Intelligent elevator cloud monitoring system based on machine vision

Abstract

The invention discloses an intelligent elevator cloud monitoring system based on machine vision. The intelligent elevator cloud monitoring system comprises a data acquisition module, a passenger state recognition module, an elevator fault monitoring module and an intelligent monitoring module. The invention relates to the technical field of elevator monitoring, in particular to an intelligent elevator cloud monitoring system based on machine vision. According to the scheme, a dynamic space-time diagram is constructed through self-adaptive trajectory similarity and scene density, multi-head edge-level attention is introduced into diagram convolution, convolution weights are distributed in combination with time domain filtering, and therefore the dynamic space-time diagram is obtained; and finally, the deep spatial-temporal features are input into a lightweight multi-layer sensor, so that high-precision and low-delay identification of the passenger state is realized, and the robustness and the real-time performance of the system in a complex environment are remarkably improved. According to the method, a self-adaptive differential weight unit is introduced to dynamically measure a change trend, a gating mechanism is combined, and multi-level time sequence modeling is carried out based on a time integral vector, so that accurate prediction of a fault probability is realized, and the sensitivity and robustness of fault detection are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of elevator monitoring technology, specifically to an intelligent elevator cloud monitoring system based on machine vision. Background Technology

[0002] In recent years, machine vision technology has made breakthroughs in video analysis, behavior recognition, and target tracking, enabling real-time perception of passenger status inside elevator cars. Meanwhile, the combination of cloud computing and IoT technologies has made real-time acquisition, storage, and analysis of large-scale heterogeneous data possible, providing new technical means for elevator status monitoring and fault prediction. However, traditional elevator passenger status recognition methods often rely on static frames or single spatiotemporal relationships, making them prone to missed or false positives in scenarios with large variations in crowd density and diverse behavioral spectra. Traditional elevator fault detection methods typically rely on single statistical features or fixed thresholds, making it difficult to capture high-order interactions and temporal dynamic changes between multidimensional sensor data. Summary of the Invention

[0003] To address the above issues and overcome the shortcomings of existing technologies, this invention provides an intelligent elevator cloud monitoring system based on machine vision. Addressing the problems of existing digital twin simulations relying heavily on static models, struggling to balance macroscopic network and microscopic finite element heat transfer coupling, and experiencing parameter drift during long-term operation leading to a disconnect between simulation results and actual operating conditions, resulting in error accumulation, this solution utilizes a thermodynamic simulation framework constructed from a BIM model. It innovatively employs thermodynamic equations coupled with finite element and air temperature balance models to predict temperature fields and heat loads, and introduces deep reinforcement learning algorithms for online negative mean square error optimization and calibration. Furthermore, it addresses the limitations of existing building energy load prediction systems in capturing long-term temporal dependencies and achieving high accuracy when operating conditions change rapidly. To address the issue of insufficient robustness, this solution constructs a three-level causal dilated convolutional network to extract multi-scale temporal dynamic features; adaptively generates a dynamic adjacency matrix and introduces a two-layer graph convolutional network to deeply characterize spatial coupling relationships; and adaptively fuses spatiotemporal features through a channel attention mechanism to achieve accurate load prediction. Addressing the problems of existing control strategies failing to balance energy consumption minimization and indoor comfort under strict temperature and load constraints, and exhibiting lag and low optimization efficiency in dynamic operating conditions, this solution is based on a hybrid optimization architecture of model predictive control and reinforcement learning. It constructs a weighted objective function that considers both energy consumption and temperature deviation, and employs mixed-integer linear programming to solve for the optimal scheduling strategy in the rolling time domain.

[0004] The technical solution adopted by the present invention is as follows: The present invention provides an intelligent elevator cloud monitoring system based on machine vision, including a data acquisition module, a passenger status recognition module, an elevator fault monitoring module and an intelligent monitoring module;

[0005] The data acquisition module specifically includes a sensor configuration unit, a data acquisition unit, and a cloud data storage unit;

[0006] The passenger state recognition module specifically includes a spatiotemporal graph construction unit, a graph attention enhancement unit, an adaptive temporal filtering unit, and a state recognition unit.

[0007] The elevator fault monitoring module specifically includes a dataset generation unit, a label setting unit, a feature interaction unit, a differential weight generation unit, an adaptive gating unit, a time integral vector generation unit, a candidate state definition unit, a hidden state update unit, and an output unit.

[0008] The intelligent monitoring module specifically acquires elevator operation video data and sensor data from the cloud in real time, inputs them into the passenger status recognition module and the elevator fault monitoring module respectively, and outputs the passenger status and elevator status. When an abnormal status occurs, a monitoring warning signal is issued.

[0009] Furthermore, the data acquisition module specifically includes the following units:

[0010] The sensor configuration unit deploys acceleration sensors, vibration sensors, temperature sensors, and pressure sensors on the elevator.

[0011] The data acquisition unit first collects historical elevator operation videos and passenger status data, including normal, crowded, and abnormal conditions. Then, it collects the elevator's acceleration in the horizontal and vertical directions using an accelerometer; it collects the vibration frequency generated during elevator operation using a vibration sensor; it collects the temperature of the motor, control cabinet, brake, and bearing components using a temperature sensor; and it collects the pressure of the cables, car, and hydraulic system using a pressure sensor. Simultaneously, it records the elevator status for each set of data, including normal and abnormal states.

[0012] The cloud-based data storage unit uses the Kafka message bus for real-time data splitting, writing time-series data to a time-series database, and writing video and image files to object storage and registering indexes in a relational database.

[0013] Furthermore, the passenger status recognition module specifically includes the following units:

[0014] Spatiotemporal graph units are constructed, and edge weights are adaptively calculated based on trajectory similarity and scene density to dynamically model the spatiotemporal relationships between nodes;

[0015] The graph attention enhancement unit employs multi-head edge-level attention, adding edge-level attention weights on top of standard graph convolution;

[0016] The adaptive temporal filtering unit allocates temporal convolution weights based on the passenger's behavioral spectrum;

[0017] The state recognition unit acquires the Lth layer output result after adaptive temporal filtering, inputs it into a lightweight multilayer perceptron, and predicts the category of passenger state.

[0018] Furthermore, the elevator fault monitoring module specifically includes the following units:

[0019] The dataset generation unit assembles the data collected by the sensors into an initial data vector, which includes six dimensions, corresponding to six features: horizontal acceleration, vertical acceleration, vibration frequency, temperature, pressure, and elevator status; and creates the dataset for model training.

[0020] The label setting unit sets the elevator status as a label for the elevator fault monitoring model;

[0021] The feature interaction unit calculates the feature interaction output value at each time step through tensor operations and weight interactions;

[0022] Generate differential weighting units to capture the changing trends and dynamic characteristics of time series data;

[0023] The adaptive gating unit introduces an update unit and a reset unit. The update unit determines how much historical information to retain, and the reset unit determines how much historical information to discard.

[0024] The time integral vector unit is generated by accumulating the data values ​​of historical time steps and applying an exponential function to generate the time integral vector.

[0025] Define candidate state units, and combine the output of the feature interaction module, the time integral vector, and the output of the reset unit to generate candidate hidden states for the current time step;

[0026] Update the hidden state unit;

[0027] The output unit maps the hidden state to the fault prediction result through a normalized exponential function.

[0028] Furthermore, the intelligent monitoring module specifically acquires elevator operation video data and sensor data from the cloud in real time. First, it inputs the elevator operation video data into the passenger status recognition module and outputs the passenger status. When the passenger status is crowded, it issues an enhanced attention prompt signal. When the passenger status is abnormal, it issues a passenger abnormality warning signal. Then, it inputs the sensor data into the elevator fault monitoring module and outputs the elevator status. When the elevator status is abnormal, it issues an elevator abnormality warning signal.

[0029] The beneficial effects achieved by the present invention using the above solution are as follows:

[0030] (1) Traditional elevator passenger status recognition methods rely on static frames or single spatiotemporal relationships, which are prone to missed or misjudged recognition in scenarios with large changes in crowd density and diverse behavioral spectra. This solution constructs a dynamic spatiotemporal graph by adaptive trajectory similarity and scene density, introduces multi-head edge-level attention on graph convolution to enhance the expression of neighborhood information, and combines temporal filtering based on behavioral spectra to allocate convolution weights. Finally, the deep spatiotemporal features are input into a lightweight multilayer perceptron, which realizes high-precision and low-latency recognition of passengers' normal, crowded and abnormal states, and significantly improves the robustness and real-time performance of the system in complex environments.

[0031] (2) In view of the problem that traditional elevator fault detection methods usually rely on a single statistical feature or fixed threshold, which makes it difficult to capture the high-order interaction and temporal dynamic changes between multi-dimensional sensor data, this scheme constructs an initial time-series vector by using six-dimensional features such as horizontal acceleration, vertical acceleration, vibration frequency, temperature and pressure. It uses tensor operation and weight interaction module to extract high-order feature combination, introduces adaptive differential weight unit to dynamically measure the changing trend, and combines the gate control mechanism of update door and reset door to selectively retain or discard historical information. It also relies on time integral vector and candidate state update unit to perform multi-level time-series modeling, and finally realizes accurate prediction of fault probability through normalized exponential output layer, thereby significantly improving the sensitivity and robustness of fault detection. Attached Figure Description

[0032] Figure 1 A schematic diagram of an intelligent elevator cloud monitoring system based on machine vision provided by the present invention;

[0033] Figure 2 This is a schematic diagram of the data acquisition module;

[0034] Figure 3 A schematic diagram of the passenger status recognition module;

[0035] Figure 4 This is a schematic diagram of an elevator fault monitoring module.

[0036] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0037] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0038] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0039] Example 1, see Figure 1 The present invention provides an intelligent elevator cloud monitoring system based on machine vision, including a data acquisition module, a passenger status recognition module, an elevator fault monitoring module and an intelligent monitoring module;

[0040] The data acquisition module deploys acceleration, vibration, temperature, and pressure sensors and cameras on the elevator to collect historical and real-time elevator operation data, passenger data, sensor data, and elevator status data. It also uses the Kafka message bus to write time-series data into a time-series database, stores video and images in object storage, and registers indexes in a relational database. Specifically, it includes a sensor configuration unit, a data acquisition unit, and a cloud data storage unit.

[0041] The passenger state recognition module identifies passenger states by constructing a dynamic spatiotemporal graph based on trajectory similarity and scene density, introducing multi-head edge-level attention-enhanced graph convolution, adaptively allocating temporal convolution weights, and inputting the fused features into a lightweight multilayer perceptron; specifically, it includes a spatiotemporal graph construction unit, a graph attention enhancement unit, an adaptive temporal filtering unit, and a state recognition unit.

[0042] The elevator fault monitoring module predicts elevator faults through multi-layer temporal modeling of tensor feature interaction, differential weight generation, adaptive gating, time integration and candidate state update, and outputs the normalized exponential output layer. Specifically, it includes a dataset generation unit, a label setting unit, a feature interaction unit, a differential weight generation unit, an adaptive gating unit, a time integration vector generation unit, a candidate state definition unit, a hidden state update unit and an output unit.

[0043] The intelligent monitoring module specifically acquires elevator operation video data and sensor data from the cloud in real time, inputs them into the passenger status recognition module and the elevator fault monitoring module respectively, and outputs the passenger status and elevator status. When an abnormal status occurs, a monitoring warning signal is issued.

[0044] Example 2, see Figure 1 and Figure 2 This embodiment is based on the above embodiment, and the data acquisition module specifically includes the following units:

[0045] The sensor configuration unit deploys acceleration sensors, vibration sensors, temperature sensors, and pressure sensors on the elevator.

[0046] The data acquisition unit first collects historical elevator operation videos and passenger status data, including normal, crowded, and abnormal conditions. Then, it uses an accelerometer to collect the elevator's acceleration in the horizontal and vertical directions; a vibration sensor to collect the vibration frequency generated during elevator operation; a temperature sensor to collect the temperature of the motor, control cabinet, brake, and bearing components; and a pressure sensor to collect the pressure of the cables, car, and hydraulic system. Simultaneously, it records the elevator status for each set of data, including normal and abnormal states.

[0047] The cloud-based data storage unit uses the Kafka message bus for real-time data splitting, writing time-series data to a time-series database, and writing video and image files to object storage and registering indexes in a relational database.

[0048] Example 3, see Figure 1 and Figure 3 This embodiment is based on the above embodiment, and the passenger status recognition module specifically includes the following units:

[0049] A spatiotemporal graph unit is constructed, and edge weights are adaptively calculated based on trajectory similarity and scene density to dynamically model the spatiotemporal relationships between nodes, as shown below:

[0050] ;

[0051] Where i and j represent the indices of nodes in the spatiotemporal graph, and u represents the index of a frame in the elevator's video feed. This represents the edge weights of node i and node j in frame u. This represents an exponential function with the natural constant as its base. This represents the trajectory similarity scaling factor, with a value range of [0,1]. The feature vector of the i-th node in the u-th frame is composed of position and velocity. This represents the feature vector of the j-th node in the u-th frame. Indicates modulo, This represents the population density suppression coefficient, with a value range of [0,1]. This represents the environmental crowd density measure between node i and node j in frame u.

[0052] The graph attention enhancement unit employs multi-head edge-level attention, adding edge-level attention weights on top of the standard graph convolution, as shown below:

[0053] ;

[0054] in, This represents the hidden representation of the (l+1)th layer and the uth frame. Represents the linear rectified function. This represents a node that is directly connected to node i in the graph structure. This represents the edge attention weight between node i and node j in the u-th frame of layer l. This represents the learnable weight matrix of the l-th layer. Let represent the hidden representation of the j-th node in the u-th frame of layer l. Indicates the index of the node. This represents the correlation score between node i and node j. This represents the correlation score between node i and node k;

[0055] The adaptive temporal filtering unit allocates temporal convolution weights based on the passenger's behavioral spectrum, as shown below:

[0056] ;

[0057] in, Let represent the hidden representation of the i-th node in the (u+1)-th frame of layer l, q and m represent the indices of the temporal convolution kernel, and Q represent the size of the temporal convolution kernel. This represents the spectral attenuation coefficient, with a value range of [0,1]. This represents the frequency of behavior of node i. This represents the hidden representation of the i-th node in the uq-th frame of layer l;

[0058] The state recognition unit acquires the Lth layer output result after adaptive temporal filtering, inputs it into a lightweight multilayer perceptron, and predicts the passenger state category, as shown below:

[0059] ;

[0060] in, This represents the passenger status of the i-th node in the u-th frame, including normal, crowded, and abnormal. This represents a multilayer perceptron. This represents the hidden representation of the i-th node in the u-th frame of layer L.

[0061] By performing the above operations, this solution addresses the problem that traditional elevator passenger status recognition methods often rely on static frames or single spatiotemporal relationships, which can easily lead to missed or false recognitions in scenarios with large variations in crowd density and diverse behavioral spectra. This solution constructs a dynamic spatiotemporal graph using adaptive trajectory similarity and scene density, introduces multi-head edge-level attention on graph convolution to enhance neighborhood information representation, and combines temporal filtering based on behavioral spectra to allocate convolution weights. Finally, the deep spatiotemporal features are input into a lightweight multilayer perceptron, achieving high-precision, low-latency recognition of passengers' normal, crowded, and abnormal states, significantly improving the system's robustness and real-time performance in complex environments.

[0062] Example 4, see Figure 1 and Figure 4 This embodiment is based on the above embodiment, and the elevator fault monitoring module specifically includes the following units:

[0063] The dataset generation unit assembles the data collected by the sensors into an initial data vector, which includes six dimensions, corresponding to six features: horizontal acceleration, vertical acceleration, vibration frequency, temperature, pressure, and elevator status; and creates the dataset for model training.

[0064] The label setting unit sets the elevator status as a label for the elevator fault monitoring model;

[0065] The feature interaction unit calculates the feature interaction output value at each time step through tensor operations and weight interactions, as shown below:

[0066] ;

[0067] Where t represents the index of the time step, This represents the output value of the feature interaction module at time step t, where c represents the index of the number of tensor operations. This represents the total number of tensor operations. This represents the tensor weights, p and v represent the indices of the data dimensions, and P represents the total number of data dimensions. and Indicates interaction weight, This represents the value of the p-th dimension of the data point at time step t. This represents the value of the v-th dimension of the data point at time step t;

[0068] Differential weighted units are generated to capture the changing trends and dynamic characteristics of time series data, as shown below:

[0069] ;

[0070] in, and Indicates the differential weights. represents the difference coefficient, with a value range of [0,1], g represents the index of the difference term, and G represents the total number of difference terms. This represents the data value at time step t. This represents the data value at the tg-th time step. Indicates the first Data values ​​at each time step Indicates the order of difference;

[0071] The adaptive gating unit introduces an update unit and a reset unit. The update unit determines how much historical information to retain, and the reset unit determines how much historical information to discard, as shown below:

[0072] ;

[0073] in, This indicates the output value of the update unit. This indicates the output value of the reset unit. This represents the sigmoid activation function. and This indicates the weight of the updated unit. and Indicates the weight of the reset unit. This indicates the bias of the update unit. This indicates the offset of the reset unit. This represents the hidden state at time step (t-1).

[0074] The time integral vector unit is generated by accumulating the data values ​​of historical time steps and applying an exponential function to generate the time integral vector, as shown below:

[0075] ;

[0076] in, This represents the time integral vector at the t-th time step. represents the integral coefficient, with a value range of [0,1], and o represents the index of the time step. This represents the data value at the 0th time step;

[0077] Define a candidate state unit, and combine the output of the feature interaction module, the time integral vector, and the output of the reset unit to generate the candidate hidden state for the current time step, as shown below:

[0078] ;

[0079] in, This represents the output value of the candidate state unit at time step t. Represents the hyperbolic tangent function. and The weights of the candidate state units are represented. Indicates the bias of the candidate state unit;

[0080] The hidden state unit is updated as follows:

[0081] ;

[0082] in, This represents the hidden state at time step t;

[0083] The output unit maps the hidden state to the fault prediction result using a normalized exponential function, as shown below:

[0084] ;

[0085] in, This indicates the output result of the output layer. Represents the normalized exponential function, and These represent the weights and biases of the output layer, respectively, and T represents the total number of time steps. This represents the hidden state at the T-th time step.

[0086] By performing the above operations, this solution addresses the problem that traditional elevator fault detection methods typically rely on only a single statistical feature or fixed threshold, making it difficult to capture high-order interactions and temporal dynamic changes between multi-dimensional sensor data. This solution constructs an initial temporal vector from six dimensions: horizontal acceleration, vertical acceleration, vibration frequency, temperature, and pressure. It then uses a tensor operation and weight interaction module to extract high-order feature combinations, introduces an adaptive differential weighting unit to dynamically measure changing trends, and combines a gating mechanism with update and reset doors to selectively retain or discard historical information. Furthermore, it relies on a time integral vector and candidate state update units to perform multi-level temporal modeling. Finally, it achieves accurate prediction of fault probability through a normalized exponential output layer, thereby significantly improving the sensitivity and robustness of fault detection.

[0087] Example 5, see Figure 1 This embodiment is based on the above embodiment. Specifically, the intelligent monitoring module obtains elevator operation video data and sensor data from the cloud in real time. First, it inputs the elevator operation video data into the passenger status recognition module and outputs the passenger status. When the passenger status is crowded, it issues an enhanced attention prompt signal. When the passenger status is abnormal, it issues a passenger abnormality warning signal. Then, it inputs the sensor data into the elevator fault monitoring module and outputs the elevator status. When the elevator status is abnormal, it issues an elevator abnormality warning signal.

[0088] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0089] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0090] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A machine vision-based intelligent elevator cloud monitoring system, characterized in that: It includes a data acquisition module, a passenger status recognition module, an elevator fault monitoring module, and an intelligent monitoring module; The data acquisition module deploys acceleration, vibration, temperature, and pressure sensors and cameras on the elevator to collect historical and real-time elevator operation data, passenger data, sensor data, and elevator status data, and stores the data in the cloud; specifically, it includes a sensor configuration unit, a data acquisition unit, and a cloud data storage unit. The passenger state recognition module identifies passenger states by constructing a dynamic spatiotemporal graph based on trajectory similarity and scene density, introducing multi-head edge-level attention-enhanced graph convolution, adaptively allocating temporal convolution weights, and inputting the fused features into a lightweight multilayer perceptron; specifically, it includes a spatiotemporal graph construction unit, a graph attention enhancement unit, an adaptive temporal filtering unit, and a state recognition unit. The elevator fault monitoring module predicts elevator faults through multi-layer temporal modeling of tensor feature interaction, differential weight generation, adaptive gating, time integration and candidate state update, and outputs the normalized exponential output layer. Specifically, it includes a dataset generation unit, a label setting unit, a feature interaction unit, a differential weight generation unit, an adaptive gating unit, a time integration vector generation unit, a candidate state definition unit, a hidden state update unit and an output unit. The intelligent monitoring module specifically acquires elevator operation video data and sensor data from the cloud in real time, inputs them into the passenger status recognition module and the elevator fault monitoring module respectively, and outputs the passenger status and elevator status. When an abnormal status occurs, a monitoring warning signal is issued.

2. The intelligent elevator cloud monitoring system based on machine vision according to claim 1, characterized in that: The data acquisition module specifically includes the following units: The sensor configuration unit deploys acceleration sensors, vibration sensors, temperature sensors, and pressure sensors on the elevator. The data acquisition unit first collects historical elevator operation videos and passenger status data, including normal, crowded, and abnormal conditions. Then, it collects the elevator's acceleration in the horizontal and vertical directions using an accelerometer; it collects the vibration frequency generated during elevator operation using a vibration sensor; it collects the temperature of the motor, control cabinet, brake, and bearing components using a temperature sensor; and it collects the pressure of the cables, car, and hydraulic system using a pressure sensor. Simultaneously, it records the elevator status for each set of data, including normal and abnormal states. The cloud-based data storage unit uses the Kafka message bus for real-time data splitting, writing time-series data to a time-series database, and writing video and image files to object storage and registering indexes in a relational database.

3. The intelligent elevator cloud monitoring system based on machine vision according to claim 1, characterized in that: The passenger status recognition module specifically includes the following units: Spatiotemporal graph units are constructed, and edge weights are adaptively calculated based on trajectory similarity and scene density to dynamically model the spatiotemporal relationships between nodes; The graph attention enhancement unit employs multi-head edge-level attention, adding edge-level attention weights on top of standard graph convolution; The adaptive temporal filtering unit allocates temporal convolution weights based on the passenger's behavioral spectrum; The state recognition unit acquires the Lth layer output result after adaptive temporal filtering, inputs it into a lightweight multilayer perceptron, and predicts the category of passenger state.

4. The intelligent elevator cloud monitoring system based on machine vision according to claim 1, characterized in that: The elevator fault monitoring module specifically includes the following units: The dataset generation unit assembles the data collected by the sensors into an initial data vector, which includes six dimensions, corresponding to six features: horizontal acceleration, vertical acceleration, vibration frequency, temperature, pressure, and elevator status; and creates the dataset for model training. The label setting unit sets the elevator status as a label for the elevator fault monitoring model; The feature interaction unit calculates the feature interaction output value at each time step through tensor operations and weight interactions; Generate differential weighting units to capture the changing trends and dynamic characteristics of time series data; The adaptive gating unit introduces an update unit and a reset unit. The update unit determines how much historical information to retain, and the reset unit determines how much historical information to discard. The time integral vector unit is generated by accumulating the data values ​​of historical time steps and applying an exponential function to generate the time integral vector. Define candidate state units, and combine the output of the feature interaction module, the time integral vector, and the output of the reset unit to generate candidate hidden states for the current time step; Update the hidden state unit; The output unit maps the hidden state to the fault prediction result through a normalized exponential function.

5. The intelligent elevator cloud monitoring system based on machine vision according to claim 1, characterized in that: The intelligent monitoring module specifically acquires elevator operation video data and sensor data from the cloud in real time. First, it inputs the elevator operation video data into the passenger status recognition module and outputs the passenger status. When the passenger status is crowded, it issues an enhanced attention prompt signal. When the passenger status is abnormal, it issues a passenger abnormality warning signal. Then, it inputs the sensor data into the elevator fault monitoring module and outputs the elevator status. When the elevator status is abnormal, it issues an elevator abnormality warning signal.

Images