Electrical fire alarm management method and system under multi-modal information fusion

Abstract

The invention discloses an electrical fire alarm management method and system under multi-modal information fusion, and relates to the technical field of electrical fire alarm, and the method comprises the steps: carrying out the multi-modal information collection of electrical equipment and lines in a production region, and obtaining multi-modal heterogeneous data; carrying out real-time monitoring on the multi-mode heterogeneous data of each electrical device and line, and judging whether an electrical fire alarm signal is generated or not according to a monitoring result; performing dynamic association analysis on different types of electrical fire alarm signals generated by each electrical device and line in a plurality of historical acquisition periods to obtain a dynamic association network corresponding to each type of electrical fire alarm signal; according to the dynamic association network corresponding to each type of electrical fire alarm signals, an electrical fire early warning model is constructed, and real-time early warning monitoring is performed on electrical equipment and lines which do not generate the electrical fire alarm signals, so that the comprehensiveness and accuracy of electrical fire alarm are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical fire alarm, and specifically to an electrical fire alarm management method and system under multi-modal information fusion. Background Art

[0002] CN104504837B "Method for Collecting Data Information of Electrical Fire Alarm System and Collector" includes: the data information collector collects the status information of the early warning device and the electrical fire alarm host, and parses out the on, off, and fault status information of the early warning device and the fault status of the electrical fire alarm host and the issued on, off, and shielding control instruction information according to the collected status information. The corresponding relationship between the status information and the on, off status, control instruction, and fault status information is preset in the data information collector; and the parsed information is sent.

[0003] CN115311811B "Method and Device for Remote Alarm Processing of Electrical Fire Based on Internet of Things" obtains the image difference degree according to the same number of pixel points and different number of pixel points; if the image difference degree is greater than the preset difference degree, then multiple acquisition times are obtained according to the current time and the preset time period, and the regional images of the areas where each substation equipment is located in the infrared image are extracted based on the acquisition times; the pixel values of the corresponding regional images at adjacent time points in the image set are processed to obtain a plurality of pixel value change trends, and the trend change difference corresponding to each image set is obtained based on the plurality of pixel value change trends; the preset trend threshold is adjusted according to the environmental information, service life, and average value of the total number of target pixel points to obtain the current trend threshold; if the trend change difference is greater than the current trend threshold, the substation equipment corresponding to the image set is used as the early warning substation equipment, and the early warning substation equipment is sent to the fire processing end.

[0004] With the wide application of electrical equipment in various fields, the incidence rate of electrical fires has gradually increased. Traditional electrical fire alarm methods often rely only on a single type of monitoring data, such as temperature or smoke concentration. This method has problems such as high false alarm rate and large risk of missed alarms. For example, in some environments, due to natural fluctuations in temperature or dust interference near the smoke sensor, false alarms may occur; and for some early electrical faults, they may not be detected in time by relying only on single-parameter monitoring, resulting in missed alarms. In addition, single-parameter monitoring cannot comprehensively reflect the operating status and fire risk of electrical equipment, and it is difficult to accurately judge the possibility and severity of a fire. Summary of the Invention

[0005] In order to solve the above technical problems, the object of the present invention is to provide an electrical fire alarm management method under multi-modal information fusion, including the following steps:

[0006] Step s1: Collect multi-modal information of electrical equipment and lines in the production area, obtain multi-modal heterogeneous data, mark the collection time, and set the collection period.

[0007] Step s2: Monitor the multi-modal heterogeneous data of each electrical equipment and line in real time, and judge whether to generate an electrical fire alarm signal according to the monitoring results.

[0008] Step s3: Conduct dynamic correlation analysis on different types of electrical fire alarm signals generated by each electrical equipment and line in several historical collection periods, and obtain the dynamic correlation network corresponding to each type of electrical fire alarm signal.

[0009] Step s4: Construct an electrical fire warning model according to the dynamic correlation network corresponding to each type of electrical fire alarm signal, and conduct real-time warning monitoring on electrical equipment and lines that have not generated electrical fire alarm signals.

[0010] Furthermore, the multi-modal heterogeneous data includes electrical data (including electrical indicators such as current and voltage), environmental data (environmental indicators such as temperature, humidity, smoke particle concentration, and concentrations of various types of gases), and image data.

[0011] Furthermore, the process of monitoring the multi-modal heterogeneous data of each electrical equipment and line in real time and judging whether to generate an electrical fire alarm signal according to the monitoring results includes:

[0012] Extract the numerical time series corresponding to each type of electrical indicator in the electrical data of each electrical equipment and line and the numerical time series corresponding to each type of environmental indicator in the environmental data, obtain the threshold intervals corresponding to each type of electrical indicator of each electrical equipment and line, and preset the threshold intervals corresponding to each type of environmental indicator of each electrical equipment and line.

[0013] Compare the numerical time series corresponding to each type of electrical indicator and each type of environmental indicator with the corresponding threshold intervals, obtain the cumulative time when each type of electrical indicator and each type of environmental indicator are not within the corresponding threshold intervals, and compare the cumulative time with the preset error upper limit. If the cumulative time of the electrical indicator or environmental indicator is greater than the error upper limit threshold, generate the electrical fire alarm signal for the electrical indicator or environmental indicator.

[0014] If the cumulative time of the electrical indicator or environmental indicator is less than or equal to the error upper limit threshold, conduct image data monitoring of the electrical indicator or environmental indicator.

[0015] Furthermore, the process of conducting image data monitoring of the electrical indicator or environmental indicator includes:

[0016] Pre-collect simulation training image data under several different scenario states and perform scenario state annotation on the simulation training image data. The scenario states include normal operation state, abnormal phenomena such as smoking and sparks, and potential fire risk factors in the surrounding environment (such as accumulation of flammable materials, chaotic electrical circuits, etc.). Perform data format preprocessing on the simulation training image data after scenario state annotation. The data format preprocessing operations include cropping, scaling, and normalization of the images. Build an image state annotation model based on deep learning. Select Faster R-CNN as the deep learning architecture. Since the goal is to distinguish the scenario states of the images, the cross-entropy loss (Binary Cross-Entropy Loss) is selected as the optimization target. Subsequently, use the simulation training image data after data format preprocessing as training data, and use the training data to train the image state annotation model to output the trained image state annotation model;

[0017] Perform data format preprocessing on the image data of each electrical device and circuit, input the image data after data format preprocessing into the image state annotation model, and obtain the scenario states of each electrical device and circuit according to the output of the image state annotation model;

[0018] If the scenario state of an electrical device or circuit is not the normal operation state, generate an electrical fire alarm signal in the scenario state according to the scenario state of the electrical device or circuit.

[0019] Furthermore, the process of obtaining the threshold intervals corresponding to various types of electrical indicators of each electrical device and circuit includes:

[0020] Obtain the historical acquisition periods when each electrical device and circuit generate electrical fire alarm signals, perform statistical analysis on the environmental data of the historical acquisition periods, and obtain the probabilities of each electrical device and circuit having electrical fires under different environmental data conditions;

[0021] The calculation process of obtaining the probabilities of having electrical fires under different environmental data conditions is:

[0022]

[0023] Among them, P (i，M) represents the probability of electrical device or circuit i having an electrical fire under environmental data M, NUM i represents the total number of times electrical device or circuit i generates an electrical fire alarm signal, NUM i (M) represents the number of times electrical device or circuit i has an electrical fire under environmental data M, and θ represents the conversion coefficient;

[0024] A preset probability comparison table, the probability comparison table includes threshold intervals of various types of electrical indicators corresponding to different probabilities, obtain the environmental data of each electrical device and circuit in the current acquisition period, obtain the probability of electrical fire occurring for each electrical device and circuit under the condition of the environmental data, and obtain the threshold intervals corresponding to various types of electrical indicators of each electrical device and circuit according to the probability and the probability comparison table.

[0025] Further, the process of dynamically associating and analyzing different types of electrical fire alarm signals generated by each electrical device and circuit in several historical acquisition periods to obtain the dynamic association network corresponding to each type of electrical fire alarm signal includes:

[0026] The different types of electrical fire alarm signals include electrical fire alarm signals of electrical indicators, electrical fire alarm signals of environmental indicators, and electrical fire alarm signals in a situational state. Obtain the historical acquisition periods of different types of electrical fire alarm signals generated by each electrical device and circuit, conduct an association analysis between various types of electrical indicators and various types of environmental indicators in the historical acquisition periods, and obtain the correlation coefficients between various types of electrical indicators and various types of environmental indicators;

[0027] The calculation process of the correlation coefficients between various types of electrical indicators and various types of environmental indicators is as follows:

[0028]

[0029] Among them, q(xy) represents the correlation coefficient between electrical indicator x and environmental indicator y, D xt represents the value of electrical indicator x at the t-th moment in the historical acquisition period, D yt represents the value of environmental indicator y at the t-th moment in the historical acquisition period, DxA represents the average value of the values of electrical indicator x in the historical acquisition period, DyA represents the average value of the values of environmental indicator y in the historical acquisition period, and n represents the total number of moments in the acquisition period;

[0030] Compare the correlation coefficients between various types of electrical indicators and various types of environmental indicators with the preset normalized association threshold interval. If the correlation coefficient between an electrical indicator and an environmental indicator is within the normalized association threshold interval, establish a connection relationship between the electrical indicator and the environmental indicator, and extract the features of the electrical indicator and the environmental indicator with the established connection relationship to obtain the fluctuation coefficients corresponding to the electrical indicator and the environmental indicator with the established connection relationship;

[0031] The calculation process of obtaining the fluctuation coefficients corresponding to the electrical indicator and the environmental indicator with the established connection relationship is as follows:

[0032] Obtain the numerical time series corresponding to electrical indicators and environmental indicators before the generation of electrical fire alarm signals during the historical acquisition period. Obtain the fluctuation coefficients corresponding to the electrical indicators and environmental indicators based on the numerical time series corresponding to the electrical indicators and environmental indicators before the generation of electrical fire alarm signals during the historical acquisition period. The calculation formula for the fluctuation coefficient is as follows:

[0033]

[0034] Among them, z represents the fluctuation coefficient, and D t represents the value at the t-th moment during the historical acquisition period;

[0035] Construct a dynamic association network based on the electrical indicators and environmental indicators with established connection relationships and the fluctuation coefficients corresponding to the electrical indicators and environmental indicators with established connection relationships.

[0036] Furthermore, the process of constructing an electrical fire warning model for real-time warning and monitoring of electrical equipment and lines includes:

[0037] Construct an electrical fire warning model based on deep learning. Use the correlation coefficients between electrical indicators and environmental indicators and the fluctuation coefficients corresponding to electrical indicators and environmental indicators in the dynamic association network corresponding to various types of electrical fire alarm signals as the training set and the test set. Input the training set into the electrical fire warning model for training until the loss function is trained stably, and save the model parameters. Test the electrical fire warning model through the test set until it meets the preset requirements, and output the electrical fire warning model;

[0038] Input the electrical data and environmental data of the current acquisition period into the electrical fire warning model. The electrical fire warning model analyzes and predicts based on the input electrical data and environmental data, judges whether there is a risk of electrical fire, and issues an electrical fire warning signal in a timely manner.

[0039] Constructing an electrical fire warning model based on deep learning is a complex process that involves multiple steps such as model selection, training, validation, and testing. The following is a detailed supplementary description of this process:

[0040] Considering the characteristics of the electrical fire warning task, in this embodiment, a multi-layer perceptron (MLP) is selected as the deep learning architecture. A multi-layer perceptron is a feedforward neural network that consists of an input layer, several hidden layers, and an output layer. In an MLP, each neuron is connected to all neurons in the previous layer, and the output of the neuron is calculated through weights and activation functions. An MLP can automatically extract features and patterns in the data by learning a large amount of sample data, thereby realizing the classification and prediction of the data. In electrical fire warning, an MLP can use the correlation coefficient between electrical indicators and environmental indicators in the dynamic association network, as well as the fluctuation coefficients corresponding to electrical indicators and environmental indicators as inputs. By learning the relationship between these features and the occurrence of electrical fires, it predicts the occurrence of electrical fires, and selects the mean squared error loss function as the optimization objective. Subsequently, the prepared training set is input into the selected deep learning model for training. During the training process, the weights are continuously updated through the backpropagation algorithm, making the loss function gradually decrease until it reaches a stable state. During this period, techniques such as Early Stopping are used to avoid overfitting. In addition to the basic training process, various parameters of the model are tuned through Grid Search at the same time. The parameters include the learning rate, batch size, regularization coefficient, etc.

[0041] When the model training is completed and the parameters are adjusted, the final evaluation is carried out through the test set to obtain the evaluation results of the model. The evaluation results include classification metrics such as accuracy, recall rate, and F1 score. According to the evaluation results on the test set, it is judged whether the model meets the expected standards. If the requirements are met, the model parameters are saved and preparation for deployment is made; if not, it is necessary to return to a previous stage to re-examine issues such as data quality, model structure, or training strategy.

[0042] An electrical fire alarm management system under multi-modal information fusion includes a monitoring center, which is communicatively connected to a data acquisition module, a real-time monitoring module, a data analysis module, and a real-time warning module;

[0043] The data acquisition module is used to collect multi-modal information of electrical equipment and lines in the production area, obtain multi-modal heterogeneous data, mark the acquisition time, and set the acquisition period;

[0044] The real-time monitoring module is used to monitor the multi-modal heterogeneous data of each electrical equipment and line in real time, and judge whether to generate an electrical fire alarm signal according to the monitoring results;

[0045] The data analysis module is used to dynamically associate and analyze different types of electrical fire alarm signals generated by each electrical equipment and line in several historical acquisition periods, and obtain the dynamic association network corresponding to each type of electrical fire alarm signal;

[0046] The real-time warning module is used to construct an electrical fire warning model according to the dynamic association network corresponding to electrical fire alarm signals of various types, and conduct real-time warning monitoring on electrical equipment and lines that have not generated electrical fire alarm signals.

[0047] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0048] 1. Improve the comprehensiveness and accuracy of monitoring: By collecting multi-modal heterogeneous data such as electrical data, environmental data, and image data, compared with traditional single-type data monitoring, it can comprehensively cover various factors that may be involved in electrical fires. For example, when the current of an electrical device abnormally increases due to an internal fault, the electrical data can quickly capture this change; at the same time, if the device heats up due to a fault and causes the surrounding environmental temperature to rise, the environmental data will also be synchronously fed back; and the image data may capture obvious abnormal phenomena such as the device smoking or sparking. The three confirm each other, making the judgment of fire hazards more accurate.

[0049] Multi-dimensional threshold judgment: When judging whether a fire alarm signal is generated, not only the comparison between the numerical time series of electrical indicators and environmental indicators and the threshold range is considered, but also the comparison between the cumulative time and the preset error upper limit is introduced. This multi-dimensional judgment method avoids misjudgment caused by instantaneous data fluctuations or missed judgment caused by the failure to detect abnormal single indicators in a timely manner. For example, when the current of an electrical device occasionally and briefly exceeds the normal threshold, but the cumulative time does not exceed the error upper limit, the system will not immediately issue an alarm, effectively reducing false alarms; while when the current is in the abnormal range for a long time and the cumulative time exceeds the error upper limit, the system can issue an alarm in a timely and accurate manner.

[0050] 2. Enhance the intelligence and reliability of the model: Conduct dynamic association analysis on different types of electrical fire alarm signals, construct a dynamic association network, and be able to discover the complex hidden relationships between electrical indicators and environmental indicators. By calculating the correlation coefficient and establishing connection relationships, as well as extracting the fluctuation coefficient, the model can capture the co-variation rules between various factors. For example, it is found that when the environmental temperature rises and the humidity increases, the fluctuation coefficient of the leakage current of the electrical device increases significantly, indicating a close relationship between them. This relationship is difficult to be discovered by traditional methods, and the dynamic association network makes the model's assessment of fire risks more scientific and reliable.

[0051] Deep learning warning model: The electrical fire warning model is built based on deep learning, and the correlation coefficient and fluctuation coefficient in the dynamic association network are used for training, so that the model has strong learning and prediction capabilities. The model is trained with training sets and tested with test sets, and is continuously optimized and adjusted. It can accurately analyze the input electrical data and environmental data to determine whether there is an electrical fire risk. Compared with the traditional fixed rule warning method, the model can learn various complex fire risk patterns based on a large amount of historical data, adapt to different operating environments and equipment states, and greatly improve the accuracy and timeliness of the warning.

[0052] Real-time early warning monitoring: The constructed electrical fire early warning model is used to conduct real-time early warning monitoring of electrical equipment and lines that have not generated alarm signals, realizing the full tracking of electrical fire risks. The system can detect potential risks in advance and issue early warning signals in a timely manner, buying valuable time for taking preventive measures. For example, when there is no obvious failure of electrical equipment, but based on the changing trend of electrical data and environmental data, the early warning model predicts that there may be an electrical fire risk, it can notify relevant personnel in advance to conduct inspections and maintenance, effectively avoiding the occurrence of fires.

[0053] 3. Improve the adaptability of threshold settings: When obtaining the threshold intervals corresponding to various types of electrical indicators, the probability of electrical fires under different environmental data conditions is taken into account, and the thresholds are determined through a preset probability comparison table. This method enables the threshold to be dynamically adjusted according to the actual environment, improving the scientificity and adaptability of threshold settings. For example, in a high temperature and high humidity environment, the probability of electrical equipment fires increases. At this time, according to the probability comparison table, the threshold intervals of the corresponding electrical indicators will be adjusted more strictly to timely detect potential fire hazards; when the environmental conditions are good, the threshold intervals can be appropriately relaxed to avoid unnecessary alarms and improve the stability of system operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 This is a schematic diagram of the electrical fire alarm management method under multimodal information fusion according to an embodiment of the present application.

[0055] Figure 2 This is a schematic diagram of an electrical fire alarm management system under multimodal information fusion according to an embodiment of the present application. DETAILED DESCRIPTION

[0056] The following is a clear and complete description of the technical solutions in the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0057] As Figure 1 shown, the electrical fire alarm management method under multimodal information fusion includes the following steps:

[0058] Step s1: Collect multimodal information of electrical equipment and lines in the production area, obtain multimodal heterogeneous data, mark the collection time, and set the collection period;

[0059] Step s2: Real-time monitor the multimodal heterogeneous data of each electrical equipment and line, and judge whether to generate an electrical fire alarm signal according to the monitoring results;

[0060] Step s3: Dynamically analyze the different types of electrical fire alarm signals generated by each electrical equipment and line in several historical collection periods, and obtain the dynamic association network corresponding to each type of electrical fire alarm signal;

[0061] Step s4: Construct an electrical fire early warning model according to the dynamic association network corresponding to each type of electrical fire alarm signal, and conduct real-time early warning monitoring on electrical equipment and lines that have not generated electrical fire alarm signals.

[0062] It should be further noted that in the specific implementation process, the multimodal heterogeneous data includes electrical data (including electrical indicators such as current and voltage), environmental data (environmental indicators such as temperature, humidity, smoke particle concentration, and various gas concentrations), and image data.

[0063] It should be further noted that in the specific implementation process, the process of real-time monitoring the multimodal heterogeneous data of each electrical equipment and line and judging whether to generate an electrical fire alarm signal according to the monitoring results includes:

[0064] Extract the numerical time series corresponding to each type of electrical indicator in the electrical data of each electrical equipment and line and the numerical time series corresponding to each type of environmental indicator in the environmental data, obtain the threshold interval corresponding to each type of electrical indicator of each electrical equipment and line, and preset the threshold interval corresponding to each type of environmental indicator of each electrical equipment and line;

[0065] Compare the numerical time series corresponding to each type of electrical indicator and each type of environmental indicator with the corresponding threshold interval, obtain the cumulative time when each type of electrical indicator and each type of environmental indicator are not within the corresponding threshold interval, and compare the cumulative time with the preset error upper limit. If the cumulative time of the electrical indicator or environmental indicator is greater than the error upper limit threshold, generate the electrical fire alarm signal of the electrical indicator or environmental indicator;

[0066] If the cumulative time of the electrical indicator or environmental indicator is less than or equal to the error upper limit threshold, conduct image data monitoring of the electrical indicator or environmental indicator.

[0067] It should be further noted that in the specific implementation process, the process of monitoring image data of electrical indicators or environmental indicators includes:

[0068] Pre-collect a number of simulated training image data under different scenario states and perform scenario state annotation on the simulated training image data. The scenario states include normal operating states, abnormal phenomena such as smoking and sparks, and potential fire risk factors in the surrounding environment (such as accumulation of flammable materials, chaotic electrical circuits, etc.). Perform data format preprocessing on the simulated training image data after scenario state annotation. The data format preprocessing operations include cropping, scaling, and normalization of the images. Build an image state annotation model based on deep learning, select Faster R-CNN as the deep learning architecture. Since the goal is to distinguish the scenario states of the images, the cross-entropy loss (Binary Cross-Entropy Loss) is selected as the optimization target. Subsequently, use the simulated training image data after data format preprocessing as training data, and use the training data to train the image state annotation model to output a trained image state annotation model;

[0069] Perform data format preprocessing on the image data of each electrical device and circuit, input the image data after data format preprocessing into the image state annotation model, and obtain the scenario states of each electrical device and circuit according to the output of the image state annotation model;

[0070] If the scenario state of an electrical device or circuit is not the normal operating state, generate an electrical fire alarm signal in the scenario state according to the scenario state of the electrical device or circuit.

[0071] It should be further noted that in the specific implementation process, the process of obtaining the threshold intervals corresponding to various types of electrical indicators of each electrical device and circuit includes:

[0072] Obtain the historical collection periods when each electrical device and circuit generates electrical fire alarm signals, perform statistical analysis on the environmental data of the historical collection periods, and obtain the probabilities of electrical fires occurring for each electrical device and circuit under different environmental data conditions;

[0073] The calculation process for obtaining the probabilities of electrical fires occurring under different environmental data conditions is:

[0074]

[0075] where P (i，M) represents the probability of an electrical fire occurring for electrical device or circuit i under environmental data M, NUM i represents the total number of times electrical device or circuit i generates an electrical fire alarm signal, NUM i(M) represents the number of electrical fires that occur in electrical equipment or circuit i under environmental data M, and θ represents the conversion coefficient;

[0076] A preset probability comparison table, the probability comparison table includes threshold intervals of various types of electrical indicators corresponding to different probabilities. Obtain the environmental data of each electrical equipment and circuit in the current acquisition period, obtain the probability of electrical fires occurring in each electrical equipment and circuit under the environmental data condition, and obtain the threshold intervals corresponding to various types of electrical indicators of each electrical equipment and circuit according to the probability and the probability comparison table.

[0077] It should be further noted that in the specific implementation process, the process of dynamically associating and analyzing different types of electrical fire alarm signals generated by each electrical equipment and circuit in several historical acquisition periods to obtain the dynamic association network corresponding to each type of electrical fire alarm signal includes:

[0078] Different types of electrical fire alarm signals include electrical fire alarm signals of electrical indicators, electrical fire alarm signals of environmental indicators, and electrical fire alarm signals in a situational state. Obtain the historical acquisition periods of different types of electrical fire alarm signals generated by each electrical equipment and circuit, conduct an association analysis between various types of electrical indicators and various types of environmental indicators in the historical acquisition periods, and obtain the correlation coefficients between various types of electrical indicators and various types of environmental indicators;

[0079] The calculation process of the correlation coefficients between various types of electrical indicators and various types of environmental indicators is as follows:

[0080]

[0081] Among them, q(xy) represents the correlation coefficient between electrical indicator x and environmental indicator y, D xt represents the value of electrical indicator x at the t-th moment in the historical acquisition period, D yt represents the value of environmental indicator y at the t-th moment in the historical acquisition period, DxA represents the average value of the values of electrical indicator x in the historical acquisition period, DyA represents the average value of the values of environmental indicator y in the historical acquisition period, and n represents the total number of moments in the acquisition period;

[0082] Compare the correlation coefficients between various types of electrical indicators and various types of environmental indicators with the preset normalized association threshold interval. If the correlation coefficient between an electrical indicator and an environmental indicator is within the normalized association threshold interval, establish a connection relationship between the electrical indicator and the environmental indicator, and perform feature extraction on the electrical indicator and the environmental indicator with the established connection relationship to obtain the fluctuation coefficients corresponding to the electrical indicator and the environmental indicator with the established connection relationship;

[0083] The calculation process for obtaining the fluctuation coefficient corresponding to the electrical index and the environmental index with the established connection relationship is as follows:

[0084] Obtain the numerical time series corresponding to the electrical index and the environmental index before the generation of the electrical fire alarm signal within the historical collection period. According to the numerical time series corresponding to the electrical index and the environmental index before the generation of the electrical fire alarm signal within the historical collection period, obtain the fluctuation coefficient corresponding to the electrical index and the environmental index. The calculation formula for the fluctuation coefficient is:

[0085]

[0086] where z represents the fluctuation coefficient, and D t represents the value at the t-th moment within the historical collection period; the above formulas are all calculated by removing the dimension and taking their numerical values. The formula is obtained by software simulation of a large amount of collected data to get a formula closest to the real situation. The preset parameters and preset thresholds in the formula are set by those skilled in the art according to the actual situation or obtained through simulation of a large amount of data;

[0087] Construct a dynamic association network based on the electrical index and the environmental index with the established connection relationship and the fluctuation coefficient corresponding to the electrical index and the environmental index with the established connection relationship.

[0088] It should be further noted that in the specific implementation process, the process of constructing an electrical fire warning model and performing real-time warning monitoring on electrical equipment and lines includes:

[0089] Construct an electrical fire warning model based on deep learning. Use the correlation coefficient between the electrical index and the environmental index in the dynamic association network corresponding to each type of electrical fire alarm signal and the fluctuation coefficient corresponding to the electrical index and the environmental index as the training set and the test set. Input the training set into the electrical fire warning model for training until the loss function is trained stably, and save the model parameters. Test the electrical fire warning model through the test set until it meets the preset requirements, and output the electrical fire warning model;

[0090] Input the electrical data and environmental data of the current collection period into the electrical fire warning model. The electrical fire warning model analyzes and predicts based on the input electrical data and environmental data, judges whether there is a risk of electrical fire, and issues an electrical fire warning signal in a timely manner.

[0091] Constructing an electrical fire warning model based on deep learning is a complex process, which involves multiple steps such as model selection, training, validation, and testing. The following is a detailed supplementary description of this process:

[0092] Considering the characteristics of the electrical fire warning task, in this embodiment, a multi-layer perceptron (MLP) is selected as the deep learning architecture. A multi-layer perceptron is a feedforward neural network composed of an input layer, several hidden layers, and an output layer. In an MLP, each neuron is connected to all neurons in the previous layer, and the output of the neuron is calculated through weights and activation functions. The MLP can automatically extract features and patterns in the data by learning a large amount of sample data, thereby realizing the classification and prediction of the data. In electrical fire warning, the MLP can take the correlation coefficient between electrical indicators and environmental indicators in the dynamic association network, as well as the fluctuation coefficients corresponding to electrical indicators and environmental indicators, as inputs. By learning the relationship between these features and the occurrence of electrical fires, it predicts the occurrence of electrical fires, and selects the mean squared error loss function as the optimization objective. Subsequently, the prepared training set is input into the selected deep learning model for training. During the training process, the weights are continuously updated through the backpropagation algorithm, making the loss function gradually decrease until it reaches a stable state. During this period, techniques such as Early Stopping are used to avoid overfitting. In addition to the basic training process, the various parameters of the model are tuned through Grid Search. The parameters include the learning rate, batch size, regularization coefficient, etc.

[0093] When the model training is completed and the parameters are adjusted, the final evaluation is carried out through the test set to obtain the evaluation results of the model. The evaluation results include classification metrics such as accuracy, recall rate, and F1 score. According to the evaluation results on the test set, it is judged whether the model meets the expected standard. If the requirements are met, the model parameters are saved and preparation for deployment is made; if not ideal, it is necessary to return to a previous stage to re-examine issues such as data quality, model structure, or training strategy.

[0094] As Figure 2 shown, the electrical fire alarm management system under multi-modal information fusion includes a monitoring center, and the monitoring center is communicatively connected to a data acquisition module, a real-time monitoring module, a data analysis module, and a real-time warning module;

[0095] The data acquisition module is used to collect multi-modal information of electrical equipment and lines in the production area, obtain multi-modal heterogeneous data, mark the acquisition time, and set the acquisition period;

[0096] The real-time monitoring module is used to monitor the multi-modal heterogeneous data of each electrical equipment and line in real time, and judge whether to generate an electrical fire alarm signal according to the monitoring results;

[0097] The data analysis module is used to perform dynamic association analysis on different types of electrical fire alarm signals generated by each electrical equipment and line in a number of historical acquisition periods, and obtain the dynamic association network corresponding to each type of electrical fire alarm signal;

[0098] The real-time warning module is used to construct an electrical fire warning model according to the dynamic association network corresponding to electrical fire alarm signals of various types, and perform real-time warning monitoring on electrical equipment and lines that have not generated electrical fire alarm signals.

[0099] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. An electrical fire alarm management method based on multimodal information fusion, characterized in that: The following steps are involved: Step s1: collect multimodal information on electrical equipment and lines in the production area, obtain multimodal heterogeneous data, mark the collection time, and set the collection cycle; Step s2: Real-time monitoring of multi-modal heterogeneous data of various electrical equipment and lines, and determining whether to generate an electrical fire alarm signal based on the monitoring results; Step s3: dynamically analyzing the different types of electrical fire alarm signals generated by various electrical equipment and lines in several historical collection cycles, and obtaining a dynamic correlation network corresponding to each type of electrical fire alarm signal; Step s4: construct an electrical fire warning model based on the dynamic association network corresponding to each type of electrical fire alarm signal, and perform real-time early warning monitoring on electrical equipment and lines that do not generate electrical fire alarm signals.

2. The electrical fire alarm management method under multimodal information fusion according to claim 1 is characterized in that: Multimodal heterogeneous data include electrical data, environmental data, and image data.

3. The electrical fire alarm management method under multimodal information fusion according to claim 2 is characterized in that: The process of real-time monitoring of multi-modal heterogeneous data of various electrical equipment and lines and determining whether to generate an electrical fire alarm signal based on the monitoring results includes: Extract the numerical time series sequence corresponding to each type of electrical indicator in the electrical data of each electrical device and line and the numerical time series sequence corresponding to each type of environmental indicator in the environmental data, obtain the threshold interval corresponding to each type of electrical indicator of each electrical device and line, and preset the threshold interval corresponding to each type of environmental indicator of each electrical device and line; Compare the numerical time series corresponding to each type of electrical indicator and each type of environmental indicator with the corresponding threshold interval, obtain the cumulative time during which each type of electrical indicator and each type of environmental indicator is not within the corresponding threshold interval, compare the cumulative time with a preset upper error limit, and if the cumulative time of the electrical indicator or the environmental indicator is greater than the upper error threshold, generate an electrical fire alarm signal of the electrical indicator or the environmental indicator; If the accumulated time of the electrical index or the environmental index is less than or equal to the upper error threshold, image data monitoring of the electrical index or the environmental index is performed.

4. The electrical fire alarm management method under multimodal information fusion according to claim 3 is characterized in that: The process of performing image data monitoring of electrical indicators or environmental indicators includes: Pre-collecting a number of simulated training image data under different scene states and annotating the scene states of the simulated training image data, performing data format preprocessing on the simulated training image data after the scene state annotating, building an image state annotation model based on deep learning, using the simulated training image data after the data format preprocessing as training data, training the image state annotation model using the training data, and outputting the trained image state annotation model; Performing data format preprocessing on the image data of each electrical device and line, inputting the image data after data format preprocessing into the image state annotation model, and outputting the situational state of each electrical device and line according to the image state annotation model; If the situational state of the electrical equipment or line is not in a normal operating state, an electrical fire alarm signal in the situational state is generated according to the situational state of the electrical equipment or line.

5. The electrical fire alarm management method under multimodal information fusion according to claim 4 is characterized in that: The process of obtaining the threshold ranges corresponding to various types of electrical indicators of various electrical devices and lines includes: Obtain the historical collection period of electrical fire alarm signals generated by each electrical device and line, perform statistical analysis on the environmental data of the historical collection period, and obtain the probability of electrical fires occurring in each electrical device and line under different environmental data conditions; A probability comparison table is preset, wherein the probability comparison table includes threshold ranges of various types of electrical indicators corresponding to different probabilities, and environmental data of various electrical equipment and lines in the current collection period are obtained, and the probability of electrical fires occurring in each electrical equipment and line under the environmental data conditions is obtained. According to the probability and the probability comparison table, the threshold ranges corresponding to various types of electrical indicators of each electrical equipment and line are obtained.

6. The electrical fire alarm management method under multimodal information fusion according to claim 5 is characterized in that: The process of dynamically analyzing the different types of electrical fire alarm signals generated by various electrical equipment and lines in several historical acquisition cycles and obtaining the dynamic association network corresponding to each type of electrical fire alarm signal includes: Obtaining historical collection periods of different types of electrical fire alarm signals generated by various electrical devices and lines, performing correlation analysis between various types of electrical indicators and various types of environmental indicators in the historical collection periods, and obtaining correlation coefficients between various types of electrical indicators and various types of environmental indicators; Compare the correlation coefficient between each type of electrical indicator and each type of environmental indicator with a preset normalized correlation threshold interval; if the correlation coefficient between the electrical indicator and the environmental indicator is within the normalized correlation threshold interval, establish a connection relationship between the electrical indicator and the environmental indicator, perform feature extraction on the electrical indicator and the environmental indicator with the established connection relationship, and obtain the fluctuation coefficient corresponding to the electrical indicator and the environmental indicator with the established connection relationship; A dynamic association network is constructed according to the electrical indicators and environmental indicators that establish a connection relationship and the fluctuation coefficients corresponding to the electrical indicators and environmental indicators that establish a connection relationship.

7. The electrical fire alarm management method under multimodal information fusion according to claim 6 is characterized in that: The process of building an electrical fire warning model and performing real-time early warning monitoring of electrical equipment and lines includes: An electrical fire warning model is constructed based on deep learning, and the correlation coefficients between electrical indicators and environmental indicators in the dynamic association network corresponding to various types of electrical fire alarm signals and the fluctuation coefficients corresponding to the electrical indicators and environmental indicators are used as training sets and test sets. The training set is input into the electrical fire warning model for training until the loss function training is stable, and the model parameters are saved. The electrical fire warning model is tested by the test set until it meets the preset requirements, and the electrical fire warning model is output; The electrical data and environmental data of the current collection cycle are input into the electrical fire warning model, and it is determined whether to issue an electrical fire warning signal according to the electrical fire warning model.

8. An electrical fire alarm management system under multimodal information fusion, specifically applied to the electrical fire alarm management method under multimodal information fusion according to any one of claims 1 to 7, characterized in that: It includes a monitoring center, which is communicatively connected with a data acquisition module, a real-time monitoring module, a data analysis module and a real-time warning module; The data acquisition module is used to collect multimodal information on electrical equipment and lines in the production area, obtain multimodal heterogeneous data, mark the acquisition time, and set the acquisition cycle; The real-time monitoring module is used to monitor the multi-modal heterogeneous data of various electrical equipment and lines in real time, and determine whether to generate an electrical fire alarm signal based on the monitoring results; The data analysis module is used to dynamically correlate and analyze different types of electrical fire alarm signals generated by various electrical equipment and lines in several historical collection cycles, and obtain the dynamic correlation network corresponding to each type of electrical fire alarm signal; The real-time warning module is used to construct an electrical fire warning model based on the dynamic association network corresponding to various types of electrical fire alarm signals, and to perform real-time warning monitoring on electrical equipment and lines that do not generate electrical fire alarm signals.

Images