Intelligent early warning fire extinguishing system and equipment based on electrical equipment

Abstract

The invention relates to the field of industrial fire fighting, and discloses an intelligent early warning fire extinguishing system and equipment based on electrical equipment. The system comprises a multi-mode sensing module, a data fusion and feature extraction module, a fire risk prediction module, a self-adaptive fire extinguishing control module, a digital twin modeling module, an intelligent alarm and linkage module and a model self-learning optimization module, the equipment comprises a protection shell, and a plurality of heat dissipation holes distributed in an array are formed in the protection shell. The fire extinguishing assembly is installed on the front side of the protection shell, the installation buckles are fixed to the left side and the right side of the protection shell correspondingly, the circuit board is installed in the protection shell, and the sensor set is arranged on the top of the circuit board. According to the invention, through a risk scoring mechanism of the convolution and the long-short-term memory network, in combination with multi-modal data input and time sequence modeling, trend prediction of the electrical fire is realized, and the problems of inaccurate early risk pre-judgment and response lag are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of industrial fire fighting technology, and in particular to an intelligent early warning fire extinguishing system based on electrical equipment and the equipment. BACKGROUND

[0002] In the fields of industrial power distribution, automation control and energy management, motors, electrical cabinets and their attached electrical equipment are widely deployed as key infrastructure. To ensure stable operation of the equipment and reduce fire risk, monitoring systems including temperature detection, smoke alarm, gas sensing and other means have been widely used in the prior art, and fixed-point injection type fire extinguishing devices are used to suppress and control the initial fire. At the same time, some systems integrate communication modules to realize remote transmission and linkage response of alarm signals, to improve information accessibility and preliminary disposal efficiency.

[0003] However, with the complication of application environment and the diversification of fire causes, the existing systems still face certain limitations in actual operation. For example, most systems use single sensing dimension or fixed threshold determination method, which is difficult to accurately identify early atypical characteristics of fire; fire extinguishing control strategies are mostly based on preset schemes, with limited flexibility and adaptability; in addition, the systems generally lack post-operation targeted optimization capability, and it is difficult to continuously improve performance with experience accumulation. SUMMARY

[0004] The purpose of the present application is to provide an intelligent early warning fire extinguishing system based on electrical equipment and the equipment, which solves the problem of early and accurate early warning and self-adaptive efficient fire extinguishing in the running environment of electrical equipment.

[0005] To achieve the above purpose, the present application is realized by the following technical solutions: The intelligent early warning fire extinguishing system based on electrical equipment comprises: A multi-modal perception module for collecting a plurality of fire characteristic parameters related to the running environment of electrical equipment, including temperature, smoke, carbon monoxide concentration, infrared image and spectral flame characteristics; A data fusion and feature extraction module connected to the multi-modal perception module, for fusion processing of the collected sensing data and extracting a fire risk feature vector; A fire risk prediction module connected to the data fusion and feature extraction module, for predicting the fire risk score of a future time period based on historical feature sequences; An adaptive fire extinguishing control module connected to the fire risk prediction module, for determining the control parameters of fire extinguishing agent injection and controlling fire extinguishing execution by solving a multi-objective optimization equation set; a digital twin modeling module connected with the adaptive fire extinguishing control module, for constructing a three-dimensional model of the target electrical space and updating the temperature distribution and fire spread path in real time; an intelligent alarm and linkage module connected with the fire risk prediction module, for determining the alarm level based on fuzzy logic and triggering corresponding multi-level response measures; a model self-learning optimization module connected with the fire risk prediction module and the adaptive fire extinguishing control module, for iteratively optimizing the prediction model and control strategy after the fire extinguishing is completed.

[0006] Preferably, the multi-modal perception module comprises: a temperature sensing unit for detecting the ambient temperature around the electrical equipment in real time; a smoke sensing unit for detecting the concentration of smoke particles in the air; a gas detection unit for detecting the concentration of fire-related gases, including carbon monoxide; an infrared imaging unit for acquiring thermal distribution images of the equipment space; a spectral analysis unit for collecting flame radiation wavelength signals and identifying fire source characteristics.

[0007] Preferably, the data fusion and feature extraction module comprises: an attention fusion unit for weighted fusion of multi-source sensor data; a feature generation unit for generating a feature vector representing the state of the fire through a deep learning model.

[0008] Preferably, the fire risk prediction module comprises: a convolutional feature extraction unit for extracting local temporal changes in the feature sequence; a long short-term memory unit for modeling long-term dependencies in the feature sequence and outputting a fire risk score; the fire risk score satisfies the following risk classification logic: wherein, is the risk score predicted by the system neural network.

[0009] Preferably, the adaptive fire extinguishing control module comprises: a control parameter generation unit for generating control parameters of the extinguishing agent injection flow rate, injection angle, and time based on the optimization results; an optimization calculation unit for constructing a target function and solving the optimal control solution using a particle swarm algorithm and a genetic algorithm; a spray control execution unit for driving the microfluidic fire extinguishing device to perform fire extinguishing operations.

[0010] Preferably, the digital twin modeling module comprises: a three-dimensional modeling unit for constructing a digital model of the spatial structure of the electrical equipment; a thermal distribution mapping unit for converting the infrared image into a spatial temperature field; a fire spread simulation unit for simulating the spread path of the fire source and feeding back to the fire extinguishing control module.

[0011] Preferably, the digital twin modeling module comprises: a three-dimensional modeling unit for constructing a digital model of the spatial structure of the electrical equipment; a thermal distribution mapping unit for converting the infrared image into a spatial temperature field; a fire spread simulation unit for simulating the spread path of the fire source and feeding back to the fire extinguishing control module.

[0012] Preferably, the intelligent alarm and linkage module comprises: a fuzzy decision unit for determining the alarm level based on the fire risk score, the temperature rise rate, the smoke concentration growth rate, etc. through fuzzy rules; a response execution unit for executing the audible and visual alarm, user notification, fire extinguishing start, and remote platform synchronization operation according to the alarm level.

[0013] Preferably, the model self-learning optimization module comprises: a sample storage unit for saving the sensing data and control response data during the fire extinguishing event; a model updating unit for iteratively training and optimally updating the fire risk prediction model and the fire extinguishing control strategy based on the historical data.

[0014] Preferably, in the optimization calculation unit, the objective function is: wherein Q i is the fire extinguishing agent flow rate of the i th nozzle; t i is the corresponding spraying time; T max (t) is the highest temperature in the current space; T safe is the safety threshold temperature; D uncovered is the area of the fire source region not covered by spraying; α, β, γ are the set weight coefficients.

[0015] The intelligent early warning and fire extinguishing equipment based on electrical equipment comprises a protective shell, a plurality of heat dissipation holes are arranged in the protective shell in an array, a fire extinguishing assembly is installed on the front side of the protective shell, mounting buckles are fixed on the left and right sides of the protective shell, a circuit board is installed in the protective shell, a sensor group is arranged on the top of the circuit board, a main control module is installed in the circuit board, a wiring module is arranged on the bottom of the main control module, and an alarm module is arranged on the top of the main control module.

[0016] In summary, the present application includes at least one of the following beneficial technical effects: 1. The present application realizes the trend prediction of electrical fire by the risk scoring mechanism of convolution and long short-term memory network, combined with multi-modal data input and time series modeling. Compared with the traditional system with static threshold alarm, the design solves the problems of inaccurate early risk prediction and response lag.

[0017] 2. The present application builds an extinguishing control parameter optimization function and introduces a joint solution method of genetic algorithm and particle swarm algorithm. The system can output the dynamic optimal solution of injection flow, time and angle. Unlike the existing technology of manually setting fixed parameters, the phenomenon of over-extinguishing or insufficient control is effectively avoided.

[0018] 3. The present application integrates a self-learning module in the system structure and continuously iteratively optimizes based on historical events, so that the system has the ability of autonomous evolution during operation. Compared with the existing scheme model update mode which depends on manual intervention, this mechanism can significantly improve the stability and accuracy of the model in long-term use.

[0019] 4. The present application identifies infrared images and flame spectra jointly, and reconstructs the temperature distribution and fire source path in the electrical space through digital twin. The system has the ability of real-time visualization of fire space dynamics. Unlike the traditional planar sensor distribution method, it solves the problems of blind area monitoring difficulty and inaccurate injection target. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 is the system module architecture diagram of the present application; Figure 2 is a multi-modal sensing module schematic diagram of the present application; Figure 3 is a data fusion and feature extraction module schematic diagram of the present application; Figure 4 is a fire risk prediction module schematic diagram of the present application; Figure 5 is an adaptive fire extinguishing control module schematic diagram of the present application; Figure 6 is a digital twin modeling module schematic diagram of the present application; Figure 7 is an intelligent alarm and linkage module schematic diagram of the present application; Figure 8 is a model self-learning optimization module schematic diagram of the present application; Figure 9 is a three-dimensional view of the device of the present application; Figure 10 is a protective shell internal structure distribution diagram of the present application.

[0021] Wherein, 1, protective shell; 2, heat dissipation hole; 3, fire extinguishing assembly; 4, mounting buckle; 5, circuit board; 6, sensor group; 7, main control module; 8, wiring module; 9, alarm module. DETAILED DESCRIPTION

[0022] The above description is made in conjunction with the accompanying drawings Figure 1 - the accompanying drawings Figure 10 The application will be further described in detail.

[0023] The application provides an intelligent early warning fire extinguishing system based on electrical equipment and equipment.

[0024] As shown in the accompanying drawings Figure 1 - the accompanying drawings Figure 8 The intelligent early warning fire extinguishing system based on electrical equipment can include: A multi-modal perception module for collecting various fire characteristic parameters related to the operating environment of electrical equipment, including temperature, smoke, carbon monoxide concentration, infrared image and spectral flame characteristics; Specifically, in the embodiments of the application, the multi-modal perception module is used to obtain fire-related physical quantities in the operating environment of electrical equipment in real time, and convert them into computable features input to subsequent analysis models.

[0025] The multi-modal perception module includes a temperature detection part, a smoke detection part, a gas detection part, an infrared imaging part and a spectral recognition part.

[0026] The smoke detection part is arranged near the top of the equipment shell or the air inlet, for sensing the concentration of smoke particles in the air and generating an analog or digital signal.

[0027] The gas detection part includes a carbon monoxide detection sensor connected to a collection controller for obtaining combustible gas or harmful gas concentration information.

[0028] The infrared imaging part includes a thermoelectric array sensor and an image preprocessing chip, the thermoelectric array sensor is arranged inside the visible window, and the infrared thermal image in the equipment space is obtained through the glass window.

[0029] The spectral recognition part includes three sections of UV / visible / near-infrared flame recognition units, which can capture the characteristic wavelength range of the flame signal, and are used to assist in identifying the ignition type and energy intensity.

[0030] Each of the above sensor units is connected to a data fusion control board, the control board integrates a signal conditioning circuit and an analog-to-digital conversion circuit, and is used for standardizing the processing of the sensor signals.

[0031] The sensor output after standardization processing constitutes a feature initial vector: X0=[T(t),S(t),C CO (t),F λ (t),I IR (t)]; Wherein, T(t) represents temperature;S(t) represents smoke concentration;C CO (t) represents carbon monoxide concentration;F λ (t) represents the distribution function of spectral intensity at a specific wavelength;I IR (t) represents infrared image tensor.

[0032] The feature initial vector is input into the attention fusion network, the network is composed of embedding coding layer, weight distribution layer and weighted aggregation layer, and is used for carrying out semantic correlation enhancement processing on multi-source information.

[0033] The multi-modal fire perception vector output after fusion is defined as: Wherein, x i represents the feature vector output by the i-th sensor;Alpha i represents the attention coefficient, and satisfies the normalization constraint.

[0034] The attention coefficient alpha i is calculated by the following formula: Wherein Wherein, W1 is a learnable weight matrix;W2 is a weight vector;B is a bias vector.

[0035] The fusion feature vector is cached to the local time sequence cache area and is transmitted to the fire risk prediction module in real time.

[0036] The above infrared image tensor is compressed and mapped through a convolutional coding network, specifically: v IR =CNN(I IR (t)); Wherein, v IR is an infrared image feature vector, and the length depends on the CNN network structure, and the feature is used for subsequent heat distribution fitting and risk assessment.

[0037] In the multi-modal perception module of the application, the data acquisition frequency is uniformly set to f=1Hz to meet the response requirement of the change of fire characteristics.

[0038] The multi-modal perception module is connected with the embedded control mainboard via high-speed I 2 C or SPI bus, and the mainboard simultaneously manages data acquisition, caching and communication transmission to downstream modules.

[0039] Through the above structural connection and algorithm fusion, synchronous perception of multiple characteristics of electrical equipment fire is realized, which has the effects of fast response speed and strong environmental adaptability, and can provide complete input data chain for subsequent fire trend prediction and control.

[0040] A data fusion and feature extraction module is connected with the multi-modal perception module, used for fusion processing of the collected sensing data and extraction of a fire risk feature vector; Specifically, the data fusion and feature extraction module of the embodiment is used for receiving multi-source heterogeneous sensing data from the multi-modal perception module, including temperature, smoke concentration, carbon monoxide concentration, infrared image data and flame spectrum wavelength signals.

[0041] The module includes an attention fusion unit and a feature generation unit, wherein the attention fusion unit is used for realizing weighted fusion processing of heterogeneous data.

[0042] The attention fusion unit includes a first input end, a fusion calculation part and an output end, and the first input end is respectively connected with the output ends of the temperature sensing unit, the smoke sensing unit, the gas detection unit, the infrared imaging unit and the spectrum analysis unit.

[0043] In order to enhance the modeling ability of the importance of multi-source signals, the fusion calculation part preferably adopts a multi-head attention mechanism to realize data weighted fusion.

[0044] The multi-head attention mechanism is calculated according to the following formula: Wherein, Q, K and V represent query matrix, key matrix and value matrix respectively; d k is a scaling factor of vector dimension.

[0045] Before entering the attention mechanism, each type of input sensing data needs to be encoded into a tensor of the same dimension through a linear mapping module.

[0046] The output fusion feature tensor is transmitted to the feature generation unit, which includes a deep neural network model used for generating a feature vector representing the current fire state.

[0047] Preferably, the neural network structure adopts a combination design of convolutional layer and fully connected layer, wherein the convolutional layer is used for extracting local feature patterns, and the fully connected layer is used for outputting feature expressions of uniform dimension.

[0048] Let the fusion tensor be Where n is the number of sensing channels, d is the feature dimension of each channel, and after convolutional layer processing, we get: F=ReLU(W c *X+b c ); Wherein, Wc is the convolution kernel weight; b c is the bias term; * indicates the convolution operation; X indicates the fused tensor.

[0049] F is then projected into a fixed-dimensional space through a fully connected layer to obtain the fire state feature vector z: z=σ(W f F+b f ); Among them, W f is the weight of the fully connected layer; b f is the bias term; σ is the activation function.

[0050] The eigenvector z will be used as the input of the fire risk prediction module for subsequent fire scoring and response judgment.

[0051] The physical connection relationship of the above modules is as follows: the output of each sensor sub-unit of the multimodal perception module is sent to the edge computing chip through the analog-to-digital signal processing circuit. The attention fusion and feature extraction logic are deployed inside the chip and connected to the fire risk prediction module through a bus.

[0052] During the implementation process, we first construct the encoding representation of various sensor input data, then dynamically weightedly fuse the multi-source data through the attention mechanism, and finally extract the risk representation vector of unified dimension through the deep neural network.

[0053] This embodiment can realize the information fusion and abstract feature expression of heterogeneous sensor data, and has the technical effect of enhancing the early identification capability of fire.

[0054] The fire risk prediction module is connected to the data fusion and feature extraction module to predict the fire risk score for future time periods based on the historical feature sequence; Specifically, the fire risk prediction module, which includes a convolutional feature extraction unit and a long-short-term memory unit, is used to predict future fire risk scores based on historical feature sequences, thereby enabling early warning triggering control.

[0055] The fire risk prediction module uses the fire risk feature vector output by the data fusion and feature extraction module as input. This feature vector includes multidimensional data representing the current fire status, such as temperature change rate, smoke concentration change, CO concentration gradient, infrared image hot spot characteristics, and flame spectrum signals.

[0056] First, establish the fire time series feature tensor Where T is the length of the time series and N is the feature dimension. The feature vector at each moment is defined as: x t =[T t ,ΔT t ,S t ,ΔSt ,CO t ,ΔCO t ,IR t ,FS t ]; where T is temperature; T is temperature rise rate; S is smoke concentration; ΔS is smoke growth rate; CO, ΔCO are carbon monoxide concentration and its change rate, respectively; IR is infrared thermal distribution feature; FS is flame spectral intensity feature. t t t t t t t t

[0057] The convolution feature extraction unit performs one-dimensional convolution operation on the input sequence to extract local dynamic change features in the time sequence. Let the convolution kernel size be b c , and the convolution result is: z t =σ(W c *x t:t+k-1 +b c ); where σ(.) represents a nonlinear activation function; W c is the convolution kernel weight; and b c is the bias term.

[0058] Subsequently, the convolution result {z1, z2,..., z T′} is input to the long short-term memory network (LSTM) in time sequence to model long-term dependencies, and the final risk score is output: h t ,c t =LSTM(z t ,h t-1 ,c t-1 )R t =W r ·h t +b r ; where h t is the current time hidden state; c t is the cell state; W r , b r are the output layer weights and bias.

[0059] The fire risk score satisfies the following risk classification logic: The fire risk score ​​​​​​​​Used to characterize the fire possibility of the current environment. The system sets the alarm classification standards based on the score as follows: Among them, θ1, θ2 and θ3 are risk thresholds set by experience, corresponding to the low, medium and high alarm levels respectively.

[0060] The fire risk prediction module physically includes a processor, memory, and neural network acceleration unit. The module is connected to the feature extraction module through a data bus and receives feature sequence input through an internal cache. The output risk score It is synchronously transmitted to downstream control units such as the intelligent alarm and linkage module and the adaptive fire extinguishing control module through the data interface.

[0061] The fire risk prediction module and the adaptive fire extinguishing control module form a tightly coupled closed-loop structure. The prediction results are not only used for alarm judgment, but also serve as input parameters for the optimization objective function and participate in the calculation of fire extinguishing control strategies.

[0062] The above structure can be deployed on an embedded AI processor or edge computing platform, and the trained model can be deployed in formats such as TensorRT and ONNX to achieve real-time reasoning and risk assessment.

[0063] This module can achieve early prediction of fire risks, has the technical effects of fast response speed, strong adaptability and wide applicability, and can significantly improve the overall response efficiency of the system.

[0064] An adaptive fire extinguishing control module, which is connected to the fire risk prediction module and is used to determine the control parameters of the fire extinguishing agent injection and control the execution of fire extinguishing by solving a multi-objective optimization equation set; Specifically, the adaptive fire extinguishing control module includes a control parameter generation unit, an optimization calculation unit and an injection control execution unit. The control parameter generation unit is connected to the optimization calculation unit, and the optimization calculation unit is connected to the injection control execution unit. The three constitute an information and control link.

[0065] The optimization calculation unit is used to construct the fire extinguishing control objective function based on the fire risk scoring results and the fire source distribution status feedback from the digital twin module, and use a hybrid intelligent algorithm to solve multi-objective constraints.

[0066] The objective function is to maximize the efficiency of fire extinguishing agent injection and optimize the heat elimination rate, and the form is as follows: F = w1·Q(T max -T safe )+w2·A uncover +w3·Σ(Δt i ·q i ); Among them, q i represents the extinguishing agent injection flow rate of the i-th nozzle; Δti denotes the continuous injection time of the nozzle; T max denotes the highest temperature measured in the current fire source area; T safe is a preset temperature safety threshold; A uncover is the area of the fire source area that has not been covered by injection in the digital twin model; w1, w2, w3 are the weighting coefficients of the thermal safety, injection coverage, and resource conservation objectives, respectively.

[0067] To solve the optimal solution in the objective function, the optimization calculation part further adopts a joint iteration strategy of particle swarm optimization (PSO) and genetic algorithm (GA).

[0068] First, initialize the particle swarm population containing the injection parameters of each nozzle, then perform fitness evaluation, and the evaluation function is the objective function F.

[0069] In each iteration, genetic algorithm is used for crossover and mutation of individuals to improve the diversity of the solution space, while particle swarm algorithm is used for convergence guidance to update the velocity vector and position vector: v i (t+1) = ω·v i (t) + c1·r1·(p_best i -x i (t)) + c2·r2·(g_best-x i (t)); x i (t+1) = x i (t) + v i (t+1); where x i (t) is the position vector (corresponding to the injection parameters) of the i-th particle in the t-th iteration; v i (t) is the velocity vector; p_best i is the historical optimal solution of the particle; g_best is the current global optimal solution of the population; ω is the inertia weight; c1, c2 are learning factors; r1, r2 are random numbers between 0 and 1.

[0070] After the optimization process converges, the control parameter generation part outputs the injection control parameter set according to the optimal result: P = {(q i , θ i , Δt i ), (q2, θ2, Δt2),..., (q n , θ n , Δt n )}; where q i is the flow rate of the i-th nozzle; θ iis the injection angle; Δt2 is the injection time.

[0071] The injection control execution part receives the above-mentioned control parameter set, and drives the micro-fluidic pump group, the electromagnetic valve assembly and the directional nozzle module to realize precise injection of the fire extinguishing agent at multiple points.

[0072] The micro-fluidic pump group is composed of a plurality of programmable flow pumps, each pump corresponding to a nozzle, and a PWM control signal is used to adjust the injection intensity; the electromagnetic valve assembly is used to instantaneously open / close the fire extinguishing agent path, and the directional nozzle module has an electrically controlled rotating base to realize angle adjustment.

[0073] After the fire extinguishing is completed, the injection control execution part feeds back the fire extinguishing process data to the model self-learning optimization module for subsequent optimization.

[0074] Through the above-mentioned control strategy, the adaptive fire extinguishing control module can dynamically adjust the injection mode, realize preferential injection of the heat concentration area and directional coverage of the fire source spreading path, and has a high intelligent response capability.

[0075] A digital twin modeling module connected with the adaptive fire extinguishing control module is used to build a three-dimensional model of the target electrical space and update the temperature distribution and fire spreading path in real time. Specifically, the digital twin modeling module includes a three-dimensional modeling unit, a heat distribution mapping unit and a fire spread simulation unit. The module is integrated in the intelligent early warning fire extinguishing system and is connected with the adaptive fire extinguishing control module for data interaction.

[0076] The three-dimensional modeling unit includes a modeling processing part, a space data input part and a structure synchronization part. The modeling processing part is connected with the main control chip, receives the shape parameters and device layout data of the electrical equipment, and is used to build a digital structure model of the internal space of the device. The modeling process is based on point cloud data or CAD drawings to realize space topology reconstruction.

[0077] The heat distribution mapping unit includes an image receiving part, a heat field calculation part and a temperature projection part. The image receiving part is connected with the infrared imaging unit and is used to obtain the real-time thermal image of the space. The heat field calculation part calculates the heat field distribution of the space according to the received infrared image and the temperature data T i The temperature is mapped by a spatial interpolation function: Where T(x, y, z) is the estimated temperature at point T(x, y, z); T i is the temperature value collected by the i th sensor; w i is the weighting coefficient corresponding to the i th sensor; and n is the total number of sensors participating in the interpolation calculation.

[0078] The fire spread simulation unit includes a fire source modeling part, a spread calculation part and a feedback interface part. The fire source modeling part is based on the output of the risk scoring module The initial fire source position and heat source intensity are determined. The fire spread calculation is simulated based on an improved heat conduction diffusion equation: wherein, is the temperature change rate per unit time; and a is the thermal diffusion coefficient; is the spatial Laplacian of temperature, representing the spatial diffusion trend; Q s (x, y, z, t) is a source term function, representing the heat release intensity of the fire source per unit volume per unit time; Q s is the heat source intensity.

[0079] The heat source intensity Q s is expressed as follows: wherein, β is an adjustment coefficient, and f(x, y, z) represents a fire source influence function, which is set to a high value around the fire source and rapidly decays away from the fire source.

[0080] The simulation process updates the temperature field distribution result every preset time step, and renders it in real time in the three-dimensional space model, realizing the synchronous mapping of the digital space and the physical fire state.

[0081] The flame spread path output by the fire simulation unit is connected with the adaptive fire extinguishing control module through the feedback interface, providing an optimized suggestion for the spraying direction. The path data is represented in the form of a three-dimensional vector set: P={(x i ,y i ,z i ,t i )∣i=1,2,...,m}; wherein, (x i ,y i ,z i ) is the position of the i-th flame front point in the three-dimensional space; t i is the predicted time when the point is covered by the flame spread; and m is the total number of simulated path points.

[0082] The above-mentioned heat distribution mapping and spread prediction model are deployed on the local processing unit of the digital twin modeling module, which has edge computing capability, ensuring the real-time performance of the simulation update.

[0083] The data input end of the digital twin modeling module is connected with the infrared imaging unit and the temperature sensing unit in the multi-modal perception module, and the control end is output to the adaptive fire extinguishing control module, forming a complete closed loop.

[0084] The structure has the following effects: real-time virtual mapping of fire space state can be realized; dynamic adjustment of fire extinguishing control strategy is supported; and response accuracy and prediction ability of the system to complex fire environment are enhanced.

[0085] Further, the three-dimensional modeling unit supports importing modularized CAD models of standard cabinets, distribution boxes or cable slots, and adapts to deployment requirements of various electrical environments. The thermal field mapping supports adaptive precision adjustment, and automatically adjusts the interpolation radius and resolution according to the number and distribution of sensors.

[0086] The present embodiment realizes high-fidelity reproduction of the actual fire dynamics by the construction of the heat distribution equation and the fire source function, and provides input basis for realizing high-precision fire extinguishing path planning.

[0087] The above embodiment discloses a complete technical chain from modeling, heat map fusion to fire prediction of the digital twin modeling module, has engineering landing feasibility, and ensures that the corresponding function modules can be deployed and developed by the person skilled in the art.

[0088] The intelligent alarm and linkage module is connected with the fire risk prediction module, and is used for judging the alarm level based on fuzzy logic and triggering corresponding multi-level response measures; Specifically, the intelligent alarm and linkage module includes a fuzzy judgment unit and a response execution unit, which are logically connected through a data bus and respectively communicate with the fire risk prediction module to receive the prediction score result and the fire dynamic index input.

[0089] The fuzzy judgment unit includes a fuzzy reasoning chip and a rule base memory, and the fuzzy reasoning chip internally integrates a fuzzy input conversion part, a membership function matching part, a rule matching part and a fuzzy output solving part, and the parts are connected according to fixed hardware logic.

[0090] The fire risk score is generated by the fire risk prediction module and is represented as a normalized value as the first input quantity of fuzzy judgment.

[0091] In addition, the fuzzy judgment unit also receives temperature rise rate input, smoke concentration growth rate input, and CO concentration growth rate input, which are respectively denoted as fuzzy input quantities x1, x2, x3.

[0092] The above four input parameters jointly constitute a fuzzy input vector: Wherein, is the current predicted fire risk score; x1 is the temperature change rate; x2 is the smoke concentration change rate; and x3 is the CO concentration change rate.

[0093] Each input parameter is converted into three levels of membership, low, medium and high, by the fuzzification component, which correspond to the semantic levels of "safe", "warning" and "danger" respectively, using the following triangular membership functions: where a1, a2, a3 are the division nodes of each fuzzy variable, which are obtained according to sample data statistics.

[0094] The rule matching component retrieves the rule that best matches the current input membership vector from the rule base. The rule is in the following form: Rule example: "If x1 is high, and x2 is medium, and x3 is low, then the alarm level is high." The fuzzy output variable is set to the alarm level L∈{L1, L2, L3}, where L1 is a normal alarm (audible and visual prompt); L2 is a moderate alarm (notification platform + voice broadcast); and L3 is a serious alarm (triggering fire extinguishing + power cut + remote push).

[0095] The fuzzy output is defuzzified by the maximum membership principle, outputting the alarm level number l∈{1, 2, 3}.

[0096] The response execution unit includes an audible and visual alarm component, a communication linkage component, and a fire extinguishing start control interface component, all of which are connected to the controller through the CAN bus and execute the following responses according to the alarm level number: The response execution unit internally judges the state and distributes instructions through a response controller, which receives the alarm level number and starts the corresponding hardware control signal.

[0097] In an exemplary embodiment, if the system detects: Risk score: Temperature rise rate: x1 = 1.2℃ / min; Smoke concentration increase rate: x2 = 0.8mg / m 3 / min; CO concentration increase rate: x3 = 0.5ppm / min; then the input fuzzy levels are "high-medium-medium-low" respectively, and the final fuzzy inference output alarm level is L3, and the response execution unit, i.e. the linkage fire extinguishing module, starts spraying, power cut, and uploads the alarm data to the remote platform.

[0098] Through the above implementation structure, the module can realize intelligent identification and linkage control of fire levels, has the effect of fast response and adaptation to multi-variable input, and improves the reliability and effectiveness of the overall system.

[0099] ​a model self-learning optimization module connected with the fire risk prediction module and the adaptive fire extinguishing control module, used for iterative optimization of the prediction model and the control strategy after the fire extinguishing is completed; Specifically, the model self-learning optimization module includes a sample storage unit and a model updating unit. The sample storage unit is used for recording the input and response output of each module in the whole process of the fire event.

[0100] The sample storage unit is connected with the fire risk prediction module and the adaptive fire extinguishing control module, and can collect and store the following data feature sequences in real time: sensor raw data, feature vectors, prediction scores, alarm levels, spraying parameters and actual fire extinguishing effect feedback.

[0101] The model updating unit is in communication connection with the sample storage unit, and is used for structural fine-tuning and parameter optimization of the neural network model based on the stored historical samples by using an incremental training method.

[0102] The model updating process is based on a supervised learning structure, and adopts a time window rolling mechanism to extract time series training data. First, a risk score prediction model is established: wherein, is the fire risk score prediction value at time t; f θ is a neural network model, and the parameter set is θ; XX t-w:t is a multi-modal input feature vector sequence in a time window w.

[0103] The optimization target is to minimize the joint loss function of the prediction error and the control error: L total = λ1·L pred + λ2·L ctrl ; wherein, LL pred is the score prediction mean square error; L ctrl is the control parameter generation error; λ1, λ2 are weighting coefficients.

[0104] In the training process, a cyclic updating method is adopted: first, a fire event sample sequence is selected; then, a small batch of samples is used to perform back propagation optimization on the current model; finally, the model error convergence is judged, and the original model is updated or retained.

[0105] The neural network structure is composed of a convolution feature extraction layer and an LSTM memory unit. The input layer receives a vector sequence with a dimension of d output by the feature extraction module, and the output layer is a single node risk score.

[0106] The training process adopts an Adam optimizer, and the initial value of the learning rate is set to 0.001. The number of training rounds is not more than 50 rounds each time to prevent overfitting, and the updating frequency is automatically triggered after each fire response event is completed.

[0107] The fire extinguishing effect feedback is provided by the digital twin modeling module, including the temperature field drop rate and the fire source spread stop time.

[0108] The target function in the control strategy is mathematically expressed as: Wherein, Q i is the fire extinguishing agent injection flow rate of the i th nozzle; t i is the injection time; T max (t) is the highest temperature of the fire field; T safe is the temperature safety threshold; D uncovered is the un-covered area of the fire extinguishing area; and α, β, and γ are weighting coefficients for system parameter adjustment.

[0109] The optimization function is used to train the fire extinguishing control strategy model, and historical control samples are selected for model updating with the goal of minimizing the JJJ value, so as to realize continuous improvement of the fire extinguishing control effect.

[0110] The model updating unit is physically connected with the main controller through an embedded deep learning computing chip, and shares control parameters and risk score results with the main control board through logical interaction. The updated strategy is written into the memory and takes effect in the next response.

[0111] Through the above structure and algorithm mechanism, the model self-learning optimization module can dynamically update the system model after each fire disposal event, effectively improving the fire prediction accuracy and fire extinguishing control response efficiency.

[0112] The intelligent early warning fire extinguishing equipment based on electrical equipment described below can be mutually corresponding with the intelligent early warning fire extinguishing system based on electrical equipment described above.

[0113] Please refer to the attached Figure 9 -attached Figure 10 The application also provides an intelligent early warning fire extinguishing equipment based on electrical equipment, which comprises a protective shell 1, a plurality of heat dissipation holes 2 are arranged in an array in the protective shell 1, a fire extinguishing assembly 3 is installed on the front side of the protective shell 1, mounting buckles 4 are fixed on the left and right sides of the protective shell 1, a circuit board 5 is installed in the protective shell 1, a sensor group 6 is arranged on the top of the circuit board 5, a main control module 7 is installed in the circuit board 5, a wiring module 8 is arranged on the bottom of the main control module 7, and an alarm module 9 is arranged on the top of the main control module 7.

[0114] The device of the embodiment can be used to execute the above-mentioned method embodiments, and has similar principles and technical effects, which will not be described here again.

[0115] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary of the principles and application of the present application. Numerous modifications and adaptions can be effected without departing from the spirit and scope of the present application, which is not limited to the exact construction and arrangement described. It is intended, therefore, to cover all modifications and adaptions that fall within the scope of the claims and their equivalents.

Claims

1. Intelligent early warning fire extinguishing system based on electrical equipment, characterized by: include: A multimodal sensing module is used to collect various fire characteristic parameters related to the operating environment of electrical equipment, including temperature, smoke, carbon monoxide concentration, infrared images, and spectral flame characteristics; A data fusion and feature extraction module, connected to the multimodal perception module, for fusing the collected sensor data and extracting fire risk feature vectors; A fire risk prediction module, connected to the data fusion and feature extraction module, for predicting fire risk scores for future time periods based on historical feature sequences; An adaptive fire extinguishing control module, connected to the fire risk prediction module, for determining the control parameters of the fire extinguishing agent injection and controlling the execution of the fire extinguishing by solving a multi-objective optimization equation group; A digital twin modeling module, connected to the adaptive fire extinguishing control module, is used to build a three-dimensional model of the target electrical space and update the temperature distribution and fire spread path in real time; An intelligent alarm and linkage module, which is connected to the fire risk prediction module and is used to determine the alarm level based on fuzzy logic and trigger corresponding multi-level response measures; The model self-learning optimization module is connected to the fire risk prediction module and the adaptive fire extinguishing control module, and is used to iteratively optimize the prediction model and control strategy after the fire extinguishing is completed.

2. The intelligent early warning fire extinguishing system based on electrical equipment according to claim 1 is characterized in that: The multimodal perception module includes: Temperature sensing unit, used to detect the ambient temperature around the electrical equipment in real time; Smoke sensing unit, used to detect the concentration of smoke particles in the air; Gas detection units to detect concentrations of fire-related gases, including carbon monoxide; Infrared imaging unit, used to obtain thermal distribution images of the equipment space; The spectrum analysis unit is used to collect flame radiation wavelength signals and identify fire source characteristics.

3. The intelligent early warning fire extinguishing system based on electrical equipment according to claim 1 is characterized in that: The data fusion and feature extraction module includes: Attention fusion unit, used for weighted fusion of multi-source sensor data; The feature generation unit is used to generate a feature vector representing the fire status through a deep learning model.

4. The intelligent early warning fire extinguishing system based on electrical equipment according to claim 1 is characterized in that: The fire risk prediction module includes: Convolutional feature extraction unit, used to extract local temporal changes of feature sequences; Long short-term memory unit, used to model long-term dependencies in feature sequences and output a fire risk score; the fire risk score The following risk classification logic is met: in, Provides a risk score for the system's neural network predictions.

5. The intelligent early warning fire extinguishing system based on electrical equipment according to claim 1 is characterized in that: The adaptive fire extinguishing control module includes: The control parameter generation unit is used to generate the control parameters of the fire extinguishing agent injection flow rate, injection angle and time according to the optimization results; the optimization calculation unit is used to construct the objective function and use the particle swarm algorithm and genetic algorithm to solve the optimal control solution; The injection control execution unit is used to drive the microfluidic fire extinguishing device to perform fire extinguishing operations.

6. The intelligent early warning fire extinguishing system based on electrical equipment according to claim 1 is characterized in that: The digital twin modeling module includes: 3D modeling unit, used to build digital models of the spatial structure of electrical equipment; Thermal distribution mapping unit, used to convert infrared images into spatial temperature fields; The fire simulation unit is used to simulate the fire spread path and feed back to the fire extinguishing control module.

7. The intelligent early warning fire extinguishing system based on electrical equipment according to claim 1 is characterized in that: The intelligent alarm and linkage module includes: The fuzzy decision unit is used to determine the alarm level through fuzzy rules based on inputs such as fire risk score, temperature rise rate, and smoke concentration growth rate; The response execution unit is used to execute sound and light alarms, user notifications, fire extinguishing activation and remote platform synchronization operations according to the alarm level.

8. The intelligent early warning fire extinguishing system based on electrical equipment according to claim 1, characterized in that: The model self-learning optimization module includes: a sample storage unit for storing sensing data and control response data during a fire extinguishing event; The model updating unit is used to iteratively train and optimize the fire risk prediction model and fire extinguishing control strategy based on historical data.

9. The intelligent early warning fire extinguishing system based on electrical equipment according to claim 1, characterized in that: In the optimization calculation unit, the objective function is: Among them, Q i is the extinguishing agent flow rate of the i-th nozzle; t i is the corresponding injection time; T max (t) is the maximum temperature in the current space; T safe is the safety threshold temperature; D uncovered is the area of ​​the fire source area not covered by the spray; α, β, γ are the set weight coefficients.

10. An intelligent early warning fire extinguishing device based on electrical equipment, comprising a protective shell (1) according to any one of claims 1 to 9, characterized in that: The protective shell (1) is provided with a plurality of heat dissipation holes (2) distributed in an array, a fire extinguishing assembly (3) is installed on the front side of the protective shell (1), mounting buckles (4) are fixed on both the left and right sides of the protective shell (1), a circuit board (5) is installed inside the protective shell (1), a sensor group (6) is provided on the top of the circuit board (5), a main control module (7) is installed inside the circuit board (5), a wiring module (8) is provided at the bottom of the main control module (7), and an alarm module (9) is provided on the top of the main control module (7).

Images