High-voltage test hall anti-electric shock early warning system and method based on multi-source data fusion

Abstract

The invention discloses a high-voltage test hall anti-electric shock early warning system and method based on multi-source data fusion, and the system comprises an electric field dynamic monitoring module which is used for collecting and calibrating electric field data through a distributed three-axis field intensity sensor array, and predicting the field intensity distribution through the combination of an ARIMA-GRU hybrid model; the UWB-IMU cooperative positioning module adopts a TOF-TDOA-AOA algorithm to fuse inertial data so as to realize the positioning of a tester; the vital sign abnormity detection module is used for monitoring body surface leakage current and physiological parameters; the three-dimensional real scene modeling module constructs a dynamic safety boundary based on laser SLAM; the fuzzy comprehensive evaluation decision module dynamically allocates weights through an LSTM-ATT model and calculates risk indexes; and the multi-stage emergency linkage control module executes graded power-off response. Through multi-source data fusion and dynamic threshold adjustment, quick power-off response and full-scene risk management and control are realized, the false alarm rate is reduced, and high-voltage test electric shock accidents are effectively prevented.

Description

Technical Field

[0001] The present invention relates to the technical field of safety protection technology in a high-voltage electrical test environment, and specifically to a high-voltage test hall anti-electric shock warning system and method based on multi-source data fusion. Background Art

[0002] Currently, the main positioning technology used for anti-electric shock warning in high-voltage test halls generally adopts infrared positioning or ultrasonic positioning systems. Personnel location monitoring is achieved through signal transmission between fixed base stations and mobile tags. The safety monitoring adopted relies on a single electric field strength sensor or static distance threshold. The emergency control device is a relay power-off device that only supports a single power-off channel. The main threshold management is based on the mapping relationship between fixed voltage and safety distance.

[0003] Among them, traditional infrared positioning or ultrasonic positioning systems are severely interfered by high-voltage electromagnetic fields, and non-line-of-sight errors lead to positioning offsets, which cannot meet the needs of high-precision safety distance control; single electric field or distance monitoring in safety monitoring ignores the physiological state of people (such as surface leakage current, heart rate variability coefficient), and does not integrate physiological parameters with dynamic environmental change data; power-off devices rely on conventional relay circuits, and the response delay is too long, making it difficult to quickly block electric shock accidents; the main threshold management fixed safety distance and weight distribution cannot adapt to the characteristics of the test phase, and the parameters cannot be dynamically adjusted according to the test phase.

[0004] Therefore, inventing a high-voltage test hall anti-electric shock warning system based on multi-source data fusion, integrating multi-source data for positioning, integrating physiological parameters and environmental dynamic data, realizing multi-dimensional risk assessment, using dual-channel power-off control, improving safety, and realizing adaptive threshold management have become technical problems that need to be solved urgently. Summary of the Invention

[0005] The purpose of the present invention is to provide a high-voltage test hall anti-electric shock warning system based on multi-source data fusion. The present invention can solve the problems of low safety distance control, failure to integrate physiological parameters and environmental dynamic change data, long power-off device response delay, and inability to dynamically adjust parameters according to the test stage.

[0006] To achieve this purpose, the present invention designs a high-voltage test hall anti-electric shock warning system based on multi-source data fusion, which includes:

[0007] The electric field dynamic monitoring module is used to compensate the electric field intensity data of the test area collected in real time for environmental temperature drift and electromagnetic interference, obtain the compensated electric field intensity data, build a spatiotemporal gated hybrid prediction model based on the compensated electric field intensity data, and input the real-time collected electric field intensity data of the test area into the spatiotemporal gated hybrid prediction model to obtain the predicted electric field intensity;

[0008] The collaborative positioning module is used to calculate the three-dimensional spatial coordinates of the experimenter's location tag by fusing the ultra-wideband positioning data sent by the ultra-wideband base station arranged in the test hall with the inertial measurement data of the inertial measurement unit in the experimenter's wearable device;

[0009] The abnormality detection module is used to monitor the body surface leakage current and physiological parameter data in real time;

[0010] The 3D real-scene modeling module constructs a digital twin model of the test hall based on a synchronous positioning and mapping method. It dynamically marks the boundaries of the safe area based on the digital twin model of the test hall and dynamically adjusts the safety distance threshold based on the test voltage level.

[0011] The fuzzy comprehensive evaluation decision module is used to assign weights to the predicted electric field strength, the three-dimensional spatial coordinates of the test personnel's position label, the surface leakage current, and the physiological parameter data according to the dynamic weight distribution model, and to obtain the electric shock risk index in combination with the safety distance threshold, and to trigger a graded warning based on the risk index.

[0012] Preferably, it also includes a multi-level emergency linkage control module for executing test power cut-off and emergency procedure operations according to graded warnings.

[0013] Preferably, the specific method for obtaining the predicted electric field strength is: using the compensated electric field strength data to train the spatiotemporal gated hybrid prediction model, parsing the periodicity and trend components of the time series through the ARIMA (Auto Regressive Integrated Moving Average) unit to generate a linear prediction value; inputting the nonlinear features output by the ARIMA into the GRU (Gated Recurrent Unit) neural network to capture the nonlinear fluctuation time series law; fusing the linear prediction value of the ARIMA and the nonlinear correction value of the GRU to predict and generate the future electric field distribution intensity.

[0014] Preferably, the fuzzy comprehensive evaluation decision module adopts a time series prediction model based on LSTM-ATT (Long Short-Term Memory with Attention, a long short-term memory network with an attention mechanism) to obtain a dynamic weight matrix; the dynamic weight distribution model adjusts the evaluation index weights of each stage in the test based on the dynamic weight matrix, and each stage includes a pre-pressurization stage, a steady-state stage and a depressurization stage. The hierarchical analysis method and the fuzzy comprehensive evaluation method are used to calculate the weight consistency ratio of each stage. When the weight consistency ratio of each stage reaches the set standard, the current dynamic weight distribution of the predicted electric field strength, the three-dimensional spatial coordinates of the test personnel position label, the surface leakage current and the physiological parameter data is obtained.

[0015] Preferably, it also includes a risk heat map generation module, which generates a risk heat map based on the electric field strength data predicted by the ARIMA model, that is, the electric field risk distribution trend, to predict whether there are areas in the high-voltage test hall that are prohibited from entry or can only be entered after protective measures are implemented; a safety report automatic generation module, the safety report automatically records the maximum field strength value, the minimum safety distance and the disposal response time.

[0016] Beneficial effects of the present invention:

[0017] This paper proposes a high-voltage test hall electric shock warning system based on multi-source data fusion. This system employs a step-by-step optimization and multi-task joint decision-making mechanism to address the problems of insufficient positioning accuracy, high false alarm rates from a single data source, delayed emergency response, and poor adaptability of static thresholds in traditional technologies. By integrating an electric field intensity sensor array, a UWB-IMU (Ultra-Wideband and Inertial Measurement Unit) collaborative positioning module, and a vital sign detection module, it integrates multi-dimensional data such as dynamic electric field monitoring, spatial trajectories of personnel, and physiological parameters. A fuzzy comprehensive evaluation decision model is adopted to dynamically adjust the weight distribution (electric field weight 40% to 20%, distance weight 30% to 50%) according to the test stage (pre-pressurization, steady state, and power-off post-processing) to achieve dynamic adaptation of risk assessment, which is significantly better than the traditional single threshold judgment method; by adopting the TOF-TDOA-AOA hybrid positioning algorithm and IMU inertial data fusion, combined with CNN gait correction technology, the positioning error is reduced, the multi-level emergency linkage module integrates the hard-wired power-off controller and the gradient lighting system, and uses dual-channel power-off control to improve the response speed and effectively block the risk of electric shock; the electric field dynamic monitoring module adopts the ARIMA-GRU hybrid model, through the Kalman filter dynamic compensation loop Environmental temperature drift and electromagnetic interference, combined with GRU neural network to capture transient pulse characteristics, to achieve the prediction of electric field distribution in the next 5 minutes (RMSE ≤ 3.2kV ​​ / m), the three-dimensional real-scene modeling platform builds a digital twin model based on laser SLAM (accuracy error ≤ 2cm), dynamically adjusts the safety distance threshold, and solves the problem that the traditional static boundary cannot adapt to voltage changes; through the step-by-step optimization strategy, the electric field calibration, positioning correction, physiological abnormality detection and other tasks are processed independently to reduce the computational complexity. At the same time, the LSTM-ATT timing model and the fuzzy comprehensive evaluation method are used to perform multi-task weight allocation to achieve accurate calculation of the risk index, trigger a three-level warning, and achieve compliance audit and strategy optimization for the entire test cycle. The present invention breaks through the bottleneck of poor environmental adaptability, delayed response and high misjudgment rate in traditional technologies through the synergistic effect of multi-source data fusion, dynamic weight allocation and intelligent response mechanism, and provides an integrated solution for high-precision space control, dual-channel power outage emergency cut-off and multi-dimensional risk assessment for high-voltage test environments, with significant technical advantages and engineering practicality. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a structural schematic diagram of the present invention;

[0019] Figure 2 Schematic diagram of the method flow of the present invention;

[0020] Figure 3 Provide a flow chart for dynamic weight assignment;

[0021] Figure 4 This is a schematic diagram of the 3D real scene modeling interface;

[0022] Figure 5 Schematic diagram of three types of gait patterns. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0024] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0025] Example 1

[0026] A high-voltage test hall anti-electric shock warning system based on multi-source data fusion, such as Figure 1 As shown, it includes:

[0027] The electric field dynamic monitoring module is used to compensate the electric field intensity data of the test area collected in real time for environmental temperature drift and electromagnetic interference, obtain the compensated electric field intensity data, build a spatiotemporal gated hybrid prediction model based on the compensated electric field intensity data, and input the real-time collected electric field intensity data of the test area into the spatiotemporal gated hybrid prediction model to obtain the predicted electric field intensity;

[0028] The collaborative positioning module is used to calculate the three-dimensional spatial coordinates of the experimenter's location tag by fusing the ultra-wideband positioning data sent by the UWB (ultra-wideband) base station arranged in the test hall with the inertial measurement data of the IMU (inertial measurement unit) in the experimenter's wearable device;

[0029] The abnormality detection module is used to monitor the body surface leakage current and physiological parameter data in real time;

[0030] The 3D real-scene modeling module constructs a digital twin model of the test hall based on a synchronous positioning and mapping method. It dynamically marks the boundaries of the safe area based on the digital twin model of the test hall and dynamically adjusts the safety distance threshold based on the test voltage level.

[0031] The fuzzy comprehensive evaluation decision module is used to assign weights to the predicted electric field strength, the three-dimensional spatial coordinates of the test personnel's position label, the surface leakage current, and the physiological parameter data according to the dynamic weight distribution model, and to obtain the electric shock risk index in combination with the safety distance threshold, and to trigger a graded warning based on the risk index.

[0032] In the above technical solution, the electric field strength data is obtained by real-time collection of the test area by a distributed field strength sensor array.

[0033] In the above technical solution, the specific arrangement method of the distributed field strength sensor array is: arrange sensors around high-voltage equipment, at the boundaries of safe areas (such as the dynamic boundaries of 1.5m for 100kV and 5m for 500kV) and along personnel activity paths (such as the operator's permanent area).

[0034] In the above technical solution, the predicted electric field strength refers to the predicted electric field distribution strength in the next 5 minutes.

[0035] In the above technical solution, the physiological parameters include heart rate variability (HRV) and blood oxygen saturation.

[0036] In the above technical solution, the digital twin model builds a three-dimensional model of the test hall based on laser SLAM (Laser-based Simultaneous Localization and Mapping, Laser SLAM for short), with an accuracy error of ≤ 2cm (the size difference between the modeled model and the actual object). The model marks the location of high-voltage equipment, the boundaries of the safe area (such as 100kV corresponding to 1.5m, 500kV corresponding to 5m) and the emergency exit. Figure 4 As shown, the safe area boundary is dynamically adjusted according to the voltage level.

[0037] In the above technical solution, the safety distance threshold is automatically updated according to the test voltage level and rendered in real time in the three-dimensional interface.

[0038] In the aforementioned technical solution, the vital signs abnormality detection module uses a wearable device with a flexible circuit substrate (bending radius ≤ 3mm) integrated with a PPG-ECG sensor. This sensor monitors surface leakage current (threshold ≤ 0.5mA), heart rate variability (HRV), and blood oxygen saturation (sampling rate ≥ 250Hz). It uses the AES-256 encrypted transmission protocol and supports Bluetooth 5.2 and LoRa dual-mode communication. The wearable device has a built-in electric shock safety device. When the leakage current exceeds the threshold (0.5mA), it disconnects the circuit within 150ms and activates the backup inductive power supply to ensure data is not lost.

[0039] In the above technical solution, the vital signs abnormality detection module includes an abnormal alarm. When HRV is abnormal (such as standard deviation > 50ms) or blood oxygen saturation is < 90%, the device triggers a local vibration alarm and uploads it to the central processing platform.

[0040] In the above technical solution, by building a core framework, the core process of multi-source data fusion (electric field, positioning, physiological parameters) is covered, solving the misjudgment problem caused by the traditional single data source.

[0041] In the above technical solution, it also includes a multi-level emergency linkage control module, which is used to execute test power cut-off and emergency procedure operations according to graded warnings.

[0042] In the above technical solution, the multi-level emergency linkage control module realizes power-off control by designing dual-channel control logic. The dual-channel control logic has a hard-wired response of the main channel of ≤120ms and a LoRa wireless control response of the auxiliary channel of ≤500ms, as shown in Table 1 below:

[0043] Channel Type Trigger Conditions Response time Action Commands Main Channel E≥90kV / m and D≤1.2m ≤120ms All circuits are powered off Auxiliary channel Abnormal HRV lasting for 10 seconds ≤500ms Reduce voltage to a safe threshold

[0044] Table 1

[0045] Where E is the electric field strength and D is the real-time distance between the test personnel and the high-voltage equipment.

[0046] In the above technical solution, the hard-wired power-off controller is a high-reliability fast power-off device directly connected through a physical line for use in a multi-level emergency linkage control module. It provides dual protection circuits through optocoupler isolation and relay direct connection circuits. The response time is ≤150ms and the grounding resistance is ≤0.1Ω.

[0047] In the above technical solution, gradient emergency lighting is a dynamically adjustable emergency lighting system used in a multi-level emergency linkage control module. The brightness adjustment range is 50lux-1000lux, and the startup delay is ≤50ms to avoid blinding caused by strong light.

[0048] In the above technical solution, dual-channel power-off control logic is configured, the main channel realizes ≤150ms fast power-off when the field strength is ≥90kV / m and the distance is ≤1.2m, and the auxiliary channel realizes voltage reduction protection based on abnormal physiological parameters.

[0049] In the above technical solution, by introducing a multi-level emergency linkage control module, differentiated emergency operations are triggered according to the risk assessment results (such as risk index R≥0.7 / 0.8 / 0.9). Compared with a single power-off action, the impact of erroneous operation on test continuity is reduced. In addition, through the coordination of the main channel (hard wiring) and the auxiliary channel (wireless control), layered protection of rapid power off (main channel ≤120ms) and secondary emergency voltage reduction (auxiliary channel ≤500ms) is achieved to ensure operational reliability in extreme scenarios.

[0050] In the above technical solution, the specific method for obtaining the predicted electric field strength is:

[0051] The compensated electric field intensity data is used to train a spatiotemporal gated hybrid prediction model (i.e., the ARIMA-GRU hybrid model). The ARIMA unit (autoregressive order p=2, difference order d=1, moving average order q=1) is used to analyze the periodic and trend components of the time series to generate linear prediction values. The nonlinear features of the ARIMA output are input into the GRU neural network to capture the time series laws of nonlinear fluctuations. The linear prediction values ​​of the ARIMA and the nonlinear correction values ​​of the GRU are integrated to generate the electric field distribution intensity for the next 5 minutes through rolling multi-step prediction, forming a staged prediction and compensation closed loop, which provides data support for the dynamic adjustment of the safety threshold.

[0052] In the above technical solution, the spatiotemporal gated hybrid prediction model is the ARIMA-GRU hybrid prediction model. The specific MATLAB code for implementing the ARIMA-GRU hybrid prediction model is as follows:

[0053] %%ARIMA-GRU hybrid prediction model

[0054] % Function: Predict the field strength distribution in the next 5 minutes (300 seconds) based on historical electric field data

[0055] % Input: E_field_history (N×1 vector, historical electric field strength data, unit kV / m)

[0056] % Output: predicted_E (300×1 vector, predicted value for the next 5 minutes)

[0057] Patented technical features: ARIMA and GRU residual compensation mechanism, RMSE ≤ 3.2kV / m

[0058] %% S1: Data preprocessing

[0059] clc;clear;close all;

[0060] % Load sample data (replace with actual electric field data)

[0061] load('E_field_sample.mat'); % contains E_field_history variable

[0062] % Parameter settings

[0063] pred_steps=300;% prediction step length (5 minutes)

[0064] train_ratio=0.8;% training set ratio

[0065] % Data normalization

[0066] [E_norm,mu,sigma]=zscore(E_field_history);

[0067] % Divide the training set and test set

[0068] train_size=floor(length(E_norm)*train_ratio);

[0069] train_data=E_norm(1:train_size);

[0070] test_data=E_norm(train_size+1:end);

[0071] %% S2: ARIMA model construction (processing linear components)

[0072] %ARIMA parameter settings (optimized according to the AIC criterion)

[0073] p=2;%autoregressive order

[0074] d=1;% difference order

[0075] q=1;% moving average order

[0076] %Create ARIMA model

[0077] Mdl=arima(p,d,q);

[0078] % Model Fit

[0079] EstMdl=estimate(Mdl,train_data,'Display','off');

[0080] %Predict training set [arima_pred,~]=forecast(EstMdl,train_size,'Y0',train_data); %Calculate ARIMA residuals (nonlinear part)

[0081] residuals=train_data-arima_pred;

[0082] %% S3: GRU network construction (processing nonlinear residuals)

[0083] %Network parameters (patented core parameters)

[0084] numFeatures=1;% Input number of features

[0085] numHiddenUnits=128;%GRU hidden layer nodes

[0086] maxEpochs=100;% training rounds

[0087] miniBatchSize=32;% batch size

[0088] %Create residual sequence training set

[0089] XTrain=cell(length(residuals)-1,1);

[0090] YTrain=cell(length(residuals)-1,1);

[0091] for i=1:length(residuals)-1

[0092] XTrain{i}=residuals(i);

[0093] YTrain{i}=residuals(i+1);

[0094] end

[0095] %Define GRU network structure

[0096] layers=[...

[0097] sequenceInputLayer(numFeatures)

[0098] gruLayer(numHiddenUnits,'OutputMode','sequence')

[0099] fullyConnectedLayer(1)

[0100] regressionLayer];

[0101] % Training options

[0102] options=trainingOptions('adam',...

[0103] 'MaxEpochs',maxEpochs,...

[0104] 'MiniBatchSize',miniBatchSize,...

[0105] 'InitialLearnRate',0.001,...

[0106] 'LearnRateDropPeriod',20,...

[0107] 'LearnRateSchedule','piecewise',...

[0108] 'GradientThreshold',1,...

[0109] 'Shuffle','every-epoch',...

[0110] 'Plots','training-progress',...

[0111] 'Verbose',0);

[0112] %Training GRU network

[0113] net_gru=trainNetwork(XTrain,YTrain,layers,options);

[0114] %%S4: Hybrid prediction

[0115] % Initialize prediction data

[0116] combined_pred=zeros(pred_steps,1);

[0117] current_data = E_norm(end-pred_steps:end); % Initialize using the last 300 points

[0118] %Multi-step rolling forecast

[0119] fort=1:pred_steps

[0120] %ARIMA forecast

[0121] [arima_step,~]=forecast(EstMdl,1,'Y0',current_data);

[0122] %GRU residual prediction

[0123] gru_input=arrayDatastore(current_data-arima_step(1:end-1));

[0124] gru_pred=predict(net_gru,gru_input);

[0125] %Combined prediction results

[0126] combined_pred(t)=arima_step(end)+gru_pred(end);

[0127] % Update input sequence

[0128] current_data=[current_data(2:end);combined_pred(t)];

[0129] end

[0130] % Denormalization

[0131] predicted_E=combined_pred*sigma+mu;

[0132] %% S5: Results visualization and evaluation

[0133] figure('Position',[100100800400])

[0134] holdon

[0135] plot(E_field_history,'b','LineWidth',1.5)

[0136] plot(length(E_field_history)+1:length(E_field_history)+pred_steps,...

[0137] predicted_E,'r--','LineWidth',1.5)

[0138] title('Electric field strength prediction results (ARIMA-GRU hybrid model)')

[0139] xlabel('time step')

[0140] ylabel('Electric field strength (kV / m)')

[0141] legend('historical data', 'predicted value')

[0142] gridon

[0143] %Calculate RMSE (need to be replaced with actual test data)

[0144] %test_pred=...;%generated during actual test

[0145] %rmse=sqrt(mean((test_data-test_pred).^2));

[0146] %fprintf('RMSE:%.2fkV / m\n',rmse);

[0147] %%Model save (for patent implementation)

[0148] save('ARIMA_GRU_Model.mat','EstMdl','net_gru','mu','sigma');

[0149] In the above process, the MATLAB working environment was initialized (clc; clear; close all;) to ensure the reliability and repeatability of the data preprocessing process, and the compensated electric field data (E_field_history) was loaded. The prediction step size (5 minutes) and the training set ratio were set to 0.8. Generate a normalized data set (E_norm) through standardization, where mu and sigma are scalars, representing the mean of the data set and the standard deviation of the original data of the data set, respectively. Divide the data into a training data set (train_data) and a test data set (test_data); build an ARIMA model and set the autoregressive order (p=2), difference order (d=1) and moving average order (q=1). Fit the model (EstMdl) parameters (where 'Display', 'off' means turning off the command line output during the model fitting process) with the training data set (train_data) and generate the linear prediction component (arim a_pred), then extract the ARIMA prediction residuals (residuals) as nonlinear features; build a GRU neural network with 128 hidden units (numHiddenUnits), set the number of features (numFeatures) to 1, use the Adam optimizer for 100 rounds of training (maxEpochs) to capture the temporal pattern in the residuals, and the number of data samples input to the model at a single time when training the neural network (miniBatchSize) is 32. Create residual sequence training sets XTrain and YTrain, where XTrain is a cell array named XTrain with the number of rows re The length of siduals is minus 1, the number of columns is 1, each cell element will be used to store a sample of the input sequence, YTrain is a cell array named YTrain, the structure is the same as XTrain, used to store the corresponding target value, and then loop from 1 to the length of residuals minus 1, residuals has train_size elements, the loop will run train_size-1 times, the i-th residual value is used as the i-th cell element of the input feature XTrain, the i+1th residual value is used as the i-th cell element of the target value YTrain, and then define the GRU network structure, define the input layer, and input features. The function receives time series data, creates a GRU (Gated Recurrent Unit) layer to capture temporal dependencies, outputs the hidden state of all time steps (if 'last', only the last step is output), sets the number of hidden units (numHiddenUnits) to 128, creates a fully connected layer (fullyConnectedLayer), defines the loss function for the regression task (regressionLayer), and sets the training options (options), including the maximum number of training rounds (maxEpochs), batch size (miniBatchSize), and initial learning rate (set to 0.001), learning rate decline cycle (set to 20), piecewise learning rate scheduling (piecewise), gradient clipping threshold (set to 1), data shuffling strategy (every-epoch), visualize training progress (training-progress) and close command line (set to 0), and finally train the GRU network (trainNetwork), map the GRU output (net_gru) to the predicted value; in the hybrid prediction stage, reserve space to store the prediction results of the next 300 steps (combined_pred), where the prediction step size is 1 and the total number of prediction steps (pred_steps) is 300. 300 historical points are intercepted from the end of the normalized data (end-pred_steps) as the initial prediction data (current_data), input the initial prediction data, ARIMA predicts the next value (arima_step, scalar), captures the linear trend, performs GRU residual prediction, and generates the input data of the GRU network (gru_input) to capture the nonlinear residual not explained by the ARIMA model. The U network predicts the correction amount of the nonlinear residual (gru_pred), and obtains the combined result (combined_pred(t)) based on the next prediction value of ARIMA (arima_step(end)) and the correction amount of GRU for nonlinear fluctuations (gru_pred(end)). Update the input sequence, remove the oldest data in the window, add the latest prediction value, keep the input length 300, perform denormalization, and obtain the actual value (predicted_E); finally, perform data visualization, set the graphic window size (figure) to 800px wide and 400px high, and enable multi-layer overlay, draw the compensated electric field data ((E_field_history)), and the predicted curve (predicted_E), where the continuous time steps after the end point of the compensated electric field data (for example, if the compensated electric field data has 1000 points, the prediction is from 1001 to 1300), and perform graphic annotation, including title (title), X-axis label (xlabel), Y-axis label (ylabel), legend (legend) and grid (grid) on), the annotation indicates that the model accuracy is evaluated by calculating the root mean square error (RMSE ≤ 3.2 kV / m), and the trained ARIMA and GRU model parameters (ARIMA_GRU_Model.mat, EstMdl, net_gru, mu, and sigma) are persisted (saved) for subsequent use.

[0150] In the above technical solution, a hybrid architecture of ARIMA (autoregressive integrated moving average model) and GRU (gated recurrent unit) is adopted to divide the work of processing the linear trend component and nonlinear transient characteristics of the electric field strength data (such as electromagnetic pulse interference in high-voltage tests). The ARIMA unit is used to analyze the linear laws such as periodicity and trend of the electric field data, and the GRU neural network is used to capture nonlinear interference such as environmental temperature drift and equipment discharge transients. The linear trend component of the electric field data is first extracted through the ARIMA model, and then the residual (that is, the nonlinear fluctuations not explained by the linear model) is modeled using the GRU network to form a closed loop of staged prediction and compensation. The hybrid model can predict the electric field distribution in the next 5 minutes, identify high-risk areas in advance, and provide effective data support for dynamically adjusting the safety threshold.

[0151] In the above technical solution, the three-dimensional spatial coordinates of the test person's location tag calculated by fusing the ultra-wideband signal with the inertial measurement unit are obtained by hybrid positioning using the time of motion-time difference of arrival-angle of arrival hybrid positioning algorithm (TOF-TDOA-AOA). The specific method is as follows:

[0152] The time-of-flight (TOF) method is used to calculate the distance between the tester's location tag and a single signal base station in the high-voltage test hall. The time difference of arrival (TDOA) method is used to eliminate clock synchronization errors. The angle of arrival (AOA) method is used to determine the azimuth of the tester's location tag. The extended Kalman filter algorithm is used to calculate and integrate the inertial measurement unit data to compensate for non-line-of-sight errors and obtain the three-dimensional spatial coordinates of the tester's location tag.

[0153] Among them, the two-way ranging formula between a single signal base station and a tag in the high-voltage test hall is:

[0154]

[0155] Where d is the distance between the test personnel and the ultra-wideband base station, c is the speed of light, and T RTT is the signal round trip time, T TAT For the tag response time delay, a hyperbola equation is constructed based on the time difference of multiple base stations to obtain the three-dimensional spatial coordinates (x, y, z) of the test person's location tag:

[0156]

[0157] Where Δt ij is the time difference of the signal arriving at the two base stations, (x i ,y i ,z i ) and (x j ,y j ,z j ) are the coordinates of the i-th and j-th base stations respectively.

[0158] In the above technical solution, a UWB (Ultra-Wideband) base station with a regular tetrahedron layout is used, and the TOF-TDOA-AOA algorithm (frequency band 3.5-6.5GHz, pulse width <2ns) is used to integrate the angular velocity / acceleration data of the IMU (Inertial Measurement Unit). The UWB and IMU compensate for the non-line-of-sight error through the extended Kalman filter (EKF) algorithm, where the process noise parameter Q = 0.01, the observation noise parameter R = 0.1, the positioning accuracy parameter ±5cm, and the coverage radius parameter ≤50m. It supports real-time positioning data transmission and calculation for 100 people or devices wearing positioning tags. In this experiment, the experimental hall is rectangular, so UWB base stations are arranged at the four corners of the experimental hall and around the high-voltage test equipment to ensure full area coverage; the CNN classifier is introduced to analyze the acceleration characteristics (Z-axis error compensation coefficient 0.76~1.00) and establish three types of gait patterns, such as Figure 5 As shown, where g is the acceleration due to gravity.

[0159] In the above technical solution, the limitations of a single technology are solved by adopting a hybrid algorithm of TOF (time of flight), TDOA (time difference of arrival) and AOA (angle of arrival), and the ability to resist electromagnetic interference is significantly improved; the extended Kalman filter (EKF) is introduced to fuse IMU data (angular velocity, acceleration) to correct positioning offsets caused by equipment occlusion or electromagnetic interference in real time, and the adaptability to dynamic scenarios is enhanced; and the system supports 100 positioning tags working simultaneously, with a coverage radius of ≤50 meters and a positioning accuracy of ±5cm, which improves positioning accuracy and meets the needs of multi-person collaborative operations and seamless coverage of the entire area in large-scale high-voltage test scenarios.

[0160] In the above technical solution, the fuzzy comprehensive evaluation decision module adopts a time series prediction model based on LSTM-ATT to obtain a dynamic weight matrix; the dynamic weight distribution model adjusts the evaluation index weights of each stage in the test based on the dynamic weight matrix. The stages include the pre-pressurization stage, the steady-state stage and the decompression stage. The hierarchical analysis method and the fuzzy comprehensive evaluation method are used to calculate the weight consistency ratio of each stage. When the weight consistency ratio of each stage reaches the set standard, the verification is passed, and the predicted electric field strength, the three-dimensional spatial coordinates of the test personnel position label, the surface leakage current and the current dynamic weight distribution of the physiological parameter data are obtained.

[0161] In this technical solution, a time series prediction model based on the LSTM-ATT (attention mechanism) was constructed. The input dimensions include 12 feature categories, including voltage level V∈[10kV,800kV], personnel role R∈{operator, safety officer}, and equipment status S∈{charging, discharging, grounding}, and the output is a dynamic weight matrix W(t). The core parameters of the LSTM-ATT network in this experiment are: 128 hidden nodes, a dropout rate of 0.2, and a learning rate of 0.001.

[0162] In the above technical solution, the three-level evaluation indicators achieve efficient integration of multi-source data and accurate quantification of risk assessment through structured decomposition of risk factors, dynamic adaptation of test scenarios, and refinement of anomaly detection granularity. This is the core design for improving the accuracy of system warnings. The constructed three-level evaluation indicators are shown in Table 2 below:

[0163] First-level indicators Secondary indicators Weight adjustment factor Electric field strength Power frequency field strength, transient pulse intensity, gradient change rate α＝0.4-0.6 spatial distance Real-time distance, moving speed, trajectory deviation β＝0.3-0.5 Physiological parameters Leakage current, HRV abnormality, blood oxygen saturation γ=0.2-0.4

[0164] Table 2

[0165] In the above technical solution, the electric field strength in Table 2 is used to quantify electric field risks, monitor power frequency field strength, transient pulse strength, and gradient change rate, and capture the steady-state and transient characteristics of electric field changes during high-voltage testing. It is used as input to the prediction model (ARIMA-GRU hybrid model) of the electric field dynamic monitoring module and serves as the basis for calculating the electric field risk weight in the fuzzy comprehensive evaluation. It is also used to predict electric field strength and field strength risk membership. Among them, power frequency field strength is used to reflect steady-state electric field strength and is used to assess electric field risks during routine testing. Transient pulse strength is used to capture transient anomalies such as electromagnetic interference and equipment discharge, improving the response capability to high-risk events such as voltage flashovers. The gradient change rate is used to monitor field strength trends and provide early warning of the risk of rapid electric field increases.

[0166] In the above technical solution, the spatial distance in Table 2 is used to assess the real-time distance, movement speed, and trajectory deviation between the test personnel and the high-voltage equipment to prevent personnel from accidentally entering high-risk areas; it is used for the collaborative positioning module (UWB-IMU fusion positioning) to calculate three-dimensional coordinates in real time, combined with the dynamic safety boundary threshold to trigger distance warnings; and it is used for real-time distance risk membership and trajectory anomaly alarms. Among them, the real-time distance is used to directly map the safety distance threshold to trigger power outages. The movement speed is used to identify the behavior of personnel quickly approaching high-risk areas (such as sudden movements) to enhance adaptability to dynamic scenarios. The trajectory deviation is used to detect abnormal paths (such as deviations from the preset safe route) to prevent the risk of misoperation.

[0167] In the above technical solution, the physiological parameters in Table 2 are used to monitor leakage current (threshold ≤ 0.5mA), heart rate variability (HRV) and blood oxygen saturation (threshold ≥ 90%) to prevent physiological abnormalities caused by electric shock; for the vital signs detection module to collect data through wearable devices; and for physiological risk membership and emergency voltage reduction instructions. Among them, the leakage current is used to directly associate with the electric shock risk threshold (0.5mA), and the equipment safety device is triggered when the leakage exceeds the limit. HRV abnormal values ​​are used to reflect the state of the autonomic nervous system (such as standard deviation > 50ms) and warn of stress-related physiological reactions. Blood oxygen saturation is used to monitor the risk of hypoxia (such as respiratory depression after electric shock) and trigger voltage reduction protection (auxiliary channel).

[0168] In the above technical solution, the input multi-source data includes electric field prediction values, personnel positions, physiological parameters and equipment status, etc. The secondary indicators are normalized into fuzzy sets through membership functions (Gaussian / triangular), and the risk index is calculated in combination with dynamic weights to output the triggered three-level warning (R≥0.7 bracelet vibration, R≥0.8 sound and light alarm, R≥0.9 linkage power off).

[0169] In the above technical solution, Gaussian membership function is used to process continuous variables, and triangular membership function is used to process discrete states, and M(·,+) operator is selected for weighted averaging to avoid information loss.

[0170] In the above technical solution, the dynamic weight adjustment model proposes a dynamic weight allocation mechanism based on the test phase (pre-pressurization / steady state / depressurization), and its dynamic weight allocation process is as follows: Figure 3 As shown in the figure, the weight factor is adjusted in real time according to the on-site status such as voltage level, personnel location, and equipment status.

[0171] In the above technical solution, the high-voltage test process is divided into three stages: pre-pressurization, steady state, and power-off post-processing. The weight distribution is shown in Table 3:

[0172] Trial phase Electric field weight Distance Weight Physiological weight Historical violation weight Pre-pressurization stage 40％ 30％ 20％ 10％ Steady-state test phase 30％ 50％ 15％ 5％ Blood pressure reduction phase 20％ 30％ 40％ 10％

[0173] Table 3

[0174] In the above technical solution, the specific method of using the analytic hierarchy process (AHP) and the fuzzy comprehensive evaluation method to calculate the weight consistency ratio of each stage is: the weight consistency ratio of each stage is calculated by AHP (analytic hierarchy process). If the consistency ratio fails to pass the test, the dynamic weight matrix needs to be adjusted, and then the fuzzy comprehensive evaluation method is used to dynamically adjust the weight matrix until the weight consistency ratio of each stage reaches the set standard and the verification passes.

[0175] In the above technical solution, the LSTM-ATT (long short-term memory network-attention mechanism) model is used to parse time series data (such as voltage level, equipment status, and personnel role) in real time, and the weight matrix is ​​dynamically generated, breaking through the scenario limitations of traditional static weights. Combined with multi-dimensional consistency verification, CR≤0.1 is forced to ensure that the weights are logically consistent, avoid manual experience bias, and improve safety and computational efficiency.

[0176] In the above technical solution, the dynamic weight matrix W(t) is:

[0177] W(t)=softmax(LSTM(xt-τ,...,xt)),

[0178] Where τ represents the time window length, τ = 10s, xt is the input feature, namely the real-time collected electric field strength, the three-dimensional spatial coordinates of the test person's location label, surface leakage current and physiological parameter data, and other dynamic parameters. Softmax is used to generate attention weights, W(t) is the dynamic weight matrix, and LSTM is used to obtain time dependency.

[0179] In the above technical solution, an LSTM network receives multi-source time series data (such as voltage level, equipment status, and personnel role) within a time window of τ = 10 seconds, capturing the temporal evolution of parameters such as electric field strength and personnel location. Input features include real-time data such as predicted electric field values, three-dimensional coordinates of test personnel, surface leakage current, and physiological parameters. The output LSTM generates a sequence of hidden states that encodes the long-term dependencies of the input data. An attention mechanism weights the hidden states of the LSTM to identify the most critical time steps or features for the current risk assessment (such as transient electric field changes or personnel rapidly approaching a high-risk area). The softmax function normalizes the attention-weighted LSTM output into a probability distribution, generating a dynamic weight matrix W(t), whose elements correspond to the weight distribution of different primary indicators (electric field strength, spatial distance, and physiological parameters).

[0180] In the above technical solution, the attention mechanism (ATT) is introduced to mathematically model the calculation of the dynamic weight matrix, and the generation logic of the dynamic weight is clearly defined to ensure the reproducibility of the algorithm and the clarity of the patent protection scope; the time window data of the input features is processed by the LSTM network to capture the temporal evolution of parameters such as the electric field strength and personnel position in the test environment, and then the softmax function is used to normalize the LSTM output to generate a dynamic weight matrix, which gives higher weights to the current risk-dominant factors, thereby achieving the effect of strengthening key features and suppressing redundant information.

[0181] In the above technical solution, the specific method for obtaining the risk index by combining the predicted electric field strength, the three-dimensional spatial coordinates of the test person's location tag, the current dynamic weight distribution of the surface leakage current and the physiological parameter data with the safety distance threshold is as follows:

[0182] The predicted electric field strength, the three-dimensional spatial coordinates of the test person's location tag, the current dynamic weight distribution of the physiological parameter data are converted into a unified fuzzy set by using the membership function. The risk index is then calculated using the weighted aggregation formula. The specific formula is:

[0183]

[0184] Where i represents the input variables of different dimensions (predicted electric field strength, three-dimensional spatial coordinates of the test person's position label, body surface leakage current and physiological parameter data), n represents the total number of input variables, and w i is the current dynamic weight of each input variable, u i (x) is the membership degree of the corresponding input variable, and R is the risk index.

[0185] In the above technical solution, the output of the risk index calculation is R∈[0,1], and the graded trigger warning is as follows: when R≥0.7, the bracelet vibrates; when R≥0.8, an audible and visual alarm sounds; and when R≥0.9, the power is cut off.

[0186] In the above technical solution, parameters of different dimensions such as electric field strength, spatial distance, leakage current, etc. are converted into a unified fuzzy set (interval [0,1]) through membership functions, which solves the problem of data incomparability, breaks through the limitations of the traditional threshold method, and improves the dynamic scene adaptability.

[0187] The above technical solution also includes a risk heat map generation module, which generates a risk heat map based on the electric field strength data predicted by the ARIMA model, that is, the electric field risk distribution trend, to predict whether there are areas in the high-voltage test hall that are prohibited from entry or can only be entered after protective measures are implemented; a safety report automatic generation module, the safety report automatically records the maximum field strength value, minimum safety distance and disposal response time.

[0188] In the above technical solution, the risk heat map can be superimposed to fuse the electric field prediction results with the personnel positioning data to generate a dynamic risk heat map (the color from green to red indicates increasing risk). Figure 4 As shown in the figure, the dotted line represents the historical movement path, the pedestrian pattern is the real-time position of the experimenter, red represents the high-risk area with a field strength of >90kV / m, and green represents the safe area.

[0189] In the above technical solution, by generating risk heat maps, the electric field prediction, personnel location, and equipment status (such as circuit breaker opening and closing) are integrated and displayed, which helps to make manual decisions; by automatically generating safety reports and recording key parameters, the time error (±1 second) problem of traditional manual recording is solved; in accident investigations, the accident process can also be restored through the heat maps and positioning data in the report, and the responsibility for human errors or equipment failures can be clarified, realizing full-scene data traceability and responsibility definition.

[0190] Example 2

[0191] A warning step for preventing electric shock in high-voltage test hall based on multi-source data fusion, such as Figure 2 As shown:

[0192] S1: Test preparation and equipment initialization:

[0193] The test personnel wear wearable devices that integrate UWB tags and physiological sensors, input the test plan and maximum voltage parameters, start the 3D real-scene modeling platform, load the hall model constructed by laser SLAM, and mark the dangerous areas of the equipment.

[0194] S2: Dynamic monitoring of electric field strength:

[0195] Distributed field strength sensors collect electric field data in real time, generate field strength heat maps after calibration using the Kalman filter algorithm, and use the ARIMA model to predict the electric field distribution in the next five minutes and adjust the boundaries of the safe area.

[0196] S3: Personnel positioning and trajectory tracking:

[0197] The UWB-IMU collaborative localization algorithm calculates the 3D coordinates of people in real time, compensating for non-line-of-sight errors. Combined with historical trajectory data, it can detect abnormal movement patterns (such as staying in high-risk areas).

[0198] S4: Dynamic risk assessment and decision-making:

[0199] The fuzzy comprehensive evaluation model dynamically adjusts the weight according to the test stage, and calculates the risk index R. The pre-pressurization stage focuses on the electric field intensity, the steady-state stage focuses on the distance threshold, and the depressurization stage focuses on physiological parameters.

[0200] S5: Multi-level emergency response execution:

[0201] When R ≥ 0.9, the hard-wired controller cuts off the power (response time ≤ 150ms) and simultaneously activates the gradient lighting. The 3D platform generates a safety report, recording the maximum electric field strength, minimum safe distance, and response time.

[0202] S6: Data encryption and remote synchronization:

[0203] Test data is uploaded to the cloud after being encrypted with AES-256, supporting compliance audits, updating the equipment risk database and emergency plan library, and optimizing subsequent test plans.

[0204] Example 3

[0205] A method for preventing electric shock in a high-voltage test hall based on multi-source data fusion, comprising:

[0206] Compensating the electric field strength data of the test area collected in real time for environmental temperature drift and electromagnetic interference to obtain compensated electric field strength data, constructing a spatiotemporal gated hybrid prediction model based on the compensated electric field strength data, and inputting the electric field strength data of the test area collected in real time into the spatiotemporal gated hybrid prediction model to obtain predicted electric field strength;

[0207] The three-dimensional spatial coordinates of the experimenter's location tag are calculated by fusing the ultra-wideband positioning data sent by the ultra-wideband base station arranged in the test hall with the inertial measurement data of the inertial measurement unit in the experimenter's wearable device;

[0208] Real-time monitoring of body surface leakage current and physiological parameter data;

[0209] A digital twin model of the test hall is constructed based on a synchronous positioning and mapping method. The boundaries of the safety zone are dynamically marked based on the digital twin model of the test hall, and the safety distance threshold is dynamically adjusted based on the test voltage level.

[0210] According to the dynamic weight distribution model, the predicted electric field strength, the three-dimensional spatial coordinates of the test person's location tag, the surface leakage current and the physiological parameter data are weighted. The electric shock risk index is obtained by combining the safety distance threshold, and a graded warning is triggered according to the risk index.

[0211] Example 4

[0212] A computer program product includes a computer program, which implements the steps of the method described in Example 3 when executed by a processor.

[0213] The contents not described in detail in this specification belong to the prior art known to professional and technical personnel in this field.

Claims

1. A high-voltage test hall anti-electric shock warning system based on multi-source data fusion, characterized in that: It includes: The electric field dynamic monitoring module is used to compensate the electric field intensity data of the test area collected in real time for environmental temperature drift and electromagnetic interference, obtain the compensated electric field intensity data, build a spatiotemporal gated hybrid prediction model based on the compensated electric field intensity data, and input the real-time collected electric field intensity data of the test area into the spatiotemporal gated hybrid prediction model to obtain the predicted electric field intensity; The collaborative positioning module is used to calculate the three-dimensional spatial coordinates of the experimenter's location tag by fusing the ultra-wideband positioning data sent by the ultra-wideband base station arranged in the test hall with the inertial measurement data of the inertial measurement unit in the experimenter's wearable device; The abnormality detection module is used to monitor the body surface leakage current and physiological parameter data in real time; The 3D real-scene modeling module constructs a digital twin model of the test hall based on a synchronous positioning and mapping method. It dynamically marks the boundaries of the safe area based on the digital twin model of the test hall and dynamically adjusts the safety distance threshold based on the test voltage level. The fuzzy comprehensive evaluation decision module is used to assign weights to the predicted electric field strength, the three-dimensional spatial coordinates of the test personnel's position label, the surface leakage current, and the physiological parameter data according to the dynamic weight distribution model, and to obtain the electric shock risk index in combination with the safety distance threshold, and to trigger a graded warning based on the risk index.

2. The high-voltage test hall anti-electric shock warning system based on multi-source data fusion according to claim 1 is characterized by: It also includes a multi-level emergency linkage control module for executing test power cut-off and emergency procedure operations according to graded warnings.

3. The high-voltage test hall anti-electric shock warning system based on multi-source data fusion according to claim 1 is characterized by: The specific method for obtaining the predicted electric field strength is: The compensated electric field intensity data is used to train the spatiotemporal gated hybrid prediction model. The ARIMA unit is used to analyze the periodic and trend components of the time series to generate linear prediction values. The nonlinear features of the ARIMA output are input into the GRU neural network to capture the nonlinear fluctuation time series rules. The linear prediction value of ARIMA and the nonlinear correction value of GRU are fused to predict the future electric field distribution intensity.

4. The high-voltage test hall anti-electric shock warning system based on multi-source data fusion according to claim 1 is characterized by: The specific method for calculating the three-dimensional space coordinates of the test person's position label is: The time-of-flight method is used to calculate the distance between the tester's location tag and a single signal base station in the high-voltage test hall. The arrival time difference method is used to eliminate clock synchronization errors. The arrival angle method is used to determine the azimuth of the tester's location tag. The extended Kalman filter algorithm is used to calculate and integrate the inertial measurement unit data to compensate for non-line-of-sight errors and obtain the three-dimensional spatial coordinates of the tester's location tag. Among them, the two-way ranging formula between a single signal base station and a tag in the high-voltage test hall is: Where d is the distance between the test personnel and the ultra-wideband base station, c is the speed of light, T RTT is the signal round trip time, T TAT For the tag response time delay, a hyperbola equation is constructed based on the time difference of multiple base stations to obtain the three-dimensional spatial coordinates (x, y, z) of the test person's location tag: Where, Δt ij is the time difference of the signal arriving at the two base stations, (x i ,y i ,z i ) and (x j ,y j ,z j ) are the coordinates of the i-th and j-th base stations respectively.

5. The high-voltage test hall anti-electric shock warning system based on multi-source data fusion according to claim 1 is characterized by: The fuzzy comprehensive evaluation decision module adopts a time series prediction model based on LSTM-ATT to obtain a dynamic weight matrix; the dynamic weight allocation model adjusts the evaluation index weights of each stage in the test based on the dynamic weight matrix. The stages include the pre-pressurization stage, the steady-state stage and the depressurization stage. The hierarchical analysis method and the fuzzy comprehensive evaluation method are used to calculate the weight consistency ratio of each stage. When the weight consistency ratio of each stage reaches the set standard, the current dynamic weight distribution of the predicted electric field strength, the three-dimensional spatial coordinates of the test person's position label, the surface leakage current and the physiological parameter data is obtained.

6. The high-voltage test hall electric shock warning system based on multi-source data fusion according to claim 5 is characterized by: The dynamic weight matrix W(t) is: W(t)=softmax(LSTM(xt-τ,...,xt)), Where τ represents the length of the time window, xt is the input feature, which includes the real-time collected electric field strength, the three-dimensional spatial coordinates of the test person's location label, surface leakage current and physiological parameter data, softmax is used to generate attention weights, W(t) is the dynamic weight matrix, and LSTM is used to obtain time dependency.

7. The high-voltage test hall electric shock warning system based on multi-source data fusion according to claim 1 is characterized by: The specific method for obtaining the risk index by combining the predicted electric field strength, the three-dimensional spatial coordinates of the test person's location tag, the current dynamic weight distribution of the surface leakage current and the physiological parameter data with the safety distance threshold is as follows: The predicted electric field strength, the three-dimensional spatial coordinates of the test person's location tag, the current dynamic weight distribution of the physiological parameter data are converted into a unified fuzzy set by using the membership function. The risk index is then calculated using the weighted aggregation formula. The specific formula is: Where i represents the input variables of different dimensions, n represents the total number of input variables, namely the predicted electric field strength, the three-dimensional spatial coordinates of the test person's position label, the surface leakage current and the physiological parameter data, and w i is the current dynamic weight of each input variable, u i (x) is the membership degree of the corresponding input variable, and R is the risk index.

8. The high-voltage test hall anti-electric shock warning system based on multi-source data fusion according to claim 1 is characterized by: It also includes a risk heat map generation module, which generates a risk heat map based on the electric field intensity data predicted by the ARIMA model, that is, the electric field risk distribution trend, to predict whether there are areas in the high-voltage test hall where entry is prohibited or entry is allowed only after protective measures are in place; Automatic safety report generation module, the safety report automatically records the maximum field strength value, minimum safety distance and disposal response time.

9. A high-voltage test hall anti-electric shock warning method based on multi-source data fusion, characterized in that: It includes the following steps: Compensating the electric field strength data of the test area collected in real time for environmental temperature drift and electromagnetic interference to obtain compensated electric field strength data, constructing a spatiotemporal gated hybrid prediction model based on the compensated electric field strength data, and inputting the electric field strength data of the test area collected in real time into the spatiotemporal gated hybrid prediction model to obtain predicted electric field strength; The three-dimensional spatial coordinates of the experimenter's location tag are calculated by fusing the ultra-wideband positioning data sent by the ultra-wideband base station arranged in the test hall with the inertial measurement data of the inertial measurement unit in the experimenter's wearable device; Real-time monitoring of body surface leakage current and physiological parameter data; A digital twin model of the test hall is constructed based on a synchronous positioning and mapping method. The boundaries of the safety zone are dynamically marked based on the digital twin model of the test hall, and the safety distance threshold is dynamically adjusted based on the test voltage level. According to the dynamic weight distribution model, the predicted electric field strength, the three-dimensional spatial coordinates of the test person's location tag, the surface leakage current and the physiological parameter data are weighted. The electric shock risk index is obtained by combining the safety distance threshold, and a graded warning is triggered according to the risk index.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to claim 9 are implemented.

Images