Industrial thunder and lightning early warning method based on multi-modal data fusion and adaptive filtering

Abstract

The invention relates to the technical field of safety protection, in particular to an industrial thunder and lightning early warning method based on multi-modal data fusion and adaptive filtering, and the method comprises the steps: collecting multi-modal data in an industrial environment, preprocessing the multi-modal data, evaluating the quality of the preprocessed multi-modal data, carrying out the adaptive power frequency notch filtering processing of the multi-modal data, and carrying out the early warning of the industrial thunder and lightning. Fusing the multi-modal data, performing multi-scale feature extraction on the multi-modal data, diagnosing a fault position, protecting the stability of fault position parameter data, performing smoothing processing on output parameter data, adaptively adjusting an industrial thunder and lightning early warning threshold value, and performing judgment to realize industrial thunder and lightning early warning. The method can effectively cope with multiple interferences and structural influences in an industrial environment, and ensures that the system still has efficient data processing capability, reliable safety early warning performance and intelligent adaptive operation characteristics under complex conditions.

Description

Technical Field

[0001] This invention relates to the field of safety protection technology, specifically to an industrial lightning early warning method and system based on multimodal data fusion and adaptive filtering. Background Technology

[0002] With the development of industrial automation and smart grids, lightning disasters pose a serious threat to the safe operation of power systems, communication facilities, and large industrial installations. To achieve effective early warning of lightning activity, existing lightning detection technologies mainly rely on single-mode detection or simple fusion methods of multi-source data.

[0003] In single-modal detection schemes, the system typically relies on a single type of sensor (such as an atmospheric electric field meter, lightning locator, or weather radar) to identify and warn of lightning activity by setting a fixed threshold. This method is simple in structure and low in implementation cost, but due to the lack of cross-validation from multi-source information, its detection accuracy and anti-interference capability are weak. Especially in complex industrial environments, it is easily affected by local electromagnetic interference, equipment noise, or changes in meteorological conditions, leading to high false alarm and false negative rates.

[0004] To improve detection reliability, some systems employ simple multi-source data fusion techniques to initially integrate data from different monitoring devices such as atmospheric electric field meters, lightning location systems, and weather radars. However, existing fusion methods often use fixed-weight weighted averaging or logic-gated combinations, failing to fully consider the inherent differences in spatiotemporal resolution, response speed, measurement accuracy, and environmental adaptability among the sensors. For example, atmospheric electric field meters are extremely sensitive to power frequency interference (such as 50Hz / 60Hz industrial power supply noise), while lightning location systems have blind spots in spatial coverage, and radar echoes are easily affected by changes in the phase state of precipitation particles. Simple fusion strategies cannot dynamically adjust the weights of each sensor, leading to a significant decrease in overall system performance under complex electromagnetic environments or severe weather conditions. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide an industrial lightning early warning method based on multimodal data fusion and adaptive filtering. This method can effectively cope with multiple interferences and structural influences in the industrial environment, and ensure that the system still has efficient data processing capabilities, reliable safety early warning performance and intelligent adaptive operation characteristics under complex conditions.

[0006] To achieve the above objectives, the present invention provides the following technical solution: An industrial lightning early warning method based on multimodal data fusion and adaptive filtering includes: Multimodal data is collected from an industrial environment, preprocessed, and its quality is evaluated. Adaptive power frequency notch filtering is applied to the multimodal data, and the multimodal data is then fused. Multi-scale feature extraction is performed on the multimodal data to diagnose fault locations and protect the stability of fault location parameter data. The output parameter data is smoothed, and the industrial lightning warning threshold is adaptively adjusted and judged to realize industrial lightning warning.

[0007] In some implementations, sensors are also used to receive data from the external environment for parameter adjustment.

[0008] In some implementations, preprocessing the multimodal data and evaluating the quality of the preprocessed multimodal data specifically includes: The multimodal data were preprocessed and quality assessed using the following 3σ criterion anomaly detection formula. , : The actual observed value at time k; : The predicted observation value at time k, where L is the autoregressive order and P is the moving average order. Based on the AIC criterion, L=3 and P=2 are determined. Signal standard deviation; : Signal mean, representing the average value of the signal within the statistical window; W: Statistical window size, the number of sampling points used to calculate the standard deviation; The actual observation value at time km is from multiple sources, including atmospheric electric field meter, lightning locator, and weather radar. The actual observed value at time k is given, where L is the autoregressive order and P is the moving average order. The model parameters are updated online using the recursive least squares method. ; : The error term at time k−j, representing the deviation between the predicted value and the actual value.

[0009] In some implementations, multi-scale feature extraction is performed on the multimodal data to diagnose fault locations and protect the stability of fault location parameter data, specifically including: Step 1: Use the following formula to diagnose the location of the fault. , : Fault indicator, where 1 indicates a fault and 0 indicates normal; Fault threshold, the fault threshold range is set to 0.5~0.7; This is the health indicator for the i-th sensor or system component. Step 2: Disturbance suppression and fault tolerance processing, including disturbance detection and quantification: , : The quantized value of the perturbation energy at time k; M: Size of the sliding statistical window, used to calculate the number of historical data points for the disturbance energy; : Actual observed value at time k; The actual observed value at time k is given, where L is the autoregressive order and P is the moving average order. The model parameters are updated online using the recursive least squares method. ; : Time decay coefficient, which controls the decay rate of the influence of historical disturbances. The larger λ is, the greater the weight of recent disturbances. t: Current absolute time, system runtime; Reference time point, set as the system startup time or the time of the last major disturbance reset; Signal standard deviation.

[0010] In some implementations, the adaptive power frequency notch filtering processing of the multimodal data specifically includes: A notch filter is constructed using a second-order infinite impulse response notch filter, and the following formula is used. , : Notch filter center frequency; Sampling frequency; r: Zero-point radius, controls the notch depth; R: Pole radius, controls the notch width; Normalized digital angular frequency, used in digital signal processing; The following power estimation function is used to estimate and update power values ​​in real time: , Interference power estimate; N: Number of sampling points; Window size; x(n): The input signal at the nth sampling point; : Indicator function, is 1 within the frequency range [f_min, f_max], and 0 otherwise; f: Frequency variable; The following formula is used to adaptively update the parameters, changing the parameter r in real time according to the signal-to-noise ratio: , : The radius of the zero point in the k-th iteration; Learning rate (step size); , , Weighting coefficients; : Hyperbolic tangent function, Mapped to the range (-1, 1); Signal-to-noise ratio; : Indicates signal power; : Indicates noise power; : Indicates total power; The following formula is used to set the frequency tracking parameters, and the center frequency value is changed in real time to adjust the interference frequency tracking parameters: , : The center frequency of the k-th iteration; : Frequency learning rate; Cost function; The partial derivative of the cost function with respect to the center frequency; Variable step size learning: , The learning rate of the i-th parameter in the k-th iteration; The change in the cost function; : Control parameters; Filtering performance metrics, used to quantitatively evaluate a filter's ability to suppress power frequency interference: , IRR: Interference Rejection Ratio, a measure of notch depth; the higher the IRR value, the better. Frequency response amplitude; : The maximum amplitude value excluding the center frequency.

[0011] In some implementations, the multimodal data fusion specifically includes: Adaptive weight calculation performs adaptive weight allocation for each sensor: , : The noise variance of the i-th sensor; : Signal-to-noise ratio of the i-th sensor; Steepness coefficient of the S-shaped function; Signal-to-noise ratio threshold; The spatiotemporal consistency constraint formula for multi-source data fusion is used to unify the time errors of various sensors used for data acquisition: ; Time alignment compensation; Environmental compensation factors; Dynamic weights; : The reference signal at time k; : The measurement value of the i-th sensor at time k−τ; Multi-objective optimization involves balancing various indicators across multiple objectives. , θ: The vector of parameters to be optimized; J(θ): Total cost function; α, β, γ: weighting coefficients; Performance cost item; Complexity cost term; Stability cost term; Stability protection is achieved using the following formula. ,, : The single-step change of the i-th parameter; : Maximum allowable variation of parameters; : Relative rate of change coefficient; The parameter value at the previous moment serves as a benchmark reference for the relative rate of change. , The fused signal estimate serves as the final early warning input. : The dynamic weight of the i-th sensor; Time alignment compensation.

[0012] In some implementations, the fused parameter data undergoes output processing and smoothing, specifically including: The following formula enables a smoothing mechanism for the output. , : The smoothed output value at time k; : The original output value at time k. : The smoothed output value from the previous time step. Smoothing factor.

[0013] In some implementations, the industrial lightning warning threshold is adaptively adjusted and determined to achieve industrial lightning warning, specifically including: , High-risk threshold, where γ is the sensitivity coefficient and ϵ(k) is the disturbance energy, which automatically increases the threshold when there is a large amount of environmental interference, thereby reducing false alarms; Low-risk threshold; : is a fixed or slowly updated base value.

[0014] In some implementations, the multimodal data is collected using at least one of an atmospheric electric field meter, a lightning locator, and a weather radar.

[0015] An electronic device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method described above.

[0016] The beneficial effects of this invention, which proposes an industrial lightning early warning method and electronic device based on multimodal data fusion and adaptive filtering, are as follows: This invention discloses an industrial lightning warning method based on multimodal data fusion and adaptive filtering. It dynamically adjusts parameters through an adaptive algorithm to address electromagnetic interference in industrial environments and integrates information from multiple sensors to improve warning accuracy. The overall architecture embodies closed-loop processing from data acquisition to decision output, ensuring stable system operation under complex conditions. Therefore, multi-source data undergoes signal preprocessing, health assessment, adaptive filtering, multimodal fusion, and output smoothing sequentially, with parameters dynamically adjusted throughout by an adaptive algorithm. This approach balances strong electromagnetic interference suppression with multi-sensor collaboration, achieving highly stable and accurate lightning warnings in complex industrial scenarios. Attached Figure Description

[0017] Figure 1 This is a flowchart of an industrial lightning early warning method based on multimodal data fusion and adaptive filtering, as described in this embodiment. Detailed Implementation

[0018] Preferred embodiments of the invention will now be described in more detail. While preferred embodiments of the invention have been shown, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0019] The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular forms “a” and “the” as used in this invention and the appended claims are intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0020] The lightning monitoring system is currently experiencing the following problems: Poor anti-interference capability: It lacks a dedicated suppression mechanism for strong power frequency interference in industrial environments, resulting in poor electric field signal quality and affecting the accuracy of threshold judgment.

[0021] The data fusion effect is not good: the data from various sensors have different dimensions and lack a unified standardized processing framework; fixed weights cannot reflect the dynamic changes in the reliability of each data source under different weather scenarios.

[0022] Insufficient environmental adaptability: The algorithm parameters are fixed and cannot be adaptively adjusted according to environmental changes (such as humidity, temperature, electromagnetic background noise, etc.), resulting in large fluctuations in early warning performance and low system stability and reliability in complex industrial scenarios.

[0023] To solve the above-mentioned technical problems, the following embodiments are disclosed: See Figure 1 , Figure 1 This embodiment discloses an industrial lightning early warning method based on multimodal data fusion and adaptive filtering. Multimodal data is collected from an industrial environment. This multimodal data comes from multiple sources of sensors, including atmospheric electric field meters, lightning location devices, and weather radar. The multimodal data is preprocessed, its quality is evaluated, adaptive power frequency notch filtering is applied, and the multimodal data is fused. Multi-scale feature extraction is performed on the multimodal data to diagnose fault locations and protect the stability of fault location parameter data. The output parameter data is smoothed, and the industrial lightning warning threshold is adaptively adjusted and judged to realize industrial lightning warning.

[0024] Signal preprocessing and quality assessment are performed in the following manner: The 3σ criterion anomaly detection formula is used to detect anomalies in the environment and reduce the false positive rate. , The actual observation value at time k comes from multiple sources, including atmospheric electric field meters, lightning locators, and weather radar. The predicted observation value at time k is given by L in the prediction formula, where L is the autoregressive order and P is the moving average order. Based on the AIC criterion, L=3 and P=2 are determined. Signal standard deviation; : Signal mean, representing the average value of the signal within the statistical window; W: Statistical window size, the number of sampling points used to calculate the standard deviation; The actual observation value at time km is from multiple sources, including atmospheric electric field meter, lightning locator, and weather radar. The actual observed value at time k is used in the prediction formula, where L is the autoregressive order and P is the moving average order. The model parameters are updated online using the recursive least squares method. ; The error term at time k-j represents the deviation between the predicted and actual values. Health and reliability assessments are conducted in the following manner: Reliability assessment includes fault detection and diagnosis, parameter change protection, and ensuring output quality. , Fault flags (1 indicates a fault, 0 indicates normal); Fault threshold (usually set to 0.5-0.7); It is the health index of the i-th sensor or system component, a quantitative parameter used to evaluate the operating status and reliability level of the sensor or system component; Disturbance suppression and fault tolerance processing, including disturbance detection and quantification: , The disturbance energy quantization value at time k includes the energy assessment of abnormal events such as sudden electromagnetic pulses and instantaneous sensor failures. It represents the degree of combined impact of external disturbances and internal anomalies on the system at the current sampling time. It is used to trigger different levels of fault tolerance processing strategies and is a key indicator for system stability assessment. M: Sliding statistical window size, used to calculate the number of historical data points for perturbation energy, reflecting the system's memory length. It needs to be balanced between real-time performance and statistical stability. If M is too small, it will lead to large estimation fluctuations, and if M is too large, it will reduce the response speed. The actual observation value at time km is obtained from multiple sources, including atmospheric electric field meters, lightning locators, and weather radar. The actual observed value at time k is used in the prediction formula, where L is the autoregressive order and P is the moving average order. The model parameters are updated online using the recursive least squares method. ; The time decay coefficient controls the decay rate of the impact of historical disturbances, reflecting the "recency effect". The larger λ is, the greater the weight of recent disturbances, and the more sensitive the system is to sudden disturbances. t: Current absolute time, system runtime; The reference time point is usually set to the system startup time or the time of the last major disturbance reset. Signal standard deviation; Adaptive filtering of signals is performed in the following manner: A second-order infinite impulse response (IIR) notch filter is used to construct a notch filter to suppress power frequency interference. , The first formula is derived from the initial continuous-time transfer function through a bilinear transformation. This formula represents the transfer function of a second-order infinite impulse response (IIR) notch filter in the z-domain. The numerator determines the location of the filter's zeros, responsible for generating notches at specific frequencies; the denominator determines the location of the poles, controlling the filter's bandwidth characteristics.

[0025] : Notch filter center frequency; Sampling frequency; r: Zero-point radius, controls the notch depth; R: Pole radius, controls the notch width; Normalized digital angular frequency, used in digital signal processing; Power estimation functions are used to estimate and update power values ​​in real time. , Interference power estimate; N: Number of sampling points; Window size; x(n): The input signal at the nth sampling point; : Indicator function, is 1 within the frequency range [f_min, f_max], and 0 otherwise; f: Frequency variable; Adaptive parameter updates are used to change the parameters of r in real time based on the signal-to-noise ratio: , The radius of the zero point in the k-th iteration; Learning rate (step size) controls the update magnitude; , , Weighting coefficients balance different influencing factors; : Hyperbolic tangent function, Mapped to the range (-1, 1); Signal-to-noise ratio; : Indicates signal power; : Indicates noise power; : Indicates total power; The center frequency value is changed in real time to adjust the interference frequency tracking parameters using frequency tracking parameters:

[0026] : The center frequency of the k-th iteration; : Frequency learning rate; Cost function, used to measure system performance; The partial derivative of the cost function with respect to the center frequency; Variable step size learning increases learning speed and accelerates system response: , The learning rate of the i-th parameter in the k-th iteration; The change in the cost function; : Control parameters, which can be adjusted as needed; Filtering performance metrics, used to quantitatively evaluate a filter's ability to suppress power frequency interference: , IRR: Interference Rejection Ratio, a measure of notch depth. The higher the IRR value, the stronger the filter's ability to suppress power frequency interference. Frequency response amplitude; The maximum amplitude value excluding the center frequency; Multimodal data fusion Adaptive weight calculation performs adaptive weight allocation for each sensor: , : The noise variance of the i-th sensor; : Signal-to-noise ratio of the i-th sensor; Steepness coefficient of the S-shaped function; Signal-to-noise ratio threshold; The spatiotemporal consistency constraint formula for multi-source data fusion unifies the time errors of various sensors: ,, Time alignment compensation solves sensor response delay; Environmental compensation factor, which corrects for environmental effects such as temperature and humidity; Dynamic weighting, based on sensor reliability allocation; The reference signal at time k (e.g., the average value from multiple sensors); The measurement value of the i-th sensor at time k−τ; Multi-objective optimization involves balancing various indicators across multiple objectives. ,, θ: Parameter vector to be optimized J(θ): Total cost function α, β, γ: Weighting coefficients Performance cost item; Complexity cost term; Stability cost term; Stability protection, as the system's last line of defense, ensures system operation under strict conditions: , The single-step change of the i-th parameter represents the adjustment range of the parameter between adjacent sampling times, reflecting the update speed of the control parameters and avoiding overshoot and oscillation. The maximum allowable variation of parameters is a hard limit based on system stability and safety requirements; The relative rate of change coefficient represents the maximum percentage change of the control parameter relative to its current value. Sensitive parameters take a small value, while robust parameters can take a large value. The parameter value at the previous moment serves as a benchmark reference for the relative rate of change. This formula ensures that parameter changes satisfy both absolute and relative constraints, preventing system instability caused by sudden parameter changes. When the parameter value is relatively small, the relative constraint plays a major role; when the parameter value is relatively large, the absolute constraint plays a major role.

[0027] ,, The fused signal estimate serves as the final early warning input; : The dynamic weight of the i-th sensor; Time alignment compensation; Output processing and smoothing The output smoothing mechanism makes the output smoother and reduces the impact of interference. , The smoothed output value at time k, after filtering, is the final system output and serves as the input signal for the early warning decision module. Although it has phase lag, it significantly reduces output fluctuations. The original output value at time k is composed of the real signal, noise, and interference, and usually follows a certain probability distribution, such as a Gaussian distribution or a heavy-tailed distribution.

[0028] The smoothed output value from the previous moment reflects the weighted cumulative effect of historical outputs.

[0029] Smoothing factor: controls the weighting of historical data and current data, and has adaptive adjustment, increasing the smoothing degree when the disturbance is large; Adaptive threshold adjustment is performed in the following manner: ,, High-risk threshold, where γ is the sensitivity coefficient and ϵ(k) is the disturbance energy, which automatically increases the threshold when there is a large amount of environmental interference, thereby reducing false alarms; Low-risk threshold, typically set as a multiple of the background noise level. : This is a fixed or slowly updated base value, usually based on historical data statistics.

[0030] This invention employs power frequency notch filtering technology and real-time parameter updates to dynamically suppress power frequency interference in industrial environments.

[0031] This embodiment achieves collaborative optimization of multi-sensor data, automatically adjusting parameters under different weather conditions. Adaptive filtering and multimodal fusion reduce signal noise, improving the early identification accuracy of lightning activity. Fault tolerance mechanisms and stability protection ensure minimal system fluctuations in complex electromagnetic environments. It can cope with power frequency interference from high-voltage equipment and the shielding effect of metal structures. Intelligent algorithms reduce false alarms and missed alarms, lowering maintenance costs and minimizing the waste of emergency resources.

[0032] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An industrial lightning early warning method based on multimodal data fusion and adaptive filtering, characterized in that, include: Multimodal data is collected from an industrial environment, preprocessed, and its quality is evaluated. Adaptive power frequency notch filtering is applied to the multimodal data, and the multimodal data is then fused. Multi-scale feature extraction is performed on the multimodal data to diagnose fault locations and protect the stability of fault location parameter data. The output parameter data is smoothed, and the industrial lightning warning threshold is adaptively adjusted and judged to realize industrial lightning warning.

2. The industrial lightning early warning method based on multimodal data fusion and adaptive filtering according to claim 1, characterized in that, It also uses sensors to receive data from the external environment, which is then transmitted to an adaptive power frequency notch filter for parameter adjustment.

3. The industrial lightning early warning method based on multimodal data fusion and adaptive filtering according to claim 1, characterized in that, Preprocessing the multimodal data and evaluating the quality of the preprocessed multimodal data specifically includes: The multimodal data were preprocessed and quality assessed using the following 3σ criterion anomaly detection formula. ， : The actual observed value at time k; : The predicted observation value at time k, where L is the autoregressive order and P is the moving average order. Based on the AIC criterion, L=3 and P=2 are determined. Signal standard deviation; : Signal mean, representing the average value of the signal within the statistical window; W: Statistical window size, the number of sampling points used to calculate the standard deviation; The actual observation value at time km is from multiple sources, including atmospheric electric field meter, lightning locator, and weather radar. The actual observed value at time k is given, where L is the autoregressive order and P is the moving average order. The model parameters are updated online using the recursive least squares method. ; : The error term at time k−j, representing the deviation between the predicted value and the actual value.

4. The industrial lightning early warning method based on multimodal data fusion and adaptive filtering according to claim 1, characterized in that, Multi-scale feature extraction is performed on the multimodal data to diagnose fault locations and protect the stability of fault location parameter data, specifically including: Step 1: Use the following formula to diagnose the location of the fault. ， : Fault indicator, where 1 indicates a fault and 0 indicates normal; Fault threshold, the fault threshold range is set to 0.5~0.7; This is the health indicator for the i-th sensor or system component. Step 2: Disturbance suppression and fault tolerance processing, including disturbance detection and quantification: ， : The quantized value of the perturbation energy at time k; M: Size of the sliding statistical window, used to calculate the number of historical data points for perturbation energy; : Actual observed value at time k; The actual observed value at time k is given, where L is the autoregressive order and P is the moving average order. The model parameters are updated online using the recursive least squares method. ; : Time decay coefficient, which controls the decay rate of the influence of historical disturbances. The larger λ is, the greater the weight of recent disturbances. t: Current absolute time, system runtime; Reference time point, set as the system startup time or the time of the last major disturbance reset; Signal standard deviation.

5. The industrial lightning early warning method based on multimodal data fusion and adaptive filtering according to claim 1, characterized in that, The adaptive power frequency notch filtering process for the multimodal data specifically includes: A notch filter is constructed using a second-order infinite impulse response notch filter, and the following formula is used. ， Notch filter center frequency; Sampling frequency; r: Zero-point radius, controls the notch depth; R: Pole radius, controls the notch width; Normalized digital angular frequency, used in digital signal processing; The following power estimation function is used to estimate and update power values ​​in real time: ， Interference power estimate; N: Number of sampling points; Window size; x(n): The input signal at the nth sampling point; : Indicator function, is 1 within the frequency range [f_min, f_max], and 0 otherwise; f: Frequency variable; The following formula is used to adaptively update the parameters, changing the parameter r in real time according to the signal-to-noise ratio: ， : The radius of the zero point in the k-th iteration; Learning rate (step size); , , Weighting coefficients; The hyperbolic tangent function, Mapped to the range (-1, 1); Signal-to-noise ratio; : Indicates signal power; : Indicates noise power; : Indicates total power; The following formula is used to set the frequency tracking parameters, and the center frequency value is changed in real time to adjust the interference frequency tracking parameters: , : The center frequency of the k-th iteration; : Frequency learning rate; Cost function; The partial derivative of the cost function with respect to the center frequency; Variable step size learning: , The learning rate of the i-th parameter in the k-th iteration; The change in the cost function; : Control parameters; Filtering performance metrics, used to quantitatively evaluate a filter's ability to suppress power frequency interference: ， IRR: Interference Rejection Ratio, a measure of notch depth; the higher the IRR value, the better. Frequency response amplitude; : The maximum amplitude value excluding the center frequency.

6. The industrial lightning early warning method based on multimodal data fusion and adaptive filtering according to claim 1, characterized in that, The multimodal data fusion specifically includes: Adaptive weight calculation performs adaptive weight allocation for each sensor: ， : The noise variance of the i-th sensor; : Signal-to-noise ratio of the i-th sensor; Steepness coefficient of the S-shaped function; Signal-to-noise ratio threshold; The spatiotemporal consistency constraint formula for multi-source data fusion is used to unify the time errors of various sensors used for data acquisition: ； Time alignment compensation; Environmental compensation factors; Dynamic weights; : The reference signal at time k; : The measurement value of the i-th sensor at time k−τ; Multi-objective optimization involves balancing various indicators across multiple objectives. ； θ: The vector of parameters to be optimized; J(θ): Total cost function; α, β, γ: weighting coefficients; Performance cost item; Complexity cost term; Stability cost term; Stability protection is achieved using the following formula: ， The single-step change of the i-th parameter; : Maximum allowable variation of the parameter; : Relative rate of change coefficient; The parameter value at the previous moment serves as a benchmark reference for the relative rate of change. ， The fused signal estimate serves as the final early warning input. : The dynamic weight of the i-th sensor; Time alignment compensation.

7. The industrial lightning early warning method based on multimodal data fusion and adaptive filtering according to claim 1, characterized in that, The output data of the fused parameters is processed and smoothed, specifically including: The following formula enables a smoothing mechanism for the output. ， : The smoothed output value at time k; : The original output value at time k. : The smoothed output value from the previous time step. Smoothing factor.

8. The industrial lightning early warning method based on multimodal data fusion and adaptive filtering according to claim 1, characterized in that, Adaptively adjust and determine the industrial lightning warning threshold to achieve industrial lightning warning, specifically including: ， High-risk threshold, where γ is the sensitivity coefficient and ϵ(k) is the disturbance energy, which automatically increases the threshold when there is a large amount of environmental interference, thereby reducing false alarms; Low-risk threshold; : is a fixed or slowly updated base value.

9. The industrial lightning early warning method based on multimodal data fusion and adaptive filtering according to claim 2, characterized in that, The sensor includes at least one of an atmospheric electric field meter, a lightning locator, and a weather radar.

10. An electronic device, characterized in that, The invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the computer program, implements the method as described in any one of claims 1 to 9.

Images