Underground space electric vehicle fire detection early warning prevention and control system and method

Abstract

The invention discloses an underground space electric vehicle fire detection early warning prevention and control system and method, and relates to the field of fire detection early warning, and the technical scheme is characterized in that the method comprises the steps: collecting the charging data of an electric vehicle, carrying out the region segmentation of an underground space, and collecting the comprehensive data of the underground space through a multi-sensor fusion technology; carrying out trend analysis on the charging data and the comprehensive data in the period time by using a segmented regression method, carrying out early warning marking, and carrying out resampling on the marked electric cars or areas to obtain a warning data set; constructing a fire early warning model, inputting the warning data set into the trained fire early warning model, and obtaining a fire analysis scheme of the underground space; and providing a visual tool to check a fire analysis scheme, marking and displaying a fire occurrence area or an electric vehicle, and performing interface interaction to realize efficient detection, early warning, prevention and control of the electric vehicle in the underground space.

Description

Technical Field

[0001] The present invention relates to the field of fire detection and early warning, and more specifically, to an underground space electric vehicle fire detection, early warning and prevention system and method. Background Art

[0002] In underground spaces, the design and implementation of electric vehicle fire detection and warning systems face many difficulties and problems. These difficulties are mainly attributed to the special environment of underground spaces, the fire characteristics of electric vehicles, and the limitations of existing technologies.

[0003] Specifically, electric vehicle fires can be divided into charging and non-charging states. Especially for electric vehicles in charging state, the probability of fire is higher. For electric vehicles in charging state, the fire may originate from inside the vehicle, especially the battery system, which may be difficult to identify through external detection means in the early stage of the fire, resulting in detection delays and difficulty in detecting the fire in the early stage. For electric vehicles in non-charging state, due to the relatively closed underground space and poor air circulation, the smoke diffusion pattern is different from that in open space. When a fire occurs, smoke is easy to accumulate in the local space, resulting in the detector being unable to perceive the fire signal in a timely and accurate manner, and may not be able to effectively detect. In addition, due to the complex environment of the underground space, factors such as temperature, humidity, and dust may interfere with the detector, increasing the possibility of false alarms and affecting the effectiveness of detection.

[0004] Therefore, based on the above problems, it is necessary to design the detection, early warning and prevention of electric vehicle fires in underground spaces. Summary of the invention

[0005] In view of the problems existing in the prior art, the purpose of the present invention is to provide a system and method for detecting, warning and preventing fires in underground electric vehicles, so as to realize efficient fire detection, warning and prevention of underground electric vehicles.

[0006] To achieve the above-mentioned purpose, the present invention provides the following technical solution: the underground space electric vehicle fire detection, early warning and prevention system comprises: Data multi-source acquisition module: collects the charging data of trams, divides the underground space into regions, and collects comprehensive data of underground space through multi-sensor fusion technology; Data processing module: Use the segmented regression method to perform trend analysis on the charging data and comprehensive data within the cycle time, make early warning marks, resample the marked trams or areas, and obtain the warning data set; Model analysis module: Build a fire warning model, input the warning data set into the trained fire warning model, and obtain the fire analysis plan for the underground space; Interface interaction module: provides visualization tools to view the fire analysis plan, mark and display the fire area or tram, and perform interface interaction.

[0007] Preferably, the charging data includes current data of each electric vehicle, and the method of charging data includes: Arrange current sensors on charging piles, connect data acquisition equipment to current sensors, convert analog current signals into digital signals through ADC, and obtain charging data of each electric vehicle; The comprehensive data of underground space includes temperature values, smoke concentration, gas data and flame data of each area; Multi-sensor fusion technology involves deploying nodes in underground spaces. The nodes are connected via wireless or wired networks to form a local monitoring network. Each node monitors the data of temperature sensors, smoke sensors, gas sensors, flame sensors and current sensors in parallel, and data is shared between nodes.

[0008] Preferably, the method for segmenting the underground space into regions comprises: Step A1: Extract the 3D area map of the underground space, randomly select a base point in the underground space, and calculate the environmental assessment index at the base point According to the horizontal and vertical directions of the three-dimensional area map, the base point is used as the origin to extend in both directions horizontally and vertically. When the stop condition is met When the two-way extension stops, each extension point is connected to form a base rectangular area, and the base rectangular area is marked on the three-dimensional area map to divide the underground space; Step A2: Randomly select a new base point outside the base rectangular area, calculate the environmental assessment index at the new base point, and extend in both directions from the new base point. When one of the extension directions is perpendicular to the edge of the base rectangular area, a vertical point is formed. When the environmental assessment index between the new base point and the vertical point meets the pause condition, the bidirectional extension stops. Connect the extension points to form a new base rectangular area, and mark the new base rectangular area on the three-dimensional area map to divide the underground space again. Step A3: repeat step A2, and when the remaining area cannot be segmented, segment it into independent areas according to the area, and stop when the three-dimensional area map is completely segmented; in, , and is the adjustment factor, Indicates the air velocity, Indicates the dust density, Indicates the temperature value, Indicates humidity value. Indicates the environmental assessment index at the base point , represents the limit threshold, Indicates the environmental assessment index at the extension point .

[0009] Preferably, the method of using a segmented regression method to perform trend analysis on charging data and comprehensive data within a cycle time and to perform early warning marking includes: For each data category of charging data and comprehensive data, the data values ​​within the cycle time are arranged in chronological order to obtain a time series data set for each data category. A window is used to slide on the time series data set to determine the breakpoints, and the time series data set is divided into data intervals by the breakpoints. For each data interval, the curve expression of each data interval is obtained by curve fitting, the slope of each curve expression and the expression value at the breakpoint are extracted, and the slopes and expression values ​​of adjacent curve expressions are compared in order of the data intervals. If the slope and the expression value are the same, it means that the curve expression is continuous at the breakpoint and there is no mutation. Otherwise, it means that the curve expression is discontinuous at the breakpoint and there is a mutation. The tram logo or area corresponding to the mutation is extracted for early warning marking.

[0010] Preferably, the method of using a window to slide on a time series data set to determine breakpoints and dividing the time series data set into data intervals by the breakpoints includes: Step B1: Define a window of fixed length, initialize the window position at the first data value, and calculate the statistical characteristics within the window ; in, and represents the weight coefficient, represents the variance of the data values ​​within the window, Represents the average value of the data in the window; Step B2: Move the window to the next data value, update the data value in the window and calculate the statistical features, and obtain the absolute difference between the statistical features of two adjacent windows. , then the first data value of the current window is determined to be a breakpoint, otherwise, it is not a breakpoint; in, Represents the statistical characteristics within the current window, Represents the statistical characteristics in the previous window, Represents the cutoff threshold, which can be set through experimental data analysis or experience; Step B3: Repeat step B2 until the window reaches the data value of the last time series data set, collect all breakpoints, divide the time series data set into data intervals through the breakpoints, and the last interval contains all data values ​​after the last breakpoint until the last data value.

[0011] Preferably, the method of resampling the marked tram or area to obtain the warning data set includes: Identify the tram signs or areas with warning marks and adjust the sampling frequency For electric vehicles in a charging state, the charging data of the electric vehicle and the comprehensive data of the area where the electric vehicle is located are collected at the same time. For electric vehicles in a non-charging state, the comprehensive data of the area is collected; in, represents the sampling frequency before adjustment, represents the adjusted sampling frequency, Indicates The data category is The slope over the data interval, Indicates the number of data categories, Indicates the number of data intervals; Collect all charging data and comprehensive data after sampling frequency adjustment to form a warning data set.

[0012] Preferably, the method for constructing the fire warning model includes: Set up shared layers and task-specific headers; The shared layer is based on a hybrid design of CNN and LSTM layers. The CNN layer first extracts the local time series features of the model input, and passes the output of the CNN layer to the LSTM layer to further extract the global dependency features of the time series. The input of the task-specific header is set as the output of the shared layer, and the task-specific header includes a charging state detection header, a fire prediction header, a fire type classification header, a fire spread range prediction header, and a warning level assessment header; The charging status detection head sets a binary classification task, including charging or not charging, to determine whether the electric vehicle is in a charging state. It includes a fully connected layer and an output layer, uses a Sigmoid activation function, and sets the loss function to a binary cross entropy loss. The fire prediction head sets the regression task, including the fully connected layer and the output layer, uses the ReLU activation function to predict the probability of fire occurrence, and sets the loss function to the mean square error; The fire type classification head is used to detect the factors that cause fires. It includes a fully connected layer and an output layer. It uses the ReLU activation function and sets the loss function to a multi-class cross entropy loss. The multi-class cross entropy loss is used to calculate the error between the predicted probability of the influencing factor and the true label. The fire spread prediction head is used to predict the spread or diffusion speed of the fire. It includes a fully connected layer and an output layer. It uses a linear activation function and sets the loss function to mean square error, which is used to measure the difference between the predicted fire spread or diffusion speed and the true value. The alert level assessment head is used to output the fire alert level. It includes a fully connected layer and an output layer. It uses the Softmax activation function and sets the loss function to a multi-class cross entropy loss to calculate the error between the predicted alert level and the true label. Set the total loss function to: ; in, , , , and represents the weight coefficient, represents the loss function of the charging state detection head, represents the loss function of the fire prediction head, represents the loss function of the fire type classification head, represents the loss function of the fire spread prediction head, Represents the loss function of the vigilance assessment head.

[0013] Preferably, the fire analysis scheme includes the charging state of the electric vehicle, the probability of fire occurrence, the influencing factors of fire occurrence, the scope or spread speed of the fire and the fire warning level.

[0014] Preferably, the method of inputting the warning data set into a trained fire warning model to obtain a fire analysis plan for the underground space includes: Step D1: Set a sample set, which includes T_U groups of samples. Each group of samples includes a warning data set and a corresponding fire analysis plan. The sample set is divided into S_J batches, which are input into the constructed fire warning model in sequence for training; Step D2: Use an adaptive learning rate strategy and an Adam or SGD optimizer to output the loss function of each task-specific head in turn. Stop training when the total loss function value no longer changes, and obtain a trained fire warning model. Step D3: Input the warning data set into the trained fire warning model to obtain the fire analysis plan for the current underground space.

[0015] A method for detecting, warning and preventing underground electric vehicle fires, comprising: Step S1: segment the underground space into regions and collect the charging data of the tram and the comprehensive data of the underground space; Step S2: Use the segmented regression method to perform trend analysis on the charging data and comprehensive data within the cycle time, and make early warning marks; Step S3: resample the marked trams or areas to obtain a warning data set; Step S4: constructing a fire warning model; Step S5: Input the current warning data set into the trained fire warning model, obtain the fire analysis plan for the underground space, and display it on the interface.

[0016] Compared with the prior art, the present invention has the following beneficial effects: Comprehensive data collection and analysis are achieved through multi-source data collection. This multi-dimensional perception can more accurately capture abnormal changes before a fire occurs, improving the timeliness and accuracy of fire warnings. The underground space is divided into multiple areas through the segmentation method of the three-dimensional area map, and each area is monitored independently. This segmentation method not only improves the precision of monitoring, but also enables the system to accurately locate the problem area when risks are discovered, making it easier to take countermeasures quickly.

[0017] The segmented regression method is used to analyze the trend of periodic data. Combined with the sliding window to determine the mutation point of the data, it can effectively detect potential fire risks and avoid the problems of underreporting and false alarms in traditional methods. After detecting an anomaly, the marked trams or areas are resampled and the data collection frequency is adjusted. More intensive high-precision data is obtained to further improve the accuracy of early warning.

[0018] By combining the hybrid design of CNN and LSTM, the system can simultaneously extract local and global time series features, capture complex data changes before a fire occurs, and further improve the accuracy of fire warnings. The multi-task design of the task-specific head (charging status detection, fire prediction, fire type classification, fire spread prediction, and alert level assessment) enables the system to simultaneously perform multiple aspects of fire analysis, not only to determine whether there is a fire, but also to analyze the source, spread, and severity of the fire, providing more comprehensive decision support.

[0019] Through trend analysis, breakpoint detection and curve fitting of different data categories and regions, the system can better distinguish between normal fluctuations and abnormal changes in data, reducing the problem of false alarms and missed alarms in fire warnings. Accurately identifying abnormal changes, combined with intelligent analysis of fire warning models, ensures that fire warnings are more reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 The present invention proposes a structural schematic diagram of an underground space electric vehicle fire detection, early warning and prevention system; Figure 2 A schematic diagram of a method for detecting, warning and preventing underground electric vehicle fires in the present invention; Figure 3 It is a structural schematic diagram of the fire warning model in the present invention. DETAILED DESCRIPTION

[0021] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0022] Embodiment 1 Reference Figure 1 , Figure 2 and Figure 3 Embodiment 1 further illustrates an underground space electric vehicle fire detection, early warning and prevention system proposed by the present invention.

[0023] Once the battery "thermal runaway" phenomenon occurs, it is often difficult to accurately predict it through conventional early warning methods. Internal battery failures may cause fires to escalate rapidly in a short period of time, and existing monitoring systems are difficult to effectively predict. Therefore, it is necessary to design the underground space electric vehicle fire detection, early warning and prevention process, the specific contents are as follows: Data multi-source acquisition module: collects the charging data of trams, divides the underground space into regions, and collects comprehensive data of underground space through multi-sensor fusion technology; The charging data includes the current data of each electric vehicle, and the charging data method includes: Arrange current sensors on charging piles, connect data acquisition equipment to current sensors, convert analog current signals into digital signals through ADC, and obtain charging data of each electric vehicle; The comprehensive data of underground space includes temperature values, smoke concentration, gas data and flame data of each area; Multi-sensor fusion technology involves deploying nodes in underground spaces. The nodes are connected via wireless or wired networks to form a local monitoring network. Each node monitors the data of temperature sensors, smoke sensors, gas sensors, flame sensors and current sensors in parallel, and data is shared between nodes.

[0024] Select a suitable current sensor, such as a Hall effect sensor or a shunt, based on the current range of the charging system. Install the current sensor on the charging line of each charging pile to measure the real-time current during the charging process.

[0025] Design a data acquisition device, including a microcontroller, an analog-to-digital converter (ADC), a wireless communication module, etc. Connect the data acquisition device to the current sensor and convert the analog current signal into a digital signal through the ADC. Deploy the data acquisition device to each charging pile and work with the current sensor. Transmit the collected current data to the data center in real time through a wireless communication module (such as Wi-Fi, 4G, etc.). Set up a receiving server in the data center to receive the current data from each charging pile and store the data in a database.

[0026] In order to distinguish different trams, an identification system, such as RFID or License Plate Recognition (LPR), can be installed to identify the unique identity of each tram.

[0027] In addition, the monitoring data of all charging piles can be centrally managed to establish a unified monitoring platform. Real-time monitoring of all charging vehicles can be achieved, and the charging status and current fluctuation of each vehicle can be viewed. Data query and statistical functions are provided to facilitate historical data analysis and trend prediction.

[0028] In underground spaces, temperature sensors are used to obtain temperature values ​​and monitor ambient temperature and temperature changes around electric vehicles, especially when charging, battery system failures may cause the temperature to rise sharply. Smoke sensors (photoelectric or ion) are used to monitor changes in smoke concentration in the air and can detect smoke produced when the battery catches fire. Gas sensors are used to collect gas data and detect characteristic gases (such as carbon monoxide, carbon dioxide, hydrogen, etc.) released when electric vehicle batteries are thermally runaway. Lithium batteries release specific gases when they catch fire, and gas sensors can detect these signs at a very early stage. Photoelectric flame sensors work based on the photoelectric effect. When there is a flame, the flame will emit visible light signals. The sensor will detect these signals and convert them into electrical signal outputs. By analyzing and processing the electrical signals, it can be determined whether there is a flame.

[0029] These multiple sensors are deployed in parallel at base points in various areas and transmitted together to the data center for synchronous fusion analysis, taking advantage of the complementary advantages of different sensors to reduce the possibility of false alarms or missed alarms from a single sensor.

[0030] Methods for segmenting underground space include: Step A1: Extract the 3D area map of the underground space. The 3D area map refers to the actual distribution map of the buildings and locations in the underground space. The map model can be constructed through 3D software or obtained based on the building map of the underground space. A base point is randomly selected in the underground space and the environmental assessment index at the base point is calculated. According to the horizontal and vertical directions of the three-dimensional area map, the base point is used as the origin to extend in both directions horizontally and vertically. When the stop condition is met When the two-way extension stops, each extension point is connected to form a base rectangular area, and the base rectangular area is marked on the three-dimensional area map to divide the underground space; Step A2: Randomly select a new base point outside the base rectangular area, calculate the environmental assessment index at the new base point, and extend in both directions from the new base point. When one of the extension directions is perpendicular to the edge of the base rectangular area, a vertical point is formed. When the environmental assessment index between the new base point and the vertical point meets the pause condition, the bidirectional extension stops. Connect the extension points to form a new base rectangular area, and mark the new base rectangular area on the three-dimensional area map to divide the underground space again. Step A3: Repeat step A2. When the remaining area cannot be segmented, segment it into independent areas according to the area until the three-dimensional area map is completely segmented. in, , and is the adjustment factor, Indicates the air velocity, Indicates the dust density, Indicates the temperature value, Indicates humidity value. Indicates the environmental assessment index at the base point , represents the limit threshold, Indicates the environmental assessment index at the extension point ; Environmental Assessment Index at Base Point It is calculated by collecting the air velocity, dust density, temperature and humidity values ​​at the base point. Similarly, the environmental assessment index at the extension point It is calculated by collecting the air velocity, dust density, temperature and humidity values ​​at the extension point. The adjustment coefficient and limit threshold can be set through historical data or experimental data analysis.

[0031] The air flow rate can be obtained through the air sensor, the dust density can be collected through the dust sensor, and the temperature and humidity value distribution can be collected through the temperature sensor and the humidity sensor.

[0032] Data processing module: Use the segmented regression method to perform trend analysis on the charging data and comprehensive data within the cycle time, and make early warning marks, resample the marked trams or areas, and obtain the warning data set.

[0033] The method of using the segmented regression method to analyze the trend of charging data and comprehensive data within the cycle time and to make early warning marks includes: For each data category of charging data and comprehensive data, the data values ​​within the cycle time are arranged in chronological order to obtain a time series data set for each data category. A window is used to slide on the time series data set to determine the breakpoints, and the time series data set is divided into data intervals by the breakpoints. For each data interval, the curve expression of each data interval is obtained by curve fitting, the slope of each curve expression and the expression value at the breakpoint are extracted, and the slopes and expression values ​​of adjacent curve expressions are compared in order of the data intervals. If the slope and the expression value are the same, it means that the curve expression is continuous at the breakpoint and there is no mutation. Otherwise, it means that the curve expression is discontinuous at the breakpoint and there is a mutation. The tram logo or area corresponding to the mutation is extracted for early warning marking.

[0034] Assuming that there is one breakpoint, the time series data set is divided into two data intervals by the breakpoint. The curve fitting curve in the first data interval is expressed as , the curve fitting curve in the second data interval is expressed as ,in, and is the residual term, which is used to capture the part that the model cannot explain. and is the intercept term, , , and Represents the slope, and then a curve is fitted to the data values ​​in each data interval.

[0035] Methods for using a window to slide on a time series data set to determine breakpoints and dividing the time series data set into data intervals by breakpoints include: Step B1: Define a window of fixed length, initialize the window position at the first data value, and calculate the statistical characteristics within the window ; The fixed length of the window can be set based on operational experience or experimental data analysis.

[0036] in, and represents the weight coefficient, represents the variance of the data values ​​within the window, Represents the average value of the data in the window; Step B2: Move the window to the next data value, update the data value in the window and calculate the statistical features, and obtain the absolute difference between the statistical features of two adjacent windows. , then the first data value of the current window is determined to be a breakpoint, otherwise, it is not a breakpoint; in, Represents the statistical characteristics within the current window, Represents the statistical characteristics in the previous window, Represents the cutoff threshold, which can be set through experimental data analysis or experience; Step B3: Repeat step B2 until the window reaches the data value of the last time series data set, collect all breakpoints, divide the time series data set into data intervals through the breakpoints, and the last interval contains all data values ​​after the last breakpoint until the last data value.

[0037] In time series data analysis, the sliding window method is used to identify breakpoints (i.e., points where data features change significantly), and based on this, the entire data set is divided into sub-intervals with different characteristics. Dividing the time series data set into a series of continuous, non-overlapping data intervals can help us better understand the changes in the internal structure of the data set.

[0038] Methods for resampling the marked trams or areas to obtain warning data sets include: Identify the tram signs or areas with warning marks and adjust the sampling frequency For electric vehicles in a charging state, the charging data of the electric vehicle and the comprehensive data of the area where the electric vehicle is located are collected at the same time. For electric vehicles in a non-charging state, the comprehensive data of the area is collected; in, represents the sampling frequency before adjustment, represents the adjusted sampling frequency, Indicates The data category is The slope over the data interval, Indicates the number of data categories, Indicates the number of data intervals; Collect all charging data and comprehensive data after sampling frequency adjustment to form a warning data set.

[0039] Model analysis module: Build a fire warning model, input the warning data set into the trained fire warning model, and obtain the fire analysis plan for the underground space.

[0040] When building a multi-task learning architecture based on a neural network, the core idea is to use a shared feature extraction layer to process the input data and design an independent task output layer (i.e., task-specific head) for each specific task. The advantage of this architecture is that multiple tasks can share the same underlying features, thereby effectively utilizing the correlation of the data and improving the overall performance of the model.

[0041] The construction method of the fire warning model includes: Set up shared layers and task-specific headers; The shared layer is based on a hybrid design of CNN and LSTM layers. The CNN layer first extracts the local time series features of the model input, passes the output of the CNN layer to the LSTM layer, and further extracts the global dependency features of the time series. This hybrid architecture helps capture local details of the data and long-term trend changes.

[0042] In the multi-task learning architecture, the task-specific head is responsible for processing the output of each specific task. After receiving the features from the shared layer, the task-specific head further processes them for the specific task. The output layer (head) of each task uses different network structures and loss functions according to the nature of the task. The design of each task-specific head is described in detail below.

[0043] The input of the task-specific header is set as the output of the shared layer, and the task-specific header includes a charging state detection header, a fire prediction header, a fire type classification header, a fire spread range prediction header, and a warning level assessment header; The charging status detection head sets a binary classification task, including charging or not charging, to determine whether the electric vehicle is in a charging state. It includes a fully connected layer and an output layer, uses a Sigmoid activation function, and sets the loss function to a binary cross entropy loss. The fully connected layer further maps the features output by the shared layer to a low-dimensional space. The ReLU activation function is used for nonlinear transformation. The output layer uses the Sigmoid activation function to output a probability between 0 and 1, indicating the probability that the vehicle is in a charging state. The binary cross-entropy loss is used to calculate the error between the predicted value and the true label. It can be expressed as ,in, represents the actual label, usually 0 or 1, indicating whether a certain event has occurred (for example, whether the charging state is "charging"), Represents the probability predicted by the model, which indicates the likelihood of an event occurring (e.g., the probability that a device is charging).

[0044] The fire prediction head sets the regression task, including the fully connected layer and the output layer, uses the ReLU activation function to predict the probability of fire occurrence, and sets the loss function to the mean square error; The fire prediction head is used to predict the probability of fire occurrence. According to the charging status (charged / uncharged), the model will predict the probability of fire occurrence separately, which is a regression task.

[0045] The fully connected layer maps the shared features to a low-dimensional space. ReLU is used for nonlinear transformation. The output layer uses a linear activation function to output the probability of fire occurrence, ranging from 0 to 1. The mean square error can be expressed as ,in, Represents the total number of samples, that is, the number of fire events involved in the prediction. The model predicts the The probability of fire occurrence of samples is in the range of [0,1]. Indicates The actual fire occurrence of a sample, usually 0 (not occurred) or 1 (occurred).

[0046] The fire type classification head is used to detect factors that lead to fires, such as battery overheating, abnormal current fluctuations, abnormal temperature rise, or excessive dust accumulation. It includes a fully connected layer and an output layer, uses the ReLU activation function, and sets the loss function to a multi-class cross entropy loss, which is used to calculate the error between the predicted probability of the influencing factor and the true label; The multi-class cross entropy loss is expressed as ,in, Represents the actual abnormal category label, usually in one-hot encoding form. For example, for the case of three categories, if the actual category is the first category, then =[1,0,0], The model predicts the The probability of a category indicates the possibility that the sample belongs to that category, usually in the range of [0,1].

[0047] The fire spread prediction head is used to predict the spread (such as area or distance) or diffusion speed of the fire. It includes a fully connected layer and an output layer. It uses a linear activation function and sets the loss function to mean square error, which is used to measure the difference between the predicted fire spread or diffusion speed and the true value. The alert level assessment head is used to output the fire alert level (such as low, medium, and high alert). It includes a fully connected layer and an output layer. It uses the Softmax activation function and sets the loss function to a multi-class cross entropy loss to calculate the error between the predicted alert level and the true label. The total loss function is set to ,in, , , , and represents the weight coefficient, represents the loss function of the charging state detection head, represents the loss function of the fire prediction head, represents the loss function of the fire type classification head, represents the loss function of the fire spread prediction head, Represents the loss function of the vigilance assessment head.

[0048] The core advantage of adopting a multi-task learning architecture is that the model can share the underlying features of different tasks, thereby improving learning efficiency and generalization ability.

[0049] The fire analysis plan includes the charging status of the tram, the probability of fire occurrence, the factors affecting the occurrence of fire, the scope or spread speed of the fire and the fire warning level.

[0050] Input the warning data set into the trained fire warning model to obtain the fire analysis plan for underground space. The specific methods include: Step D1: Set a sample set, which includes T_U groups of samples. Each group of samples includes a warning data set and a corresponding fire analysis plan. The sample set is divided into S_J batches, which are input into the constructed fire warning model in sequence for training; Step D2: Use an adaptive learning rate strategy and an Adam or SGD optimizer to output the loss function of each task-specific head in turn. Stop training when the total loss function value no longer changes, and obtain a trained fire warning model. Step D3: Input the warning data set into the trained fire warning model to obtain the fire analysis plan for the current underground space.

[0051] Interface interaction module: provides visualization tools to view the fire analysis plan, mark and display the fire area or tram, and perform interface interaction.

[0052] The visualization tool is used to provide an interface for interaction, and the interface is provided by a computer or an intelligent display device.

[0053] Electric vehicle fires, especially battery combustion, are fierce and difficult to extinguish. Lithium batteries generate uncontrollable high temperatures during combustion, which may cause explosions or secondary fires. Conventional fire-fighting measures (such as water or foam) have limited control effects on battery fires and may even introduce new dangers. Therefore, special fire-fighting equipment is required in underground spaces for fire prevention and control.

[0054] Embodiment 2 Reference Figure 1 and Figure 2 , Example 2 further illustrates an underground space electric vehicle fire detection, early warning and prevention system proposed by the present invention.

[0055] A method for detecting, warning and preventing underground electric vehicle fires, comprising the following steps: Step S1: segment the underground space into regions and collect the charging data of the tram and the comprehensive data of the underground space; Step S2: Use the segmented regression method to perform trend analysis on the charging data and comprehensive data within the cycle time, and make early warning marks; Step S3: resample the marked trams or areas to obtain a warning data set; Step S4: constructing a fire warning model; Step S5: Input the current warning data set into the trained fire warning model, obtain the fire analysis plan for the underground space, and display it on the interface.

[0056] An underground space electric vehicle fire detection, early warning and prevention system, applied to an underground space electric vehicle fire detection, early warning and prevention method, includes: Data multi-source acquisition module: collects the charging data of trams, divides the underground space into regions, and collects comprehensive data of underground space through multi-sensor fusion technology; Data processing module: Use the segmented regression method to perform trend analysis on the charging data and comprehensive data within the cycle time, make early warning marks, resample the marked trams or areas, and obtain the warning data set; Model analysis module: Build a fire warning model, input the warning data set into the trained fire warning model, and obtain the fire analysis plan for the underground space; Interface interaction module: provides visualization tools to view the fire analysis plan, mark and display the fire area or tram, and perform interface interaction; The modules are connected to each other via wired and / or wireless means.

[0057] In addition, according to the implementation of the present application, the process described in the attached figure of a system for detecting, warning and controlling fires in underground electric vehicles can be implemented as a computer software program. For example, the present application provides a non-transitory machine-readable storage medium, which stores machine-readable instructions, and the machine-readable instructions can be executed by a processor to execute instructions corresponding to the method steps provided in the present application. Of course, the architecture shown in the attached figure of a system for detecting, warning and controlling fires in underground electric vehicles is only exemplary. When implementing different devices, adaptive selection or adjustment can be made according to actual needs.

[0058] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0059] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. An underground space electric vehicle fire detection, early warning and prevention system, characterized in that: include: Data multi-source acquisition module: collects the charging data of trams, divides the underground space into regions, and collects comprehensive data of underground space through multi-sensor fusion technology; Data processing module: Use the segmented regression method to perform trend analysis on the charging data and comprehensive data within the cycle time, make early warning marks, resample the marked trams or areas, and obtain the warning data set; Model analysis module: Build a fire warning model, input the warning data set into the trained fire warning model, and obtain the fire analysis plan for the underground space; Interface interaction module: provides visualization tools to view the fire analysis plan, mark and display the fire area or tram, and perform interface interaction.

2. The underground space electric vehicle fire detection, early warning and prevention system according to claim 1 is characterized in that: The charging data includes current data of each electric vehicle, and the method of charging data includes: Arrange current sensors on charging piles, connect data acquisition equipment to current sensors, convert analog current signals into digital signals through ADC, and obtain charging data of each electric vehicle; The comprehensive data of underground space includes temperature values, smoke concentration, gas data and flame data of each area; Multi-sensor fusion technology involves deploying nodes in underground spaces. The nodes are connected via wireless or wired networks to form a local monitoring network. Each node monitors the data of temperature sensors, smoke sensors, gas sensors, flame sensors and current sensors in parallel, and data is shared between nodes.

3. The underground space electric vehicle fire detection, early warning and prevention system according to claim 2 is characterized in that: The method for segmenting the underground space into regions comprises: Step A1: Extract the 3D area map of the underground space, randomly select a base point in the underground space, and calculate the environmental assessment index at the base point According to the horizontal and vertical directions of the three-dimensional area map, the base point is used as the origin to extend in both directions horizontally and vertically. When the stop condition is met When the two-way extension stops, each extension point is connected to form a base rectangular area, and the base rectangular area is marked on the three-dimensional area map to divide the underground space; Step A2: Randomly select a new base point outside the base rectangular area, calculate the environmental assessment index at the new base point, extend in both directions from the new base point, form a vertical point when one of the extension directions is perpendicular to the edge of the base rectangular area, and stop extending in both directions when the environmental assessment index between the new base point and the vertical point meets the pause condition, connect each extension point to form a new base rectangular area, and mark the new base rectangular area on the three-dimensional area map, and divide the underground space again; Step A3: repeat step A2, and when the remaining area cannot be segmented, segment it into independent areas according to the area, and stop when the three-dimensional area map is completely segmented; in, , and is the adjustment factor, Indicates the air velocity, Indicates the dust density, Indicates the temperature value, Indicates humidity value. Indicates the environmental assessment index at the base point , represents the limit threshold, Indicates the environmental assessment index at the extension point .

4. The underground space electric vehicle fire detection, early warning and prevention system according to claim 3 is characterized in that: The method of using the segmented regression method to perform trend analysis on the charging data and comprehensive data within the cycle time and to perform early warning marking includes: For each data category of charging data and comprehensive data, the data values ​​within the cycle time are arranged in chronological order to obtain a time series data set for each data category. A window is used to slide on the time series data set to determine the breakpoints, and the time series data set is divided into data intervals by the breakpoints. For each data interval, the curve expression of each data interval is obtained by curve fitting, the slope of each curve expression and the expression value at the breakpoint are extracted, and the slopes and expression values ​​of adjacent curve expressions are compared in order of the data intervals. If the slope and the expression value are the same, it means that the curve expression is continuous at the breakpoint and there is no mutation. Otherwise, it means that the curve expression is discontinuous at the breakpoint and there is a mutation. The tram logo or area corresponding to the mutation is extracted for early warning marking.

5. The underground space electric vehicle fire detection, early warning and prevention system according to claim 4 is characterized in that: The method of using a window to slide on a time series data set to determine breakpoints and dividing the time series data set into data intervals by the breakpoints includes: Step B1: Define a window of fixed length, initialize the window position at the first data value, and calculate the statistical characteristics within the window ; in, and represents the weight coefficient, represents the variance of the data values ​​within the window, Represents the average value of the data in the window; Step B2: Move the window to the next data value, update the data value in the window and calculate the statistical features, and obtain the absolute difference between the statistical features of two adjacent windows. , then the first data value of the current window is determined to be a breakpoint, otherwise, it is not a breakpoint; in, Represents the statistical characteristics within the current window, Represents the statistical characteristics in the previous window, Represents the cutoff threshold, which is set through experimental data analysis or experience; Step B3: Repeat step B2 until the window reaches the data value of the last time series data set, collect all breakpoints, divide the time series data set into data intervals by the breakpoints, and the last interval contains all data values ​​after the last breakpoint until the last data value.

6. The underground space electric vehicle fire detection, early warning and prevention system according to claim 5 is characterized in that: The method of resampling the marked tram or area to obtain the warning data set includes: Identify the tram signs or areas with warning marks and adjust the sampling frequency For electric vehicles in a charging state, the charging data of the electric vehicle and the comprehensive data of the area where the electric vehicle is located are collected at the same time. For electric vehicles in a non-charging state, the comprehensive data of the area is collected; in, represents the sampling frequency before adjustment, represents the adjusted sampling frequency, Indicates The data category is The slope over the data interval, Indicates the number of data categories, Indicates the number of data intervals; Collect all charging data and comprehensive data after sampling frequency adjustment to form a warning data set.

7. The underground space electric vehicle fire detection, early warning and prevention system according to claim 6 is characterized in that: The method for constructing the fire warning model includes: Set up shared layers and task-specific headers; The shared layer is based on a hybrid design of CNN and LSTM layers. The CNN layer first extracts the local time series features of the model input, and passes the output of the CNN layer to the LSTM layer to further extract the global dependency features of the time series. The input of the task-specific header is set as the output of the shared layer, and the task-specific header includes a charging state detection header, a fire prediction header, a fire type classification header, a fire spread range prediction header, and a warning level assessment header; The charging status detection head sets a binary classification task, including charging or not charging, to determine whether the electric vehicle is in a charging state. It includes a fully connected layer and an output layer, uses a Sigmoid activation function, and sets the loss function to a binary cross entropy loss. The fire prediction head sets the regression task, including the fully connected layer and the output layer, uses the ReLU activation function to predict the probability of fire occurrence, and sets the loss function to the mean square error; The fire type classification head is used to detect the factors that cause fires. It includes a fully connected layer and an output layer. It uses the ReLU activation function and sets the loss function to a multi-class cross entropy loss. The multi-class cross entropy loss is used to calculate the error between the predicted probability of the influencing factor and the true label. The fire spread prediction head is used to predict the spread or diffusion speed of the fire. It includes a fully connected layer and an output layer. It uses a linear activation function and sets the loss function to mean square error, which is used to measure the difference between the predicted fire spread or diffusion speed and the true value. The alert level assessment head is used to output the fire alert level. It includes a fully connected layer and an output layer. It uses the Softmax activation function and sets the loss function to a multi-class cross entropy loss to calculate the error between the predicted alert level and the true label. Set the total loss function to: ； in, , , , and represents the weight coefficient, represents the loss function of the charging state detection head, represents the loss function of the fire prediction head, represents the loss function of the fire type classification head, represents the loss function of the fire spread prediction head, Represents the loss function of the vigilance assessment head.

8. The underground space electric vehicle fire detection, early warning and prevention system according to claim 7 is characterized in that: The fire analysis scheme includes the charging state of the electric vehicle, the probability of fire occurrence, the influencing factors of fire occurrence, the scope of fire spread or the speed of fire spread and the fire warning level.

9. The underground space electric vehicle fire detection, early warning and prevention system according to claim 8 is characterized in that: The warning data set is input into the trained fire warning model to obtain the fire analysis plan for the underground space. The specific method includes: Step D1: Set a sample set, which includes T_U groups of samples. Each group of samples includes a warning data set and a corresponding fire analysis plan. The sample set is divided into S_J batches and input into the constructed fire warning model for training in sequence. Step D2: Use an adaptive learning rate strategy and an Adam or SGD optimizer to output the loss function of each task-specific head in turn. Stop training when the total loss function value no longer changes, and obtain a trained fire warning model. Step D3: Input the warning data set into the trained fire warning model to obtain the fire analysis plan for the current underground space.

10. A method for detecting, warning and preventing underground electric vehicle fires, characterized in that: The underground space electric vehicle fire detection, early warning and prevention system used to implement any one of claims 1 to 9 comprises: Step S1: Segment the underground space into regions and collect the charging data of the tram and the comprehensive data of the underground space; Step S2: Use the segmented regression method to perform trend analysis on the charging data and comprehensive data within the cycle time, and make early warning marks; Step S3: resampling the marked trams or areas to obtain a warning data set; Step S4: constructing a fire warning model; Step S5: Input the current warning data set into the trained fire warning model, obtain the fire analysis plan for the underground space, and display it on the interface.

Images