, acceleration and temperature. The operating parameter data is associated and matched with a pre-set elevator equipment knowledge graph, and the operating parameter data is mapped to the embedding space corresponding to the elevator equipment knowledge graph through the knowledge graph embedding technology to obtain a node vector corresponding to the operating parameter data. The node vector is embedded and learned through a fault diagnosis model pre-set in the cloud server to obtain an alarm embedding vector corresponding to the node vector. For each node in the elevator equipment knowledge graph, the semantic similarity between the alarm embedding vector and each node is calculated, and the three nodes with the highest semantic similarity are selected to extract the corresponding semantic label information and use it as the input feature of the fault diagnosis model;

[0105] Based on the extracted semantic label information and the input features, the fault diagnosis model models the fault diagnosis task as a graph-based Markov decision process, defines the state space and action space based on the elevator equipment knowledge graph, sets a reward function and combines the discount factor and the transition probability, solves the fault diagnosis task according to a reinforcement learning algorithm based on a graph neural network, repeatedly solves and updates the parameters in the policy network and the value function network in the fault diagnosis model according to minimizing the temporal difference loss, and obtains the fault diagnosis report;

[0106] Based on the fault diagnosis report, a search is performed in a preset historical elevator fault case library through an intelligent optimization algorithm based on case reasoning, and the case similarity between each historical elevator fault case and the fault diagnosis report is calculated. The historical elevator fault cases are sorted in descending order according to the case similarity, and the rescue actions corresponding to the top three historical elevator fault cases are selected to obtain an initial rescue plan by combination. For the initial rescue plan, the initial rescue plan is encoded into a chromosome through a genetic algorithm, and a fitness function is defined. For each chromosome, a child chromosome is generated through a crossover operation and a mutation operation and the fitness value corresponding to the child chromosome is calculated. The child chromosomes are sorted in descending order based on the fitness value, and the child chromosomes in the top 10% are selected and the crossover operation and mutation operation are performed again. The iteration is repeated until the preset maximum number of iterations is reached to obtain the optimal rescue plan.

[0107] The optimal rescue plan is backed up and sent to the security management department.

[0108] The node vector is a vector used to represent the features or embedded representation of the nodes in the graph, and is used to capture the relationship and feature information between the nodes. The alarm embedding vector is a feature representation of the alarm information, and is used to convert the alarm information into a vector form that can be processed by the model for further analysis and processing. The minimization of the temporal difference loss is the goal in reinforcement learning. By optimizing the loss function, the estimation of the strategy and value function is improved, and the performance of the agent in the environment is enhanced.

[0109] Based on the pre-acquired safety level corresponding to the current elevator, determine whether the current safety level is greater than the preset alarm threshold. For example, the safety level can be divided into 1-5 levels, where 5 represents the most dangerous and 1 represents the safest. The alarm threshold is set to 3. The safety level of the current elevator is 4, which is greater than the alarm threshold 3, then the alarm mechanism is triggered;

[0110] After the alarm mechanism is triggered, the edge computing gateway sends an alarm message to the cloud server. The alarm message includes the elevator operation parameter data that triggered the alarm, such as speed, acceleration, and temperature. For example, the speed is 2m / s and the acceleration is 0.5m / s. 2 , the temperature is 35℃, the operating parameter data is associated and matched with the pre-set elevator equipment knowledge graph, the knowledge graph contains entities such as elevator components, fault types, operating parameters and their relationships, and the operating parameter data is mapped to the embedding space of the knowledge graph through knowledge graph embedding technology, such as TransE algorithm, to obtain the corresponding node vector, for example, speed, acceleration, and temperature are mapped to 100-dimensional vectors respectively;

[0111] Use the fault diagnosis model pre-deployed on the cloud server to embed node vectors for learning. Through multi-layer graph convolution operations, convert the input node vector into an alarm embedding vector. For each node in the knowledge graph, calculate the cosine similarity between it and the alarm embedding vector, and select the three nodes with the highest similarity. For example, the three selected nodes are "motor overheating", "brake system failure" and "door lock failure". Extract the corresponding semantic label information as the input feature of the fault diagnosis model. Based on the extracted semantic label information and input features, the fault diagnosis model models the diagnosis task as a graph-based Markov decision process, with the nodes in the knowledge graph as the state space and the relationship between nodes as the action space. Set the reward function, such as giving a reward of +1 for correctly diagnosing the fault and a penalty of -1 for incorrect diagnosis. At the same time, introduce a discount factor of 0.9 and define the state transition probability based on the structure of the knowledge graph;

[0112] A reinforcement learning algorithm based on graph neural network is used to solve the decision-making process. Specifically, the Actor-Critic framework is used, in which both the Actor network and the Critic network adopt graph convolutional network structure. Through multiple iterative training, the temporal difference loss is minimized, and the network parameters are continuously updated to finally obtain a fault diagnosis report, which may include diagnostic results such as "brake system failure caused by motor overheating";

[0113] Based on the fault diagnosis report, an intelligent optimization algorithm based on case reasoning is used to search and match in the historical elevator fault case library. A large number of historical fault cases are stored in the case library. Each case contains information such as fault description, diagnosis results, and rescue plan. The similarity between each historical case and the current fault diagnosis report is calculated. Text similarity calculation methods such as TF-IDF can be used to sort the historical cases in descending order according to the calculated case similarity, and select the top three cases with the highest similarity. For example, the three selected cases are "XX Community Elevator Fault Case in May 2022", "XX Building Elevator Fault Case in August 2021", and "XX Shopping Mall Elevator Fault Case in January 2023". The rescue actions corresponding to the cases are extracted, such as "cut off power", "manually release the brake", "evacuate trapped personnel", etc., and combined to form an initial rescue plan;

[0114] To optimize the initial rescue plan, a genetic algorithm is used. First, the rescue plan is encoded into chromosomes, for example, a 0 / 1 sequence is used to indicate whether a rescue action is performed. A fitness function is defined, which can take into account factors such as rescue efficiency and safety. The chromosomes are crossover and mutation operations are performed to generate offspring chromosomes, and the fitness value of the offspring is calculated.

[0115] The offspring chromosomes are sorted in descending order according to the fitness value, and the top 10% of the excellent individuals are selected for retention. These individuals are crossover and mutation operations are performed again to generate a new generation of populations. This process is repeated until the preset maximum number of iterations, such as 100 times, is reached. The rescue plan corresponding to the optimal individual finally obtained is the optimal rescue plan. The optimal rescue plan is backed up and can be stored in the cloud database. At the same time, the plan is sent to the relevant security management department, such as the elevator maintenance company, property management office, etc., through a secure communication channel, so that corresponding rescue measures can be taken in time.

[0116] In an optional embodiment,

[0117] The parameters in the strategy network and the value function network in the fault diagnosis model are updated according to minimizing the temporal difference loss as shown in the following formula:

[0118] ;

[0119] in, L ( θ , φ ) represents the time difference loss value, is the expected value operation, which represents the experience replay buffer The samples in the statistical average are s t Represents the time step t The state of a t Represents the time step t The action of time, r t Represents the time step t Instant rewards when s t+1 Represents the time step t+ The state at 1 o'clock, γ represents the discount factor, Indicates the next state s t+1 Next, according to the current strategy π θ After selecting the action Q Value estimation, For the current Q Value estimation, indicating that in state s t Next action a t Expected cumulative reward.

[0120] In this embodiment, the parameters of the two networks can be updated simultaneously by minimizing the loss function, thereby realizing the coordinated optimization of strategy learning and value estimation. The synchronous update method helps to improve the accuracy and efficiency of fault diagnosis decisions. A discount factor is introduced in the temporal difference loss to balance the importance of immediate rewards and future rewards. The time scale of fault diagnosis decisions can be controlled, and the short-term effects and long-term interests can be weighed. By minimizing the temporal difference loss, value-based strategy optimization is realized. In summary, this embodiment combines a number of key technologies in reinforcement learning, effectively improves the performance and robustness of the fault diagnosis model, and has important technical effects.

[0121] Figure 2 FIG. 1 is a schematic diagram of the structure of an intelligent analysis and early warning system for elevator emergency rescue data according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0122] The first unit is used to collect the operating parameter data of the elevator in real time during operation through multiple sensors set in the elevator car, upload it to the edge computing gateway through the industrial Ethernet bus and pre-process it, and perform local calculation on the pre-processed operating parameter data through the elevator safety assessment model based on the adaptive reinforcement learning algorithm pre-set in the edge computing gateway, learn the nonlinear mapping relationship between the elevator safety state and the operating parameter data in combination with the interaction process between the intelligent agent and the environment, output the safety assessment score and determine the safety level based on the pre-set safety level threshold, generate a reward signal and dynamically adjust the hyperparameters of the elevator safety assessment model;

[0123] The second unit is used to determine whether to trigger the early warning mechanism based on the safety level. If the early warning mechanism is triggered, the early warning information is sent through the edge computing gateway. The cloud server receives multiple early warning information and classifies and summarizes the early warning information according to the elevator number to obtain a historical early warning data sequence. The historical early warning data sequence corresponding to each elevator is subjected to feature learning through a multivariate time series analysis method to generate deep features and construct an elevator fault early warning model. The fault probability distribution corresponding to the current elevator is predicted by combining Monte Carlo simulation and confidence interval estimation. Based on the fault probability distribution, key predictive factors are identified and an elevator fault early warning report and preventive maintenance strategy are generated.

[0124] The third unit is used to determine whether to trigger the alarm mechanism based on the safety level. If the alarm mechanism is triggered, the alarm information is sent through the edge computing gateway, the operating parameter data in each alarm information is extracted and associated with the pre-set elevator equipment knowledge graph, and representation learning is performed through the fault diagnosis model deployed on the cloud. The semantic label information corresponding to the elevator fault is extracted, and the fault diagnosis task is modeled as a graph-based Markov decision process and solved to obtain a fault diagnosis report. Based on the fault diagnosis report, the historical elevator fault case library is searched and matched through the case-based reasoning intelligent optimization algorithm, the historical elevator fault cases are sorted in descending order according to the similarity, and the rescue action combination corresponding to the first three historical elevator fault cases is selected to obtain an initial rescue plan, and the initial rescue plan is optimized by a genetic algorithm to obtain the optimal rescue plan and send it to the safety management department.

[0125] According to a third aspect of the embodiments of the present invention,

[0126] An electronic device is provided, comprising:

[0127] processor;

[0128] a memory for storing processor-executable instructions;

[0129] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0130] A fourth aspect of the embodiments of the present invention is:

[0131] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0132] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0133] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. Intelligent analysis and early warning method for elevator emergency rescue data, characterized in that: include: The operating parameter data of the elevator during operation is collected in real time by multiple sensors installed in the elevator car, and uploaded to the edge computing gateway through the industrial Ethernet bus for preprocessing. For the preprocessed operating parameter data, the elevator safety assessment model based on the adaptive reinforcement learning algorithm pre-set in the edge computing gateway is used for local calculation, and the nonlinear mapping relationship between the elevator safety state and the operating parameter data is learned in combination with the interaction process between the intelligent agent and the environment. The safety assessment score is output and the safety level is determined based on the pre-set safety level threshold, and a reward signal is generated and the hyperparameters of the elevator safety assessment model are dynamically adjusted; Based on the safety level, it is determined whether to trigger the early warning mechanism. If the early warning mechanism is triggered, the early warning information is sent through the edge computing gateway. The cloud server receives multiple early warning information and classifies and summarizes the early warning information according to the elevator number to obtain a historical early warning data sequence. The historical early warning data sequence corresponding to each elevator is subjected to feature learning through a multivariate time series analysis method to generate deep features and construct an elevator fault early warning model. The fault probability distribution corresponding to the current elevator is predicted by combining Monte Carlo simulation and confidence interval estimation. Based on the fault probability distribution, key predictive factors are identified and an elevator fault early warning report and preventive maintenance strategy are generated. Whether to trigger the alarm mechanism is determined based on the safety level. If the alarm mechanism is triggered, the alarm information is sent through the edge computing gateway, the operating parameter data in each alarm information is extracted and associated with the pre-set elevator equipment knowledge graph, and representation learning is performed through the fault diagnosis model deployed in the cloud to extract the semantic label information corresponding to the elevator fault. The fault diagnosis task is modeled as a graph-based Markov decision process and solved to obtain a fault diagnosis report. Based on the fault diagnosis report, a search and match is performed in the historical elevator fault case library through an intelligent optimization algorithm based on case reasoning. The historical elevator fault cases are sorted in descending order according to the similarity and the rescue action combinations corresponding to the first three historical elevator fault cases are selected to obtain an initial rescue plan. The initial rescue plan is optimized through a genetic algorithm to obtain the optimal rescue plan and send it to the safety management department.

2. The method according to claim 1, characterized in that The operating parameter data of the elevator during operation is collected in real time by multiple sensors set in the elevator car, uploaded to the edge computing gateway through the industrial Ethernet bus and preprocessed. For the preprocessed operating parameter data, the elevator safety assessment model based on the adaptive reinforcement learning algorithm pre-set in the edge computing gateway is used for local calculation, and the nonlinear mapping relationship between the elevator safety state and the operating parameter data is learned in combination with the interaction process between the intelligent agent and the environment. The safety assessment score is output and the safety level is determined based on the pre-set safety level threshold. The reward signal is generated and the hyperparameters of the elevator safety assessment model are dynamically adjusted, including: Multiple sensors are set up in the elevator car to collect the acceleration, speed, temperature and vibration frequency of the current elevator as operating parameter data in real time, and a bus pre-set in the industrial Ethernet is connected to each sensor for data transmission, and the operating parameter data is uploaded to the edge computing gateway; For the operating parameter data uploaded to the edge computing gateway, the neighbor data points of each data point are determined by a density-based local outlier factor algorithm, the first local density corresponding to the current data point and the second local density corresponding to the neighbor data point are calculated, the local outlier value is calculated based on the first local density and the second local density, the abnormal points are identified and removed based on the pre-set outlier threshold and the local outlier value, the maximum and minimum normalization method is used to perform normalization processing and the principal component analysis algorithm is combined to remove redundant information in the operating parameter data, and the preprocessed operating parameter data is obtained; An elevator safety assessment model is constructed in the edge computing gateway based on an adaptive reinforcement learning algorithm, and the preprocessed operating parameter data is evaluated by the asynchronous dominant actor critic algorithm in the elevator safety assessment model to generate a safety assessment score and add random noise for exploration, and the safety assessment score is compared with the valuation and immediate reward of the critic network in the elevator safety assessment model, and the parameters of the critic network are updated in combination with an adaptive optimization algorithm, and the update is repeated until convergence, and the preset safety level threshold is updated according to the updated critic network and the elevator safety state is determined, and the operating parameter data is used as the agent action, and the elevator safety state is used as the environment state, and the nonlinear mapping relationship between the elevator safety state and the operating parameter data is determined by interacting with the environment, and the safety level corresponding to the current safety assessment score is determined based on the nonlinear mapping relationship and the updated safety level threshold; Based on the safety level, a reward signal is generated and an action-value function is approximated by a neural network to dynamically adjust the hyperparameters of the elevator safety assessment model.

3. The method according to claim 2, characterized in that The local outlier factor is shown in the following formula: ； in, LOF k ( p ) represents a data point p The local outlier factor of N k ( p ) is the data point p of k -Neighborhood, representing the data point p The nearest k A set of neighbor points, w ( p , r ) represents a data point p and neighbors r The distance weight between w ( o , q ) indicates neighbor points o and neighbors q The distance weight between N ( o ) indicates the neighboring point o Adjacent k A set of neighbor points, Represents data points p Point with neighbors r The distance between Neighbor point o and neighbors q The distance between o Represents data points p Neighborhood point.

4. The method according to claim 1, characterized in that Based on the safety level, it is determined whether to trigger the early warning mechanism. If the early warning mechanism is triggered, the early warning information is sent through the edge computing gateway. The cloud server receives multiple early warning information and classifies and summarizes the early warning information according to the elevator number to obtain a historical early warning data sequence. The historical early warning data sequence corresponding to each elevator is subjected to feature learning through a multivariate time series analysis method to generate deep features and construct an elevator fault early warning model. The fault probability distribution corresponding to the current elevator is predicted by combining Monte Carlo simulation and confidence interval estimation. Based on the fault probability distribution, key predictive factors are identified and an elevator fault early warning report and preventive maintenance strategies are generated, including: Based on the pre-acquired safety level corresponding to the current elevator, determine whether the current safety level is greater than the pre-set warning threshold. If so, trigger the warning mechanism; otherwise, do not trigger the warning mechanism; If the warning mechanism is triggered, the warning information is sent to the cloud server through the edge computing gateway, wherein the warning information includes the current elevator number, the current safety level and the warning timestamp. The cloud server summarizes the received warning information and constructs the warning information corresponding to each elevator into a time series. The cloud server groups the warning information according to the station number to obtain the historical warning data sequence corresponding to each elevator. For each elevator's historical warning data sequence, forward propagate the historical warning data sequence in multiple causal convolution blocks through a multivariate time series analysis method, extract the output of the last causal convolution block as a deep feature, determine the corresponding fault label for each deep feature, classify in combination with a support vector machine and set the objective function corresponding to the support vector machine, solve the objective function to obtain the decision function corresponding to the support vector machine, add the deep feature to the decision function, perform multiple Monte Carlo simulations to add random disturbances to the decision function, calculate the fault probability in each simulation, repeat the calculation and average the fault probabilities obtained from multiple Monte Carlo simulations to obtain the fault probability estimate corresponding to the current elevator, calculate the confidence interval corresponding to the fault probability estimate, and obtain the fault probability distribution corresponding to the current elevator; Based on the fault probability distribution, the contribution of each feature in the warning information to the fault probability is calculated by the Shapley value method, the features in the warning information are sorted in descending order based on the contribution, the top 3 data are selected as the key predictive factors, and the elevator fault warning report and the preventive maintenance strategy are generated.

5. The method according to claim 4, characterized in that The contribution of each feature in the warning information to the fault probability is calculated by the Shapley value method as shown in the following formula: ； in, Represents the Shapley value, used to quantify features w ij Contribution to the probability of elevator failure, S Representation feature set F i A subset of w ij represents the first j Warning information features, | S | indicates a subset S The number of features in | F i | represents feature set F i The number of features in v i ( S ) is the failure probability estimation function, which means that when the feature set F i Restricted to a subset S At that time, i The probability of failure of an elevator, is the marginal contribution, indicating the S Based on this, we add the features w ij The change in the probability of failure after Indicates in the collection F i In the w ij All subsets after S .

6. The method according to claim 1, characterized in that Based on the safety level, it is determined whether to trigger the alarm mechanism. If the alarm mechanism is triggered, the alarm information is sent through the edge computing gateway, the operating parameter data in each alarm information is extracted and associated with the preset elevator equipment knowledge graph, and the representation learning is performed through the fault diagnosis model deployed in the cloud, and the semantic label information corresponding to the elevator fault is extracted. The fault diagnosis task is modeled as a graph-based Markov decision process and solved to obtain a fault diagnosis report. Based on the fault diagnosis report, the historical elevator fault case library is searched and matched through the case-based reasoning intelligent optimization algorithm, the historical elevator fault cases are sorted in descending order according to the similarity, and the rescue action combination corresponding to the first three historical elevator fault cases is selected to obtain the initial rescue plan, and the initial rescue plan is optimized by the genetic algorithm to obtain the optimal rescue plan and send it to the safety management department, including: Based on the pre-acquired safety level corresponding to the current elevator, determine whether the current safety level is greater than the preset alarm threshold. If so, trigger the alarm mechanism; otherwise, do not trigger the alarm mechanism; If the alarm mechanism is triggered, an alarm message is sent to the cloud server through the edge computing gateway, wherein the alarm message includes the operating parameter data corresponding to the elevator that triggered the alarm mechanism, and the operating parameter data includes speed, acceleration and temperature. The operating parameter data is associated and matched with a pre-set elevator equipment knowledge graph, and the operating parameter data is mapped to the embedding space corresponding to the elevator equipment knowledge graph through the knowledge graph embedding technology to obtain a node vector corresponding to the operating parameter data. The node vector is embedded and learned through a fault diagnosis model pre-set in the cloud server to obtain an alarm embedding vector corresponding to the node vector. For each node in the elevator equipment knowledge graph, the semantic similarity between the alarm embedding vector and each node is calculated, and the three nodes with the highest semantic similarity are selected to extract the corresponding semantic label information and use it as the input feature of the fault diagnosis model; Based on the extracted semantic label information and the input features, the fault diagnosis model models the fault diagnosis task as a graph-based Markov decision process, defines the state space and action space based on the elevator equipment knowledge graph, sets a reward function and combines the discount factor and the transition probability, solves the fault diagnosis task according to a reinforcement learning algorithm based on a graph neural network, repeatedly solves and updates the parameters in the policy network and the value function network in the fault diagnosis model according to minimizing the temporal difference loss, and obtains the fault diagnosis report; Based on the fault diagnosis report, a search is performed in a preset historical elevator fault case library through an intelligent optimization algorithm based on case reasoning, and the case similarity between each historical elevator fault case and the fault diagnosis report is calculated. The historical elevator fault cases are sorted in descending order according to the case similarity, and the rescue actions corresponding to the top three historical elevator fault cases are selected to obtain an initial rescue plan by combination. For the initial rescue plan, the initial rescue plan is encoded into a chromosome through a genetic algorithm, and a fitness function is defined. For each chromosome, a child chromosome is generated through a crossover operation and a mutation operation and the fitness value corresponding to the child chromosome is calculated. The child chromosomes are sorted in descending order based on the fitness value, and the child chromosomes in the top 10% are selected and the crossover operation and mutation operation are performed again. The iteration is repeated until the preset maximum number of iterations is reached to obtain the optimal rescue plan. The optimal rescue plan is backed up and sent to the security management department.

7. The method according to claim 6, characterized in that The parameters in the strategy network and the value function network in the fault diagnosis model are updated according to minimizing the temporal difference loss as shown in the following formula: ； in, L ( θ , φ ) represents the time difference loss value, is the expected value operation, which represents the experience replay buffer The samples in the statistical average are s t Represents the time step t The state of a t Represents the time step t The action of time, r t Represents the time step t Instant rewards when s t+1 Represents the time step t+ The state at 1 o'clock, γ represents the discount factor, Indicates the next state s t+1 Next, according to the current strategy π θ After selecting the action Q Value estimation, For the current Q Value estimation, indicating that in state s t Next action a t Expected cumulative reward.

8. An intelligent analysis and early warning system for elevator emergency rescue data, used to implement the method according to any one of claims 1 to 7, characterized in that: include: The first unit is used to collect the operating parameter data of the elevator in real time during operation through multiple sensors set in the elevator car, upload it to the edge computing gateway through the industrial Ethernet bus and pre-process it, and perform local calculation on the pre-processed operating parameter data through the elevator safety assessment model based on the adaptive reinforcement learning algorithm pre-set in the edge computing gateway, learn the nonlinear mapping relationship between the elevator safety state and the operating parameter data in combination with the interaction process between the intelligent agent and the environment, output the safety assessment score and determine the safety level based on the pre-set safety level threshold, generate a reward signal and dynamically adjust the hyperparameters of the elevator safety assessment model; The second unit is used to determine whether to trigger the early warning mechanism based on the safety level. If the early warning mechanism is triggered, the early warning information is sent through the edge computing gateway. The cloud server receives multiple early warning information and classifies and summarizes the early warning information according to the elevator number to obtain a historical early warning data sequence. The historical early warning data sequence corresponding to each elevator is subjected to feature learning through a multivariate time series analysis method to generate deep features and construct an elevator fault early warning model. The fault probability distribution corresponding to the current elevator is predicted by combining Monte Carlo simulation and confidence interval estimation. Based on the fault probability distribution, key predictive factors are identified and an elevator fault early warning report and preventive maintenance strategy are generated. The third unit is used to determine whether to trigger the alarm mechanism based on the safety level. If the alarm mechanism is triggered, the alarm information is sent through the edge computing gateway, the operating parameter data in each alarm information is extracted and associated with the pre-set elevator equipment knowledge graph, and representation learning is performed through the fault diagnosis model deployed on the cloud. The semantic label information corresponding to the elevator fault is extracted, and the fault diagnosis task is modeled as a graph-based Markov decision process and solved to obtain a fault diagnosis report. Based on the fault diagnosis report, the historical elevator fault case library is searched and matched through the case-based reasoning intelligent optimization algorithm, the historical elevator fault cases are sorted in descending order according to the similarity, and the rescue action combination corresponding to the first three historical elevator fault cases is selected to obtain an initial rescue plan, and the initial rescue plan is optimized by a genetic algorithm to obtain the optimal rescue plan and send it to the safety management department.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images