A Fire and Explosion Early Warning Method for Chemical Industrial Parks Based on Multimodal Data Fusion

Abstract

This invention belongs to the field of fire early warning technology, specifically disclosing a fire and explosion early warning method for chemical industrial parks based on multimodal data fusion. The method includes: collecting time-series temperature data and concentration time-series data of at least one combustible or toxic gas within a monitoring area of ​​the chemical industrial park; generating synchronized temperature and gas concentration sequences after spatiotemporal alignment; calculating the temperature change rate sequence and gas concentration gradient distribution sequence; aligning the two within the same sliding time window, calculating the correlation, and determining positive or negative coupling based on the correlation, thereby determining the weights of the first and second input vectors respectively; calculating a comprehensive explosion risk index based on the weighted two vectors, and triggering an early warning when the index reaches an early warning threshold. This invention, by quantifying the coupling relationship between temperature changes and concentration gradient changes, can distinguish between fire precursors and normal process temperature rise, effectively reducing false alarm and missed alarm rates.

Description

Technical Field

[0001] This invention belongs to the field of fire early warning technology, and specifically relates to a fire and explosion early warning method for chemical industrial parks based on multimodal data fusion. Background Technology

[0002] Chemical industrial parks typically deploy temperature sensors, combustible / toxic gas sensors, and video surveillance equipment to collect process parameters, environmental concentrations, and on-site images, respectively. Each sensor channel is independently configured with fixed threshold alarm conditions. When the measured value of any channel exceeds its threshold, the corresponding alarm is triggered, thereby ensuring the fire safety of the park.

[0003] However, in the early stages of leaks or smoldering in chemical industrial parks, there are physical spatiotemporal differences between heat transfer and gas diffusion. Thermal response typically precedes concentration response, and rising temperature accelerates gas diffusion, with gas concentration distribution influenced by the temperature field. Currently, temperature and gas sensors are deployed independently, with varying sampling frequencies and spatial locations. This makes it impossible to simultaneously acquire the evolution of the temperature and gas concentration fields under the same spatiotemporal reference. Consequently, the positive coupling relationship between temperature and gas diffusion cannot be identified and utilized, while the negative decoupling relationship—where there is only a temperature change without a gas response—cannot be detected. This results in normal temperature rises being falsely reported as fire alarms, while actual leaks are missed because both signals are below the threshold. Summary of the Invention

[0004] In view of this, in order to solve the above problems, a fire and explosion early warning method based on multimodal data fusion is proposed for chemical industrial parks.

[0005] The objective of this invention can be achieved through the following technical solution: This invention provides a method for early warning of fire and explosion in chemical industrial parks based on multimodal data fusion. The method includes: collecting time-series temperature data and time-series concentration data of at least one combustible gas or toxic gas within the monitoring area of ​​the chemical industrial park, and performing spatiotemporal alignment to generate a spatiotemporally synchronized temperature sequence and gas concentration sequence.

[0006] The temperature sequence is differentially calculated to obtain the temperature change rate sequence, and the gas concentration sequence is spatially gradient calculated to obtain the gas concentration gradient distribution sequence.

[0007] The temperature change rate sequence and the gas concentration gradient distribution sequence are aligned within the same sliding time window and used as the first input vector and the second input vector, respectively. The correlation between the two within the time window is calculated, and the coupling type is determined based on the correlation. The coupling type is either positive coupling or negative coupling, and the weights of the first input vector and the second input vector are determined based on the coupling type.

[0008] Based on the weighted first and second input vectors, a comprehensive explosion risk index is calculated. If the comprehensive risk index is greater than or equal to the warning threshold, a fire and explosion warning command is triggered.

[0009] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention achieves accurate synchronous characterization of temperature field and gas concentration field by aligning temperature time series data and gas concentration time series data in time and space, thereby providing a reliable data foundation for subsequent multimodal data fusion analysis and avoiding misjudgment of coupling relationship caused by time and space misalignment.

[0010] (2) This invention quantifies the correlation strength between temperature change and concentration gradient change by calculating the correlation between the temperature change rate sequence and the gas concentration gradient distribution sequence and determining positive or negative coupling based on the correlation. This solves the problem that the positive coupling relationship between temperature and gas diffusion cannot be identified and utilized. It can accurately identify the negative decoupling relationship of normal process temperature rise, and realize the accurate distinction between fire precursors and normal process fluctuations and the collaborative processing between multiple physical fields.

[0011] (3) This invention assigns differentiated weights to the temperature change rate and gas concentration gradient according to different coupling types, and then integrates them into a comprehensive explosion risk index as an early warning criterion. Compared with the traditional approach of setting fixed thresholds for each sensor independently, this integrated early warning mechanism can promptly capture potential risks based on abnormal changes in the coupling relationship between the two fields when a single signal has not yet reached the alarm threshold. At the same time, it avoids misjudging normal process temperature rise as a fire, thereby reducing the false alarm rate and effectively reducing the missed alarm rate, thus improving the overall reliability and response timeliness of fire and explosion early warning in chemical industrial parks. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the overall implementation process of the method of the present invention; Figure 2 This is a schematic diagram of the process for determining the coupling type of the present invention; Figure 3 This is a schematic diagram of the calculation process for the comprehensive explosion risk index of this invention. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] Currently, temperature data and gas concentration data are usually treated as two independent variables and judged separately, with fixed threshold alarm conditions set for each. When facing a scenario where the thermal response of a leak or smoldering initial stage precedes the concentration response, the temperature rise will accelerate gas diffusion, and the two types of signals will produce positively coupled fluctuations with similar shapes. Existing technology may misjudge them as two independent events exceeding the threshold or neither event exceeding the threshold, resulting in false alarms or missed alarms and a high false detection rate.

[0015] Based on this, this invention discloses a method for early warning of fire and explosion risks in chemical industrial parks based on multimodal data fusion, which can effectively improve the accuracy of early warning of fire and explosion risks.

[0016] Please refer to the details. Figure 1 As shown, the present invention provides a method for early warning of fire and explosion in chemical industrial parks based on multimodal data fusion. The method includes: S1, collecting time-series temperature data and time-series concentration data of at least one combustible gas or toxic gas within the monitoring area of ​​the chemical industrial park, and performing spatiotemporal alignment to generate a spatiotemporally synchronized temperature sequence and gas concentration sequence.

[0017] Temperature time-series data refers to the sequence of temperature values ​​over time, continuously collected by multiple temperature sensors deployed within the monitoring area of ​​the chemical industrial park at their respective sampling frequencies. Gas concentration time-series data refers to the sequence of concentration values ​​over time, continuously collected by multiple gas concentration sensors deployed within the same monitoring area at their respective sampling frequencies, for at least one combustible or toxic gas.

[0018] Understandably, the combustible gases include, but are not limited to, hydrocarbon combustible gases such as methane, hydrogen, ethane, ethylene, propylene, and acetylene. The toxic gases include, but are not limited to, carbon monoxide, hydrogen sulfide, chlorine, ammonia, hydrogen chloride, and sulfur dioxide. In specific implementation, appropriate types of gas sensors can be deployed within the above range based on the actual process characteristics and hazardous substance list of the chemical industrial park.

[0019] Since the temperature sensor and the gas concentration sensor have different sampling frequencies and spatial locations, spatiotemporal alignment of the two types of data is required before data fusion.

[0020] During time alignment, the lower sampling frequency of the temperature sensor and the gas concentration sensor is used as the target sampling frequency, and interpolation resampling is performed on the temperature time series data and the gas concentration time series data respectively.

[0021] Intelligibly, resampling refers to downsampling data with a sampling frequency higher than the target sampling frequency and upsampling data with a sampling frequency lower than the target sampling frequency, so that the two types of data have a unified sampling time in the time dimension.

[0022] It should be noted that downsampling refers to extracting data values ​​from the original data sequence at the time points corresponding to the target sampling frequency. Upsampling refers to supplementing data values ​​at intermediate time points between adjacent sampling points in the original data sequence through linear interpolation.

[0023] Furthermore, during spatial alignment, a planar grid of the monitoring area is first constructed. This involves obtaining the spatial coordinates of all sensors, taking the minimum value of the horizontal coordinates of all sensors as the left boundary and the maximum value as the right boundary, and taking the minimum value of the vertical coordinates of all sensors as the lower boundary and the maximum value as the upper boundary. The above boundaries form a rectangular area covering all sensors as the initial boundary. The initial boundary is then extended outward by a distance of one grid step as the final boundary. Within the final boundary range, the monitoring area is divided into several grid units according to a preset grid step. All grid units together constitute a planar grid, and the grid step is taken as 0.3 to 0.5 times the average spacing between the sensors.

[0024] Subsequently, for each sampling time, the temperature data and gas concentration data of each sensor corresponding to that time after resampling are mapped to each grid node of the planar grid using the inverse distance weighted interpolation method.

[0025] The specific method of mapping to each grid node of the planar grid using the inverse distance weighted interpolation method is as follows: For each grid node, a sensor acquisition point within a preset search radius is found based on its spatial coordinates. The inverse distance weighted interpolation method is then used to calculate the temperature interpolation data and gas concentration interpolation data for that grid node at that sampling time. The exponent of the inverse distance weighted interpolation is set to 2, meaning that the weight of each sensor acquisition point participating in the interpolation is inversely proportional to the square of the distance from that point to the grid node; the search radius is 1.5 times the average spacing between the sensors. This process is repeated for all sampling times, ensuring that each grid node obtains the corresponding temperature and gas concentration values ​​at each sampling time.

[0026] Finally, after all grid nodes have been interpolated, the temperature and gas concentration values ​​of the same grid node at the same sampling time are extracted, arranged in chronological order, and spatiotemporally synchronized temperature and gas concentration sequences are generated respectively.

[0027] It is important to note that when no sensor data points are found within the preset search radius of a grid node, the interpolated data value for that node is determined using the following extrapolation strategy: Using the current grid node as the center, the search radius is gradually expanded to 1.5 times, then 2 times the preset search radius, until at least one sensor data point is found within the search radius. If no sensor is found after expanding the search radius to 4 times the average sensor spacing, the arithmetic mean of the measurements taken by all sensors within the entire monitoring area at the same time is taken as the interpolated data value for that grid node. For grid nodes located in the boundary region of a planar grid, if their preset search radius partially exceeds the boundary of the monitoring area, only sensor data points inside the boundary are used in the interpolation calculation when calculating the distance. If no internal sensor is found within the search radius of the boundary node, the interpolated data value for that boundary node is assigned the interpolated data value of the nearest grid node inside the boundary.

[0028] This step achieves precise synchronous characterization of the temperature field and the gas concentration field by aligning the temperature time series data and the gas concentration time series data in time and space. This provides a reliable data foundation for subsequent multimodal data fusion analysis and avoids misjudgment of coupling relationship caused by time and space misalignment.

[0029] S2. Perform differential calculation on the temperature sequence to obtain the temperature change rate sequence, and perform spatial gradient calculation on the gas concentration sequence to obtain the gas concentration gradient distribution sequence.

[0030] It should be noted that chemical industrial park fires and explosions are usually preceded by abnormal temperature rises and gas leakage and diffusion. The rate of temperature change characterizes the dynamic trend of the thermal field, while the gas concentration gradient characterizes the degree of spatial non-uniformity in gas concentration distribution. In the actual development of a fire, temperature rise accelerates gas diffusion, and the spatial distribution of gas concentration is strongly correlated with temperature changes in both time and space. However, in non-fire scenarios such as normal process temperature rises, there is only a temperature change without a corresponding gas concentration gradient response, indicating a negative decoupling relationship between the two. This invention distinguishes between fire precursors and non-fire abnormal temperature rises by quantifying the degree of coupling between the rate of temperature change and the gas concentration gradient.

[0031] Specifically, the temperature change rate sequence is obtained as follows: For each moment in the spatiotemporally synchronized temperature sequence, the temperature difference between that moment and the previous moment is calculated, and divided by the corresponding time interval to obtain the temperature change rate at that moment. After traversing all moments, all temperature change rates are arranged in chronological order to form a temperature change rate sequence.

[0032] Specifically, the gas concentration gradient distribution sequence is obtained through the following steps: S21, obtain the spatial coordinates of all gas sensors in the chemical industrial park and their real-time concentration measurements, and divide the monitoring area into high-risk densification zone, medium-risk normal deployment zone and low-risk sparse zone according to the fire and explosion risk level of each sensor location. Among them, the high-risk level corresponds to the high-risk densification zone, the medium-risk level corresponds to the medium-risk normal deployment zone, and the low-risk level corresponds to the low-risk sparse zone.

[0033] Understandably, the fire and explosion risk levels are predetermined as follows: the locations of each sensor are spatially associated with risk units in the risk classification and control list of the chemical industrial park. This risk classification and control list is pre-defined based on the types of flammable and explosive materials, material inventory, equipment operating temperature, and pressure levels involved in each area, and is divided into high-risk, medium-risk, and low-risk units. The risk unit level corresponding to the sensor's location is the fire and explosion risk level of that sensor.

[0034] S22. Take each gas sensor as the target sensor in turn, and determine the set of adjacent sensors based on the type of the region to which the current target sensor is located.

[0035] Understandably, different area types are pre-associated with adjacent distance thresholds. Among them, the adjacent distance threshold corresponding to the high-risk encrypted area is the smallest, followed by the medium-risk normal deployment area, and the low-risk sparse area is the largest. As a preferred example, the adjacent distance threshold is 5 meters for the high-risk encrypted area, 10 meters for the medium-risk normal deployment area, and 20 meters for the low-risk sparse area.

[0036] S23. Calculate the concentration difference between the target sensor and each gas sensor in the adjacent sensor set in sequence, and divide it by the Euclidean distance between the installation position of the target sensor and the corresponding gas sensor installation position in the adjacent sensor set to obtain the spatial concentration gradient component in that direction.

[0037] S24. Perform vector synthesis of the spatial concentration gradient components in all directions to obtain the total concentration gradient vector at the target sensor, calculate the amplitude of the total vector, and use it as the gas concentration gradient value of the current target sensor.

[0038] The formula for calculating the magnitude of the total vector is: .

[0039] In the formula, The magnitude of the total vector represents the rate of change of concentration at the target sensor location in the direction of the most dramatic spatial variation, serving as the gas concentration gradient value for that target sensor. (Symbol) This indicates the operation of taking the magnitude of a vector. This represents the total number of sensors in the adjacent sensor set. For the first Spatial concentration gradient components in each direction.

[0040] S25. Arrange the gas concentration gradient values ​​of all target sensors according to the sensor installation location to generate a gas concentration gradient distribution sequence.

[0041] It should be added that, in the specific arrangement, the gas concentration gradient values ​​of all target sensors are arranged in ascending order according to the X-axis coordinate of the sensor installation location, and if the X-axis coordinates are the same, they are arranged in ascending order according to the Y-axis coordinate.

[0042] S3. Align the temperature change rate sequence and the gas concentration gradient distribution sequence within the same sliding time window, and use them as the first input vector and the second input vector, respectively. Calculate the correlation between the two within the time window, determine the coupling type based on the correlation, and determine whether the coupling type is positive coupling or negative coupling. Then, determine the weights of the first input vector and the second input vector based on the coupling type.

[0043] The length of the sliding time window is determined as follows: A1. Obtain the response time constant of the temperature sensor and the response time constant of the gas concentration sensor, respectively, and denoted as the first time constant and the second time constant. The larger of the two values ​​is used as the initial length of the sliding time window. The response time constant characterizes the sensor’s response speed to a step change in the input signal, and its value is provided by the sensor manufacturer.

[0044] A2. Calculate the variance of the temperature change rate sequence and the variance of the gas concentration gradient sequence within the time window intercepted with the initial length as the window length and the current sampling time as the window end.

[0045] A3. If the variance of the temperature change rate sequence exceeds the preset first fluctuation threshold, the average of the first time constant and the second time constant will be used as the length of the sliding time window.

[0046] A4. If the variance of the temperature change rate sequence does not exceed the first fluctuation threshold while the variance of the gas concentration gradient sequence exceeds the preset second fluctuation threshold, then the sum of the first time constant and the second time constant shall be used as the length of the sliding time window.

[0047] A5. If the variances of both do not exceed the corresponding fluctuation thresholds, then the larger of the first time constant and the second time constant is taken as the length of the sliding time window.

[0048] It should be added that the first and second fluctuation thresholds are predetermined as follows: Temperature change rate sequences and gas concentration gradient sequences from multiple historical time periods within the chemical industrial park, assuming no sensor alarms are triggered. For the temperature change rate sequence, the sequence is divided into time periods equal to the length of the window to be determined, with adjacent time periods being sequentially truncated using a sliding step of 50% of the window length. The variance of the temperature change rate within each interval is calculated to form a temperature change rate variance distribution, and the 95th quantile of this distribution is taken as the first fluctuation threshold. Similarly, the same operation is performed on the gas concentration gradient sequence, and the 95th quantile of the variance distribution is taken as the second fluctuation threshold.

[0049] Understandably, the historical time period should be selected based on the continuous normal operation of the park without triggering any alarms, and the data collection period should cover at least one complete production scheduling cycle of the chemical industrial park.

[0050] After determining the window length, the current sampling time is taken as the end point of the window, and the time corresponding to the time before the sampling time minus the window length is taken as the start point of the window. Data within this time interval are extracted from the synchronized temperature change rate sequence and gas concentration gradient sequence to form the first input vector and the second input vector, respectively.

[0051] After determining the first input vector and the second input vector, calculate their correlation within the sliding time window according to the following steps: B1, extract the numerical sequence of temperature change rate and gas concentration gradient within the sliding time window respectively.

[0052] B2. Calculate the Pearson correlation coefficient between the numerical sequence of temperature change rate and the numerical sequence of gas concentration gradient within the sliding time window.

[0053] B3. Since the Pearson correlation coefficient only reflects the degree of linear correlation between the two sequences as a whole, and does not consider the synchronicity and consistency of the trend of temperature change and concentration gradient change over time, it needs to be corrected. That is, after calculating the Pearson correlation coefficient, the peak time of the temperature change rate sequence and the peak time of the gas concentration gradient sequence within the window are obtained, the time difference between the corresponding peak times is calculated, and the absolute value of the time difference is taken as the peak time difference. At the same time, the least squares method is used to perform univariate linear regression fitting on the temperature change rate sequence data points and the gas concentration gradient sequence data points within the window, respectively. The regression slope is taken as the rising slope of the temperature change rate sequence and the rising slope of the gas concentration gradient sequence within the window, respectively. The ratio of the two slopes is calculated and recorded as the rising slope ratio.

[0054] The peak time refers to the sampling time corresponding to the global maximum value reached by the sequence value within the time interval of the corresponding sliding time window. If there are multiple identical maximum values ​​in the sequence within the window, the sampling time corresponding to the earliest occurrence of that maximum value in chronological order is taken as the peak time. If the sequence does not show an upward trend throughout the window and the maximum value appears at the beginning of the window, the peak time is taken as the beginning time of the window.

[0055] B4. Calculate the ratio of the peak time difference to the window length, denoted as the time offset ratio. Meanwhile, the ratio of the rising slopes is denoted as .

[0056] B5. Calculate the correction factor based on the time offset ratio and the rising slope ratio. , .

[0057] It should be understood that, The closer the value is to 0, the closer the peak times of temperature change and concentration gradient change are, and the better the temporal synchronization between the two sequences. The closer the value is to 1, the closer the rates of increase of the two sequences are and the better the consistency of their changing trends. Used to measure the quality of time synchronization. Used to measure the quality of trend consistency; the smaller the values ​​of both, the worse the performance in the corresponding aspect. When the value is negative, it indicates that the rate of temperature change is opposite to the trend of the gas concentration gradient. The correction factor is from The decision is made. At this point, when trends are opposite, time synchronicity becomes the primary factor influencing correlation.

[0058] It's worth noting that using the smaller of the two values ​​as the correction factor ensures that the degree of reduction is determined by the one with poorer performance in terms of synchronicity and consistency. If the two sequences are asynchronous in time or have significantly different rates of change, the correction factor is less than 1, and the correlation coefficient is adjusted downwards accordingly. Conversely, if both synchronicity and consistency are good, the correction factor is close to 1, and the correlation coefficient remains essentially unchanged.

[0059] B6. Multiply the correction factor by the Pearson correlation coefficient to obtain the corrected Pearson correlation coefficient. Use the corrected Pearson correlation coefficient as the correlation degree. The correlation degree takes values ​​in the range of [-1, 1], where a positive value indicates a positive correlation between the temperature change rate and the gas concentration gradient, and a negative value indicates a negative correlation.

[0060] Further, please refer to Figure 2As shown, the coupling type is determined according to the following rules: the correlation is compared with a preset positive correlation threshold. If the correlation is greater than the preset positive correlation threshold, the coupling type is determined to be positive coupling; otherwise, the coupling type is determined to be negative coupling.

[0061] It should be noted that the preset positive correlation threshold is set as follows: the correlation of the chemical industrial park in multiple historical time periods under the condition that no sensor alarms are triggered is obtained to form a normal operation correlation distribution, and the 95th percentile of this distribution is taken as the preset positive correlation threshold.

[0062] It should also be noted that the positive coupling characterizes the physical process of accelerated gas diffusion caused by heat during the initial stage of leakage or smoldering in the chemical industrial park. At this time, the rate of temperature change and the gas concentration gradient show an increasing trend in the same direction. The negative coupling characterizes the normal process heating condition in the chemical industrial park without leakage. At this time, the rate of temperature change increases but the gas concentration gradient does not respond significantly.

[0063] Furthermore, after obtaining the coupling type, the specific process of determining the weights of the first input vector and the second input vector is as follows: when the coupling type is positive coupling, the first input vector and the second input vector adopt equal weights, so that when the temperature change rate and the gas concentration gradient rise synchronously in the early stage of leakage or smoldering, the contributions of both are amplified equally, and the comprehensive risk index can accumulate two weak signals, thereby improving the sensitivity of early warning. For example, the weights of the first input vector and the second input vector are both 0.5.

[0064] When the coupling type is negative coupling, the weight of the first input vector is less than the weight of the second input vector. This suppresses the contribution of the temperature change rate to the comprehensive risk index when there is only a temperature rise and no gas diffusion during normal process heating, keeping the comprehensive risk index at a low level and thus avoiding false triggering of warnings. The sum of the weights of the first and second input vectors is 1.

[0065] This step quantifies the correlation strength between temperature change and concentration gradient change by calculating the correlation between the temperature change rate sequence and the gas concentration gradient distribution sequence and determining positive or negative coupling based on the correlation. This solves the problems that the positive coupling relationship between temperature and gas diffusion cannot be identified and utilized, and the negative decoupling relationship of normal process temperature rise cannot be judged. It enables accurate differentiation between fire precursors and normal process fluctuations and collaborative processing between multiple physical fields.

[0066] S4. Based on the weighted first and second input vectors, calculate the comprehensive explosion risk index. If the comprehensive risk index is greater than or equal to the warning threshold, trigger the fire and explosion warning command.

[0067] Specifically, please refer to Figure 3As shown, the specific process for calculating the comprehensive explosion risk index is as follows: obtain the first and second input vectors after weighting within the current sliding time window.

[0068] For each sampling moment within the window, the weighted rate of temperature change is multiplied by the weighted magnitude of the gas concentration gradient to quantify the strength of the interaction between the two fields at that moment. When the temperature increase and concentration gradient increase occur simultaneously at the same moment, the product is positive and large, indicating enhanced coupling between the two fields. If only the temperature increases without a response from the concentration gradient, the product approaches zero or is small, indicating decoupling between the two fields. The product is recorded as the instantaneous coupling strength, and the instantaneous coupling strengths at all sampling moments are arranged in chronological order to form an instantaneous coupling strength sequence.

[0069] The comprehensive explosion risk index is obtained by summing all the values ​​in the instantaneous coupling strength sequence and then dividing by the number of sampling points in the window.

[0070] It should be understood that the product of the rate of temperature change and the magnitude of the gas concentration gradient is not a direct physical quantity calculation, but rather a numerical representation of the strength of the synergistic effect between the two fields. The larger the product value, the more pronounced the trend of simultaneous enhancement of temperature change and concentration gradient change at the same moment.

[0071] It should also be understood that the comprehensive explosion risk index characterizes the average coupling strength between the temperature field and the gas concentration field within the current window. The higher the value, the stronger the synergistic effect between the two fields, and the higher the risk of fire and explosion. By calculating this index, potential risks can be detected in a timely manner based on abnormal changes in the coupling relationship between the two fields, even when a single signal does not reach the threshold.

[0072] Considering that the temperature rise and gas diffusion in a real fire and explosion accident in a chemical industrial park is a continuous physical process rather than an instantaneous event, the positive coupling relationship between the rate of temperature change and the gas concentration gradient should have a certain duration over time. However, signal fluctuations caused by interference factors such as sensor noise and environmental disturbances usually exhibit instantaneous characteristics and lack consistency over a longer time scale.

[0073] Based on this, the present invention uses two sliding time windows of different lengths to determine coupling stability, in order to distinguish between continuous physical processes and transient disturbances.

[0074] Specifically, the process for determining coupling stability is as follows: At least two sliding time windows of different lengths are set, denoted as the short window and the long window. The coupling type of the temperature change rate sequence and the gas concentration gradient distribution sequence within the short window and the long window is determined respectively, as described above. Simultaneously, the comprehensive explosion risk index for the short window and the comprehensive explosion risk index for the long window are calculated.

[0075] When both the short window and the long window are determined to be of the positive coupling type and the comprehensive explosion risk index calculated by the two windows exceeds the warning threshold, the current coupling is determined to be stable, and a fire and explosion warning command is triggered. Otherwise, the temperature time series data and the concentration time series data of at least one combustible gas or toxic gas in the monitoring area of ​​the chemical industrial park continue to be collected. The window slides forward with the newly collected data, and the steps of coupling type determination and comprehensive explosion risk index calculation are repeated.

[0076] It should be noted that the warning threshold is predetermined in the following way: A temperature change rate sequence and gas concentration gradient sequence for at least 30 consecutive natural days are obtained from the chemical industrial park under normal operating conditions without triggering any alarms. All window positions within the collection period are traversed using the same window length and sliding step size as in step S4. The comprehensive explosion risk index for each window is calculated according to the methods described in steps S3 and S4, forming a normal operating index distribution. The sample size of this distribution is no less than 500, and the 99th percentile of this distribution is taken as the warning threshold. If the park has been in operation for less than 30 natural days, data from at least 20 consecutive normal operating days can be obtained, and the sample size is expanded to at least 500 using a bootstrapping method before taking the 99th percentile.

[0077] Furthermore, the fire and explosion early warning command includes sending an alarm signal to the central control system of the chemical industrial park, which includes the trigger time, the coupling type within the trigger window, and the comprehensive explosion risk index value, and simultaneously issuing an audible and visual alarm prompt.

[0078] After the warning command is triggered, the system will continue to monitor until the alarm is automatically lifted by manual confirmation or the risk index falls below the warning threshold for a preset period of time.

[0079] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. A method for early warning of fires and explosions in chemical industrial parks based on multimodal data fusion, characterized in that, The method includes: Collect time-series temperature data and time-series concentration data of at least one combustible or toxic gas within the monitoring area of ​​the chemical industrial park, and perform spatiotemporal alignment to generate spatiotemporally synchronized temperature and gas concentration sequences. The temperature sequence is differentially calculated to obtain the temperature change rate sequence, and the spatial gradient of the gas concentration sequence is calculated to obtain the gas concentration gradient distribution sequence. The temperature change rate sequence and the gas concentration gradient distribution sequence are aligned within the same sliding time window and used as the first input vector and the second input vector, respectively. The correlation between the two within the time window is calculated, and the coupling type is determined based on the correlation. The coupling type is either positive coupling or negative coupling, and the weights of the first input vector and the second input vector are determined based on the coupling type. Based on the weighted first and second input vectors, a comprehensive explosion risk index is calculated. If the comprehensive risk index is greater than or equal to the warning threshold, a fire and explosion warning command is triggered.

2. The method for early warning of fires and explosions in chemical industrial parks based on multimodal data fusion as described in claim 1, characterized in that: The specific steps for performing spatiotemporal alignment are as follows: Obtain the data sampling frequencies of the temperature sensor and the gas concentration sensor respectively, and use the lower of the two sampling frequencies as the target sampling frequency; The temperature time series data and gas concentration time series data are interpolated and resampled according to the target sampling frequency. At the same time, a planar grid of the monitoring area is constructed based on the spatial coordinates of the temperature sensor and gas concentration sensor in the chemical industrial park. The resampled temperature and gas concentration sequences are mapped to each grid node of the planar grid using inverse distance weighted interpolation; Temperature and gas concentration values ​​of the same grid node at the same sampling time are extracted, arranged in chronological order, and spatiotemporally synchronized temperature and gas concentration sequences are generated respectively.

3. The method for early warning of fires and explosions in chemical industrial parks based on multimodal data fusion as described in claim 1, characterized in that: The specific process of calculating the spatial gradient of a gas concentration sequence includes: The spatial coordinates and real-time concentration measurements of all gas sensors in the chemical industrial park are obtained, and the monitoring area is divided into high-risk dense zone, medium-risk normal deployment zone and low-risk sparse zone according to the fire and explosion risk level of each sensor location. Each gas sensor is taken as the target sensor in turn, and the set of adjacent sensors is determined based on the type of the region to which the current target sensor is centered. The concentration difference between the target sensor and each gas sensor in the adjacent sensor set is calculated sequentially and divided by the Euclidean distance between the installation location of the target sensor and the corresponding gas sensor installation location in the adjacent sensor set to obtain the spatial concentration gradient component in that direction. Vector synthesize the spatial concentration gradient components in all directions and calculate the magnitude of the synthesized total vector as the gas concentration gradient value of the current target sensor. Arrange the gas concentration gradient values ​​of all target sensors according to the sensor installation location to generate a gas concentration gradient distribution sequence.

4. The method for early warning of fires and explosions in chemical industrial parks based on multimodal data fusion as described in claim 1, characterized in that: The length of the sliding time window is determined by the following steps: The response time constants of the temperature sensor and the gas concentration sensor are obtained respectively and denoted as the first time constant and the second time constant, respectively. The larger of the two values ​​is used as the initial length of the sliding time window. Calculate the variance of the temperature change rate sequence and the variance of the gas concentration gradient sequence within a time window that is truncated with the initial window length as the window length and the current sampling time as the window end. If the variance of the temperature change rate sequence exceeds the preset first fluctuation threshold, the mean of the first time constant and the second time constant will be used as the length of the sliding time window. If the variance of the temperature change rate sequence does not exceed the first fluctuation threshold while the variance of the gas concentration gradient sequence exceeds the preset second fluctuation threshold, then the sum of the first time constant and the second time constant will be used as the length of the sliding time window. If the variances of both do not exceed the corresponding fluctuation thresholds, then the larger of the first time constant and the second time constant is taken as the length of the sliding time window.

5. The method for early warning of fires and explosions in chemical industrial parks based on multimodal data fusion as described in claim 1, characterized in that: The specific calculation process for the relevance is as follows: The numerical sequences of temperature change rate and gas concentration gradient within the sliding time window were extracted respectively. Calculate the Pearson correlation coefficient between the numerical sequence of temperature change rate and the numerical sequence of gas concentration gradient within the sliding time window; Obtain the peak time of the temperature change rate sequence and the peak time of the gas concentration gradient sequence within the window, calculate the time difference between the corresponding peak times, and take the absolute value of the time difference as the peak time difference. At the same time, calculate the rising slope of the temperature change rate sequence and the rising slope of the gas concentration gradient sequence within the window through linear regression, calculate the ratio of the rising slopes of the two, and obtain the rising slope ratio. The Pearson correlation coefficient, calculated based on the peak time difference and the ratio of rising slope, is corrected, and the corrected result is used as the correlation degree.

6. The method for early warning of fires and explosions in chemical industrial parks based on multimodal data fusion as described in claim 5, characterized in that: The specific correction process for the calculated Pearson correlation coefficient includes: The ratio of the peak time difference to the window length is calculated and denoted as the time offset ratio. Meanwhile, the ratio of the rising slopes is denoted as ; The correction factor is calculated based on the time offset ratio and the rise slope ratio. , ; Multiply the correction factor by the Pearson correlation coefficient to obtain the corrected Pearson correlation coefficient.

7. The method for early warning of fires and explosions in chemical industrial parks based on multimodal data fusion as described in claim 5, characterized in that: Coupling type is determined according to the following rules: The correlation is compared with a preset positive correlation threshold; If the correlation is greater than the preset positive correlation threshold, the coupling type is determined to be positive coupling; otherwise, the coupling type is determined to be negative coupling.

8. The method for early warning of fires and explosions in chemical industrial parks based on multimodal data fusion as described in claim 1, characterized in that: The specific process for determining the weights of the first and second input vectors is as follows: If the coupling type is positive coupling, the weights of the first input vector and the weights of the second input vector have the same value; If the coupling type is negative coupling, set the first input vector as the first weight value, set the weight of the second input vector as the second weight value, and set the first weight value to be less than the second weight value; The sum of the weights of the first input vector and the second input vector is 1.

9. The method for early warning of fires and explosions in chemical industrial parks based on multimodal data fusion as described in claim 1, characterized in that: The specific calculation process for the comprehensive explosion risk index includes: Obtain the weighted first and second input vectors within the current sliding time window; For each sampling moment within the window, the temperature change rate corresponding to that moment in the weighted first input vector is multiplied by the gas concentration gradient magnitude corresponding to that moment in the weighted second input vector to obtain the instantaneous coupling strength at that moment. The instantaneous coupling strengths of all sampling moments are arranged in chronological order to form an instantaneous coupling strength sequence. The comprehensive explosion risk index is obtained by summing all the values ​​in the instantaneous coupling strength sequence and then dividing by the number of sampling points in the window.

10. The method for early warning of fires and explosions in chemical industrial parks based on multimodal data fusion as described in claim 1, characterized in that: Before triggering a fire or explosion warning command, a coupling stability assessment is also performed. Set at least two sliding time windows of different lengths, denoted as short window and long window, and determine the coupling type of temperature change rate sequence and gas concentration gradient distribution sequence in short window and long window respectively. At the same time, calculate the comprehensive explosion risk index of short window and comprehensive explosion risk index of long window. When both the short window and the long window are determined to be of the positive coupling type and the comprehensive explosion risk index calculated by the two windows exceeds the warning threshold, the current coupling is determined to be stable, and a fire and explosion warning command is triggered. Otherwise, the temperature time series data and the concentration time series data of at least one combustible gas or toxic gas in the monitoring area of ​​the chemical industrial park continue to be collected.

Images