Flood control early warning decision-making system and method based on artificial intelligence

Abstract

The invention discloses a flood control early warning decision-making system and method based on artificial intelligence, and relates to the technical field of flood control data monitoring, and the system comprises a target monitoring area marking module, a target sensor marking module, a reconstruction sensor group generation module, a priority analysis module, and an early warning response switching module. The target monitoring area marking module is used for marking a flood control monitoring area which stores and records flood events and is provided with no redundant sensors in the arranged sensor devices as a target monitoring area; the target sensor marking module is used for outputting and marking the sensor with physical damage as a target sensor; the reconstructed sensor group generation module is used for generating a reconstructed sensor group taking a target sensor as a main sensor; the early warning response switching module is used for carrying out replacement reminding when the isolated sensor responds to damage abnormity in the early warning event; and when the to-be-updated sensor responds to damage abnormity in the early warning event, carrying out optimal-level redundant sensor function switching.

Description

Technical Field

[0001] This invention relates to the field of flood control data monitoring technology, specifically to a flood control early warning decision-making system and method based on artificial intelligence. Background Technology

[0002] Floods, as a common and devastating natural disaster worldwide, cause enormous losses to human society every year. Traditional flood warning systems mainly rely on a centralized data processing model, where data collected by sensors is transmitted to a data center for centralized processing and analysis. Traditional warning systems often rely on data from a single type of sensor, such as water level sensors or rain gauges, lacking the fusion and analysis of multi-source data, and failing to fully utilize various types of information to improve the accuracy and reliability of warnings. Although existing technologies mention modular reconfiguration of sensors, there is not much data support regarding the accuracy of reconfigurable sensors and the selection of reconfiguration methods for flood warning applications, thus hindering the effective verification of the safety, rationality, and effectiveness of the application. Summary of the Invention

[0003] The purpose of this invention is to provide a flood control early warning decision-making system and method based on artificial intelligence to solve the problems raised in the prior art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a flood prevention early warning decision-making method based on artificial intelligence, the method comprising the following steps: Step S100: Mark the flood monitoring area where the stored flood event records and the deployed sensor devices do not have redundant sensors as the target monitoring area; a flood event refers to an event that responds to the system's early warning mechanism, and a redundant sensor refers to a sensor that performs a function switch after the original sensor is damaged, based on the modular reconfigurable design; extract the monitoring data of historical flood events recorded in the target monitoring area, analyze and output the data, and mark the sensors with physical damage as target sensors; The purpose of this application in analyzing the target monitoring area is to determine whether a single-function sensor can be reconfigured to switch functions when it may be damaged after a flood event, thereby avoiding the surge in sensor replacement workload caused by large-scale damage and a series of safety issues caused by untimely replacement. Step S200: Determine the target sensor's functional type, output the redundant sensors for which reconfigurable design can be implemented and their corresponding functional types; combine the target sensor and the corresponding redundant sensors one-to-one to generate a reconfigurable sensor group with the target sensor as the main sensor; Step S300: Extract the monitoring data recorded by the reconfigurable sensor group, analyze whether the target sensor can switch the function of the redundant sensors when responding to damage anomalies; and when there is a reconfigurable sensor group consisting of multiple different redundant sensors corresponding to the same main sensor, output the priority of the switchable redundant sensors; Step S400: Mark the target sensors that failed to form a reconfigurable sensor group in step S200 and those that did not have redundant sensor switching based on monitoring data analysis in step S300 as isolated sensors; the target sensors corresponding to the other cases are sensors to be updated; when an isolated sensor responds to damage anomaly in an early warning event, a replacement reminder is issued; when a sensor to be updated responds to damage anomaly in an early warning event, the optimal level redundant sensor function is switched.

[0005] Furthermore, step S100, which analyzes the output and marks sensors with physical damage as target sensors, includes the following steps: Step S110: Monitoring data refers to various physical data recorded and transmitted by the sensors, as well as the location data of each sensor; the sensors in the target monitoring area are initially divided according to their corresponding functional types, and sensors of the same type are marked as a monitoring group. Based on spatial mapping, state coding and interaction design, a visualization model corresponding to the same monitoring group is generated. The visualization model marks the location data of each sensor in the same monitoring group and establishes a network topology and communication link layer. Adjacent sensors are connected by arrows to show the communication relationship of the self-organizing network. Step S120: Extract the hardware data recorded by the sensors in the same monitoring group at the first moment. The hardware data includes the power supply voltage, signal-to-noise ratio, and probe response time. Set the parameter thresholds for the corresponding data. When the value of the corresponding parameter in the hardware data exceeds the corresponding parameter threshold for n consecutive sampling periods, it is marked as feature A. Obtain the probability P(A|fault) of feature A occurring when the sensor is faulty and the probability P(A|normal) of feature A occurring when the sensor is normal. Calculate P(fault|A) using Bayes' theorem: P(fault|A) = [P(A|fault)]. P(fault)] / [P(A|fault)] P(Fault) + P(A|Normal) P(normal)], where P(fault) is the prior probability of sensor fault in the initial state, P(normal) = 1 - P(fault); Step S130: Extract the parameter values ​​corresponding to the hardware data at the second moment. If the parameter values ​​are fixed within a preset time period, and the difference between the monitoring values ​​recorded by the closest adjacent sensors in the same monitoring group that have a communication relationship is greater than the first threshold, mark them as feature B; obtain the probability P(B|fault) of feature B occurring when the sensor is faulty and the probability P(B|normal) of feature B occurring when the sensor is normal, and calculate the updated confidence level P(fault|A,B)=[P(B|fault)]. P(fault|A)] / [P(B|fault)] P(Fault|A) + P(B|Normal) P(Normal|A)], P(Normal|A) = 1 = P(Fault|A); Step S140: Set the confidence threshold P0. When P(fault|A,B)>P0, determine that the corresponding sensor is a target sensor with physical damage.

[0006] The above method can quickly and accurately identify sensors that are physically damaged during or after flooding events, without the need for external devices for inspection or manual verification. Damage can be effectively determined simply by uploading and analyzing the sensor's own data, thus reducing the resource cost of physical sensor detection and improving detection efficiency.

[0007] Furthermore, step S200 includes the following specific steps: Reconfigurable design refers to the dynamic reconfiguration of sensor functions by using modular hardware design and cross-physical quantity sensing algorithms to combine redundant sensors with target sensors. Cross-physical quantity sensing algorithms include algorithms built based on physical formulas and algorithms built based on implicit correlations between physical quantities; Based on the physical relationship established by the cross-physical quantity sensing algorithm, the sensor with the corresponding function of the physical relationship with the target sensor is regarded as a redundant sensor that can be reconfigured.

[0008] Furthermore, step S300 includes the following: Step S310: Mark the data monitoring area of ​​the target sensor, extract the monitoring data recorded by the redundant sensor in the reconstructed sensor group to which the target sensor belongs within the data monitoring area as the first monitoring data, and calculate the derived target data obtained from the first monitoring data based on the cross-physical quantity sensing algorithm between the redundant sensor and the target sensor. Step S320: Identify the nearest neighboring sensors in the visualization model where the target sensor is located, which have network communication connections, as key sensors. Using key sensors as core sensors, find other sensors that have network communication connections with the core sensors to form the sensor group to be analyzed. Generate independent elements for each sensor in the sensor group to be analyzed and record the monitoring data of each independent element at the same monitoring time as the dependent variable. The data difference refers to the monitoring value recorded by the core sensor minus the corresponding value of other sensors. Extract all independent variable factors that affect the changes in the monitoring data recorded by the corresponding function type of the core sensor, and calculate the numerical difference of each independent variable factor in the same independent element according to the relationship between the data difference and the calculation object as the independent variable of the independent element. Based on the data group composed of independent and dependent variables of each independent element record of the same core sensor, construct the linear regression equation for the corresponding core sensor. Step S330: Set the monitoring data of the target sensor as an unknown x, and extract all independent variables obtained from the target sensor and the core sensor based on the same independent variable factors. Substitute these variables into the linear regression equation of the core sensor to obtain the x value, and use x as the theoretical target data. Calculate the difference between the theoretical target data and the derived target data. When the difference is less than or equal to a preset difference threshold, mark the corresponding redundant sensor so that the redundant sensor can switch functions when the target sensor responds to abnormal damage. When the target sensor has physical damage, the acquired data cannot be used as a valid basis for judgment. Therefore, this application constructs a network topology diagram of the same type of sensor in the monitoring area in a visualization model to realize data analysis, effectively filling the data gaps. Step S340: Traverse all redundant sensors corresponding to the same target sensor, executing steps S310 to S330, and mark redundant sensors that can be functionally switched as redundant sensors to be calibrated; calculate the difference between the theoretical target data and the derived target data in the analysis events corresponding to the same redundant sensor to be calibrated under different first monitoring data or theoretical target data acquisition conditions, forming a difference analysis set for the same redundant sensor to be calibrated; based on the difference analysis set, use the formula: F={[∑(f1-f0)} 2 ] / u} 1 / 2 Calculate the fluctuation coefficient F for each redundant sensor to be calibrated, where f1 represents the value of each element recorded in the difference analysis set, f0 represents the average value of all elements in the difference analysis set, u represents the number of elements in the difference analysis set, and [∑(f1-f0)] represents the value of each element recorded in the difference analysis set. 2 The expression represents summing the squares of the differences between each element in the set and the mean. Step S350: Extract the maximum difference f between the theoretical target data and the derived target data recorded by all redundant sensors to be calibrated for the same target sensor. max Using the formula: Z=0.55 f max +0.45 F; Calculate the priority index Z for each redundant sensor to be calibrated in the corresponding target sensor; based on the priority index, sort the redundant sensors to be calibrated in ascending order of priority index value to generate the priority of switchable redundant sensors.

[0009] Analyzing the difference between derived data and theoretical data helps determine the reliability of data after switching using redundant sensors. When multiple redundant sensors can be switched, reliability can be further combined with a comprehensive analysis from a stability perspective to select redundant sensors that can be switched in the event of damage to the target sensor. This improves the adaptability to various scenarios and provides strong security for flood warning systems in the direction of sensor anomaly monitoring.

[0010] Furthermore, step S400 includes the following steps: The system retrieves the function type of the sensor to be updated that has a damaged or abnormal response. The system automatically activates the redundancy resource pool, which stores all switchable redundant sensors recorded in the sensor to be updated. The system then selects the redundancy sensor with the highest priority as the optimal redundant sensor and executes the corresponding conversion process. If the optimal redundant sensor is damaged or the conversion fails, the system continues to select the redundant sensor with the next highest priority until the conversion is successful.

[0011] An artificial intelligence-based flood warning and decision-making system includes a target monitoring area marking module, a target sensor marking module, a reconfigurable sensor group generation module, a priority analysis module, and an early warning response switching module. The target monitoring area marking module is used to mark flood monitoring areas where the stored and recorded flood events are located and where none of the deployed sensor devices are redundant as target monitoring areas; The target sensor marking module is used to output and mark sensors with physical damage as target sensors; The reconfigurable sensor group generation module is used to combine the target sensor with the corresponding redundant sensor one-to-one to generate a reconfigurable sensor group with the target sensor as the main sensor. The priority analysis module is used to output the priority of switchable redundant sensors; The early warning response switching module is used to provide a replacement reminder when an isolated sensor responds abnormally to an early warning event; and to switch the optimal redundant sensor function when a sensor to be updated responds abnormally to an early warning event.

[0012] Furthermore, the target sensor marking module includes a monitoring group division unit, a visualization model construction unit, and a confidence analysis unit; The monitoring group division unit is used to initially divide the sensors within the target monitoring area according to their corresponding functional types, marking sensors of the same type as a monitoring group. The visualization model building unit is used to generate visualization models corresponding to the same monitoring group based on spatial mapping, state coding, and interaction design. The confidence analysis unit is used to analyze the confidence level of the updated data in conjunction with the visualization model; and to set a confidence threshold. When the confidence level is greater than the confidence threshold, the corresponding sensor is determined to be a target sensor with physical damage.

[0013] Furthermore, the priority analysis module includes a derivation target data calculation unit, a theoretical target data calculation unit, a volatility coefficient calculation unit, and a priority index calculation unit; The target data derivation calculation unit is used to calculate the target data derived from the first monitoring data; The theoretical target data calculation unit is used to construct the linear regression equation of the corresponding core sensor. The monitoring data of the target sensor is set as the unknown, and all independent variables obtained by the target sensor and the core sensor on the basis of the same independent variable factors are extracted and substituted into the linear regression equation of the core sensor to obtain the unknown value. The unknown is used as the theoretical target data. The fluctuation coefficient calculation unit is used to calculate the difference between the theoretical target data and the derived target data in the analysis event corresponding to different first monitoring data or theoretical target data acquired by the same redundant sensor to be calibrated, thus forming a difference analysis set for the same redundant sensor to be calibrated; based on the difference analysis set, the fluctuation coefficient F of each redundant sensor to be calibrated is calculated. The priority index calculation unit is used to extract the maximum difference between the theoretical target data and the derived target data recorded by all redundant sensors to be calibrated for the same target sensor; calculate the priority index Z of each redundant sensor to be calibrated in the corresponding target sensor; and based on the priority index, sort the redundant sensors to be calibrated in ascending order of the priority index value to generate the priority of switchable redundant sensors.

[0014] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention analyzes flood events recorded by the flood warning system to identify abnormally damaged sensors and extracts the monitoring data recorded by the sensors. It further monitors and judges the real-time sensor status from the data level, reducing the resource costs of manual and equipment work. 2. This application modularly reconstructs damaged sensors, enabling timely switching of redundant sensor functions in emergency situations to ensure uninterrupted data transmission and effective data support for the overall flood control system. This avoids instability caused by sensor damage and increased errors in data analysis. Simultaneously, it conducts further detailed data analysis on various sensors capable of functional conversion, enabling reasonable priority selection to adapt sensor configurations to different environmental parameters and flood control data, thus achieving intelligent and diversified early warning systems. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the structure of a flood early warning decision-making method based on artificial intelligence according to the present invention. Detailed Implementation

[0016] Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0017] Example: Figure 1 As shown, this invention provides a flood prevention early warning decision-making method based on artificial intelligence, the method comprising the following steps: Step S100: Mark the flood monitoring area where the stored flood event records and the deployed sensor devices do not have redundant sensors as the target monitoring area; a flood event refers to an event that responds to the system's early warning mechanism, and a redundant sensor refers to a sensor that performs a function switch after the original sensor is damaged, based on the modular reconfigurable design; extract the monitoring data of historical flood events recorded in the target monitoring area, analyze and output the data, and mark the sensors with physical damage as target sensors; The purpose of this application in analyzing the target monitoring area is to determine whether a single-function sensor can be reconfigured to switch functions when it may be damaged after a flood event, thereby avoiding the surge in sensor replacement workload caused by large-scale damage and a series of safety issues caused by untimely replacement. Step S200: Determine the target sensor function type, output the redundant sensors that can be reconfigured and their corresponding function types; combine the target sensor and the corresponding redundant sensor one-to-one to generate a reconfigurable sensor group with the target sensor as the main sensor; For example, a pressure sensor (redundant sensor 21) and a water level sensor (target sensor 2) can form a reconfigurable sensor group. The water level sensor serves as the main sensor, and the pressure sensor serves as the redundant sensor. When the water level sensor is damaged, the system responds by switching the sensor group to the pressure sensor. The water level data is estimated by performing a physical calculation, P=ρgh, based on the pressure data measured by the pressure sensor. The above is based on a physical formula. There are also cross-physical quantity mapping algorithms. For example, in a flood environment, changes in water level directly affect the surrounding temperature and humidity (e.g., when the water level rises, the near-surface humidity will increase significantly, and the temperature will be lower than the air temperature due to water evaporation). The two are strongly correlated. A mapping relationship model between temperature / humidity and water level can be trained using historical data to achieve data prediction. The temperature / humidity sensor is the redundant sensor 22 in the figure. Target sensors 1, 2, 3, and 4 are sensors of the same functional type. Step S300: Extract the monitoring data recorded by the reconfigured sensor group, analyze whether the target sensor can switch the function of the redundant sensor when responding to the damage anomaly; and when there is a reconfigured sensor group consisting of multiple different redundant sensors corresponding to the same main sensor, output the priority of the switchable redundant sensor. Step S400: Mark the target sensors that failed to form a reconfigurable sensor group in step S200 and those that did not have redundant sensor switching based on monitoring data analysis in step S300 as isolated sensors; the target sensors corresponding to the other cases are sensors to be updated; when an isolated sensor responds to damage anomaly in an early warning event, a replacement reminder is issued; when a sensor to be updated responds to damage anomaly in an early warning event, the optimal level redundant sensor function is switched.

[0018] Step S100, which analyzes the output and marks sensors with physical damage as target sensors, includes the following steps: Step S110: Monitoring data refers to various physical data recorded and transmitted by the sensors, as well as the location data of each sensor; the sensors in the target monitoring area are initially divided according to their corresponding functional types, and sensors of the same type are marked as a monitoring group. Based on spatial mapping, state coding and interaction design, a visualization model corresponding to the same monitoring group is generated. The visualization model marks the location data of each sensor in the same monitoring group and establishes a network topology and communication link layer. Adjacent sensors are connected by arrows to show the communication relationship of the self-organizing network. Step S120: Extract the hardware data recorded by the sensors in the same monitoring group at the first moment. The hardware data includes the power supply voltage, signal-to-noise ratio, and probe response time. Set the parameter thresholds for the corresponding data. When the value of the corresponding parameter in the hardware data exceeds the corresponding parameter threshold for n consecutive sampling periods, it is marked as feature A. Obtain the probability P(A|fault) of feature A occurring when the sensor is faulty and the probability P(A|normal) of feature A occurring when the sensor is normal. Calculate P(fault|A) using Bayes' theorem: P(fault|A) = [P(A|fault)]. P(fault)] / [P(A|fault)] P(Fault) + P(A|Normal) P(normal)], where P(fault) is the prior probability of sensor fault in the initial state, P(normal) = 1 - P(fault); As shown in the example: Feature A is defined as "the water level sensor responds to water level changes in 200ms, exceeding the threshold by 100ms". Fault data: 100 records of confirmed sensor faults in history, of which feature A appeared 80 times; Normal data: 1000 records of normal sensor operation (marked "normal"), of which feature A appeared 20 times; then P(A|fault) = 80 / 100 = 80%; P(A|normal) = 20 / 1000 = 2%.

[0019] Step S130: Extract the parameter values ​​corresponding to the hardware data at the second moment. If the parameter values ​​are fixed within a preset time period, and the difference between the monitoring values ​​recorded by the closest adjacent sensors in the same monitoring group that have a communication relationship is greater than the first threshold, mark them as feature B; obtain the probability P(B|fault) of feature B occurring when the sensor is faulty and the probability P(B|normal) of feature B occurring when the sensor is normal, and calculate the updated confidence level P(fault|A,B)=[P(B|fault)]. P(fault|A)] / [P(B|fault)] P(Fault|A) + P(B|Normal) P(Normal|A)], P(Normal|A) = 1 = P(Fault|A); Step S140: Set the confidence threshold P0. When P(fault|A,B)>P0, determine that the corresponding sensor is a target sensor with physical damage.

[0020] The above method can quickly and accurately identify sensors that are physically damaged during or after flooding events, without the need for external devices for inspection or manual verification. Damage can be effectively determined simply by uploading and analyzing the sensor's own data, thus reducing the resource cost of physical sensor detection and improving detection efficiency.

[0021] Step S200 includes the following specific steps: Reconfigurable design refers to the dynamic reconfiguration of sensor functions by using modular hardware design and cross-physical quantity sensing algorithms to combine redundant sensors with the target sensor. The redundant sensor and the original sensor have different functions, but their structures are integrated into a single sensor. Cross-physical quantity sensing algorithms include algorithms built based on physical formulas and algorithms built based on implicit correlations between physical quantities; Based on the physical relationship established by the cross-physical quantity sensing algorithm, the sensor with the corresponding function of the physical relationship with the target sensor is regarded as a redundant sensor that can be reconfigured.

[0022] As shown in the example: pressure sensor → water level monitoring, replacing the damaged water level gauge; at this time, the target sensor is the water level gauge, and the redundant sensor is the pressure sensor; this reconfiguration design is an algorithm built based on the physical formula, P=ρgh; Temperature and humidity sensor → water level monitoring, replacing the damaged water level gauge; at this time, the target sensor is the water level gauge, and the redundant sensor is the temperature and humidity sensor; this reconfiguration design is based on the algorithm built on the implicit correlation between physical quantities, and the "temperature and humidity-water level" mapping relationship is trained through machine learning (such as random forest, LSTM).

[0023] Step S300 includes the following: Step S310: Mark the data monitoring area of ​​the target sensor, extract the monitoring data recorded by the redundant sensor in the reconstructed sensor group to which the target sensor belongs within the data monitoring area as the first monitoring data, and calculate the derived target data obtained from the first monitoring data based on the cross-physical quantity sensing algorithm between the redundant sensor and the target sensor. Step S320: Identify the nearest neighboring sensors in the visualization model where the target sensor is located, which have network communication connections, as key sensors. Using key sensors as core sensors, find other sensors that have network communication connections with the core sensors to form the sensor group to be analyzed. Generate independent elements for each sensor in the sensor group to be analyzed and record the monitoring data of each independent element at the same monitoring time as the dependent variable. The data difference refers to the monitoring value recorded by the core sensor minus the corresponding value of other sensors. Extract all independent variable factors that affect the changes in the monitoring data recorded by the corresponding function type of the core sensor, and calculate the numerical difference of each independent variable factor in the same independent element according to the relationship between the data difference and the calculation object as the independent variable of the independent element. Based on the data group composed of independent and dependent variables of each independent element record of the same core sensor, construct the linear regression equation for the corresponding core sensor. Step S330: Set the monitoring data of the target sensor as an unknown x, and extract all independent variables obtained from the target sensor and the core sensor based on the same independent variable factors. Substitute these variables into the linear regression equation of the core sensor to obtain the x value, and use x as the theoretical target data. Calculate the difference between the theoretical target data and the derived target data. When the difference is less than or equal to a preset difference threshold, mark the corresponding redundant sensor so that the redundant sensor can switch functions when the target sensor responds to abnormal damage. When the target sensor has physical damage, the acquired data cannot be used as a valid basis for judgment. Therefore, this application constructs a network topology diagram of the same type of sensor in the monitoring area in a visualization model to realize data analysis, effectively filling the data gaps. Step S340: Traverse all redundant sensors corresponding to the same target sensor, executing steps S310 to S330, and mark redundant sensors that can be functionally switched as redundant sensors to be calibrated; calculate the difference between the theoretical target data and the derived target data in the analysis events corresponding to the same redundant sensor to be calibrated under different first monitoring data or theoretical target data acquisition conditions, forming a difference analysis set for the same redundant sensor to be calibrated; based on the difference analysis set, use the formula: F={[∑(f1-f0)} 2 ] / u} 1 / 2 Calculate the fluctuation coefficient F for each redundant sensor to be calibrated, where f1 represents the value of each element recorded in the difference analysis set, f0 represents the average value of all elements in the difference analysis set, u represents the number of elements in the difference analysis set, and [∑(f1-f0)] represents the value of each element recorded in the difference analysis set. 2 The expression represents summing the squares of the differences between each element in the set and the mean. Step S350: Extract the maximum difference f between the theoretical target data and the derived target data recorded by all redundant sensors to be calibrated for the same target sensor. max Using the formula: Z=0.55 f max +0.45 F; Calculate the priority index Z for each redundant sensor to be calibrated in the corresponding target sensor; based on the priority index, sort the redundant sensors to be calibrated in ascending order of priority index value to generate the priority of switchable redundant sensors.

[0024] Analyzing the difference between derived data and theoretical data helps determine the reliability of data after switching using redundant sensors. When multiple redundant sensors can be switched, reliability can be further combined with a comprehensive analysis from a stability perspective to select redundant sensors that can be switched in the event of damage to the target sensor. This improves the adaptability to various scenarios and provides strong security for flood warning systems in the direction of sensor anomaly monitoring.

[0025] As shown in the example: the target sensor is a water level gauge, and the redundant sensor is a pressure sensor; when the water level gauge is damaged, the pressure data corresponding to the water level gauge monitoring area is obtained as the first monitoring data. If there is no real-time pressure data, the nearest pressure sensor can be found through the network topology to construct a linear regression equation and obtain the corresponding value. The cross-physical quantity sensing algorithm between the water level gauge and the pressure sensor is P=ρgh, and the corresponding target data h1 can be obtained through this formula; The visualization model identifies water level gauge 2 as the core sensor. It's crucial to ensure the key sensor is functioning correctly; otherwise, the search continues until the key sensor is determined. Based on water level gauge 2, the sensor group to be analyzed is identified, including sensor 3 and sensor 4. Independent variables are generated as (water level gauge 2, water level gauge 2), (water level gauge 2, water level gauge 4). The dependent variable in the independent variables is the water level difference, and the independent variables are factors affecting water level such as flow velocity, terrain difference, and spatial distance. These data should be acquired when there is no dynamic rainfall and the data tends to be stable. A linear regression equation corresponding to water level gauge 2 is generated, and the theoretical target data h2 can be obtained by inputting the independent variables. If the difference h1-h2 is less than the difference threshold, it indicates that the water level data obtained through redundant sensor analysis has a certain standard and can be temporarily used as substitute data for damaged sensors. This reduces data loss caused by sudden sensor failure and inability to replace them in time, thus mitigating the risks and difficulties brought to monitoring.

[0026] Step S400 includes the following steps: The system retrieves the function type of the sensor to be updated that has a damaged or abnormal response. The system automatically activates the redundancy resource pool, which stores all switchable redundant sensors recorded in the sensor to be updated. The system then selects the redundancy sensor with the highest priority as the optimal redundant sensor and executes the corresponding conversion process. If the optimal redundant sensor is damaged or the conversion fails, the system continues to select the redundant sensor with the next highest priority until the conversion is successful.

[0027] The conversion process is as follows: A "function switching command" is sent to the temperature and humidity sensor via a bus protocol (such as CANopen), switching its physical interface from "temperature and humidity mode" to "water level monitoring mode"; for example, closing the vent valve of the humidity probe (to prevent water ingress) and activating the probe's waterproof protection circuit. The system dynamically loads the water level monitoring driver (replacing the original temperature and humidity driver) and calls the "cross-physical quantity mapping model" to convert the original data. Simultaneously, an "emergency alternative data" label is added to the data output terminal to indicate to the decision-making system that the data is an estimate (to reduce the risk of misjudgment).

[0028] An artificial intelligence-based flood warning and decision-making system includes a target monitoring area marking module, a target sensor marking module, a reconfigurable sensor group generation module, a priority analysis module, and an early warning response switching module. The target monitoring area marking module is used to mark flood monitoring areas where the stored and recorded flood events are located and where none of the deployed sensor devices are redundant as target monitoring areas; The target sensor marking module is used to output and mark sensors with physical damage as target sensors; The reconfigurable sensor group generation module is used to combine the target sensor with the corresponding redundant sensor one-to-one to generate a reconfigurable sensor group with the target sensor as the main sensor. The priority analysis module is used to output the priority of switchable redundant sensors; The early warning response switching module is used to provide a replacement reminder when an isolated sensor responds abnormally to an early warning event; and to switch the optimal redundant sensor function when a sensor to be updated responds abnormally to an early warning event.

[0029] The target sensor marking module includes a monitoring group division unit, a visualization model construction unit, and a confidence analysis unit; The monitoring group division unit is used to initially divide the sensors within the target monitoring area according to their corresponding functional types, marking sensors of the same type as a monitoring group. The visualization model building unit is used to generate visualization models corresponding to the same monitoring group based on spatial mapping, state coding, and interaction design. The confidence analysis unit is used to analyze the confidence level of the updated data in conjunction with the visualization model; and to set a confidence threshold. When the confidence level is greater than the confidence threshold, the corresponding sensor is determined to be a target sensor with physical damage.

[0030] The priority analysis module includes a derivation target data calculation unit, a theoretical target data calculation unit, a volatility coefficient calculation unit, and a priority index calculation unit; The target data derivation calculation unit is used to calculate the target data derived from the first monitoring data; The theoretical target data calculation unit is used to construct the linear regression equation of the corresponding core sensor. The monitoring data of the target sensor is set as the unknown, and all independent variables obtained by the target sensor and the core sensor on the basis of the same independent variable factors are extracted and substituted into the linear regression equation of the core sensor to obtain the unknown value. The unknown is used as the theoretical target data. The fluctuation coefficient calculation unit is used to calculate the difference between the theoretical target data and the derived target data in the analysis event corresponding to different first monitoring data or theoretical target data acquired by the same redundant sensor to be calibrated, thus forming a difference analysis set for the same redundant sensor to be calibrated; based on the difference analysis set, the fluctuation coefficient F of each redundant sensor to be calibrated is calculated. The priority index calculation unit is used to extract the maximum difference between the theoretical target data and the derived target data recorded by all redundant sensors to be calibrated for the same target sensor; calculate the priority index Z of each redundant sensor to be calibrated in the corresponding target sensor; and based on the priority index, sort the redundant sensors to be calibrated in ascending order of the priority index value to generate the priority of switchable redundant sensors.

[0031] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An artificial intelligence-based flood warning decision-making method, characterized in that: The method comprises the following steps: Step S100: Mark the flood control monitoring area where the sensor device recording the flood event and the layout does not have redundant sensor settings as the target monitoring area; the flood event refers to the event in response to the system warning mechanism; the redundant sensor refers to the sensor corresponding to the function switching implemented after the original sensor function is damaged based on the module reconfigurable design; extract the monitoring data of the target monitoring area recording the historical flood event, analyze and output, and mark the target sensor with physical damage as the target sensor; Step S200: Determine the function type of the target sensor, output the redundant sensor that can implement the reconfigurable design and the corresponding function type; combine the target sensor with the corresponding redundant sensor one by one to generate a reconfigured sensor group with the target sensor as the main sensor; Step S300: Extract the monitoring data recorded by the reconfigured sensor group, analyze whether the target sensor can switch the function of the redundant sensor in response to damage abnormalities; and when there is a reconfigured sensor group composed of the same main sensor and multiple different redundant sensors, output the priority of the switchable redundant sensor; Step S400: Mark the target sensor in step S200 that fails to form a reconfigured sensor group and the target sensor in step S300 that does not have redundant sensor switching based on monitoring data analysis as an isolated sensor; the remaining target sensors are updated sensors; when the isolated sensor responds to damage abnormalities in the warning event, it is replaced; when the updated sensor responds to damage abnormalities in the warning event, the optimal redundant sensor function switching is performed. 2.The flood early warning decision-making method of artificial intelligence according to claim 1, characterized in that: The step S100 of analyzing and outputting the target sensor with physical damage includes the following steps: Step S110: The monitoring data refers to various physical data recorded and transmitted by the sensor and the position data of each sensor; the sensors in the target monitoring area are preliminarily divided according to the corresponding function type, the same type of sensor is marked as a monitoring group, and a visualization model corresponding to the same monitoring group is generated based on spatial mapping, state coding and interaction design; the visualization model marks the position data of each sensor in the same monitoring group and establishes a network topology and communication link layer to connect adjacent sensors with arrows to show the communication relationship of the self-organizing network; Step S120: extracting hardware data recorded by the sensor in the same monitoring group at the first time, the hardware data including power supply voltage, signal noise ratio and probe response duration; setting parameter threshold of corresponding data, when there is hardware data corresponding parameter value in continuous n sampling periods, marking as feature A; obtaining probability P(A|fault) of feature A appearing when the sensor fails and probability P(A|normal) of feature A appearing when normal, calculating P(fault|A) by using Bayes formula, P(fault|A)=[P(A|fault) P(fault)] / [P(A|fault) P(fault)+P(A|normal) P(normal)], wherein P(fault) is prior probability of setting sensor failure in initial state, P(normal)=1-P(fault); Step S130: Extract the parameter value corresponding to the second time hardware data, which is a fixed value within a preset time length, and the monitoring value difference recorded by the adjacent sensors with communication relationship and closest distance in the same monitoring group in the visualization model is greater than the first threshold value, and mark it as feature B; obtain the probability P(B|failure) of feature B appearing when the historical storage sensor fails and the probability P(B|normal) of feature B appearing when normal, and calculate the updated confidence P(failure|A,B)=[P(B|failure) P(failure|A)] / [P(B|failure) P(failure|A)+P(B|normal) P(normal|A)], P(normal|A)=1=P(failure|A); Step S140: Set a confidence threshold P0, when P(failure|A, B) > P0, determine that the corresponding sensor is a target sensor with physical damage.

3. The flood warning decision-making method of artificial intelligence according to claim 2, characterized in that: The step S200 includes the following specific steps: The reconfigurable design refers to the dynamic reconfiguration of sensor functions through modular hardware design and cross-physical quantity perception algorithms for redundant sensors and target sensors; The cross-physical quantity perception algorithm includes an algorithm based on physical formulas and an algorithm based on the hidden relationship between physical quantities; Based on the physical relationship built by the cross-physical quantity perception algorithm, the sensors with the same function as the target sensor in the physical relationship are considered as redundant sensors that can implement the reconfigurable design.

4. The flood warning and decision-making method of artificial intelligence according to claim 3, characterized in that: The step S300 includes the following: Step S310: Mark the data monitoring area of the target sensor, extract the monitoring data recorded by the redundant sensor in the reconstruction sensor group to which the target sensor belongs within the data monitoring area as first monitoring data, and calculate the derived target data obtained from the first monitoring data based on the cross-physical quantity sensing algorithm between the redundant sensor and the target sensor; Step S320: Obtain the nearest adjacent sensor existing network communication connection in the visualization model where the target sensor is located as the key sensor, take the key sensor as the core sensor, find other sensors existing network communication connection with the core sensor to form a sensor group to be analyzed; store the data difference of each independent element recording monitoring data at the same monitoring time as the dependent variable, wherein the data difference refers to the monitoring value recorded by the core sensor minus the corresponding value of the other sensor; extract all independent variable factors affecting the change of the monitoring data recorded by the core sensor corresponding to the functional type, and calculate the value difference of each independent variable factor in the same independent element according to the data difference corresponding to the calculation object relationship as the independent variable of the independent element; based on the data group composed of the independent variable and the dependent variable recorded by each independent element of the same core sensor, a linear regression equation corresponding to the core sensor is constructed; Step S330: Set the monitoring data of the target sensor as an unknown x, and extract all independent variables obtained by the target sensor and the core sensor based on the same independent variable factor, and substitute them into the linear regression equation of the core sensor to obtain the value of x, which is taken as the theoretical target data; Calculate the difference value between the theoretical target data and the derived target data, and when there is a difference value less than or equal to a preset difference threshold, mark that the corresponding redundant sensor can perform function switching of the redundant sensor when the target sensor responds to damage abnormality; Step S340: performing steps S310 to S330 on all redundant sensors corresponding to the same target sensor, marking the redundant sensors that can be switched as the to-be-corrected redundant sensors; calculating the difference between the theoretical target data and the derived target data in the corresponding analysis event of the same to-be-corrected redundant sensor under different conditions of obtaining different first monitoring data or different theoretical target data, to form a difference analysis set of the same to-be-corrected redundant sensor; based on the difference analysis set, calculating the fluctuation coefficient F of each to-be-corrected redundant sensor by using the formula: F = {[∑(f1-f0) 2 ] / u} 1 / 2 , wherein f1 represents the numerical value of each element recorded in the difference analysis set, f0 represents the average value of all elements contained in the difference analysis set; u represents the number of elements in the difference analysis set; and [∑(f1-f0) 2 ] represents the sum of the squares of the difference between each element in the set and the average value. Step S350: Extract the maximum value f of the difference between the theoretical target data and the derived target data of all the redundant sensor records to be corrected corresponding to the same target sensor max using the formula: Z = 0.55 f max +0.45 F; Calculate the priority index Z of each redundant sensor to be corrected in the corresponding target sensor; based on the priority index, the redundant sensors to be corrected are sorted from small to large according to the value of the priority index to generate the priority of the switchable redundant sensor.

5. The flood warning decision-making method of artificial intelligence according to claim 4, characterized in that: The step S400 includes the following steps: Obtain the functional type of the sensor to be updated in response to damage abnormality, automatically activate the redundant resource pool, and store all switchable redundant sensors recorded by the sensor to be updated in the redundant resource pool; call the redundant sensor with the first priority as the optimal redundant sensor; and execute the corresponding conversion process; when the optimal redundant sensor is damaged or conversion is abnormal, the redundant sensor corresponding to the priority is postponed until the conversion is successful.

6. An artificial intelligence flood warning decision system, like the artificial intelligence flood warning decision method of any one of claims 1-5, characterized in that: The system includes a target monitoring area marking module, a target sensor marking module, a reconstruction sensor group generation module, a priority analysis module, and a early warning response switching module; The target monitoring area marking module is used to mark the flood monitoring area where the sensor device recording flood events and arranged without redundant sensor setting as the target monitoring area; The target sensor marking module is used to output and mark the sensor existing physical damage as the target sensor; The reconstruction sensor group generation module is configured to combine the target sensor with the corresponding redundant sensor one by one to generate a reconstruction sensor group with the target sensor as a primary sensor; The priority analysis module is configured to output the priority of the switchable redundant sensor; The early warning response switching module is configured to perform replacement prompting when the isolated sensor responds to damage abnormality in the early warning event; The optimal level redundant sensor function switching is performed when the to-be-updated sensor responds to damage abnormality in the early warning event.

7. The flood warning and decision-making system of artificial intelligence according to claim 6, characterized in that: The target sensor marking module comprises a monitoring group division unit, a visualization model construction unit, and a confidence degree analysis unit; The monitoring group division unit is configured to preliminarily divide the sensors in the target monitoring area into groups according to corresponding function types, and mark sensors of the same type as a monitoring group, The visualization model construction unit is configured to generate a visualization model corresponding to the same monitoring group based on space mapping, state coding, and interaction design, The confidence degree analysis unit is configured to analyze the data confidence degree after updating in combination with the visualization model, and set a confidence degree threshold, and determine that the corresponding sensor is a target sensor with physical damage when the confidence degree is greater than the confidence degree threshold.

8. The flood warning and decision-making system of artificial intelligence according to claim 7, characterized in that: The priority analysis module comprises a derived target data calculation unit, a theoretical target data calculation unit, a fluctuation coefficient calculation unit, and a priority index calculation unit; The derived target data calculation unit is configured to calculate derived target data obtained from the first monitoring data; The theoretical target data calculation unit is configured to construct a linear regression equation of the corresponding core sensor, set the monitoring data of the target sensor as an unknown number, extract all independent variables obtained on the basis of the same independent variable factors of the target sensor and the core sensor, substitute the independent variables into the linear regression equation of the core sensor to obtain the unknown number, and take the unknown number as the theoretical target data; The fluctuation coefficient calculation unit is configured to calculate the difference between the theoretical target data and the derived target data in the corresponding analysis event under different conditions of obtaining different first monitoring data or theoretical target data of the same to-be-corrected redundant sensor, to form a difference analysis set of the same to-be-corrected redundant sensor, and calculate the fluctuation coefficient F of each to-be-corrected redundant sensor based on the difference analysis set; The priority index calculation unit is configured to extract the maximum value of the difference between the theoretical target data and the derived target data recorded by all to-be-corrected redundant sensors corresponding to the same target sensor, calculate the priority index Z of each to-be-corrected redundant sensor in the corresponding target sensor, and sort the to-be-corrected redundant sensors according to the values of the priority indexes from small to large to generate the priority of the switchable redundant sensor based on the priority indexes.

Images