AI-based mosquito outbreak prediction and quarantine decision-making method and computer program
An AI-based method using a Random Forest Regressor model predicts mosquito outbreaks and determines control decisions, addressing the lack of automated prediction and verification in existing systems, improving the efficiency and accuracy of mosquito control.
Patent Information
- Application Number
- KR1020260132947
- Authority / Receiving Office
- KR · KR
- Patent Type
- Patents
- Current Assignee / Owner
- Priority Date
- 2026-04-30
- Filing Date
- 2026-07-20
- Publication Date
- 2026-11-06
- Estimated Expiration
- 2046-07-20
AI Technical Summary
Existing mosquito control measures lack efficient methods for predicting future outbreaks and making proactive, automated decisions based on AI, relying heavily on human judgment and lacking verification of control effectiveness.
An AI-based method using a Random Forest Regressor model to predict mosquito outbreaks by combining actual data from traps with past and future weather information, calculating a comprehensive risk index, and automatically determining control decisions, with verification of control effectiveness through time-series analysis.
Enables automated prediction and decision-making for mosquito control, improving outbreak prediction and verification of control effectiveness, enhancing the efficiency and accuracy of pest control measures.
Smart Images

Figure 112026088018489-PAT00016_ABST
Abstract
Description
Technology Field
[0001] The present invention relates to a method and computer program for predicting mosquito outbreaks and making decisions regarding pest control, and more specifically, to an AI-based method and computer program for predicting future mosquito outbreaks and making decisions regarding pest control that automatically predicts future mosquito outbreaks based on AI, automatically determines a decision regarding pest control in advance based on the prediction, and automatically verifies the effectiveness of pest control after implementing pest control according to the determined decision. Background Technology
[0003] There are 9 genera and 56 species of mosquitoes recorded occurring in Korea, and there are currently 14 species known to transmit infectious diseases in Korea or to transmit infectious diseases introduced from overseas.
[0004] Looking at the representative mosquitoes inhabiting Korea, the genus Culex includes *Culex tritaeniorhynchus*, which transmits Japanese encephalitis; *Culex pipiens complx*, which is not related to disease transmission but has a high population density and bothers people by invading residential areas; and *Culex pipiens molestus*, which has recently emerged after adapting to urban environments. The genus Aedes includes *Aedes vexans nipponii*, which is active both day and night and commonly attacks people in residential areas, gardens, and parks; and *Aedes albopictus*, a diurnal mosquito that is primarily active during the day and feeds on blood, frequently found in hills, urban green spaces such as parks, grassy areas, and ancient palaces. Additionally, the genus Anopheles includes *Anopheles*, which transmits malaria, a disease for which cases have recently been reported in Korea. There are sinensis, etc.
[0005] Therefore, activities to monitor, predict, and control the occurrence of mosquitoes that transmit infectious diseases are necessary; however, in the past, mosquito control measures were mainly implemented at the individual or household level, or local governments carried out control measures uniformly by considering only the mosquito season.
[0006] Recently, local governments and the Korea Disease Control and Prevention Agency have been taking more proactive measures regarding the occurrence of mosquitoes that transmit infectious diseases. For example, the Seoul Metropolitan Government has implemented a mosquito forecasting system that forecasts the risk of mosquito outbreaks in four stages by installing digital mosquito measuring devices (DMS) in waterfront areas, residential areas, and parks to predict the average daily mosquito population and calculating a mosquito index that comprehensively considers environmental factors favorable for mosquito breeding and the mosquito growth period. However, this mosquito forecasting system in Seoul is provided to inform citizens of behavioral guidelines based on whether they are active in waterfront areas, residential areas, or parks.
[0007] In addition, the Korea Disease Control and Prevention Agency helps local governments make decisions regarding mosquito control by collecting and providing actual mosquito capture data in cooperation with local government public health centers; however, specific predictions regarding future mosquito outbreaks and the establishment of corresponding control plans (what, when, and how to control) relied on the experience and arbitrary judgment of officials in charge at local governments. Prior art literature
[0009] Korean Registered Patent Publication No. 10-2059859 Korean Registered Patent Publication No. 10-1900570 The problem to be solved
[0010] The present invention takes into account the problems of the prior art as described above and aims to provide an AI-based method and computer program for predicting future mosquito outbreaks and making pest control decisions, which can efficiently improve future mosquito outbreak prediction and corresponding pest control by automatically predicting future mosquito outbreaks based on AI, automatically determining a pest control decision in advance based on the prediction, and automatically verifying the effectiveness of pest control after implementing pest control according to the determined pest control decision. means of solving the problem
[0012] The AI-based mosquito occurrence prediction and control decision-making method according to the present invention comprises: a first step of collecting actual data related to the number of mosquitoes captured from mosquito traps installed at one or more observation stations; a second step of combining and analyzing external information including at least past weather information and future weather forecast information with stored data in a database including at least the history of mosquito occurrence for each observation station to produce analysis information; a third step of using an AI learning model to predict the future amount of mosquito occurrence using the actual data collected in the first step and the analysis information produced in the second step as input variables; a fourth step of calculating a comprehensive risk index based on the prediction result produced in the third step and the analysis information produced in the second step; a fifth step of automatically determining a control decision, including whether to implement control and the control method, based on the comprehensive risk index produced in the fourth step; and a sixth step of notifying a terminal of the control decision determined in the fifth step.
[0013] In the third step of the AI-based mosquito outbreak prediction and control decision-making method according to the present invention, the AI learning model is characterized as being a Random Forest Regressor-based machine learning model.
[0014] In the third step of the AI-based mosquito occurrence prediction and control decision-making method according to the present invention, when predicting the future mosquito occurrence amount after implementing control measures, the future mosquito occurrence amount is calculated by applying a control effect attenuation coefficient for each control method.
[0015] In the third step of the AI-based mosquito outbreak prediction and control decision-making method according to the present invention, when two or more control methods are implemented in parallel, the individual effect attenuation coefficients of each control method are calculated independently, and then the total control effect attenuation coefficients are calculated by summing them.
[0016] In the fourth step of the AI-based mosquito outbreak prediction and disease control decision-making method according to the present invention, the comprehensive risk index is calculated according to a multiple weighted sum formula.
[0017] The fifth step of the AI-based mosquito outbreak prediction and control decision-making method according to the present invention is characterized by classifying the risk level into two or more predefined levels according to the value of the comprehensive risk index calculated in the fourth step, and automatically determining a control decision including a pre-mapped control method and a control agent for each risk level.
[0018] The AI-based mosquito occurrence prediction and control decision-making method according to the present invention is characterized by further including a seventh step of verifying the control effect by analyzing the time-series difference by comparing the actual number of mosquitoes collected in the first step with the amount of mosquitoes predicted in the third step during a verification period after the implementation of control.
[0019] In the seventh step of the AI-based mosquito occurrence prediction and control decision-making method according to the present invention, the control effect is verified by calculating a difference ratio by comparing the time series of the predicted mosquito occurrence amount (predicted value) with the time series of the collected actual number of mosquitoes (actual value), wherein the difference ratio is defined as (actual value - predicted value) / predicted value.
[0020] In the seventh step of the AI-based mosquito outbreak prediction and control decision-making method according to the present invention, the time-series pattern of the difference ratio is automatically classified into one of a normal pattern, a rebound pattern, and a stagnation pattern. The normal pattern is a pattern in which the difference ratio maintains a trend within a set error range during the period after the implementation of control, which is judged as indicating that the control is working effectively. The rebound pattern is a pattern in which the difference ratio is maintained within a set error range until a certain point in time after the implementation of control, but after that point, the actual value rebounds and the difference ratio gradually increases, causing the difference ratio to exceed the set error range, which is judged as a situation in which the primary control effect resulting from the implementation of control has been manifested, but additional outbreaks from untreated sources within the control area are suspected thereafter. The stagnation pattern is a pattern in which the actual value decreases but the difference ratio gradually increases until a certain point in time after the implementation of control, but after that point, the actual value stagnates within a certain range while the difference ratio gradually expands and exceeds the set error range, which is judged as a situation in which the existence of an untreated large source outside the control area is suspected.
[0021] A computer program according to the present invention is a computer program stored on a computer-readable recording medium to enable a computer to perform the above-described method. Effects of the invention
[0023] According to the present invention, future mosquito outbreaks can be automatically predicted based on AI, and a decision on pest control can be automatically made in advance based on the prediction.
[0024] In addition, according to the present invention, the disinfection effect can be verified after disinfection is carried out according to an automatically determined disinfection decision, thereby efficiently improving disinfection. Brief explanation of the drawing
[0026] FIG. 1 is a conceptual diagram for explaining an AI-based mosquito outbreak prediction and disease control decision-making automation system (hereinafter referred to as the "system") according to the present invention. FIG. 2 is a schematic diagram illustrating the configuration of a system according to the present invention. FIG. 3 is a schematic diagram illustrating the configuration of a mosquito trap according to the present invention. FIG. 4 is a diagram illustrating the prediction of future mosquito populations through a Random Forest Regressor-based machine learning model in the system according to the present invention. Figure 5 is a graph showing the attenuation pattern of effectiveness by disease control method. Figure 6 is a graph showing a pattern for determining the disinfection effect according to the amount of mosquitoes captured after disinfection treatment in the system according to the present invention. FIG. 7 is a diagram illustrating disease control decision-making based on a comprehensive risk index in a system according to the present invention. FIG. 8 is a schematic diagram illustrating the configuration of an integrated control system according to the present invention. FIG. 9 is a flowchart illustrating an AI-based method for predicting mosquito outbreaks and automating pest control decision-making by the system of the present invention. FIG. 10 is a diagram illustrating the flow of feeding back results produced by the system of the present invention for model retraining. Figure 11 is a graph showing the future mosquito outbreak prediction for each observation station calculated by the system of the present invention. FIG. 12 is a diagram showing the risk level for each observation station calculated by the system of the present invention. FIG. 13 is a diagram showing the quarantine intention for each observation station calculated by the system of the present invention. Figure 14 is a graph showing the prediction of mosquito occurrence after disinfection at each observation station, calculated by the system of the present invention. Specific details for implementing the invention
[0027] Hereinafter, an AI-based mosquito outbreak prediction and disease control decision-making automation system according to the present invention (hereinafter referred to as "system (10)") will be described in detail with reference to the drawings.
[0028] FIG. 1 is a conceptual diagram for explaining a system according to the present invention, FIG. 2 is a schematic diagram for explaining the configuration of a system according to the present invention, FIG. 3 is a schematic diagram for explaining the configuration of a mosquito trap according to the present invention, FIG. 4 is a diagram for explaining the prediction of future mosquito occurrence through a Random Forest Regressor-based machine learning model in a system according to the present invention, FIG. 5 is a graph showing the effect attenuation pattern by pest control method, FIG. 6 is a graph showing the pattern for judging the pest control effect according to the amount of mosquitoes captured after pest control treatment in a system according to the present invention, FIG. 7 is a diagram for explaining pest control decision-making based on a comprehensive risk index in a system according to the present invention, FIG. 8 is a schematic diagram for explaining the configuration of an integrated control system according to the present invention, FIG. 9 is a flowchart for explaining an AI-based method for predicting mosquito occurrence and automating pest control decision-making by a system according to the present invention, FIG. 10 is a diagram for explaining the flow of feedback of results calculated by a system according to the present invention for model retraining, FIG. 11 is a graph showing the prediction of future mosquito occurrence for each observation station calculated by a system according to the present invention, and FIG. Figure 12 is a diagram showing the risk level for each observation station calculated by the system of the present invention, Figure 13 is a diagram showing the intention to disinfect each observation station calculated by the system of the present invention, and Figure 14 is a graph showing the prediction of mosquito occurrence after disinfection for each observation station calculated by the system of the present invention.
[0029] First, the system (10) according to the present invention, as shown in FIGS. 1 and 2, has each observation station (P1, P2, P3, P4···P nA data collection unit (100) that collects data and / or information, such as the number of mosquitoes captured in a mosquito trap (20) installed in a ), an information analysis unit (200) that combines and analyzes external information provided from an external information provision system (40), information provided from a database (600), and data provided by the data collection unit (100), an occurrence prediction unit (300) that predicts mosquito occurrence based on the data collected by the data collection unit (100) and the analysis information analyzed by the information analysis unit (200), a disease control decision unit (400) that decides on disease control based on the number of mosquitoes predicted by the occurrence prediction unit (300) and the information analyzed by the information analysis unit (200), and a disease control effect that verifies the disease control effect by comparing the number of mosquitoes predicted by the occurrence prediction unit (300) after disease control with the actual number of mosquitoes collected by the data collection unit (100) after disease control. It includes a verification unit (500) and a database (600) that stores information and data produced by the data collection unit (100), information analysis unit (200), occurrence amount prediction unit (300), quarantine decision unit (400), and quarantine effect verification unit (500), as well as data such as quarantine implementation history input through a terminal (30).
[0030] First, each observation station (P1, P2, P3, P4···P) where the above-mentioned mosquito trap (20) is provided n It is desirable that ) be provided in appropriate locations and in appropriate numbers, taking into account the density of people and the distance from the source of the mosquitoes to be captured.
[0031] Mosquitoes tend not to travel very far from their breeding grounds, but the distance traveled varies depending on the species and environment. For example, most house mosquitoes are mainly active within approximately 100 to 300 meters of the breeding ground, but depending on the species, they can fly over 1 km. In the case of Aedes albopictus, the range of movement is short, so it is mainly active within 50 to 100 meters of the breeding ground, but it can travel hundreds of meters or more. Additionally, Anopheles mosquitoes can travel up to several kilometers, so control measures are necessary within 1 km of the breeding ground.
[0032] That is, it is desirable to install the above-mentioned mosquito traps (20) in appropriate locations and in appropriate numbers, taking into consideration whether mosquitoes can fly to residential areas, parks, walking paths, etc., where people gather due to mosquito outbreaks, or whether mosquitoes cannot fly to residential areas, parks, walking paths, etc., where people gather even if mosquitoes occur.
[0033] Additionally, as shown in FIG. 3, the mosquito trap (20) includes a CO2 generation module (21) for generating CO2 as an attractant to lure surrounding mosquitoes into the mosquito trap (20), a counting module (22) for distinguishing whether the individuals captured by the mosquito trap (20) are mosquitoes or non-mosquitoes and counting the number of mosquitoes, an environmental sensor module (24) for collecting environmental information such as temperature and humidity of the area where the mosquito trap (20) is installed, a self-diagnosis module (25) for diagnosing whether there is an operational abnormality of the mosquito trap (20), a communication module (26) for communicating with the system (10), and a control module (27) for controlling the operation of each module of the mosquito trap (20), and may further include a species identification module (23) for identifying the species of mosquitoes captured by the mosquito trap (20).
[0034] Typically, the mosquito trap (20) comprises an inlet passage (not shown) through which mosquitoes enter, a receiving section (not shown) for capturing and receiving mosquitoes that have entered through the inlet passage, and a fan (not shown) for forming an airflow from the inlet passage to the receiving section. The CO2 generating module (21) generates CO2 and sprays CO2 around the entrance of the inlet passage to attract mosquitoes to the mosquito trap (20). At this time, the control module (27) can control the amount of CO2 generated by the CO2 generating module (21) or control the time of CO2 generation (e.g., from 6 PM to 6 AM the next day, when mosquito activity is high) according to the risk level provided by the system (10).
[0035] Additionally, the CO2 generation module (21) is intended to attract mosquitoes to the mosquito trap (20), and can be replaced or used in combination with a light source attraction module (not shown) or an attractant generation module (not shown).
[0036] Additionally, the counting module (22) is intended to count the number of mosquitoes by distinguishing whether the object entering and being captured by the mosquito trap (20) is a mosquito or a non-mosquito. The counting method may include a photoelectric counting method that calculates the number of objects by measuring the change in light intensity of an object passing through a light sensor inside the mosquito trap (20), or a non-photoelectric counting method that detects mosquitoes and calculates the number of objects by utilizing camera images and image processing technology.
[0037] However, as long as it is possible to distinguish mosquitoes and count their population, there are no specific restrictions on the method; nevertheless, a photoelectric counting method may be preferred in terms of cost and energy consumption, and even in the case of the photoelectric counting method, competitiveness compared to non-photoelectric counting methods can be secured by improving accuracy through automatic relearning based on accumulated data from the operating environment.
[0038] Additionally, the species identification module (23) is for identifying which species (e.g., Culex pipiens, Aedes albopictus, Aedes aureus, Anopheles pipiens, Aedes aureus, etc.) of the mosquito entering the mosquito trap (20), and if the counting module (22) adopts a photoelectric counting method, the species identification module (23) may be further included.
[0039] At this time, methods for identifying species include image recognition methods that identify individual mosquitoes by analyzing camera images, acoustic recognition methods that detect the number and type of mosquitoes by recognizing the flight sound waves of mosquitoes, or impedance measurement methods that automatically identify the type, species, and number of mosquitoes in real time by utilizing the wingbeat frequency and internal electrical characteristics (impedance) of mosquitoes.
[0040] In this way, if the species identification module (23) is further included, the species identification module (23) identifies the mosquitoes that are introduced into and captured by the mosquito trap (20) by species (e.g., Culex pipiens, Aedes albopictus, Aedes aureus, Anopheles spp., Aedes spp., etc.), and the counting module (22) can count the number of individuals for each species of mosquito identified by the species identification module (23).
[0041] Additionally, the environment sensor module (24) is intended to collect information related to mosquito habitat in the area where the mosquito trap (20) is installed (e.g., temperature, humidity, wind speed, precipitation, etc.), and may include a temperature sensor, a humidity sensor, a wind speed sensor, a precipitation sensor, etc.
[0042] Additionally, the self-diagnosis module (25) can diagnose whether there is an operational abnormality of the mosquito trap (20), for example, whether the counting module (22) and the species identification module (23) are operating normally, whether each sensor of the environment sensor module (24) is detecting normally, whether the CO2 generation module (21) is generating CO2 normally in accordance with the control of the control module (27), or whether CO2 is generated normally at a set time and the fan is operating normally.
[0043] The communication module (26) is composed of wired and / or wireless communication modules that transmit information and data collected or produced in each module to the data collection unit (100) of the system (10) in real-time or near-real-time, and any one or more of LTE, 5G, LoRa, NB-IoT, or Wi-Fi may be used as wireless communication protocols, and an appropriate protocol may be selected according to the communication environment and data transmission frequency of the mosquito trap (20).
[0044] Here, 'real-time' transmission means transmitting information and / or data to the data collection unit (100) of the system (10) as soon as it is collected or produced, and 'semi-real-time' transmission means transmitting the collected or produced information and / or data to the data collection unit (100) of the system (10) at predetermined time intervals (e.g., 1 hour, 3 hours, 6 hours, 12 hours, 24 hours, etc.).
[0045] The control module (27) includes a microcontroller or a single-board computer, processes control commands received from the system (10), and controls the operation of each sub-module.
[0046] The above data collection unit (100) includes each observation station (P1, P2, P3, P4···P nIt is communicationally connected to the mosquito trap (20) installed in the ) and receives and collects information and data produced from each module of each mosquito trap (20).
[0047] At this time, the information and data produced by each module are for each observation station (P1, P2, P3, P4···P n Data related to the type and / or number of mosquitoes captured by each of the mosquito traps (20) installed in the above locations, and each observation station (P1, P2, P3, P4···P) where the mosquito traps (20) are installed. n It may include environmental information of ), information regarding whether there is an operational abnormality of the mosquito trap (20) diagnosed by the self-diagnosis module (25), etc.
[0048] The data collected from each of the observation stations (P1, P2, P3, P4···P) at the above data collection unit (100) n Data and environmental information regarding the number of mosquitoes captured by each ) are provided to the generation prediction unit (300) to be used for predicting future mosquito generation, and are also provided to the information analysis unit (200) and database (600) to be used for calculating analysis information and retraining the model of the system (10), and at the same time, notifications can be provided to the administrator through the terminal (30).
[0049] In addition, information regarding whether there is an operational abnormality of the mosquito trap (20) collected by the data collection unit (100) is provided to the administrator as an immediate notification through the terminal (30), and at the same time, is provided to the generation amount prediction unit (300) to be used for correcting the predicted amount of mosquitoes in the future, and is provided to the information analysis unit (200) and database (600) to calculate analysis information and manage the failure history of each mosquito trap (20).
[0050] At this time, data provided from the data collection unit (100) to the generation amount prediction unit (300), information analysis unit (200), database (600), etc. is converted into a standard data format predefined in the data collection unit (100) and transmitted.
[0051] Here, the 'standard data format' can be defined as a structure including fields such as an observation station identifier, a measurement time (year-month-day-hour-minute), a mosquito population count, a measurement method code, environmental variables (temperature, humidity, etc.), and a data quality indicator, and thereby the system (10) of the present invention can integrally process data provided from the mosquito trap (20) of various manufacturers.
[0052] In addition, the data collection unit (100) includes each observation station (P1, P2, P3, P4···P n Anomalies can be automatically identified by applying a predefined statistical threshold-based anomaly detection method or a machine learning-based anomaly detection algorithm to the data collected from the mosquito trap (20) installed in the above location. For example, machine learning algorithms such as whether a threshold is exceeded based on the average value and standard deviation over a preset period of time, correlation analysis with adjacent observation station data, or Isolation Forest anomaly detection and single-class SVM anomaly detection may be applied.
[0053] At this time, for data determined to be an outlier, a notification is sent to the terminal (30), and at the same time, data transmission to the subsequent information analysis unit (200) and occurrence amount prediction unit (300) is blocked and replaced with a predefined correction value (e.g., the average value of adjacent observation stations at the same time, the average value of the same observation station at the same time for N days immediately preceding, etc.).
[0054] In addition, the data collection unit (100) includes each observation station (P1, P2, P3, P4···P nIf the reception of data from some of the mosquito traps (20) installed in the area is interrupted for a predetermined period of time or longer, it is desirable to notify the terminal (30) of the possibility of communication failure or operation stoppage of the said mosquito trap (20), and at the same time, the generation amount prediction unit (300) and the disease control decision unit (400) are replaced with data from adjacent observation stations or data from the same past period according to a predetermined missing data processing rule, so that the operation of the entire system is not interrupted due to the failure of some of the mosquito traps (20).
[0055] Additionally, the data collection unit (100) may additionally collect user report data. User report data may include field reports from public health centers, citizen reports, complaints regarding external mosquitoes, reports via a proprietary app, etc., and may be received through various input channels such as messenger channels (e.g., KakaoTalk, Naver Line, Instagram, etc.), text messages, proprietary apps, web forms, etc.
[0056] At this time, the user report data collected by the data collection unit (100) is structured into fields such as location information (GPS coordinates or administrative district), time of report, content of the report, and informant identifier (can be anonymized) and stored in the database (600), and can be used as an auxiliary input variable for the occurrence amount prediction unit (300) and the quarantine decision unit (400).
[0057] The above terminal (30) is a computing device capable of connecting to the above system (10), and includes a desktop, laptop, laptop, smartphone, tablet PC, wearable device, etc., and is equipped with a communication module such as Bluetooth, NFC, Wi-Fi, etc., so as to be able to communicate with the above system (10) in various network environments.
[0058] Next, the information analysis unit (200) combines external information provided from the external information provision system (40), data and information stored in the database (600), and data and information collected from the data collection unit (100) to produce analysis information.
[0059] At this time, the information analysis unit (200) combines external information provided by the external information provision system (40) according to a predefined data combination rule. The data combination rule may include time synchronization (e.g., matching the time unit of mosquito-related data with the time unit of weather-related data), spatial synchronization (e.g., mapping the coordinates of an observation station where a mosquito trap (20) is installed with the nearest weather station to that observation station), unit conversion (e.g., converting the unit of precipitation from 'mm / hr' to 'mm / day'), data interpolation (e.g., filling in data for missing time periods), data normalization (e.g., converting data into a '0 to 1 range' or 'standard normal distribution'), etc.
[0060] The above external information provision system (40) is a general term for an external system that the system (10) of the present invention can access to receive necessary information and data, and includes not only information provision systems of public institutions but also all external systems that the system (10) can access to receive necessary information and data, such as digital platforms like Google, Naver, and Kakao.
[0061] The external information provided by the above external information provision system (40) includes, for example, weather information provided by the Korea Meteorological Administration, water system information provided by the National Water Resources Management Information System (including future water system prediction information such as river water level, reservoir water level, heavy rain warning, and heavy rain preliminary special warning), geographic information provided by map services such as Google, Naver, and Kakao (administrative districts, road networks, building information, etc.), and mosquito outbreak history information provided by the Disease Control and Prevention Agency's Geographic Information System (GIS), and all publicly available data that can be accessed and utilized by the system (10) of the present invention.
[0062] In addition, the above external information includes both past information prior to the reference point (present) and future forecast information for a certain period after the reference point.
[0063] In other words, the weather information provided by the Korea Meteorological Administration includes both past weather information prior to the reference point (present) (temperature, humidity, precipitation, wind speed, sunshine hours, etc.) and future weather forecast information for a certain period after the reference point (predicted temperature, predicted humidity, predicted precipitation, predicted probability of precipitation, predicted wind speed, predicted sunshine hours, etc.).
[0064] At this time, the information analysis unit (200) receives and combines and analyzes not only past weather information but also future weather forecast information as external information. It combines and processes the past weather information and future weather forecast information into a single continuous time series along a time axis, and assigns actual weather data to the data of past points in time and weather forecast data to the data of future points in time after the reference point, respectively, with the reference point as the boundary. At this time, reflecting the fact that the reliability of future weather forecast data decreases with the elapsed time from the forecast point, a reliability weight or uncertainty indicator may be assigned to the forecast data.
[0065] Additionally, the information analysis unit (200) can calculate a spatiotemporal influence matrix that reflects the distance between the source and the observation station, the possible flight distance of mosquitoes depending on the wind direction, and topographic information (e.g., altitude difference, vegetation, etc.) by combining location information of the mosquito source estimated from the external information or pre-stored in the database (600) (e.g., stagnant water system, rainwater reservoir, waste tire storage area, flooded area, etc.) with actual measurement data from the data collection unit (100).
[0066] By providing the calculated spatiotemporal influence matrix as an input variable or correction variable to the generation amount prediction unit (300), it becomes possible to make a prediction that reflects the spatial relationship with adjacent observation stations and sources, rather than just the data of a single observation station.
[0067] The above-mentioned occurrence prediction unit (300) predicts future mosquito occurrences through an AI learning model based on actual data collected by the above-mentioned data collection unit (100) and analysis information calculated by the above-mentioned information analysis unit (200).
[0068] In the embodiments of the present invention, a Random Forest Regressor-based machine learning model was used as the AI learning model.
[0069] As shown in FIG. 4, the above Random Forest Regressor is an ensemble regression model that individually trains multiple decision trees and then takes the average value of the results predicted by each decision tree to predict a final continuous number. Each decision tree is trained on a subset randomly extracted from the training data, and determines the optimal branching condition by considering only some input variables randomly selected at each branching point.
[0070] The technical reason for adopting the Random Forest Regressor in this invention is as follows.
[0071] First, it can automatically learn the non-linear combination relationships between time-series data (e.g., fluctuations in mosquito populations) and multivariate environmental data (e.g., weather, water systems, geography, etc.). In other words, it can automatically learn the non-linear combination relationships between complex time-series data, such as mosquito outbreak predictions, and multivariate environmental data without the need for separate, complex mathematical formula definitions.
[0072] Second, domain experts can interpret the basis of the model's predictions by calculating the feature importance of the input variables. In other words, the basis of the model's predictions can be interpreted because it is calculated by averaging how much each variable reduces Gini impurity when splitting the decision tree nodes. For example, when predicting mosquito populations, the relative influence of each variable can be grasped at a glance, such as when it is shown that the 'average temperature two weeks ago' and the 'regional variable within 100m of the water system' contributed the most to node splitting.
[0073] Third, it is robust against overfitting, enabling stable prediction even in environments with limited training data. That is, stable prediction is possible even in situations where the sample size (number of data points) itself is small due to climate change or budget limitations, such as with mosquito population data, or where data is insufficient compared to the types of variables (weather, geography, etc.).
[0074] Fourth, gradual relearning using newly accumulated operational data is facilitated. That is, as illustrated in FIG. 10, gradual relearning is facilitated by updating the learning data by reflecting the data produced during the operation process of the system (10).
[0075] The input variables (features) of the Random Forest Regressor model according to an embodiment of the present invention include the following, as illustrated in FIG. 4:
[0076] (a) Time series variables for daily or hourly measured mosquito populations over the preceding M days (variables classified by mosquito species if identification information by species is available), (b) Past weather variables for the same period—variables such as average, maximum, and minimum temperature, average humidity, total precipitation, average wind speed, and sunshine duration, (c) Future weather forecast variables for the next N days after the reference point—variables such as predicted average, maximum, and minimum temperature, predicted humidity, predicted precipitation, predicted probability of precipitation, predicted wind speed, and predicted sunshine duration, (d) Nearby water system information variables (river water level, reservoir status, flooded area, etc.) and water system prediction variables (preliminary heavy rain advisories, water level fluctuation predictions based on precipitation forecasts, etc.), (e) Disinfection history variables for the preceding K days (disinfection method, pesticide, spraying amount, spraying date), (f) Geographic information variables (observation station elevation, vegetation index, urbanization index, population density, etc.), (g) User report variables (number of reports in the preceding period, classification of report patterns, etc.) etc.
[0077] In the present invention, unlike conventional mosquito occurrence prediction systems, not only past weather variables but also future weather forecast variables after a reference point are used simultaneously as input variables.
[0078] This is intended to reflect the fact that the developmental stages of mosquitoes, from egg-laying to adult emergence, take approximately 7 to 14 days or more depending on the species, and that current weather conditions not only affect the population volume in the near future but also determine the fluctuations in the population volume during the same period. For example, if precipitation is forecast within a certain period in the future, changes in the water environment of mosquito breeding grounds and changes in blood-feeding activity are expected, and these changes cannot be adequately predicted using only past weather data.
[0079] In other words, by integrally utilizing past weather variables and future weather forecast variables as input variables for the same model, the present invention enables the prediction of occurrence amounts that reflect the past, present, and near future time axes.
[0080] In addition, the application period of the future weather forecast variable is linked to the output period (N) of the above-mentioned generation amount prediction unit (300), and, for example, can be determined according to the operating environment within a range of 1 to 14 days.
[0081] At this time, future weather forecast variables may be received from the above external information, such as the official API of the Korea Meteorological Administration, short-term forecasts (within about 3 days), medium-term forecasts (within about 10 days), or forecast data from multiple public or accredited institutions. When using two or more forecast sources, the information analysis unit (200) may calculate an integrated forecast value through a weighted average based on the past accuracy of each forecast source.
[0082] In addition, the information analysis unit (200) includes each observation station (P1, P2, P3, P4···P n Weather data received from adjacent stations within a predefined radius based on the coordinates of the target station can be combined into the input variables of the target station.
[0083] In this way, when combining weather data from adjacent stations with the input variables of the target station, spatial weights reflecting the influence based on distance to adjacent stations, bearing, and wind direction can be applied, thereby enabling the integration of wide-area weather change patterns into the model input that are difficult to capture with data from a single station alone.
[0084] This is intended to reflect the fact that mosquitoes can typically fly a certain distance to enter adjacent areas by combining weather data from adjacent stations with the input variables of the target station, and that weather changes in adjacent areas affect the future occurrence volume of the target station with a time delay.
[0085] In addition, if the mosquito trap (20) further includes the species identification module (23) to utilize species identification information, as shown in FIG. 4, the Random Forest Regressor-based machine learning model of the occurrence prediction unit (300) is configured to have a separate set of model parameters or a separate model instance for each mosquito species (e.g., Culex pipiens, Aedes albopictus, Anopheles mosquito, etc.).
[0086] This is intended to automatically incorporate into the model different domain knowledge, such as the fact that preferred sources of occurrence differ depending on the mosquito species (e.g., Culex pipiens mosquitoes in stagnant water systems, Aedes albopictus mosquitoes in artificial waters, Anopheles mosquitoes in rice paddies and wetlands, etc.), activity times differ (e.g., Culex pipiens mosquitoes at night, Aedes albopictus mosquitoes during the day, Anopheles mosquitoes at dawn or dusk, etc.), and types of vector-borne diseases differ (e.g., Culex pipiens mosquitoes are vectors of Japanese encephalitis, Aedes albopictus mosquitoes are vectors of dengue and Zika, Anopheles mosquitoes are vectors of malaria, etc.).
[0087] The training of the Random Forest Regressor model according to an embodiment of the present invention is performed as follows.
[0088] The training dataset consists of time-series data spanning the past M years for each observation station and corresponding environmental variable data. It is split into a training set, a validation set, and a test set through a temporal split that maintains chronological order. Model parameters are trained using the training set, hyperparameters (such as the number of trees, maximum depth, minimum split sample size, and number of randomly selected features) are tuned using the validation set, and the performance of the final model is evaluated using the test set.
[0089] The hyperparameters of the Random Forest Regressor model can be determined by methods such as grid search or Bayesian optimization, and in the present invention, the range of hyperparameters can be set in the range of the number of trees from 100 to 500, the range of the maximum depth from 10 to 30, and the range of the minimum number of partitioned samples from 2 to 20.
[0090] In other words, if the number of trees is reduced too low, the ensemble effect decreases, increasing prediction variability; conversely, while prediction accuracy improves as the number of trees increases, the model becomes heavier, slowing down training and prediction speeds. Additionally, if the maximum depth (thickness, which determines the maximum number of steps each tree can extend downward) value is lowered too low, underfitting occurs, resulting in poor performance due to insufficient learning from the data; conversely, if the value is raised too high, overfitting occurs, where the model fits the training data perfectly but fails to fit the actual data. Furthermore, the minimum split sample count is a criterion that determines the minimum number of data points that must remain at the current node for a tree node (branch point) to split further into lower branches, and it is set as an integer between 2 and 20.
[0091] In addition, the Random Forest Regressor model according to an embodiment of the present invention has a self-learning structure, as shown in FIG. 10.
[0092] That is, the error between the actual data and the prediction result stored in the database (600) is received as feedback, and the model is automatically retrained at a preset interval (e.g., daily, weekly, or monthly) or when the cumulative data threshold is reached. In addition, the verification result of the quarantine effect verification unit (500) described later (especially when a suspected omission of the source is detected) can also be used as a trigger signal for model retraining.
[0093] In addition, the Random Forest Regressor model according to an embodiment of the present invention is configured to have separate sets of model parameters or separate model instances for each observation station, region, or season, thereby enabling precise prediction that reflects regional specificities (e.g., differences between urban and rural areas, differences between coastal and inland areas) and seasonal specificities (e.g., differences between the wet and dry seasons, differences between spring and autumn).
[0094] In addition, the generation amount prediction unit (300) of the present invention may produce a prediction result by combining one or more auxiliary models, such as a decision tree, gradient boosting, a neural network model (e.g., time series neural network such as LSTM, GRU, etc.), a Bayesian network, and a time series prediction model (e.g., ARIMA, Prophet, etc.), in addition to a Random Forest Regressor model, into an ensemble using a weighted averaging method or a stacking method. In this case, in the weighted averaging method, the prediction values are combined with weights proportional to the test set performance of each model, and in the stacking method, a meta-model is trained using the prediction values of each model as input to produce a final prediction.
[0095] In the Random Forest Regressor model of the generation amount prediction unit (300) of the present invention, the output variable is the daily or hourly predicted mosquito generation amount for the next N days, as shown in FIG. 4. Here, the past time series length (M) and the future prediction period (N) can be determined according to the operating environment, for example, the past time series length (M) can be set in the range of 7 to 30 days, and the future prediction period (N) can be set in the range of 1 to 14 days.
[0096] In addition, the mosquito generation amount prediction result of the generation amount prediction unit (300) is, as shown in FIG. 11, each observation station (P1, P2, P3, P4···P nIt can be calculated by ) and displayed on the terminal (30).
[0097] In addition, the Random Forest Regressor model of the above-mentioned occurrence amount prediction unit (300) can provide the user with the basis for the prediction by calculating and outputting the feature importance or SHAP (SHapley Additive exPlanations) value of each input variable in addition to the prediction result. That is, if the predicted occurrence amount for a specific date is high, the user can be provided with a visualized representation of whether the main cause is due to an increase in precipitation in the previous period, a rise in temperature, or environmental changes near a specific source.
[0098] In addition, the prediction result of the above-mentioned occurrence amount prediction unit (300) is not limited to a single prediction value. That is, it may include pre-set confidence interval or quantile information (e.g., 10th percentile, 50th percentile, 90th percentile), and through this, the disease control decision unit (400) can make a decision considering the uncertainty of the prediction. For example, if the 90th percentile of the prediction value exceeds a risk threshold, a conservative disease control decision can be automatically recommended even if the average prediction value is below the threshold.
[0099] In addition, when pest control is carried out, in predicting the future amount of mosquitoes after pest control is carried out, the above-mentioned amount prediction unit (300) needs to predict the amount of mosquitoes by considering the pest control effect.
[0100] In other words, when predicting future mosquito populations, the effectiveness of pest control measures must be considered; it is necessary to predict mosquito populations by reflecting the phenomenon where the effect of previous measures gradually diminishes over time.
[0101] Accordingly, in the present invention, the generation amount prediction unit (300) applies a damping coefficient of the disinfection effect for each disinfection method when predicting the future generation amount of mosquitoes after disinfection is carried out.
[0102] That is, when disinfection is carried out using one disinfection method, and the date of implementation is D and the number of days elapsed since the date of implementation of disinfection is N, the above-mentioned generation amount prediction unit (300) can calculate the predicted generation amount of mosquitoes on the Nth day after implementation of disinfection according to the following equation (1).
[0103] Predicted occurrence amount (D+N) = Weather-based expected occurrence amount (D+N) × (1 - Attenuation coefficient of disease control effect (D+N))… Equation (1)
[0104] Here, the weather-based expected occurrence amount (D+N) is a standard occurrence amount based on weather variables without considering the disinfection effect when N days have passed since the disinfection implementation date (D), and is a predicted value of mosquito occurrence calculated by the Random Forest Regressor model of the occurrence amount prediction unit (300) with the disinfection history variable set to 0 (no disinfection assumption).
[0105] At this time, the above weather-based expected occurrence amount (D+N) is calculated by inputting not only past weather variables up to the reference point (D) but also future weather forecast variables for days N after the reference point (D).
[0106] In addition, the above-mentioned attenuation coefficient (D+N) is a function value that changes over time (N) for each disinfection method and is defined within the range of 0 (no disinfection effect) to 1 (complete disinfection effect). That is, the above-mentioned attenuation coefficient (D+N) is a function value that changes as a ratio of the remaining disinfection effect of each disinfection method over time (N) based on the complete disinfection effect. When the remaining disinfection effect of each disinfection method is the complete disinfection effect, the above-mentioned attenuation coefficient (D+N) becomes 1, and when the remaining disinfection effect of each disinfection method is 0, the above-mentioned attenuation coefficient (D+N) becomes 0.
[0107] That is, the disinfection effect attenuates differently depending on the disinfection method. As shown in FIG. 5, for each disinfection method, such as biological larva control agent (Bacillus thuringiensis israelensis, hereinafter referred to as "Bti"), fogging or smoke, insect growth regulator (hereinafter referred to as "IGR"), and residual spraying, the time at which the effect appears, the time at which the effect reaches its maximum, and the time at which the effect ends are defined as different time functions modeled over time after the disinfection is carried out.
[0108] In addition, Figure 5 is a graph plotted over time showing the relative values of the expected mosquito population according to the effectiveness of each control method, with the expected mosquito population under no-control conditions set to 1, under conditions of a temperature of 25–28°C, a wind speed of 0.5 m / s or less, humidity of 60% or more, and no precipitation. In other words, Figure 5 represents the characteristics of the control effectiveness of each method under ideal conditions, and each observation station (P1, P2, P3, P4···P n Based on the weather analysis results (temperature, humidity, wind speed, precipitation, etc.) by ) and the learning of the disease control effectiveness of each disease control method, the graph in Fig. 5 shows each observation station (P1, P2, P3, P4···P n It can be corrected or updated by )
[0109] When examining the effectiveness of each control method in terms of the number of mosquitoes, the fogging or smoke screen mentioned above is a control method in which insecticide is sprayed on adult mosquitoes in accordance with the activity time of adult mosquitoes. Since it shows maximum effect immediately after spraying (approximately 1 to 2 days after spraying) and rapidly diminishes within a short period (approximately 3 days after spraying), the number of mosquitoes decreases rapidly immediately after spraying (1 to 2 days after spraying), then increases again after approximately 3 days, and the effect of the control disappears after approximately 7 days.
[0110] In addition, the above residual spray is a pest control method that involves applying a long-lasting insecticide to places where mosquitoes inhabit or move, killing mosquitoes whenever they come into contact with the insecticide. With a single application, the control effect lasts for a long period, and the number of mosquitoes decreases in the initial stages of application (about 3 days after application), after which the number of mosquitoes generally remains the same.
[0111] In addition, Bti is a biological, eco-friendly control agent that selectively kills only mosquito larvae. Because it selectively kills only larvae, it exhibits an effect that manifests immediately after application and lasts for several days before gradually declining. In the initial stages of control, since it kills only mosquito larvae, the number of adult mosquitoes remains constant for a certain period immediately after application (approximately 3 days after application). Subsequently, the number of adult mosquitoes decreases due to the larval killing effect, and then the number of adult mosquitoes increases from the point when the larval killing effect diminishes (approximately 7 days after application).
[0112] In addition, IGR is a pest control agent that kills mosquito larvae by disrupting their growth process so that they cannot emerge as adults. Its effect manifests after a certain period of time has elapsed since spraying and has a pattern of lasting for a long period (approximately 2 weeks to 1 month after spraying). Given that IGR targets only mosquito larvae (not adults) and its effect manifests after a certain period of time has elapsed since spraying, the number of adult mosquitoes remains unchanged during the initial period of control (approximately 7 days after spraying), but subsequently, due to the long-term control effect against mosquito larvae, the number of adult mosquitoes decreases over a long period.
[0113] As mentioned above, since the fog, smoke, and residual spray are mainly effective for controlling adult mosquitoes, and Bti and IGR are mainly effective for controlling mosquito larvae, it is necessary to implement two or more control methods in parallel to control both adult mosquitoes and larvae.
[0114] In this way, when two or more disease control methods are implemented in parallel, the individual effect attenuation coefficients of each disease control method can be calculated independently, and then the total disease control effect attenuation coefficients can be calculated by summing them according to the following equation (2).
[0115]
[0116] Here, i is the index of the quarantine method implemented in parallel, ∏ is the multiplication operator for all i, and m is the number of quarantine methods implemented in parallel. Additionally, Equation (2) is a formula for calculating the combined effect under the assumption that the attenuation effect of each quarantine method acts independently, ensuring that the attenuation effect relative to the simple arithmetic sum (sum of individual effect attenuation coefficients) does not exceed 1, while satisfying the natural boundary condition that when all individual effect attenuation coefficients are 0, the sum effect attenuation coefficient is also 0.
[0117] Accordingly, when two or more disease control methods are carried out in parallel, and the date of implementation is D and the number of days elapsed since the date of implementation of disease control is N, the above-mentioned generation prediction unit (300) can calculate the predicted generation amount of mosquitoes on the Nth day after implementation of disease control according to the following equation (3).
[0118]
[0119] In addition, as shown in FIG. 10, the effectiveness attenuation coefficient for each disease control method is automatically relearned and updated by receiving feedback from the cumulative actual measurement data of the operating environment and the verification results of the disease control effectiveness verification unit (500) at preset intervals or when a preset verification result threshold is reached. In addition, the update history is stored in the database (600) along with the update time, values before and after the update, and update trigger information (such as the arrival of a cycle or the reaching of a verification threshold), enabling post-hoc tracking and regression analysis.
[0120] In addition, for non-disinfection sections where no disinfection is carried out for a certain period, the disinfection effect attenuation coefficient (N) is treated as 0, and (1 - disinfection effect attenuation coefficient (N)) in Equation (1) becomes 1. In this case, the predicted occurrence amount is calculated using only the weather-based expected occurrence amount (D+N), and this can be used as a result to separately predict the trend of mosquito occurrence in natural conditions. In addition, the prediction results for non-disinfection sections can be used as decision support data or as a comparison standard for virtual simulations of disinfection.
[0121] In addition, for the time of occurrence of a vector-borne infectious disease, the time of a sudden surge in precipitation, or a special management time pre-set by the user (e.g., large-scale outdoor events, military outdoor training, etc.), the above-mentioned occurrence prediction unit (300) may apply a separate risk weighting coefficient to further increase the predicted occurrence amount. At this time, the above-mentioned risk weighting coefficient is set to a value of 1 or greater (e.g., a value in the range of 1.1 to 2.0), but can be determined according to the operating environment, thereby enabling conservative decision-making during the special management period.
[0122] In addition, the mosquito occurrence prediction result of the above occurrence prediction unit (300) after the disinfection is, as shown in FIG. 14, each observation station (P1, P2, P3, P4···P n It can be calculated by ) and displayed as a graph on the terminal (30).
[0123] Next, the disease control decision unit (400), based on the number of mosquitoes predicted by the occurrence prediction unit (300) and the analysis information analyzed by the information analysis unit (200), determines each observation station (P1, P2, P3, P4···P n After deciding on the decision regarding whether to carry out disinfection and the method of disinfection, etc., for each case, the manager is notified to recommend the implementation of disinfection.
[0124] First, the above-mentioned disease control decision unit (400), as shown in FIG. 7, based on the number of mosquitoes predicted to occur by the occurrence prediction unit (300) and the analysis information analyzed by the information analysis unit (200), each observation station (P1, P2, P3, P4···P n A comprehensive risk index is calculated for each ) according to the following equation (4) to determine whether to implement quarantine:
[0125] Composite Risk Index = Weather Score × W1 + Occurrence Score × Trend Weight × W2 + Environmental Correction Value … Equation (4)
[0126] Here, the weather score is a score calculated in the range of 0 to 100 according to a predefined mapping function from the weather analysis results (temperature, humidity, wind speed, precipitation, etc.) of the information analysis unit (200), the occurrence score is a score calculated in the range of 0 to 100 according to a predefined mapping function from the prediction results of the occurrence prediction unit (300), the trend weight is a correction weight determined according to the occurrence trend (increase / maintenance / decrease) of the previous N days, the environment correction value is a correction value determined according to the environmental characteristics of the area where the observation station is installed (e.g., population density, urbanization index, history of vector-borne infectious disease occurrence, etc.), and W1 and W2 are pre-set weighting coefficients that can be determined according to the operating environment in the range of 0.3 to 0.7.
[0127] That is, the above weather score is a score calculated in the range of 0 to 100 from the weather analysis results (temperature, humidity, wind speed, precipitation, etc.) of the information analysis unit (200) according to a predefined mapping function after learning through an AI learning model the effect of weather variables (temperature, humidity, wind speed, precipitation, etc.) on the activity, egg-laying, lifespan, blood-feeding activity, etc. of mosquitoes.
[0128] In addition, the above-mentioned occurrence score is a score calculated in the range of 0 to 100 from the prediction result of the above-mentioned occurrence prediction unit (300) according to a predefined mapping function after learning past mosquito occurrence data (e.g., mosquito occurrence amount, occurrence trend, etc. during the same period of past years) and vector-borne infectious disease occurrence through an AI learning model.
[0129] The above trend weights are assigned values of 1+α if the amount of occurrence during the previous N days is in an increasing trend, 1 if the amount of occurrence during the previous N days is in a maintaining trend with no change, and 1-β if the amount of occurrence during the previous N days is in a decreasing trend, wherein α and β are set within the range of 0<α,β<1 depending on the slope of the trend line indicating the increasing trend or decreasing trend.
[0130] In addition, W1 and W2 represent the relative weights determining whether to place more weight on the weather score or the occurrence score in judging the comprehensive risk index, so W1 + W2 = 1; typically, W1 = W2 is set to 0.5, but for each observation station (P1, P2, P3, P4···P n It may be determined differently depending on the operating environment of ).
[0131] In addition, the above environmental correction values are for each observation station (P1, P2, P3, P4···P n It is an additional correction value that reflects the environmental characteristics of the area where it is installed, and can have the effect of lowering the risk criteria for disease control decisions judged by objective indicators (weather score, occurrence score) for areas with a high risk of disease transmission due to mosquito outbreaks, such as areas with high population density or a history of vector-borne diseases.
[0132] In other words, in general cases, the above environmental correction value is 0, and for special areas such as regions with a high risk of transmitting infectious diseases due to mosquito outbreaks, a score calculated according to a predefined mapping function is added.
[0133] Additionally, the above-mentioned disease control decision unit (400) classifies the risk level into two or more predefined stages according to the value of the comprehensive risk index calculated according to Equation (4).
[0134] At this time, the classification of risk levels is intended to establish standards for the implementation of disease control measures in advance. For example, it can be classified into two levels, such as 'continued monitoring' and 'recommendation to implement disease control,' or into three levels, such as 'continued monitoring', 'recommendation to implement disease control' and 'need for urgent disease control.'
[0135] In an embodiment of the present invention, as shown in FIG. 7, the risk level may be classified into four stages of ‘low’, ‘caution’, ‘alert’, and ‘severe’ according to the priority of disease control measures, such as ‘strengthening monitoring’ → ‘preventive environmental maintenance’ → ‘recommendation for disease control’ → ‘need for emergency disease control’, but is not limited thereto.
[0136] In other words, it is desirable to classify risk levels in relation to disease control measures established based on disease prevention preparations, implementation status, and urgency, and the threshold values for each risk level can also be adjusted according to the operating environment of each observation station.
[0137] In addition, the comprehensive risk index and risk grade calculated by the above-mentioned disease control decision unit (400) are, as shown in FIG. 12, each observation station (P1, P2, P3, P4···P n It can be calculated by ) and displayed on the terminal (30).
[0138] In addition, the above-mentioned disease control decision-making unit (400) includes each observation station (P1, P2, P3, P4···P n Based on the assessed risk level for each ) station (P1, P2, P3, P4···P n Decides whether to implement quarantine measures by )
[0139] That is, each observation station (P1, P2, P3, P4···P nFor observation stations with a risk level of 'low', 'strengthened monitoring' may be recommended; for observation stations with a risk level of 'caution', 'preventive environmental maintenance (e.g., removing standing water in puddles, filling or removing puddles where water may accumulate, cutting or removing surrounding bushes)' may be recommended; for observation stations with a risk level of 'warning' or higher, 'implementation of disease control' may be recommended; and in particular, for any observation station with a risk level of 'severe', emergency disease control may be recommended. In this case, 'disinfection' in 'implementation of disease control' means 'chemical disease control'.
[0140] In other words, chemical pest control involves using chemicals containing insecticidal components to eliminate adult insects or larvae. Methods for eliminating adult insects include smoke disinfection, which involves mixing insecticides with kerosene and vaporizing them with heat; fogging, which involves mixing insecticides with water and spraying them in the form of fine particles; and residual spraying, which involves diluting residual insecticides with water and spraying them onto walls or structures to maintain the effect for a long period. For residual spraying and fogging, agents such as Deltamethrin, Permethrin, Phenothrin, and Lambda-cyhalothrin are mainly used, while for smoke disinfection, agents such as Cypermethrin are mainly used.
[0141] In addition, the method for eliminating larvae involves spraying chemicals into stagnant water where the larvae live to prevent them from developing into adult mosquitoes at the source, and microbial insecticides (Bti) and insect growth regulators (IGR) are primarily used.
[0142] Bti is a microbial culture component called Bacillus thuringiensis israelensis; when larvae consume this microorganism, their digestive organs are destroyed and eliminated, and it is harmless to fish and humans while specifically targeting mosquito larvae. Additionally, IGR is represented by the component Diflubenzuron, which interferes with the synthesis of chitin necessary for the molting process of larvae, causing them to fail to develop into adults and die.
[0143] In addition, each observation station (P1, P2, P3, P4···P n For each ), quarantine zones are established in advance centered around the observation stations, and decisions regarding quarantine methods, etc., are made for each observation station (P1, P2, P3, P4···P n It is determined by considering the environmental factors (geographical environment, social environment, climatic environment, etc.) of each quarantine zone.
[0144] At this time, the quarantine zone may be established by considering the mosquito breeding source and the mosquito's travel distance, and may be changed according to the results of the verification of the quarantine effectiveness described later, and each observation station (P1, P2, P3, P4···P n Information regarding environmental factors of ) is updated according to changes in environmental factors and stored in the database (600).
[0145] Here, geographical environment information refers to geographical information related to the habitat of adult mosquitoes or larvae, such as information about puddles or wetlands where water may accumulate within the quarantine zone, bushy areas, residential areas, urban areas, parks, etc.; social environment information refers to information related to human life or institutions, such as information about densely populated areas, drinking water source protection zones, ecological protection zones, chemical control restricted zones, eco-friendly agricultural areas, etc.; and climatic environment information refers to forecast information regarding expected future temperature, humidity, wind speed, etc.
[0146] That is, the above-mentioned disease control decision-making unit (400) is each observation station (P1, P2, P3, P4···P n For each risk level, pre-mapped quarantine methods, timing, agents, intensity, and priorities are automatically determined and output, and the quarantine methods, timing, agents, intensity, and priorities, etc., are for each observation station (P1, P2, P3, P4···P n It varies depending on environmental factors of the quarantine zone and can be updated by newly entered additional information.
[0147] In addition, the disease control method, etc. determined by the disease control decision unit (400) is, as shown in FIG. 13, each observation station (P1, P2, P3, P4···P n It can be calculated by ) and displayed on the terminal (30).
[0148] Furthermore, if there are two or more alternative pest control methods available for a specific observation station's control zone, it is desirable to calculate and output at least one of the following for each method in a comparative format: estimated effectiveness, estimated cost, estimated manpower requirements, and environmental burden score. For example, if both 'fogging' and 'smoke' pest control are optionally available for the same observation station's control zone, comparing and outputting the estimated mosquito density reduction rate, estimated pesticide cost, required manpower time, and environmental burden score for each method can support the person in charge in making a final decision.
[0149] In this way, the quarantine decision made by the quarantine decision unit (400) is notified to the person in charge through the terminal (30), and after the person in charge performs quarantine according to the quarantine decision made by the quarantine decision unit (400) and inputs information about the quarantine history through the terminal (30), the quarantine history information is stored in the database (600).
[0150] The above-mentioned disease prevention effect verification unit (500) is each observation station (P1, P2, P3, P4···Pn After each disinfection is carried out, the disinfection effect is verified by comparing the number of mosquitoes predicted by the above-mentioned occurrence prediction unit (300) with the actual number of mosquitoes collected by the above-mentioned data collection unit (100) after the disinfection is carried out.
[0151] That is, the above-mentioned disease control effect verification unit (500) compares the time series of the actual number of mosquitoes (actual value) collected by the above-mentioned data collection unit (100) during a pre-set verification period after disease control is implemented with the time series of the predicted occurrence amount (predicted value) calculated by the above-mentioned occurrence amount prediction unit (300), and calculates the difference ratio on a daily or hourly basis. At this time, the difference ratio is defined as (actual value - predicted value) / predicted value, is accumulated and analyzed over time, and can be specifically expressed by the following equation (5).
[0152] … Equation (5)
[0153] That is, a difference ratio of (+) indicates a case where the actual number of mosquitoes captured in the mosquito trap (20) is greater than the number predicted by the generation amount prediction unit (300), and (-) indicates a case where the actual number of mosquitoes captured in the mosquito trap (20) is less than the number predicted by the generation amount prediction unit (300).
[0154] At this time, the verification of the disease control effect in the above-mentioned disease control effect verification unit (500) places great emphasis on finding out whether there are any areas where disease control was omitted during the disease control process, rather than verifying the disease control method itself.
[0155] Accordingly, the above-mentioned disease prevention effect verification unit (500) is each observation station (P1, P2, P3, P4···P n Based on the time-series pattern of the difference ratio calculated during a predetermined period (hereinafter referred to as the "verification period") for each, it is automatically classified into one of the patterns: a "normal pattern," a "rebound pattern," or a "stagnation pattern." At this time, the verification period may be set within a range of 1 to 2 weeks after the implementation of the quarantine, taking into consideration the sustainability of the quarantine effect of the implemented quarantine method.
[0156] The above 'normal pattern' is a pattern in which the difference ratio maintains a trend within a set error range during the verification period after the implementation of quarantine, as illustrated in Figure 6 (a), and can be seen as indicating that the quarantine is working effectively. That is, if the time series of the actual value maintains a trend within the allowable error range with respect to the time series of the predicted amount of occurrence reflecting the attenuation effect of each quarantine method, it can be determined that the 'quarantine is normal'.
[0157] At this time, the error range is set in advance by taking into account the fact that the actual number of mosquitoes captured in the mosquito trap (20) cannot change regularly and the possibility of unforeseen environmental changes at the time of prediction.
[0158] For example, if the temperature in the area under the jurisdiction of a particular observation station is unexpectedly high or low on a specific day during the verification period (e.g., the temperature is 35°C or higher or 20°C or lower), if the wind speed is strong (e.g., the wind speed is 3 m / s or higher), or if the humidity is low (e.g., the humidity is 50% or lower), the activity of mosquitoes decreases significantly, causing the difference ratio of the observation station to increase to (-), which can be seen as a difference unrelated to the effectiveness of disease control.
[0159] In other words, the setting of the error range is for each observation station (P1, P2, P3, P4···P n It can be set by considering the possibility of unexpected environmental changes, and in this case, it is desirable to set the (+) direction small and the (-) direction large. For example, the error range can be set to +10 to -25%, but is not limited thereto.
[0160] The above 'rebound pattern' is a pattern in which the difference ratio is maintained within a set error range until a certain point in time after disinfection (e.g., three days after disinfection), as illustrated in FIG. 6 (b), and then the actual value rebounds and the difference ratio gradually increases, exceeding the set error range. This can be judged as a situation where the primary disinfection effect resulting from the disinfection has been achieved, but there is suspicion of additional outbreaks from untreated sources within the disinfection area.
[0161] In other words, if the time series of predicted mosquito outbreaks reflecting the attenuation effect of each control method continues to show a decreasing trend even after a certain point in time, while the time series of actual values rebounds to an increasing trend after that point, causing the difference ratio to gradually widen, this is determined to be an omission of control measures for some mosquito sources within the control zone and is classified as 'suspected omission of outbreak sources within the control zone.'
[0162] As such, if the difference ratio remains within the set error range up to a certain point in time after the disinfection is implemented, it indicates that the disinfection is effective. However, if the time series of the actual values rebounds sharply into an increasing trend after a certain point in time, it can be seen that disinfection was not handled or was insufficiently handled for any source among the sources within the disinfection zone set based on the relevant observation station. Therefore, the manager is notified immediately so that the unhandled source can be identified and additional disinfection can be carried out.
[0163] Third, as illustrated in (C) of FIG. 6, the above 'stagnant pattern' is a pattern in which the actual value decreases until a certain point in time after disinfection (e.g., three days after disinfection), but the difference ratio gradually increases, and then after that point, the actual value stagnates within a certain range while the difference ratio gradually expands and exceeds the set error range. This can be judged as a situation in which, despite the disinfection effect within the disinfection zone following the disinfection, the existence of an untreated large source of infection outside the disinfection zone is suspected.
[0164] In other words, if the difference ratio gradually increases despite the actual values decreasing up to a certain point after the implementation of quarantine, it may indicate that quarantine within the quarantine zone alone is insufficient. Conversely, if the actual values stagnate within a certain range after a certain point while the difference ratio gradually expands, it is determined that there is an untreated large-scale source outside the designated quarantine zone, leading to a continuous influx of adult mosquitoes from outside, and this is classified as "suspected untreated large-scale source outside the quarantine zone."
[0165] This 'stagnation pattern' occurs when adult mosquitoes at the observation station originate from sources outside the quarantine zone rather than within it. In such cases, while control measures were effective against adult mosquitoes active within the designated zone, they were ineffective against mosquitoes introduced from sources outside the zone. Consequently, the manager is notified to expand the quarantine zone to include areas outside the designated zone relative to the observation station, in order to search for the source and implement additional control measures.
[0166] In addition, the above rebound pattern and stagnation pattern can act as triggers for the AI learning model, as shown in FIG. 9.
[0167] That is, FIG. 9 shows a feedback flowchart of the model retraining results according to an embodiment of the present invention, illustrating a feedback loop leading to pattern classification of the quarantine effect verification unit (500) → storage in the database (600) → evaluation of retraining trigger conditions → operation of the model retraining trigger → update of Random Forest model parameters → new prediction cycle.
[0168] At this time, the model retraining trigger may be activated when a pre-set period (day / week / month) is reached, when accumulated data reaches a threshold, or when suspicion of missing source is detected, such as the rebound pattern or stagnation pattern mentioned above.
[0169] That is, the AI learning model of the present invention is automatically retrained by receiving feedback on the error between the actual data and the prediction result stored in the database (600) and the verification result of the quarantine effect verification unit (500), and when the quarantine effect verification unit (500) classifies it as the rebound pattern or the stagnation pattern, it may initiate retraining of the AI learning model by using this as a suspected signal of missing source.
[0170] Next, the database (600) stores information and data produced by the data collection unit (100), information analysis unit (200), occurrence amount prediction unit (300), quarantine decision unit (400), and quarantine effect verification unit (500), as well as data such as quarantine implementation history entered through the terminal (30).
[0171] At this time, the database (600) can store master data including location information by observation station, surrounding geographical information, special area classification information, past disease control history information, and past mosquito outbreak history information, and actual measurement data collected in real time separately.
[0172] Master data is static or semi-static data with a low frequency of change and can be stored in relational databases or document databases, etc., while actual data has time-series characteristics and can be stored in time-series databases or large-scale distributed storage, etc., and the two sets of separately stored data can be combined and queried through an integrated query interface.
[0173] In addition, the database (600) stores the training data, model parameters (tree structure, branching conditions, leaf node values, etc.), model version history, and model-specific performance evaluation indicators (MAE, RMSE, R², etc.) of the Random Forest Regressor model of the generation amount prediction unit (300). The model version history includes the time of model training, the range of data used for training, hyperparameters, and performance indicators, and provides a model history management function to enable regression analysis with the model at a specific time point upon the user's request. For example, the model improvement effect can be verified by comparing the prediction results of the model version at that time and the current model version for a specific past case.
[0174] In addition, it is desirable that differential access control be applied to the data stored in the above database (600) based on user authority.
[0175] In this case, user permissions can be classified into administrators, operators, analysts, external collaborators, etc., and for each permission, the scope of data that can be viewed, the scope of data that can be modified, and whether data can be exported externally are predefined. In addition, all access history, data change history, and reasons for changes are automatically recorded in the audit log, and it is advisable to store the audit log in a separate storage area to ensure integrity.
[0176] In addition, the above The database (600) stores an anonymized copy in a separate storage area for data that corresponds to user identification information (citizen reporter information, etc.) and personal information among the stored data, after undergoing a predefined anonymization process, so that it can be used for statistical analysis, model training, or external research cooperation. At this time, personal information is collected after obtaining consent in accordance with the Personal Information Protection Act, and the anonymization process can be performed through techniques such as identifier masking (name, phone number, etc.), generalization of location information into administrative district units (e.g., specific address → administrative district), k-anonymity, or differential privacy.
[0177] Next, a mosquito trap integrated control system can be configured to automatically control each module of the mosquito trap (20) according to the decision result of the system (10). At this time, the system (10) further includes a control command generation unit (700) for producing and transmitting control commands to regulate the operation of each module of the mosquito trap (20), as shown in FIG. 8.
[0178] That is, according to the risk grade or comprehensive risk index calculated by the disease control decision unit (400) of the system (10), the control command generation unit (700) of the system (10) transmits a control command to the control module (27) of the mosquito trap (20).
[0179] At this time, the control command transmitted from the system (10) may include a command to automatically adjust the operating mode and mosquito attraction intensity of the mosquito trap (20).
[0180] That is, the above control command includes whether the CO2 generation module (21) is operating, the operating time and amount of CO2 generated, whether the fan is operating and the rotation speed, etc., and the control module (27) automatically controls the operation of each sub-module according to the control command received from the control command generation unit (700) of the system (10).
[0181] Next, an AI-based mosquito outbreak prediction and control decision-making method according to the AI-based mosquito outbreak prediction and control decision-making automation system of the present invention (hereinafter referred to as the "System") is described. However, repetition of parts described in the above-described System is minimized, and parts omitted in the AI-based mosquito outbreak prediction and control decision-making method follow the description of the above-described System. Furthermore, when repeating formulas described in the above-described System, the formula number shall be the same as the formula number described in the System.
[0182] The AI-based mosquito outbreak prediction and disease control decision-making method according to the present invention includes each of the following steps, as illustrated in FIG. 9.
[0183] Step 1 involves collecting actual data regarding the number of mosquitoes captured from mosquito traps installed at one or more observation stations (S10).
[0184] That is, each observation station (P1, P2, P3, P4···P n Actual data regarding the number of mosquitoes captured by each of the mosquito traps (20) installed in the above area is collected, and if the species of mosquitoes captured by the mosquito traps (20) can be identified, actual data regarding the number of mosquitoes is collected by classifying them by species (e.g., Culex pipiens, Aedes albopictus, Aedes aureus, Aedes annuus, Aedes albopictus, etc.).
[0185] In addition, each observation station (P1, P2, P3, P4···P) where the above-mentioned mosquito trap (20) is installed n When environmental information (e.g., temperature, humidity, wind speed, precipitation, etc.) is measured in the above-mentioned mosquito trap (20), the environmental information is collected.
[0186] The second step involves combining and analyzing external information, including at least past weather information and future weather forecast information, and stored data stored in a database to produce analysis information (S20).
[0187] At this time, the external information may further include water system information, map information, and mosquito outbreak history information, in addition to at least past weather information and future weather forecast information, and is combined according to predefined data combination rules, and said data combination rules may include time synchronization, spatial synchronization, unit conversion, data interpolation, data normalization, etc.
[0188] Step 3 predicts the future mosquito population using an AI learning model with the data collected in Step 1 and the analysis information produced in Step 2 as input variables (S30).
[0189] In other words, a Random Forest Regressor-based machine learning model is used to predict future mosquito populations using data collected in the first stage and analysis information generated in the second stage as input variables.
[0190] However, the future mosquito population after the implementation of pest control is predicted by applying the attenuation coefficient of pest control effectiveness for each pest control method.
[0191] That is, when pest control is carried out using a single pest control method, the predicted amount of mosquitoes on the Nth day from the pest control implementation date (D) is calculated according to the following formula (1), and
[0192] Predicted occurrence amount (D+N) = Weather-based expected occurrence amount (D+N) × (1 - Attenuation coefficient of disease control effect (D+N))… Equation (1)
[0193] Here, the weather-based expected occurrence amount (D+N) is a predicted mosquito occurrence value calculated by the AI learning model with the disease control history variable set to 0 (no disease control assumption), and the disease control effect attenuation coefficient (D+N) is a function value that changes over time (N) for each disease control method, and is defined within the range of 0 to 1.
[0194] In addition, when two or more disease control methods are implemented in parallel, the individual effect attenuation coefficients of each disease control method can be calculated independently, and then the total disease control effect attenuation coefficients can be calculated by summing them according to the following equation (2).
[0195]
[0196] Here, i is the index of the parallel quarantine method, ∏ is the multiplication operator for all i, and m is the number of parallel quarantine methods.
[0197] Therefore, when D is the date on which two or more disease control methods are carried out in parallel and N is the number of days elapsed since the disease control was carried out, the predicted amount of mosquitoes on the Nth day after disease control can be calculated according to the following equation (3).
[0198]
[0199] Step 4 calculates a comprehensive risk index based on the prediction results calculated in Step 3 and the analysis information analyzed in Step 2 (S40).
[0200] First, based on the number of mosquitoes predicted by the generation amount prediction unit (300) and the analysis information analyzed by the information analysis unit (200), each observation station (P1, P2, P3, P4···P n A comprehensive risk index is calculated according to the following equation (4) for each )
[0201] Composite Risk Index = Weather Score × W1 + Occurrence Score × Trend Weight × W2 + Environmental Correction Value … Equation (4)
[0202] Here, the weather score is a score calculated according to a predefined mapping function from the weather analysis result of the information analysis unit (200), and the generation amount score is a score calculated according to a predefined mapping function from the prediction result of the generation amount prediction unit (300). The trend weight is a correction weight determined according to the generation amount trend of the previous N days, and a value of 1+α is assigned if the generation amount of the previous N days is in an increasing trend, 1 if the generation amount of the previous N days is in a maintaining trend with no change, and 1-β if the generation amount of the previous N days is in a decreasing trend, wherein α and β are set in the range of 0<α,β<1 depending on the slope of the increasing trend or decreasing trend, and the environment correction value is a correction value reflecting the environmental characteristics of the region where the observation station is installed, and W1 and W2 are pre-set weighting coefficients determined according to the operating environment in the range of 0.3 to 0.7, where W1+W2=1.
[0203] Step 5 automatically determines whether to implement quarantine measures, the method of quarantine, etc., based on the comprehensive risk index calculated in Step 4 and the characteristic information of the area where the observation station is installed (S50).
[0204] In other words, if the comprehensive risk index calculated in the fourth stage exceeds the threshold for determining the decision regarding the implementation of disease control, a decision is made to implement disease control, and the disease control method, timing, agents, intensity, and priority are determined based on characteristic information of the area where the observation station is installed.
[0205] Step 6 transmits the quarantine decision determined in Step 5 to the terminal (S60).
[0206] In other words, the decision on disease control determined in Stage 5 is transmitted to a terminal to notify the person in charge of recommendations regarding the implementation of disease control, methods, timing, agents, intensity, and priorities.
[0207] At this time, the terminal (30) is a computing device capable of connecting to the system (10), including a desktop, laptop, laptop, smartphone, tablet PC, wearable device, etc., and there are no special limitations as long as it is equipped with a communication module such as Bluetooth, NFC, Wi-Fi, etc., and can communicate with the system (10) in various network environments.
[0208] Step 7 involves calculating the difference ratio by comparing the actual number of mosquitoes collected in Step 1 (actual value) and the predicted mosquito population in Step 3 (predicted value) in a time series during the verification period after the implementation of the disinfection, and verifying the effectiveness of the disinfection by automatically classifying it into one of a normal pattern, a rebound pattern, or a stagnation pattern based on the time series pattern of the calculated difference ratio (S70).
[0209] At this time, the difference ratio is defined as (actual value - predicted value) / predicted value, is accumulated and analyzed over time series, and can be specifically expressed by the following equation (5).
[0210] … Equation (5)
[0211] In addition, the above 'normal pattern' is a pattern in which the difference ratio during the verification period after the quarantine is implemented remains within a set error range, as shown in Figure 4 (a), and can be judged as indicating that the quarantine is working effectively.
[0212] In addition, the above 'rebound pattern' is a pattern in which the difference ratio is maintained within a set error range until a certain point in time after disinfection (e.g., three days after disinfection), as illustrated in FIG. 4(b), and then the actual value rebounds and the difference ratio gradually increases, causing the difference ratio to exceed the set error range. This can be judged as a situation where the primary disinfection effect resulting from the disinfection has been achieved, but there is suspicion of additional outbreaks from untreated sources within the disinfection area.
[0213] The above 'stagnant pattern' is a pattern in which, as illustrated in (C) of FIG. 4, the actual value decreases until a certain point in time after disinfection (e.g., three days after disinfection), but the difference ratio gradually increases, and then after that point, the actual value stagnates within a certain range while the difference ratio gradually expands and exceeds the set error range. This can be judged as a situation in which, despite the disinfection effect within the disinfection zone following the disinfection, the existence of an untreated large source of infection outside the disinfection zone is suspected.
[0214] Step 8 transmits the results of the quarantine effect verification determined in Step 7 to the terminal (S80).
[0215] In other words, if the verification result of the disease control effectiveness at Stage 7 shows a normal pattern, a message stating 'disease control is normal' can be notified; if it shows a rebound pattern, a message recommending 'search for unhandled sources and implement additional disease control' can be notified along with a judgment that 'a source of infection within the disease control zone is suspected'; and if it shows a stagnation pattern, a recommendation to 'expand the disease control zone and implement additional disease control' can be made along with a judgment that 'a large-scale source of infection outside the disease control zone is suspected of being unhandled'.
[0216] The method according to the present invention described above may be implemented in the form of SaaS (Software as a Service) in a cloud computing environment and may be provided in a multi-tenant manner to two or more competent authorities.
[0217] In addition, the method according to the present invention described above may be implemented as a program and provided to servers or devices, and each device may connect to the server or device where the program is stored and download the program.
[0218] In addition, the method according to the present invention described above may be implemented as a program and provided by storing it on a readable recording medium. Explanation of the symbols
[0220] 10 AI-based mosquito outbreak prediction and control decision-making automation system 100 Data Collection Department 200 Information Analysis Department 300 Outbreak Prediction Unit 400 Disease Control Decision Unit 500 Disease Control Effectiveness Verification Department 600 Database 700 Control command generation unit 20 mosquito traps 21 CO2 Generation Module 22 Counting Module 23 Species Identification Module 24 Environmental Sensor Module 25 Self-diagnosis module 26 Communication module 27 Control Module 30 terminals 40 External Information Provision System
Claims
Claim 1 The method comprises: a first step of collecting actual data related to the number of mosquitoes captured from mosquito traps installed at one or more observation stations; a second step of combining and analyzing external information, including at least past weather information and future weather forecast information, with stored data stored in a database to produce analysis information; a third step of using an AI learning model to predict the future mosquito population using the actual data collected in the first step and the analysis information produced in the second step as input variables; a fourth step of calculating a comprehensive risk index based on the prediction result produced in the third step and the analysis information analyzed in the second step; a fifth step of automatically determining a decision on disease control, including whether to implement disease control and the method of control, based on the comprehensive risk index produced in the fourth step; a sixth step of notifying a terminal of the decision on disease control determined in the fifth step; and a seventh step of verifying the effectiveness of disease control by analyzing the time-series difference between the actual number of mosquitoes collected in the first step and the predicted mosquito population in the third step during a verification period after the implementation of disease control. In the seventh step, the predicted mosquito population The effectiveness of pest control is verified by calculating the difference ratio by comparing the time series (predicted value) with the time series (actual value) of the collected actual mosquito population, wherein the difference ratio is defined as (actual value - predicted value) / predicted value, and the time series pattern of the difference ratio is automatically classified into one of a normal pattern, a rebound pattern, or a stagnation pattern. The normal pattern is a pattern in which the difference ratio maintains a trend within a set error range during the period following pest control implementation, and is judged to indicate that pest control is working effectively. The rebound pattern is a pattern in which the difference ratio remains within a set error range until a certain point in time following pest control implementation, but after that point, the actual value rebounds and the difference ratio gradually increases, exceeding the set error range; this is judged to be a situation where the primary pest control effect has been manifested but additional outbreaks from untreated sources within the pest control area are suspected. The stagnation pattern is a pattern in which the actual value decreases until a certain point in time following pest control implementation, but the difference ratio gradually increases,AI-based mosquito outbreak prediction and control decision-making method characterized by determining a situation in which the existence of an untreated large-scale source of outbreak outside the quarantine zone is suspected, based on a pattern in which the actual value stagnates within a certain range after a certain point in time while the difference ratio gradually expands and exceeds a set error range. Claim 2 An AI-based mosquito outbreak prediction and control decision-making method according to claim 1, characterized in that, in the third step above, the AI learning model is a Random Forest Regressor-based machine learning model. Claim 3 In claim 1, when predicting the future mosquito occurrence amount after implementing a single mosquito control method in the third step, the AI-based mosquito occurrence prediction and control decision-making method is characterized by applying a control effect attenuation coefficient for each control method and calculating the future mosquito occurrence amount according to the following equation (1): Predicted occurrence amount (D+N) = Weather-based expected occurrence amount (D+N) × (1 - Control effect attenuation coefficient (D+N))… Equation (1) (where D is the date of implementation of control and N is the number of days elapsed since the date of implementation of control, the weather-based expected occurrence amount is a predicted value of the standard occurrence amount based on weather variables without considering the control effect, and the control effect attenuation coefficient is a function value in which the control effect changes over time for each control method.) Claim 4 In the third paragraph, when two or more disease control methods are implemented in parallel in the third step, the AI-based mosquito outbreak prediction and disease control decision-making method is characterized by independently calculating the individual effect attenuation coefficient of each disease control method and then summing the total disease control effect attenuation coefficient according to the following equation (2): (Here, i is the index of the parallel quarantine method, ∏ is the multiplication operator for all i, and m is the number of parallel quarantine methods.) Claim 5 In claim 1, the AI-based mosquito outbreak prediction and control decision-making method is characterized in that, in the fourth step, the comprehensive risk index is calculated according to the following equation (4): Comprehensive Risk Index = Weather Score × W1 + Outbreak Score × Trend Weight × W2 + Environmental Correction Value … Equation (4) (wherein the weather score is a score calculated in the range of 0 to 100 according to a predefined mapping function from the weather analysis results, the outbreak score is a score calculated in the range of 0 to 100 according to a predefined mapping function from the mosquito outbreak prediction results, the trend weight is a correction weight determined according to the outbreak trend of the previous N days, the environmental correction value is a correction value determined according to the environmental characteristics of the area where the observation station is installed, and W₁ and W₂ are pre-set weighting coefficients.) Claim 6 In claim 5, the AI-based mosquito outbreak prediction and control decision-making method is characterized in that the fifth step classifies the risk level into two or more predefined levels according to the value of the comprehensive risk index calculated in the fourth step, and automatically determines a control decision including a pre-mapped control method and a control agent for each risk level. Claim 7 delete Claim 8 delete Claim 9 delete Claim 10 A computer program stored on a computer-readable recording medium to enable a computer to perform the method of any one of paragraphs 1 through 6.
Citation Information
Patent Citations
Monitoring disease vectors
US20200170234A1
Combined confidence metric-based determination of input suitability for inputs generated by complex systems
US20260104530A1
Wireless sensor system for mosquito population growth analysis, logging, and reporting
WO2016064735A1
