Multi-source heterogeneous data disaster early warning processing method based on dynamic space-time weight

Abstract

The invention relates to the technical field of disaster early warning, in particular to a multi-source heterogeneous data disaster early warning processing method based on dynamic space-time weight, which comprises the following steps: for a single prediction area, inputting multi-source data to predict each parameter, determining the ambiguity of each prediction parameter, and determining the ambiguity of each prediction parameter; dividing each prediction parameter into strong credible data or weak credible data; determining a prediction risk probability based on each prediction parameter fitting; determining the risk tendency of the single prediction area based on the fitted stability tendency value; when it is determined that the single prediction area is the strong risk area, the distance weight and the time weight of the single prediction area are adjusted; determining whether to correct the data processing resource configuration based on the data stability index; and performing disaster prediction based on the prediction data. The weight of each piece of data is adaptively adjusted and determined in combination with the spatio-temporal dynamic characteristics of the data, so that the data is dynamically processed, and the early warning efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of disaster early warning technology, and in particular to a disaster early warning processing method based on dynamic spatiotemporal weights for multi-source heterogeneous data. Background Technology

[0002] Currently, natural disaster early warning mainly relies on a single data source or a few data sources for data collection and analysis. This approach easily leads to incomplete information and fails to accurately reflect the true situation of the disaster. Moreover, existing early warning methods typically use fixed weighting coefficients to process different data, without fully considering the spatiotemporal dynamic changes of the data, thus limiting the accuracy and timeliness of the early warning results.

[0003] Chinese Patent Publication No. CN112863132A discloses a natural disaster early warning system and method, including the following steps: Step S1, an environmental profile construction unit collects environmental data of the target area in real time and establishes an environmental profile data chain for recording and representing the target disaster prediction of each target area; Step S2, a prediction model construction unit receives the environmental profile data chain and constructs a disaster prediction model based on the environmental profile data chain. This invention performs frame-by-frame quantization of the environmental logs of the target area and extracts keyframes from the environmental logs using the similarity between adjacent frames. However, the above technical solution has the following problems: it does not consider the spatiotemporal dynamic characteristics of the data and uses a relatively fixed method to process the data, affecting the early warning efficiency. Summary of the Invention

[0004] To address this issue, the present invention provides a disaster early warning processing method based on dynamic spatiotemporal weights for multi-source heterogeneous data, which overcomes the problem that existing technologies do not consider the spatiotemporal dynamic characteristics of data and process data in a relatively fixed manner, thus affecting the efficiency of early warning.

[0005] To achieve the above objectives, this invention provides a method for disaster early warning processing based on multi-source heterogeneous data with dynamic spatiotemporal weights, comprising: For a single prediction region, input multi-source data to predict each parameter separately; Determine the ambiguity of each prediction parameter; Based on the degree of ambiguity, each prediction parameter is divided into strongly reliable data or weakly reliable data; The probability of predicted risk is determined based on the fitting of each prediction parameter; The risk propensity of a single prediction region is determined based on the fitted stable propensity value. When a single prediction region is identified as a high-risk region, the distance weight and time weight of the single prediction region are adjusted. Determine whether to adjust the data processing resource allocation based on the data stability index; Disaster prediction is based on various forecast data.

[0006] Further, the stability tendency value is determined, including: The data stability index is obtained by calculating the ratio of the number of highly reliable data points to the total number of predicted data points. For a single predicted data point, calculate the elasticity difference between the corresponding preset critical alarm value and the predicted data, and calculate the elasticity ratio between the elasticity difference and the predicted data. The average value of the elasticity ratio of each highly reliable data point is calculated to obtain the tolerance ratio. Determine the probability of safety based on the predicted probability of risk; The stability tendency value is obtained by multiplying the effective data ratio, tolerance ratio, and safety probability by their respective weight coefficients and summing them.

[0007] Furthermore, the process of determining the risk propensity of a single prediction region based on the fitted stable propensity value includes: When the stability tendency value is less than or equal to the preset stability tendency value, a single prediction area is identified as a high-risk area, and the time weights of each input multi-source data are adjusted based on the historical change magnitude characterization value. When the stability tendency value is greater than the preset stability tendency value, a single prediction area is identified as a low-risk area, and disaster prediction is carried out based on each prediction data.

[0008] Furthermore, the time weights of each input multi-source data are adjusted based on the historical change magnitude representation values, whereby... Obtain the historical actual monitoring data set corresponding to a single predicted data point to plot the time-domain curve of the actual monitoring data; The average slope of the time-domain curve of the actual monitoring data is calculated to obtain the average slope of the abnormal segment. The average slope of the time-domain curves of each actual monitoring data point under the condition that a single predicted data point is identified as a low-risk area is obtained by averaging the average slope of the stable segment. The ratio of the average slope of the abnormal segment to the average slope of the stable segment is calculated to obtain the historical variation amplitude value. The increase in the recent time weight coefficients of the input data group corresponding to a single predicted data is positively correlated with the historical change magnitude representation value.

[0009] Furthermore, when correcting the time weights for each input multi-source data, the distance weights for the corresponding ranges are corrected based on the span reference value, whereby... For a single range, determine the span coefficient between it and its adjacent ranges far from the center point; The average value of each span coefficient within a single prediction region is calculated to obtain the span reference value; The increase in the distance weight of the adjacent range of the potential hazard area is positively correlated with the span reference value.

[0010] Furthermore, when adjusting the weights for each distance, the data processing resource allocation is adjusted based on the data stability index, including: When the data stability index is less than or equal to the preset data stability index, the data volume of each input multi-source data is adjusted to the corresponding value based on the data stability index. When the data stability index is greater than the preset data stability index, the amount of data that the database buffer can hold for a single prediction region will be adjusted to the corresponding value based on the stability tendency value.

[0011] Furthermore, based on the data stability index, the data volume of each input multi-source data is adjusted to the corresponding value, wherein, The increase in data volume is negatively correlated with the data stability index.

[0012] Furthermore, based on the stability propensity value, the data capacity of the database buffer corresponding to a single prediction region is adjusted to a corresponding value, wherein, The increase in the amount of data that can be accommodated is negatively correlated with the stability propensity value.

[0013] Furthermore, when adjusting the data capacity of the database buffer corresponding to a single prediction region, the weight ratio of each strong confidence data point is determined based on the fluctuation rate of the weak confidence data, including: Obtain the ambiguity of each weakly reliable data point, calculate the difference between the maximum and minimum values ​​of each ambiguity, and obtain the fluctuation rate of the weakly reliable data. When the fluctuation rate of weakly reliable data is less than or equal to the preset weakly reliable data fluctuation rate, disaster prediction is performed based on the redefined prediction data. When the fluctuation rate of weakly trusted data is greater than the preset fluctuation rate of weakly trusted data, the weight ratio of each strongly trusted data is adjusted based on the fluctuation rate of weakly trusted data.

[0014] Furthermore, the weight ratio of each piece of highly reliable data is adjusted based on the fluctuation rate of the weakly reliable data. The increase in the weighting percentage of each highly reliable data point is positively correlated with the fluctuation rate of the less reliable data point.

[0015] Compared with existing technologies, the beneficial effects of this invention lie in determining a stability tendency value, which comprehensively reflects the risk status of the predicted area by integrating the data stability index, tolerance ratio, and safety probability. The data stability index is the proportion of highly reliable data to the total predicted data, reflecting the overall reliability and stability of the data. The tolerance ratio is the average of the elasticity ratios of highly reliable data, reflecting the degree of deviation and tolerable range of each predicted data point from its corresponding preset critical alarm value. The safety probability reflects the level of safety under the current condition. The overall stability and risk level of the predicted area are comprehensively assessed through the stability tendency value. The risk tendency is determined by the stability tendency value, providing a basis for subsequent weight adjustments and resource allocation. This improves the efficiency of disaster early warning.

[0016] Furthermore, when a single prediction area is identified as a high-risk area, the distance and time weights of that area are adjusted. High-risk areas experience more drastic data changes, so the spatiotemporal weights are dynamically adjusted based on historical data to more accurately capture data trends. The historical change magnitude characterization value measures the difference in data change magnitude between abnormal and stable periods, reflecting the degree of drastic change at different stages. When historical change magnitude is large, recent data has a greater impact on the current prediction, and the increase in the recent time weight coefficient is positively correlated with the historical change magnitude characterization value. The span reference value reflects the distance span between different ranges. The adjacent ranges of a potential hazard are more susceptible to the hazard's influence; a larger span reference value indicates a larger regional span between ranges, and the increase in the distance weight of the adjacent ranges of a potential hazard is positively correlated with the span reference value. By dynamically adjusting the spatiotemporal weights, greater attention is paid to areas of drastic change and recent data, improving the accuracy of disaster prediction. More emphasis is placed on recent data in time and on the surrounding areas of potential hazards in space, thus improving the efficiency of disaster early warning.

[0017] Furthermore, the data stability index is used to determine whether to adjust the data processing resource allocation. Different levels of data stability result in different demands for data processing resources. Data processing resources need to be allocated rationally based on the data stability index to improve resource utilization efficiency. When the data stability index is less than or equal to the preset data stability index, the data is unstable and contains many anomalous data. In this case, the amount of data from each input multi-source source is adjusted to reduce the impact of anomalous data. When the data stability index is greater than the preset data stability index, the data is relatively stable. In this case, a single prediction area is identified as being in a high-risk situation. The data capacity of the database buffer is adjusted according to the stability tendency value to accelerate data storage and processing, enabling timely response. The data stability index reflects the overall stability of the data. Dynamically adjusting the data processing resource allocation based on the data stability level can both reduce interference from anomalous data when the data is unstable and rationally utilize the database buffer when the data is stable and the prediction area is at high risk, thus improving data processing efficiency and the reliability of disaster early warning.

[0018] Furthermore, the amount of input multi-source data is adjusted based on the data stability index, which reflects the overall stability of the data. When the data is unstable, there are many anomalies, requiring more data to more accurately capture data change patterns and trends, thereby improving the accuracy of disaster prediction. Conversely, when the data is stable, a smaller amount of data is sufficient for effective analysis and prediction; excessive data may increase processing costs. The lower the data stability index, the greater the data fluctuation and the more unstable the data. In this case, increasing the amount of data compensates for the information gaps caused by data instability. The increase in data volume is negatively correlated with the data stability index; that is, the smaller the data stability index, the greater the increase in data volume. Unstable data can lead to unclear local data characteristics. Increasing the amount of data makes the data distribution more complete, thus more clearly showing the true characteristics and trends of the data. Dynamically adjusting the amount of data based on the degree of data stability allows for the acquisition of sufficient information to improve prediction accuracy when the data is unstable, while avoiding unnecessary data collection and processing when the data is stable, saving computing resources. This improves the efficiency of disaster early warning.

[0019] Furthermore, the data capacity of the database buffer is adjusted based on the stability propensity value, which comprehensively reflects the overall stability and risk level of the prediction area. When the stability propensity value is low, the risk in that area is high, and data changes are more drastic, requiring a larger database buffer to store more data for deeper analysis and processing, and timely handling of various data processing needs for high-risk areas. The lower the stability propensity value, the higher the risk and the more drastic the data changes. To cope with potentially large amounts of data and complex situations, the data capacity of the database buffer is increased. The increase in the data capacity is negatively correlated with the stability propensity value; that is, the smaller the stability propensity value, the greater the increase in the data capacity. The data capacity of the database buffer is the maximum capacity of the buffer used for temporary data storage. A low stability propensity value means that the data will fluctuate and change significantly, requiring a larger buffer to store these changing data to ensure data integrity and continuity, facilitating subsequent analysis and processing. Dynamically adjusting the data capacity of the database buffer based on the stability propensity value: 1. Rationally utilizes database resources, avoiding data loss or untimely processing due to an insufficient buffer, while also avoiding resource waste caused by an excessively large buffer. This improves the efficiency of disaster early warning.

[0020] Furthermore, the weighting of each piece of highly reliable data is adjusted based on the fluctuation rate of the weakly reliable data, which reflects the degree of volatility of the data. When the weakly reliable data fluctuates significantly, it has a substantial impact on the overall disaster prediction. In this case, adjusting the weighting of highly reliable data balances the uncertainty of the weakly reliable data. Conversely, when the weakly reliable data fluctuates less, the impact on the overall prediction is relatively small, and disaster prediction can be directly based on the redefined prediction data. The fluctuation rate of the weakly reliable data is calculated by the difference between the maximum and minimum ambiguities of each piece of data, thus measuring the volatility of the weakly reliable data. When the fluctuation rate exceeds a preset value, it indicates significant volatility. The fluctuation rate reflects the range of ambiguity fluctuations in the weakly reliable data, embodying its instability. Large fluctuations in the weakly reliable data can interfere with the disaster prediction results. Adjusting the weighting of highly reliable data highlights its role and reduces the negative impact of its fluctuations. Deciding whether to adjust the weighting of highly reliable data based on its volatility allows for flexible responses to different data situations, improving the accuracy and reliability of disaster prediction and enhancing the efficiency of disaster early warning. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the steps of the multi-source heterogeneous data disaster early warning processing method based on dynamic spatiotemporal weights according to an embodiment of the present invention. Figure 2 This is a logical decision diagram for determining the risk tendency of a single prediction region based on the fitted stable tendency value in an embodiment of the present invention. Figure 3 This is a logic diagram illustrating how the data processing resource configuration is determined based on a data stability index in an embodiment of the present invention. Figure 4 This is a logic diagram illustrating how an embodiment of the present invention determines whether to adjust the weight ratio of each strongly trusted data point based on the fluctuation rate of weakly trusted data. Detailed Implementation

[0022] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0023] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0024] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0025] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0026] Please see Figure 1 The diagram shows a flowchart of the steps in the multi-source heterogeneous data disaster early warning processing method based on dynamic spatiotemporal weights according to an embodiment of the present invention. The present invention provides a multi-source heterogeneous data disaster early warning processing method based on dynamic spatiotemporal weights, comprising: S1, For a single prediction region, input multi-source data to predict each parameter separately, determine the ambiguity of each prediction parameter, and classify each prediction parameter into strongly reliable data or weakly reliable data. S2, determine the predicted risk probability based on the fitting of each prediction parameter; S3, determine the risk propensity of a single prediction region based on the fitted stable propensity value; When a single prediction region is identified as a high-risk region, the distance weight and time weight of the single prediction region are adjusted; and the data processing resource allocation is determined based on the data stability index. S4, disaster prediction based on various forecast data.

[0027] Specifically, the process of inputting multi-source data to predict each parameter involves: performing spatiotemporal alignment and standardization preprocessing on the input heterogeneous multi-source data; training a time-series prediction model using its corresponding historical data for each type of prediction parameter; and inputting the current and historical multi-source data into the corresponding time-series prediction model to obtain the prediction data for the next time node. The various prediction parameters include rainfall, soil moisture content, and displacement velocity.

[0028] Multi-source data includes structured monitoring data, unstructured or semi-structured data; Structured monitoring data includes rainfall and temperature data from meteorological stations, displacement and stress data from geological sensors, and water level and flow velocity data from hydrological stations. Unstructured or semi-structured data includes satellite remote sensing imagery, drone aerial photography data, and social media text information after natural language processing to extract disaster-related keywords and geographic locations.

[0029] The process of determining the predicted risk probability based on fitting various prediction parameters includes: comparing each predicted data point with its corresponding preset critical alarm value, and using the calculated exceedance as the basic risk factor. A machine learning model is then used, with the basic risk factor of each predicted data point as input features. The initial weights of each feature can be set based on expert knowledge. The model output is a value between 0 and 1, which represents the comprehensive predicted risk probability for the region.

[0030] Specifically, the process of determining the ambiguity of an individual prediction parameter and classifying it into strongly reliable or weakly reliable data includes: Obtain the historical prediction data set and historical actual monitoring data set corresponding to a single prediction data set to obtain several sets of prediction-actual comparison data; a single set of prediction-actual comparison data includes the actual detection data at a single time point and the prediction data at the previous time point; Calculate the average prediction deviation of each group of predicted-actual comparison data to obtain the ambiguity for a single prediction parameter; calculate the absolute value of the difference between the actual detection data at a single time point and the predicted data at the previous time point to obtain the prediction difference value; calculate the ratio of the prediction difference value to the actual detection data at a single time point to obtain the prediction deviation of a single group of predicted-actual comparison data.

[0031] When the ambiguity is less than or equal to the preset ambiguity, a single prediction parameter is classified as strongly reliable data; when the ambiguity is greater than the preset ambiguity, a single prediction parameter is classified as weakly reliable data.

[0032] The preset ambiguity is selected within the range [0.2, 0.3]. The preset ambiguity can be determined by statistically analyzing historical data and finding the average prediction deviation of each prediction parameter during non-disaster stable periods. Those skilled in the art can adjust it according to data quality requirements. In this embodiment, preferably, the preset ambiguity is 0.2.

[0033] Specifically, each predicted data point has a pre-set critical alarm value. The pre-set critical alarm value can be determined by combining a historical disaster case database, statistically analyzing the actual numerical distribution of each parameter at the time of the disaster, taking a certain 90th percentile as an empirical threshold, and taking into account the experience of experts in the field.

[0034] Specifically, the prediction area is divided according to a geographic information system. Each prediction area has a defined center point, which can be the geographical center of gravity of the prediction area, a key hidden danger point, or the location of an important protection target. The area is divided into several concentric ring-shaped ranges based on the distance from the center point. In a single embodiment, each range includes a core area, a buffer area, and a monitoring area. Each range contains several data collection points.

[0035] The database buffer is a high-speed storage area used for temporary storage and fast processing of real-time data flowing into the prediction region. The amount of data it can hold affects the real-time processing performance of data within a single prediction region.

[0036] Specifically, determining the stability propensity value includes: The data stability index is obtained by calculating the ratio of the number of highly reliable data points to the total number of predicted data points. For a single predicted data point, calculate the elasticity difference between the corresponding preset critical alarm value and the predicted data, and calculate the elasticity ratio between the elasticity difference and the predicted data. Solve for the average value of the elasticity ratio of each highly reliable data point to obtain the tolerance ratio. Calculate the difference between 1 and the predicted risk probability to obtain the safety probability; The stability tendency value is obtained by multiplying the effective data ratio, tolerance ratio, and safety probability by their respective weight coefficients and summing them.

[0037] Specifically, the weighting coefficients for the data stability index, tolerance ratio, and safety probability are set to 0.2, 0.3, and 0.5, respectively. Safety probability directly reflects the probability of risk and is the most crucial indicator, therefore it is given the highest weight. Tolerance ratio reflects the buffer space between the current state and the danger line, and is an important indicator of risk sensitivity, thus it is given a medium weight. The data stability index reflects data quality and is the foundation of reliability, but it is not a direct measure of risk; therefore, it is given a basic weight of 0.2. These weights can be fine-tuned according to the actual application scenario.

[0038] Specifically, a stability propensity value is determined, which integrates the data stability index, tolerance ratio, and safety probability to comprehensively reflect the risk status of the predicted area. The data stability index is the proportion of highly reliable data to the total number of predicted data, reflecting the overall reliability and stability of the data. The tolerance ratio is the average of the elasticity ratios of highly reliable data, reflecting the degree of deviation and tolerable range of each predicted data point from its corresponding preset critical alarm value. The safety probability reflects the level of safety under the current conditions. The stability propensity value comprehensively assesses the overall stability and risk level of the predicted area. It also determines the risk tendency, providing a basis for subsequent weight adjustments and resource allocation, thus improving the efficiency of disaster early warning.

[0039] Please see Figure 2The diagram shown illustrates the logical decision-making process for determining the risk tendency of a single prediction region based on a fitted stable propensity value, according to an embodiment of the present invention. The process of determining the risk tendency of a single prediction region based on a fitted stable propensity value includes: When the stability tendency value is less than or equal to the preset stability tendency value, a single prediction area is identified as a high-risk area, and the time weights of each input multi-source data are adjusted based on the historical change magnitude characterization value. When the stability tendency value is greater than the preset stability tendency value, a single prediction area is identified as a low-risk area, and disaster prediction is carried out based on each prediction data.

[0040] Specifically, when a single predicted area is identified as a high-risk area, the disaster category is determined based on various predicted parameters, including: determining the main potential disaster categories based on a pre-established parameter anomaly pattern-disaster type mapping knowledge base. In a single embodiment, extremely high short-term rainfall intensity and a rapid increase in soil saturation are mapped as flash floods; accelerated surface displacement and a sudden drop in groundwater levels are mapped as landslides. Once a predicted area is determined to be a high-risk area, several parameters with the smallest elasticity ratios in its highly reliable data are identified, matched with patterns in the knowledge base, and the disaster type with the highest matching degree is determined as the main potential disaster category of the predicted area.

[0041] Specifically, the preset stability tendency value is selected within the range [0.71, 0.75]. Those skilled in the art can select and determine it themselves. It can be determined by analyzing historical data and statistically analyzing the distribution of stability tendency values ​​in all early warning cases where no disaster ultimately occurred. In this embodiment, the preset stability tendency value is preferably 0.71.

[0042] Specifically, the time weights of each input multi-source data are adjusted based on historical change magnitude representation values, whereby... Obtain the historical actual monitoring data set corresponding to a single predicted data point to plot the time-domain curve of the actual monitoring data; The average slope of the time-domain curve of the actual monitoring data is calculated to obtain the average slope of the abnormal segment. The average slope of the time-domain curves of each actual monitoring data point under the condition that a single predicted data point is identified as a low-risk area is obtained by averaging the average slope of the stable segment. The ratio of the average slope of the abnormal segment to the average slope of the stable segment is calculated to obtain the historical variation amplitude value. The increase in the recent time weight coefficients of the input data group corresponding to a single predicted data is positively correlated with the historical change magnitude representation value.

[0043] Input multi-source data includes input data sets from several different data sources.

[0044] Specifically, the time weighting coefficients for the input data set corresponding to a single prediction parameter are not equal. Data closer to the current time has a greater impact on predicting future states, and therefore is given a higher time weight. In this embodiment, the recent time window is defined as 30% of the data samples in the data set whose timestamps are closest to the current time, and the weighting coefficient of these samples is determined as the recent time weighting coefficient.

[0045] In this embodiment, optionally, The historical change amplitude representation value is compared with the first preset change amplitude representation value and the second preset change amplitude representation value; When the historical change amplitude representation value is less than or equal to the first preset change amplitude representation value, the recent time weight coefficients of each input data group corresponding to a single predicted data are adjusted to 1.11 times the corresponding recent time weight coefficient; When the historical change amplitude representation value is less than or equal to the second preset change amplitude representation value and greater than the first preset change amplitude representation value, the recent time weight coefficients of each input data group corresponding to a single predicted data are adjusted to 1.18 times the corresponding recent time weight coefficient. When the historical change magnitude representation value is greater than the second preset change magnitude representation value, the recent time weight coefficients of each input data group corresponding to a single predicted data are adjusted to 1.26 times the corresponding recent time weight coefficient. The first preset value for the magnitude of change is 1.05, and the second preset value for the magnitude of change is 1.15.

[0046] Specifically, when correcting the time weights for each input multi-source data, the distance weights for the corresponding ranges are corrected based on the span reference value. For a single range, determine the span coefficient between it and its adjacent ranges far from the center point; The average value of each span coefficient within a single prediction region is calculated to obtain the span reference value; The increase in the distance weight of the adjacent range of the potential hazard area is positively correlated with the span reference value.

[0047] Specifically, the forecast data with the smallest elasticity ratio among all forecast data is identified as the potential hazard data, and the single range corresponding to the monitoring point of the potential hazard data is identified as the potential hazard range.

[0048] The span coefficient is obtained by considering the maximum span distance between a single range and its adjacent ranges far from the center point.

[0049] In this embodiment, optionally, The span reference value is compared with the first preset span reference value and the second preset span reference value; When the span reference value is less than or equal to the first preset span reference value, the distance weight between the potential hazard range and each adjacent range is adjusted to 1.12 times the corresponding current distance weight; When the span reference value is less than or equal to the second preset span reference value and greater than the first preset span reference value, the distance weight between the hidden danger range and each adjacent range is adjusted to 1.21 times the corresponding current distance weight; When the span reference value is greater than the second preset span reference value, the distance weight between the hidden danger range and each adjacent range is adjusted to 1.26 times the corresponding current distance weight; The first preset span reference value is 300m, and the second preset span reference value is 600m. The first and second preset span reference values ​​can be determined according to the geographical division scheme of each prediction area.

[0050] Specifically, when a single prediction area is identified as a high-risk area, the distance and time weights of that area are adjusted. High-risk areas experience more drastic data changes, so the spatiotemporal weights are dynamically adjusted based on historical data to more accurately capture data trends. The historical change magnitude characterization value measures the difference in data change magnitude between abnormal and stable periods, reflecting the degree of drastic change at different stages. When historical change magnitude is large, recent data has a greater impact on the current prediction, and the increase in the recent time weight coefficient is positively correlated with the historical change magnitude characterization value. The span reference value reflects the distance span between different ranges. The adjacent ranges of a potential hazard are more susceptible to its influence; a larger span reference value indicates a larger regional span between ranges, and the increase in the distance weight of the adjacent ranges of a potential hazard is positively correlated with the span reference value. By dynamically adjusting the spatiotemporal weights, greater attention is paid to areas of drastic change and recent data, improving the accuracy of disaster prediction. More emphasis is placed on recent data in time and on the surrounding areas of potential hazards in space, thus improving the efficiency of disaster early warning.

[0051] Please see Figure 3 The diagram shown illustrates the logic for determining whether to adjust data processing resource configuration based on a data stability index, according to an embodiment of the present invention. The process of determining whether to adjust data processing resource configuration based on the data stability index when adjusting the distance weights includes: When the data stability index is less than or equal to the preset data stability index, the data volume of each input multi-source data is adjusted to the corresponding value based on the data stability index. When the data stability index is greater than the preset data stability index, the amount of data that the database buffer can hold for a single prediction region will be adjusted to the corresponding value based on the stability tendency value.

[0052] Specifically, the preset data stability index is selected within the range [0.72, 0.78]. Those skilled in the art can select and determine it themselves. Several time periods with high data quality and accurate predictions in historical data can be statistically analyzed, and the data stability index in each time period can be calculated. The average value of these data periods can be taken as the preset data stability index. In this embodiment, the preset data stability index is preferably 0.75.

[0053] Specifically, the data stability index is used to determine whether to adjust the data processing resource allocation. Different levels of data stability result in different resource requirements. Data processing resources need to be allocated rationally based on the data stability index to improve resource utilization efficiency. When the data stability index is less than or equal to the preset index, the data is unstable and contains many anomalous data. In this case, the amount of data from each input multi-source source is adjusted to reduce the impact of anomalous data. When the data stability index is greater than the preset index, the data is relatively stable. In this case, a single prediction area is identified as high-risk. The data capacity of the database buffer is adjusted according to the stability tendency value to accelerate data storage and processing, enabling timely response. The data stability index reflects the overall stability of the data. Dynamically adjusting the data processing resource allocation based on the data stability level can both reduce interference from anomalous data when the data is unstable and rationally utilize the database buffer when the data is stable and the prediction area is high-risk, thus improving data processing efficiency and the reliability of disaster early warning.

[0054] Specifically, based on the data stability index, the amount of data from each input multi-source data source is adjusted to the corresponding value, whereby... The increase in data volume is negatively correlated with the data stability index.

[0055] In this embodiment, optionally, The data stability index is compared with the first preset stability index and the second preset stability index; When the data stability index is less than or equal to the first preset stability index, the data volume of each input multi-source data is adjusted to 1.29 times the initial data volume; When the data stability index is less than or equal to the second preset stability index and greater than the first preset stability index, the data volume of each input multi-source data is adjusted to 1.19 times the initial data volume; When the data stability index is greater than the second preset stability index, the amount of each input multi-source data is adjusted to 1.08 times the initial data amount; The first preset stability index is 0.56, and the second preset stability index is 0.62.

[0056] Specifically, the amount of data from each input multi-source data source can be increased by adjusting the data time window used for the current predictive analysis.

[0057] Specifically, the amount of input multi-source data is adjusted based on the data stability index, which reflects the overall stability of the data. When the data is unstable, there are more anomalies, requiring more data to more accurately capture data change patterns and trends, thereby improving the accuracy of disaster prediction. Conversely, when the data is stable, a smaller amount of data is sufficient for effective analysis and prediction; excessive data may increase processing costs. The lower the data stability index, the greater the data fluctuation and the more unstable the data. In this case, increasing the amount of data compensates for the information gaps caused by data instability. The increase in data volume is negatively correlated with the data stability index; that is, the smaller the data stability index, the greater the increase in data volume. Unstable data can lead to unclear local data characteristics. Increasing the amount of data makes the data distribution more complete, thus more clearly showing the true characteristics and trends of the data. Dynamically adjusting the amount of data based on the degree of data stability allows for the acquisition of sufficient information to improve prediction accuracy when the data is unstable, while avoiding unnecessary data collection and processing when the data is stable, saving computing resources. This improves the efficiency of disaster early warning.

[0058] Specifically, based on the stability propensity value, the data capacity of the database buffer corresponding to a single prediction region is adjusted to a corresponding value, wherein, The increase in the amount of data that can be accommodated is negatively correlated with the stability propensity value.

[0059] In this embodiment, optionally, The stability tendency value is compared with the first preset stability tendency value and the second preset stability tendency value; If the stability tendency value is less than or equal to the first preset stability tendency value, the amount of data that the database buffer corresponding to a single prediction region can hold will be adjusted to 1.23 times the initial amount of data that can be held. If the stability tendency value is less than or equal to the second preset stability tendency value and greater than the first preset stability tendency value, then the amount of data that the database buffer corresponding to a single prediction region can hold will be adjusted to 1.17 times the initial amount of data that can be held. If the stability tendency value is greater than the second preset stability tendency value, the amount of data that the database buffer corresponding to a single prediction region can hold will be adjusted to 1.06 times the initial amount of data that can be held. The first preset stability tendency value is 0.49, and the second preset stability tendency value is 0.61.

[0060] Specifically, the data capacity of the database buffer is adjusted based on the stability propensity value, which comprehensively reflects the overall stability and risk level of the prediction area. When the stability propensity value is low, the risk in the area is high, and data changes are more drastic, requiring a larger database buffer to store more data for deeper analysis and processing, and timely handling of various data processing needs in high-risk areas. The lower the stability propensity value, the higher the risk and the more drastic the data changes. To cope with potentially large amounts of data and complex situations, the data capacity of the database buffer is increased. The increase in the data capacity is negatively correlated with the stability propensity value; that is, the smaller the stability propensity value, the greater the increase in the data capacity. The data capacity of the database buffer is the maximum capacity of the buffer used for temporary data storage. A low stability propensity value means that the data will fluctuate and change significantly, requiring a larger buffer to store these changing data to ensure data integrity and continuity, facilitating subsequent analysis and processing. Dynamically adjusting the data capacity of the database buffer based on the stability propensity value rationally utilizes database resources, avoiding data loss or untimely processing due to an insufficient buffer, while also avoiding resource waste caused by an excessively large buffer. This improves the efficiency of disaster early warning.

[0061] Please see Figure 4 The diagram shown illustrates the logic for determining whether to adjust the weight ratio of each strongly reliable data point based on the fluctuation rate of weakly reliable data in an embodiment of the present invention. When adjusting the amount of data that the database buffer corresponding to a single prediction region can hold, the process of determining whether to adjust the weight ratio of each strongly reliable data point based on the fluctuation rate of weakly reliable data includes: Obtain the ambiguity of each weakly reliable data point, calculate the difference between the maximum and minimum values ​​of each ambiguity, and obtain the fluctuation rate of the weakly reliable data. When the fluctuation rate of weakly reliable data is less than or equal to the preset weakly reliable data fluctuation rate, disaster prediction is performed based on the redefined prediction data. When the fluctuation rate of weakly trusted data is greater than the preset fluctuation rate of weakly trusted data, the weight ratio of each strongly trusted data is adjusted based on the fluctuation rate of weakly trusted data.

[0062] Specifically, the preset weakly reliable data fluctuation rate is selected within the range [0.19, 0.25]. Those skilled in the art can select and determine it themselves. The preset weakly reliable data fluctuation rate can be determined by analyzing the periods in historical data where the actual value of weakly reliable data deviates significantly from the prediction, calculating the fluctuation range of the ambiguity of weakly reliable data during these periods, and subtracting the maximum value from the minimum value. In this embodiment, the preset weakly reliable data fluctuation rate is preferably 0.2.

[0063] Specifically, the weighting of highly reliable data is adjusted based on the fluctuation rate of weakly reliable data, which reflects the degree of volatility of this data. When the fluctuation of weakly reliable data is large, it significantly impacts overall disaster prediction. In this case, adjusting the weighting of highly reliable data balances the uncertainty of the weakly reliable data. Conversely, when the fluctuation of weakly reliable data is small, the impact on overall prediction is relatively small, and disaster prediction can be directly based on the redefined prediction data. The fluctuation rate of weakly reliable data is calculated by the difference between the maximum and minimum ambiguities of each weakly reliable data point, thus measuring its volatility. When the fluctuation rate exceeds a preset value, it indicates significant volatility. The fluctuation rate reflects the range of ambiguity fluctuations in weakly reliable data, embodying its instability. Large fluctuations in weakly reliable data can interfere with disaster prediction results. Adjusting the weighting of highly reliable data highlights its role and reduces the negative impact of its volatility. Deciding whether to adjust the weighting of highly reliable data based on its volatility allows for flexible responses to different data situations, improving the accuracy and reliability of disaster prediction and enhancing disaster early warning efficiency.

[0064] Specifically, the weighting of each piece of highly reliable data is adjusted based on the fluctuation rate of the weakly reliable data. The increase in the weighting percentage of each highly reliable data point is positively correlated with the fluctuation rate of the less reliable data point.

[0065] In this embodiment, optionally, Compare the weakly reliable data fluctuation rate with the first preset fluctuation rate comparison value and the second preset fluctuation rate comparison value; If the fluctuation rate of weakly trusted data is less than or equal to the first preset fluctuation rate, then the weight ratio of each strongly trusted data will be adjusted to 1.12 times the corresponding weight ratio. If the fluctuation rate of weakly trusted data is less than or equal to the second preset fluctuation rate and greater than the first preset fluctuation rate, then the weight ratio of each strongly trusted data will be adjusted to 1.21 times the corresponding weight ratio. If the fluctuation rate of weakly trusted data is greater than the second preset fluctuation rate, then the weight ratio of each strongly trusted data will be adjusted to 1.29 times the corresponding weight ratio; The first preset floating rate comparison value is 0.27, and the second preset floating rate comparison value is 0.31.

[0066] After adjusting the weighting of each highly reliable data point, disaster prediction is performed based on the newly determined prediction data.

[0067] Specifically, after adjusting the weights of each highly reliable data point, the weights of the less reliable data points are reduced according to their original weight ratios to ensure that the overall weights remain unchanged.

[0068] The specific process of disaster prediction based on the redefined forecast data includes: Data fusion involves using dynamically adjusted time weights to weight the historical sequences of each data source for the current prediction area, and using spatial distance weights to weight the data of the adjacent range of potential hazard points. By fusing multi-source data, a set of spatiotemporally weighted feature parameter values ​​reflecting the latest and most recent regional status are obtained.

[0069] Risk calculation involves inputting the weighted feature parameter values ​​into the pre-trained disaster prediction model. During the model inference phase, the weights of each input feature are dynamically assigned: for parameters marked as highly reliable, their feature weights use their initial weights or the weights corrected in the preceding steps; for weakly reliable parameters, the weights are adaptively adjusted according to a pre-defined ratio to ensure that the sum of the weights of all reliable data is 1. The model outputs two core results: disaster type, disaster risk level, and time. The disaster risk level and timeframe outputs a comprehensive risk level and an estimated time window for disaster occurrence. The comprehensive risk level includes blue, yellow, orange, and red alerts. The alert information is then pushed to the alert platform.

[0070] The time-series forecasting model takes a single parameter, such as historical time-series data of rainfall, as input. This historical time-series data can be structured monitoring data or processed remote sensing data. The time-series forecasting model outputs the predicted value of this parameter at the next time point.

[0071] For each type of parameter that needs to be predicted, including rainfall, soil moisture content, and displacement velocity, train a separate model. ARIMA, LSTM, or Prophet time-series prediction algorithms can be used, with supervised learning training utilizing historical data for that parameter.

[0072] The risk probability fitting model takes as input all prediction parameters and the base risk factor obtained by comparing these predictions with their respective preset critical alarm values. The output is a comprehensive predicted risk probability for the prediction area. The risk probability fitting model can be a machine learning classification model or a regression model. It is trained using historical data, with standardized prediction values ​​as input features and a label indicating whether a disaster occurred during the same historical period.

[0073] The disaster prediction model is used to determine the type and risk level of a disaster. Its input consists of multi-source feature parameter values ​​fused with dynamic spatiotemporal weights, i.e., data from each monitoring point processed with adjusted time and distance weights. The output includes the main potential disaster type, disaster risk level, and estimated occurrence time window. This disaster prediction model is a multi-task model. A classification model is used to determine the disaster type, with the input being a combination of anomalous patterns from highly reliable parameters. The risk level and time window can be determined based on the output of the risk probability fitting model, combined with a rule base.

[0074] The time-series forecasting model provides input to the risk probability fitting model. The risk probability, intermediate data, ambiguity, and stability tendency values ​​output by the risk probability fitting model are used to drive dynamic adjustment rules, adjusting time weights, distance weights, and data processing resource allocation. These adjustments are applied to the raw or forecast data, forming new fusion features, which are then input into the disaster prediction model for final accurate early warning. The entire process forms a closed loop from prediction to preliminary assessment to dynamic adjustment strategies to final accurate assessment.

[0075] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0076] 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. A disaster early warning and processing method based on multi-source heterogeneous data with dynamic spatiotemporal weights, characterized in that, include: For a single prediction region, input multi-source data to predict each parameter separately; Determine the ambiguity of each prediction parameter; Based on the degree of ambiguity, each prediction parameter is divided into strongly reliable data or weakly reliable data; The probability of predicted risk is determined based on the fitting of each prediction parameter. The risk propensity of a single prediction region is determined based on the fitted stable propensity value. When a single prediction region is identified as a high-risk region, the distance weight and time weight of the single prediction region are adjusted. The data stability index is used to determine whether to adjust the allocation of data processing resources. Disaster prediction is based on various forecast data.

2. The multi-source heterogeneous data disaster early warning processing method based on dynamic spatiotemporal weights according to claim 1, characterized in that, Determining the stability propensity value includes: The data stability index is obtained by calculating the ratio of the number of highly reliable data points to the total number of predicted data points. For a single predicted data point, calculate the elasticity difference between the corresponding preset critical alarm value and the predicted data, and calculate the elasticity ratio between the elasticity difference and the predicted data. The average value of the elasticity ratio of each highly reliable data point is calculated to obtain the tolerance ratio. Determine the probability of safety based on the predicted probability of risk; The stability tendency value is obtained by multiplying the effective data ratio, tolerance ratio, and safety probability by their respective weight coefficients and summing them.

3. The multi-source heterogeneous data disaster early warning processing method based on dynamic spatiotemporal weights according to claim 2, characterized in that, The process of determining the risk propensity of a single prediction region based on the fitted stable propensity value includes: When the stability tendency value is less than or equal to the preset stability tendency value, a single prediction area is identified as a high-risk area, and the time weights of each input multi-source data are adjusted based on the historical change magnitude characterization value. When the stability tendency value is greater than the preset stability tendency value, a single prediction area is identified as a low-risk area, and disaster prediction is carried out based on each prediction data.

4. The multi-source heterogeneous data disaster early warning processing method based on dynamic spatiotemporal weights according to claim 3, characterized in that, The time weights of each input multi-source data are adjusted based on historical change magnitude representation values, whereby... Obtain the historical actual monitoring data set corresponding to a single predicted data point to plot the time-domain curve of the actual monitoring data; The average slope of the time-domain curve of the actual monitoring data is calculated to obtain the average slope of the abnormal segment. The average slope of the time-domain curves of each actual monitoring data point under the condition that a single predicted data point is identified as a low-risk area is obtained by averaging the average slope of the stable segment. The ratio of the average slope of the abnormal segment to the average slope of the stable segment is calculated to obtain the historical variation amplitude value. The increase in the recent time weight coefficients of the input data group corresponding to a single predicted data is positively correlated with the historical change magnitude representation value.

5. The multi-source heterogeneous data disaster early warning processing method based on dynamic spatiotemporal weights according to claim 4, characterized in that, When correcting the time weights for each input multi-source data, the distance weights for the corresponding ranges are corrected based on the span reference value. For a single range, determine the span coefficient between it and its adjacent ranges far from the center point; The average value of each span coefficient within a single prediction region is calculated to obtain the span reference value; The increase in the distance weight of the adjacent range of the potential hazard area is positively correlated with the span reference value.

6. The multi-source heterogeneous data disaster early warning processing method based on dynamic spatiotemporal weights according to claim 5, characterized in that, When adjusting the weights for each distance, the data processing resource allocation is adjusted based on the data stability index, including: When the data stability index is less than or equal to the preset data stability index, the data volume of each input multi-source data is adjusted to the corresponding value based on the data stability index. When the data stability index is greater than the preset data stability index, the amount of data that the database buffer can hold for a single prediction region will be adjusted to the corresponding value based on the stability tendency value.

7. The multi-source heterogeneous data disaster early warning processing method based on dynamic spatiotemporal weights according to claim 6, characterized in that, Based on the data stability index, the data volume of each input multi-source data is adjusted to the corresponding value, where, The increase in data volume is negatively correlated with the data stability index.

8. The disaster early warning and processing method based on dynamic spatiotemporal weights for multi-source heterogeneous data according to claim 7, characterized in that, Based on the stable tendency value, the data capacity of the database buffer corresponding to a single prediction region is adjusted to the corresponding value, where, The increase in the amount of data that can be accommodated is negatively correlated with the stability propensity value.

9. The multi-source heterogeneous data disaster early warning processing method based on dynamic spatiotemporal weights according to claim 8, characterized in that, When adjusting the data capacity of the database buffer corresponding to a single prediction region, the weighting of each strong confidence data point is adjusted based on the fluctuation rate of the weak confidence data, including: Obtain the ambiguity of each weakly reliable data point, calculate the difference between the maximum and minimum values ​​of each ambiguity, and obtain the fluctuation rate of the weakly reliable data. When the fluctuation rate of weakly reliable data is less than or equal to the preset weakly reliable data fluctuation rate, disaster prediction is performed based on the redefined prediction data. When the fluctuation rate of weakly trusted data is greater than the preset fluctuation rate of weakly trusted data, the weight ratio of each strongly trusted data is adjusted based on the fluctuation rate of weakly trusted data.

10. The multi-source heterogeneous data disaster early warning processing method based on dynamic spatiotemporal weights according to claim 9, characterized in that, The weighting of each piece of highly reliable data is adjusted based on the fluctuation rate of the weakly reliable data. The increase in the weighting percentage of each highly reliable data point is positively correlated with the fluctuation rate of the less reliable data point.

Images