Factory fire hazard potential safety hazard prediction method and system based on MoE large model
Through the factory fire safety hazard prediction method based on MoE large model, combined with multimodal data and gated network, the problem that existing systems cannot predict fire hazards is solved, and high-precision and real-time response fire warning is achieved, which significantly improves the factory safety management and emergency response capabilities.
Patent Information
- Application Number
- CN202411089656.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-09
- Publication Date
- 2024-12-06
AI Technical Summary
Existing industrial safety monitoring systems are unable to effectively predict potential fire hazards and are inefficient in processing large-scale and complex data, making it difficult to meet the needs of modern factories for high-precision and real-time response.
Using the factory fire safety hazard prediction method based on MoE large model, by collecting multimodal data (environmental monitoring data and equipment operation log), using the attention network model to extract deep features, and build a normal feedforward neural network and time convolution The MoE model of the network dynamically adjusts the sub-model weight based on the gated network to generate prediction results.
It improves the accuracy and timeliness of fire hazard prediction, provides a comprehensive safety monitoring solution, and greatly improves the factory's safety management capabilities and emergency response efficiency.
Smart Images

Figure CN119090053A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of safety early warning, and in particular to a method and system for predicting factory fire safety hazards based on a MoE large model. Background Art
[0002] In the existing field of industrial safety monitoring, although traditional fire prevention systems can detect the occurrence of fires to a certain extent, such as through smoke detectors and temperature sensors, these systems generally lack the ability to predict potential fire hazards and therefore cannot issue early warnings. In addition, traditional methods are often inefficient in processing and analyzing large-scale and complex data, especially equipment operation log information, and are difficult to meet the needs of modern factories for high precision and real-time response. Therefore, it is necessary to provide a factory fire safety hazard prediction method and system based on the MoE large model to solve the above problems. Summary of the invention
[0003] In view of the deficiencies in the prior art, the purpose of the present invention is to provide a method and system for predicting factory fire safety hazards based on the MoE large model to solve the problems existing in the above-mentioned background technology.
[0004] The present invention is implemented as follows: a method for predicting factory fire safety hazards based on the MoE large model, the method comprising the following steps:
[0005] Collecting multimodal data, the multimodal data including environmental monitoring data and equipment operation logs, the equipment operation logs recording the equipment's operating efficiency, historical maintenance information, and fault anomaly indicators;
[0006] Extract multimodal data to obtain deep features, perform feature fusion through the attention network model, and output deep fusion features;
[0007] Build a MoE model that includes a common feedforward neural network sub-model and a temporal convolutional network sub-model, and dynamically select and adjust the weights of each common feedforward neural network sub-model and temporal convolutional network sub-model based on the gating network;
[0008] The deep fusion features are input into the MoE model for inference to generate prediction results.
[0009] As a further solution of the present invention: the multimodal data is collected by multiple sensors, the ambient temperature is monitored by a temperature sensor, the smoke concentration is detected by a smoke detector, and real-time visual data inside the factory is obtained by video monitoring to capture abnormal behavior, flame and smoke characteristics.
[0010] As a further solution of the present invention: the step of extracting the multimodal data to obtain the deep features specifically includes:
[0011] Based on LSTM, the multimodal data collected by temperature sensors and smoke detectors are analyzed to identify abnormal patterns of temperature and smoke concentration and extract corresponding time series features.
[0012] The edge feature detection function based on convolutional neural network extracts flame and smoke features from visual data.
[0013] As a further solution of the present invention: the step of performing feature fusion through the attention network model and outputting deep fusion features specifically includes:
[0014] Outputting mutually weighted attention scores between features of each modality based on an attention network model, wherein the attention network model is a multi-layer neural network;
[0015] A synchronization mechanism is introduced into the attention network model to ensure the temporal consistency of features before fusion, handle the temporal misalignment of data of different modalities, and output deep fusion features.
[0016] As a further solution of the present invention: the ordinary feedforward neural network sub-model includes multiple levels of fully connected layers, wherein the first layer is the input layer, which directly receives the raw data from the sensor; the second layer is the hidden layer, which is used to extract high-level features in the input data, and each layer uses the ReLU activation function to enhance the network's ability to handle nonlinear problems; the last layer is the output layer, which converts the output of the hidden layer into the corresponding prediction result.
[0017] As a further solution of the present invention: the temporal convolutional network sub-model is used to analyze data with time dependence, and each extended convolutional layer is followed by a batch normalization layer and a ReLU activation function; the temporal convolutional network sub-model also includes a global average pooling layer, and the global average pooling layer is used to aggregate features in the time dimension.
[0018] As a further solution of the present invention: the gating network is composed of a fully connected layer, and outputs the weight of each sub-model's contribution to the final prediction, and the weight is calculated by a softmax function to ensure that the sum of all weights is 1.
[0019] Another object of the present invention is to provide a factory fire safety hazard prediction system based on the MoE large model, the system comprising:
[0020] A multimodal data acquisition module is used to collect multimodal data, wherein the multimodal data includes environmental monitoring data and equipment operation logs, wherein the equipment operation logs record the equipment's operating efficiency, historical maintenance information, and fault anomaly indicators; the multimodal data is acquired through a variety of sensors, such as a temperature sensor to monitor the ambient temperature, a smoke detector to detect the smoke concentration, and video monitoring to obtain real-time visual data inside the factory, capturing abnormal behavior, flames, and smoke characteristics;
[0021] The feature extraction and fusion module is used to extract deep features from multimodal data, perform feature fusion through the attention network model, and output deep fusion features;
[0022] The MoE large model building module is used to build a MoE model that includes a common feedforward neural network sub-model and a temporal convolutional network sub-model. The weights of each common feedforward neural network sub-model and temporal convolutional network sub-model are dynamically selected and adjusted based on the gating network.
[0023] The feature inference prediction module is used to input the deep fusion features into the MoE model, perform inference, and generate prediction results.
[0024] As a further solution of the present invention: the feature extraction and fusion module includes:
[0025] The time series feature unit is used to analyze the multimodal data collected by the temperature sensor and the smoke detector based on LSTM, identify the abnormal patterns of temperature and smoke concentration, and extract the corresponding time series features;
[0026] The edge feature detection unit is used to extract flame and smoke features in visual data based on the edge feature detection function of the convolutional neural network.
[0027] As a further solution of the present invention: the feature extraction and fusion module also includes:
[0028] A weighted attention score unit, configured to output a mutual weighted attention score between features of each modality based on an attention network model, wherein the attention network model is a multi-layer neural network;
[0029] The time misalignment processing unit is used to introduce a synchronization mechanism into the attention network model to ensure the temporal consistency of features before fusion, process the temporal misalignment of data of different modalities, and output deep fusion features.
[0030] Compared with the prior art, the present invention has the following beneficial effects:
[0031] The present invention combines the diversified processing capabilities of the MoE model and the long time series analysis advantages of TCN, which improves the prediction accuracy of factory fire hazards and significantly improves the timeliness of early warning. It provides a comprehensive safety monitoring solution for modern factories and greatly improves the factory's safety management capabilities and emergency response efficiency. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] Figure 1 It is a flow chart of a method for predicting factory fire safety hazards based on the MoE large model;
[0033] Figure 2 A flowchart for determining fusion features in a factory fire safety hazard prediction method based on the MoE large model;
[0034] Figure 3 This is a structural diagram of a factory fire safety hazard prediction system based on the MoE large model. DETAILED DESCRIPTION
[0035] In order to make the purpose, technical solution and advantages of the present invention clearer, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0036] The specific implementation of the present invention is described in detail below in conjunction with specific embodiments.
[0037] like Figure 1 As shown, an embodiment of the present invention provides a method for predicting factory fire safety hazards based on the MoE large model, and the method includes the following steps:
[0038] S100, collecting multimodal data, where the multimodal data includes environmental monitoring data and equipment operation logs, where the equipment operation logs record equipment operation efficiency, historical maintenance information, and fault anomaly indicators;
[0039] S200, extracting multimodal data to obtain deep features, performing feature fusion through an attention network model, and outputting deep fusion features;
[0040] S300, building a MoE model including a common feedforward neural network sub-model and a temporal convolutional network sub-model, and dynamically selecting and adjusting the weights of each common feedforward neural network sub-model and temporal convolutional network sub-model based on a gating network;
[0041] S400, input the deep fusion features into the MoE model, perform inference, and generate prediction results.
[0042] It should be noted that:
[0043] In an embodiment of the present invention, the multimodal data is collected by a variety of sensors, the ambient temperature is monitored by a temperature sensor, the smoke concentration is detected by a smoke detector, and real-time visual data inside the factory is obtained by video monitoring to capture abnormal behavior, flame and smoke characteristics. The multimodal data is formatted and noise filtered through an efficient preprocessing algorithm to obtain a preliminary feature representation. Then it is necessary to extract the multimodal data to obtain deep features, perform feature fusion through the attention network model, and output deep fusion features. Then it is necessary to build a MoE model that includes a normal feedforward neural network (FFN) submodel and a temporal convolutional network (TCN) submodel, and dynamically select and adjust the weights of each normal feedforward neural network submodel and temporal convolutional network submodel based on the gating network;
[0044] The FFN sub-model is specialized in processing real-time data from devices such as temperature sensors and smoke detectors, which usually do not contain time series information. The FFN structure includes multiple levels of fully connected layers. Specifically, the first layer is the input layer, which directly receives raw data from sensors, such as temperature values or smoke concentrations. The following hidden layers are used to extract high-level features from these input data, and each layer uses the ReLU activation function to enhance the network's ability to handle nonlinear problems. The last layer is the output layer, which converts the output of the hidden layer into corresponding prediction results, such as fire risk assessment. The advantage of this structure of FFN is that it can process and respond quickly, and can instantly identify sudden abnormal data, which is crucial for immediate warning systems. The TCN sub-model is used to analyze time-dependent data, such as equipment operation logs. These logs usually record the operating status and historical behavior of the equipment, and their time series analysis is crucial for predicting equipment failures and potential fire risks. The key feature of TCN is the use of dilated convolutions, which can expand the network's receptive field without significantly increasing computational complexity. Each dilated convolution layer is usually followed by a batch normalization layer and ReLU activation function to stabilize the training process and enhance the model's prediction ability. The end of TCN is usually a global average pooling layer that aggregates features in the time dimension, and the final output layer converts these features into specific prediction outputs.
[0045] The gating network is the core of the MoE model, which is responsible for dynamically selecting appropriate sub-models and adjusting their output weights. The gating network itself is composed of a fully connected layer, the input is the features from the FFN and TCN sub-models, and the output is the weight of each sub-model's contribution to the final prediction. These weights are calculated by the softmax function to ensure that the sum of all weights is 1. The output weight of the gating network determines the degree of influence of each sub-model output on the final prediction result, so that the model can flexibly adjust the strategy according to the characteristics of the current data. The present invention processes simple real-time data features through FFN, while TCN deeply analyzes complex time series data. The gating network adjusts the weights of each sub-model according to the real-time data status. The entire system can monitor and predict the fire hazards of the factory in real time in multiple dimensions. This comprehensive utilization of multi-modal and multi-technical levels not only greatly improves the accuracy of the prediction, but also enhances the system's adaptability and responsiveness to various scenarios, providing strong technical support for factory safety management.
[0046] Finally, by constructing the obtained MoE model, the input features are inferred to achieve the prediction effect; the mixed expert model (MoE) contains multiple FFN and TCN sub-models, and the outputs of these sub-models are integrated through a highly flexible gating network. The gating network not only dynamically selects and adjusts the weights of each sub-model, but also optimizes these weights in real time according to the changes in the real-time data stream to ensure that the model output always reflects the latest environmental conditions. This design enables the MoE model to not only accurately predict potential fire hazards, but also adapt to various sudden changes, improving the robustness and generalization ability of the model in complex industrial environments. In addition, it also contains an advanced real-time monitoring and response module. This module continuously monitors environmental data collected from various sensors, such as temperature fluctuations and changes in smoke concentration. Based on the real-time reasoning results of the MoE model, the module can dynamically adjust the monitoring strategy and response parameters. For example, if the model predicts that the fire risk in a specific area is rising, the monitoring module will automatically increase the scanning frequency of sensors in the area, enhance the density of data collection, or adjust the alarm threshold of the early warning system. In the case of determining that the fire risk is high, the system can automatically trigger an audible and visual alarm, and quickly notify factory managers and emergency response teams through an automated communication system to make corresponding disaster prevention preparations.
[0047] The embodiment of the present invention greatly improves the efficiency of fire prevention and response through an intelligent dynamic adjustment mechanism, ensuring the safety of the factory and the continuity of production. Through this highly automated monitoring and response system, the factory can effectively respond to potential fire hazards at the initial stage, thereby significantly reducing the risk of disasters and possible losses.
[0048] like Figure 2As shown, as a preferred embodiment of the present invention, the step of extracting multimodal data to obtain deep features specifically includes:
[0049] S201, analyzing the multimodal data collected by the temperature sensor and the smoke detector based on LSTM, identifying abnormal patterns of temperature and smoke concentration, and extracting corresponding time series features;
[0050] S202, extracting flame and smoke features from visual data based on the edge feature detection function of the convolutional neural network.
[0051] The step of performing feature fusion through the attention network model and outputting deep fusion features specifically includes:
[0052] S203, outputting a mutual weighted attention score between each modality feature based on an attention network model, wherein the attention network model is a multi-layer neural network;
[0053] S204, introduce a synchronization mechanism into the attention network model to ensure the temporal consistency of features before fusion, process the temporal misalignment of data of different modalities, and output deep fusion features.
[0054] In the embodiment of the present invention, in order to fully explore and utilize the rich information of multimodal data, deep feature learning is performed on the input features: the purpose is to extract high-level features that are helpful for fire hazard prediction from each modal data. For example, features are extracted from different data sources: temperature sensor data, smoke detector data and equipment operation logs, using a long short-term memory network (LSTM). LSTM is a network that is particularly suitable for processing and predicting important events with long intervals and delays in time series data. LSTM is used to analyze data from temperature sensors and smoke detectors. The data provided by these sensors usually include temperature and smoke concentration changes recorded by timestamps. LSTM can learn the time dependencies in these data and identify abnormal patterns of temperature and smoke concentration, which often indicate potential fires. LSTM effectively manages the long-term dependencies of information through its internal state and gating mechanism, which makes it very effective in capturing dynamic changes in time series data. It can identify trends of rising temperature or sudden increase in smoke concentration from historical data, which are important indicators of potential fires. Use a long short-term memory network (LSTM) to extract corresponding time series features. For video surveillance data: use a convolutional neural network (CNN) to extract visual features of flames and smoke. For video surveillance data, CNNs are used to extract visual features, especially visual signs of flames and smoke. CNNs are able to effectively process image data through their multi-layered structure, from basic edge detection to more complex shape and object recognition. In fire monitoring and prediction, CNNs can accurately identify and locate flames and smoke in videos by learning visual features at different levels. For example, the first layer of a CNN may recognize simple shapes and edges, while deeper networks can parse complex objects such as the dynamic changes of flames and the spread pattern of smoke.
[0055] Subsequently, the extracted deep features are weighted and fused through the attention mechanism. In the present invention, the deep features extracted from different modalities are integrated and fused through the cross-modal attention mechanism to form a multi-dimensional feature representation. The attention mechanism uses a multi-layer neural network, the input is the features from each modality, and the output is the mutually weighted attention score between the features of each modality. The features of each modality are weighted and combined according to the learned attention weights, so that the features that are more critical to the current task are emphasized. For example, when the quality of visual data is affected by light or occlusion, the system can automatically increase the weights of other sensor features. In addition, in order to deal with the temporal misalignment of different modal data, a synchronization mechanism is introduced to ensure the temporal consistency of features before fusion. In this way, the fused features are not only rich and global, but also significantly improve the model's overall understanding of the scene. Thus, comprehensive high-level features are generated to improve the accuracy and robustness of fire hazard prediction.
[0056] like Figure 3As shown, the embodiment of the present invention also provides a factory fire safety hazard prediction system based on the MoE large model, and the system includes:
[0057] The multimodal data acquisition module 100 is used to collect multimodal data, wherein the multimodal data includes environmental monitoring data and equipment operation logs, wherein the equipment operation logs record the equipment's operating efficiency, historical maintenance information, and fault anomaly indicators; the multimodal data is acquired through a variety of sensors, such as a temperature sensor to monitor the ambient temperature, a smoke detector to detect the smoke concentration, and video monitoring to obtain real-time visual data inside the factory, capturing abnormal behavior, flames, and smoke characteristics;
[0058] The feature extraction and fusion module 200 is used to extract deep features from multimodal data, perform feature fusion through an attention network model, and output deep fusion features;
[0059] The MoE large model construction module 300 is used to build a MoE model including a common feedforward neural network sub-model and a temporal convolutional network sub-model, and dynamically select and adjust the weights of each common feedforward neural network sub-model and the temporal convolutional network sub-model based on a gating network;
[0060] The feature inference prediction module 400 is used to input the deep fusion features into the MoE model, perform inference, and generate prediction results.
[0061] As a preferred embodiment of the present invention, the feature extraction and fusion module 200 includes:
[0062] The time series feature unit is used to analyze the multimodal data collected by the temperature sensor and the smoke detector based on LSTM, identify the abnormal patterns of temperature and smoke concentration, and extract the corresponding time series features;
[0063] The edge feature detection unit is used to extract flame and smoke features in visual data based on the edge feature detection function of the convolutional neural network.
[0064] As a preferred embodiment of the present invention, the feature extraction and fusion module 200 further includes:
[0065] A weighted attention score unit, configured to output a mutual weighted attention score between features of each modality based on an attention network model, wherein the attention network model is a multi-layer neural network;
[0066] The time misalignment processing unit is used to introduce a synchronization mechanism into the attention network model to ensure the temporal consistency of features before fusion, process the temporal misalignment of data of different modalities, and output deep fusion features.
[0067] The above only describes in detail the preferred embodiments of the present invention, which is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
[0068] It should be understood that, although each step in the flow chart of each embodiment of the present invention is shown in sequence according to the indication of the arrow, these steps are not necessarily performed in sequence according to the order indicated by the arrow. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be performed in other orders. Moreover, at least a portion of the steps in each embodiment may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.
[0069] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
[0070] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the disclosure in the specification and examples. This application is intended to cover any variations, uses or adaptations of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the claims.
Claims
1. The method for predicting factory fire safety hazards based on the MoE large model is characterized by: The method comprises the following steps: Collecting multimodal data, the multimodal data including environmental monitoring data and equipment operation logs, the equipment operation logs recording the equipment's operating efficiency, historical maintenance information, and fault anomaly indicators; Extract multimodal data to obtain deep features, perform feature fusion through the attention network model, and output deep fusion features; Build a MoE model that includes a common feedforward neural network sub-model and a temporal convolutional network sub-model, and dynamically select and adjust the weights of each common feedforward neural network sub-model and temporal convolutional network sub-model based on the gating network; The deep fusion features are input into the MoE model for inference to generate prediction results.
2. The method for predicting factory fire safety hazards based on the MoE large model according to claim 1 is characterized in that: The multimodal data is collected through a variety of sensors, the ambient temperature is monitored by a temperature sensor, the smoke concentration is detected by a smoke detector, and real-time visual data inside the factory is obtained through video monitoring to capture abnormal behavior, flame and smoke characteristics.
3. The method for predicting factory fire safety hazards based on the MoE large model according to claim 2 is characterized in that: The step of extracting the multimodal data to obtain the deep features specifically includes: Based on LSTM, the multimodal data collected by temperature sensors and smoke detectors are analyzed to identify abnormal patterns of temperature and smoke concentration and extract corresponding time series features. The edge feature detection function based on convolutional neural network extracts flame and smoke features from visual data.
4. The method for predicting factory fire safety hazards based on the MoE large model according to claim 2 is characterized in that: The step of performing feature fusion through the attention network model and outputting deep fusion features specifically includes: Outputting mutually weighted attention scores between features of each modality based on an attention network model, wherein the attention network model is a multi-layer neural network; A synchronization mechanism is introduced into the attention network model to ensure the temporal consistency of features before fusion, handle the temporal misalignment of data of different modalities, and output deep fusion features.
5. The method for predicting factory fire safety hazards based on the MoE large model according to claim 1 is characterized in that: The ordinary feedforward neural network sub-model includes multiple levels of fully connected layers, among which the first layer is the input layer, which directly receives the raw data from the sensor; the second layer is the hidden layer, which is used to extract high-level features in the input data, and each layer uses the ReLU activation function to enhance the network's ability to handle nonlinear problems; the last layer is the output layer, which converts the output of the hidden layer into the corresponding prediction result.
6. The method for predicting factory fire safety hazards based on the MoE large model according to claim 1 is characterized in that: The temporal convolutional network sub-model is used to analyze data with time dependency, and each extended convolutional layer is followed by a batch normalization layer and a ReLU activation function; the temporal convolutional network sub-model also includes a global average pooling layer, which is used to aggregate features in the time dimension.
7. The method for predicting factory fire safety hazards based on the MoE large model according to claim 1 is characterized in that: The gating network consists of a fully connected layer, which outputs the weight of each sub-model's contribution to the final prediction. The weight is calculated by the softmax function to ensure that the sum of all weights is 1.
8. The factory fire safety hazard prediction system based on the MoE large model is characterized by: The system comprises: A multimodal data acquisition module is used to collect multimodal data, wherein the multimodal data includes environmental monitoring data and equipment operation logs, wherein the equipment operation logs record the equipment's operating efficiency, historical maintenance information, and fault anomaly indicators; the multimodal data is acquired through a variety of sensors, such as a temperature sensor to monitor the ambient temperature, a smoke detector to detect the smoke concentration, and video monitoring to obtain real-time visual data inside the factory, capturing abnormal behavior, flames, and smoke characteristics; The feature extraction and fusion module is used to extract deep features from multimodal data, perform feature fusion through the attention network model, and output deep fusion features; The MoE large model building module is used to build a MoE model that includes a common feedforward neural network sub-model and a temporal convolutional network sub-model. The weights of each common feedforward neural network sub-model and temporal convolutional network sub-model are dynamically selected and adjusted based on the gating network. The feature inference prediction module is used to input the deep fusion features into the MoE model, perform inference, and generate prediction results.
9. The factory fire safety hazard prediction system based on the MoE large model according to claim 8 is characterized in that: The feature extraction and fusion module includes: The time series feature unit is used to analyze the multimodal data collected by the temperature sensor and the smoke detector based on LSTM, identify the abnormal patterns of temperature and smoke concentration, and extract the corresponding time series features; The edge feature detection unit is used to extract flame and smoke features in visual data based on the edge feature detection function of the convolutional neural network.
10. The factory fire safety hazard prediction system based on the MoE large model according to claim 8 is characterized in that: The feature extraction and fusion module also includes: A weighted attention score unit, configured to output a mutual weighted attention score between features of each modality based on an attention network model, wherein the attention network model is a multi-layer neural network; The time misalignment processing unit is used to introduce a synchronization mechanism into the attention network model to ensure the temporal consistency of features before fusion, process the temporal misalignment of data of different modalities, and output deep fusion features.
Citation Information
Cited By
Plasma stove abnormity monitoring method, device, equipment and medium
CN119885042A
Method and device for identifying fire protection hidden danger through AI visual analysis technology
CN120123829A
