Mosquito device intelligent system

Abstract

This invention relates to the field of IoT intelligent control technology and discloses an intelligent system for mosquito control equipment, including a time synchronization and control module, a dual-count fusion module, a multi-dimensional environmental perception module, a prediction and control module, a health assessment module, a regional collaborative optimization module, and a visualization management module. The time synchronization and control module obtains device control commands based on the fusion of BeiDou and GPS dual-channel time signals; the dual-count fusion module combines a weighing sensor and image recognition algorithm to obtain the accurate number of mosquitoes captured; the multi-dimensional environmental perception module collects environmental parameters and performs noise reduction processing; the prediction and control module uses transfer learning and incremental learning to build an activity prediction model; the health assessment module performs local prediction and optimizes the data upload strategy; the regional collaborative optimization module uses weighted Kriging interpolation and particle swarm optimization algorithms; and the visualization management module constructs a real-time map display. This invention enables intelligent management and precise control of mosquito control.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things (IoT) intelligent control technology, and more specifically, to an intelligent system for mosquito control devices. Background Technology

[0002] In the field of urban property management, mosquito control in large residential areas faces numerous challenges. Traditional mosquito trapping devices generally use a simple timer-based control method, which cannot adaptively adjust according to mosquito activity patterns and environmental conditions. This results in the devices wasting energy by running idle during periods when mosquitoes are not active, and failing to provide sufficient trapping intensity during periods when mosquitoes are active, leading to unsatisfactory control results.

[0003] Existing mosquito counting methods mainly rely on periodic manual inspections or single-sensor detection. Manual inspections are inefficient and cannot acquire data in real time, while single-sensor detection is susceptible to environmental interference and errors, making it difficult to accurately count the number of mosquitoes captured. Furthermore, the lack of coordination mechanisms between distributed deployments of multiple devices means that independent operation of each device cannot form a regionally optimized control strategy, resulting in unreasonable resource allocation.

[0004] Furthermore, the insufficient monitoring and fault warning capabilities of the equipment prevent managers from promptly detecting equipment anomalies, leading to equipment operating with faults or failing to provide timely maintenance after shutdown, thus impacting the overall mosquito control effectiveness. Therefore, an intelligent mosquito control system is needed that enables intelligent management, precise control, and regional collaborative optimization. Summary of the Invention

[0005] This invention provides an intelligent system for mosquito control equipment, which solves the technical problems in related technologies such as the inability of mosquito control equipment to adaptively adjust working parameters, insufficient mosquito counting accuracy, lack of equipment coordination mechanisms, and insufficient fault early warning capabilities.

[0006] This invention provides an intelligent mosquito control system, comprising: The timing and control module performs weighted fusion processing on dual-channel time signals acquired from Beidou and GPS satellites to obtain a time-driven sequence of device control commands and the device's working status after execution. The dual counting fusion module acquires the working status of the equipment, collects container weight change data based on the weighing sensor and calculates the preliminary count value, and uses an image recognition algorithm to visually count mosquitoes to obtain the number of mosquitoes captured. The multi-dimensional environmental sensing module obtains the number of mosquitoes captured, collects multi-dimensional environmental parameters based on temperature and humidity sensors, light intensity sensors and infrared sensing modules, performs noise reduction processing on the raw data, and obtains a complete monitoring data package containing environmental feature vectors and equipment operating status. The prediction and control module, based on the complete monitoring data package, uses transfer learning and incremental learning to build a mosquito activity prediction model, calculates adaptive operating parameters, obtains equipment control commands with adaptive operating parameters, and dynamically optimizes the equipment operating status after executing the commands. The health assessment module performs local predictions on the device based on dynamically optimized device operating status and optimizes data upload strategies to obtain device health assessment results and predictive maintenance warnings. The regional collaborative optimization module constructs a mosquito density distribution model based on the equipment health assessment results and all equipment data streams using the weighted kriging spatial interpolation method, identifies high-incidence areas and obtains regional collaborative prevention and control results. The visualization management module uses the Leaflet map library to build a real-time map display of the device status and visualize the spatial distribution of mosquito density, based on the results of regional collaborative prevention and control and all device data.

[0007] In a preferred embodiment, the timing and control module includes: It receives navigation signals from BeiDou and GPS satellites, extracts time information, and outputs BeiDou timestamps, GPS timestamps, and corresponding signal quality factors. The validity of BeiDou time data and GPS time data is verified, signal quality factors are extracted and normalized, Kalman filtering algorithm is used to fuse the two time signals, and the observation noise covariance matrix is ​​dynamically adjusted according to the signal quality score to obtain the standard time for accurate synchronization. The precise synchronization standard time is compared with the device's internal real-time clock. When the time deviation exceeds the preset deviation threshold, a clock synchronization calibration operation is triggered. The work schedule is read based on the device's local time, and a sequence of device control commands is generated according to the work mode identifier.

[0008] In a preferred embodiment, the dual-counting fusion module includes: A strain gauge force sensor is used to sense the weight change of the collection container in real time. The resistance change is converted into a voltage signal through a Wheatstone bridge circuit. The sampled data is recorded in time series and the average weight value is calculated. The weight increment information is extracted by differential operation on the time series weight data sequence. The count conversion is performed using the preset average weight parameter of a single mosquito. The miniature camera module automatically triggers image acquisition according to the preset shooting cycle, uploads digital images to the cloud server, and uses an improved YOLOv8 architecture mosquito target detection model for intelligent recognition processing, counting and statistically analyzing detection boxes with confidence scores higher than the preset confidence threshold. The relative error between the weighing count and the image recognition count is calculated. When the relative error is less than a preset error threshold, a simple averaging fusion strategy is adopted. When the relative error is greater than or equal to the preset error threshold, a weighted fusion strategy is adopted.

[0009] In a preferred embodiment, the multidimensional environment perception module includes: The temperature and humidity sensor communicates with the device microcontroller via the I2C bus interface and outputs ambient temperature and relative humidity values. The light intensity sensor outputs digital light intensity values. The infrared sensing module detects the movement of heat sources within the field of view and outputs a trigger signal. The sliding window averaging filter method is used to smooth the temperature, humidity and light intensity data, and the arithmetic mean of the continuous sampling point data within the preset time window is calculated. The statistical indicators of environmental parameters are calculated using a preset statistical window to form an environmental feature vector that includes the average temperature, average humidity, average light intensity, temperature change rate, humidity change rate, light intensity change rate, and number of infrared triggers. Collect equipment operating status parameters, package and encapsulate them with environmental feature vectors, and serialize them using JSON format.

[0010] In a preferred embodiment, the prediction and control module includes: Based on publicly available pest activity research datasets, a basic prediction model was constructed using the LightGBM gradient boosting tree algorithm. The model hyperparameters were set and iterative training was performed. Historical monitoring data collected by locally deployed devices is extracted from cloud databases. The number of mosquitoes captured is standardized into an activity score as the target label, and the basic prediction model is then fine-tuned in a domain-adaptive manner. When the number of accumulated new data records reaches a preset threshold, the model incremental training task is automatically triggered, and the new data is mixed with the historical training data in a preset ratio for model retraining. The environmental feature vector is input into the continuously optimized mosquito activity prediction model for inference calculation. Fuzzy control logic is used to calculate the adaptive adjustment target value of the device's operating parameters, which is then encapsulated into device control commands and sent to the target device for execution.

[0011] In a preferred embodiment, the health assessment module includes: The edge computing module integrated on the device comes pre-installed with a simplified anomaly detection rule engine, which monitors the device's operating parameters in real time and substitutes the parameter values ​​into the detection rules for judgment. When any rule is triggered, a local warning sign is generated. The data upload strategy is dynamically adjusted based on local prediction results. When no abnormal signs are detected, the normal data upload frequency is maintained. When abnormal signs are detected, the data upload frequency is increased and a warning flag field is added to the data packet. Real-time monitoring of communication signal strength indicators; different data compression and transmission strategies are adopted according to the signal strength; breakpoint resume mechanism is implemented to cope with data transmission interruption. The cloud server performs real-time monitoring and quality assessment of the data stream, uses time-series-based trend prediction methods for in-depth analysis, constructs a quantitative equipment health scoring system, sets graded early warning thresholds, and pushes early warning messages.

[0012] In a preferred embodiment, the regional collaborative optimization module includes: The data quality weight is calculated by combining the equipment health score and the data integrity rate, and a corresponding quality weight label is attached to the monitoring data records of all equipment. The spatial distribution model of mosquito density is modeled using the weighted kriging spatial interpolation method. A spatial coordinate system is constructed and the spatial covariance function is calculated. Mass weights are introduced to weight the observation data. The entire region is divided into regular grids and kriging interpolation is performed at the center point of each grid. An improved DBSCAN density clustering algorithm is used to identify mosquito-prone areas. An adaptive neighborhood radius mechanism is introduced to dynamically adjust the clustering radius based on the local device density. When the number of grid points in a cluster exceeds the preset minimum cluster size, it is identified as a mosquito-prone area. The optimal operating parameter configuration for each device is calculated using the particle swarm optimization algorithm. A dual-objective optimization function is constructed and constraints are set. The optimal operating parameter configuration scheme is distributed in batches using a phased and gradual adjustment strategy.

[0013] In a preferred embodiment, the visualization management module includes: A cloud server is used to build a web-based visual management platform. The front end integrates the Leaflet open-source map library for the visualization of geospatial data. It calls the backend API interface to obtain the real-time location coordinates and operating status data of all devices, and selects different colored icons to distinguish them according to the operating status of the devices. A mosquito density heatmap layer is overlaid on the map display of the management platform. The density estimate and coordinate position of grid points are extracted from the spatial distribution raster data of mosquito density. The rendering parameters of the heatmap are configured and automatic refresh is supported. Develop functions for comparing and analyzing the capture effect of equipment, analyzing mosquito activity trends, and analyzing abnormal events. Use a front-end chart library to create bar charts, line charts, and pie charts. The system has been developed to automatically generate daily, weekly, and monthly reports. The system automatically extracts data for the corresponding time period from the database for statistical summary, and the reports can be exported as PDF or Excel format.

[0014] In a preferred embodiment, the prediction and control module further includes: Define the fuzzy input variable as the mosquito activity score, divide it into a preset number of fuzzy subsets, define a membership function for each fuzzy subset, and substitute the precise activity score into the membership function to calculate the membership degree of the score to each fuzzy subset; The fuzzy control rule base is defined to contain multiple control rules in the form of IF-THEN. Each rule describes the adjustment coefficient of the equipment operating parameters under a specific activity level. The membership degree of the input variable to the fuzzy subset of the rule's antecedent is taken as the activation strength of the rule. Defuzzing is performed using a weighted average method. The adjustment coefficient of each rule output is multiplied by the corresponding activation intensity. The weighted output of all rules is summed and then divided by the sum of activation intensities to obtain the final working parameter adjustment coefficient. The adjustment coefficient is then multiplied by the baseline value of each working parameter to calculate the specific value.

[0015] In a preferred embodiment, the regional collaborative optimization module further includes: The optimization objective is defined as maximizing the total number of mosquitoes captured in the area while minimizing the total energy consumption in the area. A dual-objective optimization function is constructed. The two objective functions are normalized and weighted to obtain a comprehensive objective function. The constraint condition is set that the adjustment range of the working parameters of each device does not exceed the preset adjustment range of the standard value. The initial particle swarm contains a preset number of particles, each representing a set of equipment operating parameter configuration schemes. The position and velocity of each particle are randomly initialized, the comprehensive objective function value corresponding to each particle is calculated, and the individual optimal position and global optimal position of each particle are recorded. The velocity and position of the particles are updated iteratively. The velocity update formula includes inertial terms, individual cognition terms, and social learning terms. Boundary processing is performed on the updated positions. The objective function values ​​of each particle are recalculated and the individual optimum and global optimum are updated. The iteration is terminated early when the global optimum objective function value no longer improves after a preset number of consecutive rounds.

[0016] The beneficial effects of this invention are as follows: By combining a dual counting fusion module with a weighing sensor and an image recognition algorithm for cross-validation, an adaptive fusion strategy is adopted to obtain an accurate number of mosquitoes captured, thus improving counting accuracy. The prediction and control module uses transfer learning and incremental learning techniques to build a mosquito activity prediction model, and combines it with fuzzy control logic to calculate adaptive operating parameters, realizing environmental adaptive adjustment of the equipment's operating mode, improving capture efficiency and reducing energy consumption.

[0017] The regional collaborative optimization module uses weighted kriging spatial interpolation to construct a mosquito density distribution model, identifies high-incidence areas, and uses particle swarm optimization algorithm to calculate the optimal working parameter configuration for regional collaboration, thus realizing collaborative optimization of multiple devices. The health assessment module performs local prediction and optimizes data upload strategy at the device end. Combined with cloud-based health assessment and trend prediction, predictive maintenance early warning is realized, improving system reliability and management efficiency. Attached Figure Description

[0018] Figure 1 This is a block diagram of the intelligent mosquito control system of the present invention; Figure 2 This is a module sub-diagram of the intelligent mosquito control system of the present invention. Detailed Implementation

[0019] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0020] At least one embodiment of the present invention discloses an intelligent system for mosquito control devices, such as... Figures 1 to 2 As shown, it includes: The timing and control module performs weighted fusion processing on dual-channel time signals acquired from Beidou and GPS satellites to obtain a time-driven sequence of device control commands and the device's working status after execution. S11 integrates a Beidou and GPS dual-mode timing module on the main control circuit board of the mosquito trapping device. This timing module includes a Beidou signal receiving unit and a GPS signal receiving unit. The two receiving units receive navigation signals from Beidou satellites and GPS satellites respectively through independent radio frequency front-ends and signal processing circuits, extracting precise time information from the navigation signals. The Beidou signal receiving unit outputs a Beidou timestamp and signal quality factor, and the GPS signal receiving unit outputs a GPS timestamp and signal quality factor. The two time signals are transmitted to the device microcontroller through a serial communication interface to obtain Beidou time data and GPS time data.

[0021] The signal quality factor includes a satellite signal strength index and a positioning accuracy factor. The satellite signal strength index reflects the power level of the received satellite signal, while the positioning accuracy factor reflects the impact of satellite geometric distribution on positioning accuracy. Both indices jointly characterize the reliability of the time signal. When the signal quality of a certain signal deteriorates due to environmental obstruction or insufficient satellite visibility, the system can identify and reduce the fusion weight of that signal.

[0022] S12, after receiving the BeiDou time data and GPS time data obtained in step S11, the device microcontroller first verifies the validity of the two time data streams, checking the integrity of the timestamp format and the rationality of the values. If the timestamp format is incorrect or the value exceeds the reasonable range, the data stream is deemed invalid and discarded, and only the other valid data stream is used. When both data streams are valid, the BeiDou signal quality factor and GPS signal quality factor are extracted, and the signal strength index and positioning accuracy factor are normalized. The normalization method is to divide each index value by the preset maximum value of the index to obtain a normalized quality score between zero and one. The normalized scores of the signal strength index and positioning accuracy factor are then arithmetically averaged to obtain the comprehensive quality score of the BeiDou signal and the comprehensive quality score of the GPS signal.

[0023] S13. Based on the comprehensive quality scores of the BeiDou signal and GPS signal obtained in step S12, a Kalman filter algorithm is used to fuse the two time signals. The state variable of the Kalman filter is the real time, and the observation variables are the BeiDou timestamp and the GPS timestamp. The system noise covariance matrix and the observation noise covariance matrix are dynamically adjusted according to the signal quality score. The specific adjustment rules are as follows: the observation noise covariance of the BeiDou signal is set to the reciprocal of the comprehensive quality score of the BeiDou signal, and the observation noise covariance of the GPS signal is set to the reciprocal of the comprehensive quality score of the GPS signal. The higher the quality score, the smaller the corresponding observation noise, and the greater the weight of that signal in the fusion. The Kalman filter is calculated iteratively through prediction and update steps. The prediction step predicts the time estimate of the current moment according to the system state equation, and the update step corrects the time estimate according to the observation equation and the Kalman gain matrix. After multiple iterations, the fused high-precision timestamp is obtained. The synchronization error of the fused timestamp is controlled within 30 seconds, resulting in a standard time for accurate synchronization.

[0024] S14, the device microcontroller compares the precise synchronization standard time obtained in step S13 with the current time of the device's internal real-time clock, calculates the time deviation between the two, and when the absolute value of the time deviation exceeds ten seconds, triggers a clock synchronization calibration operation to adjust the time value of the device's internal real-time clock to the precise synchronization standard time, thereby achieving synchronization between the device's local time and the standard time; this clock synchronization calibration operation is automatically performed once per hour to ensure that the device time remains accurate over a long period of time, obtaining the device's local time synchronized with the standard time.

[0025] S15, based on the device local time obtained in step S14, the device microcontroller reads the work schedule pre-stored in the non-volatile memory. The work schedule contains configuration records for multiple work periods. Each record includes a period number, start time, end time, and work mode identifier. The work mode identifier is divided into three states: on, off, and standby. The microcontroller matches the current device local time with the start and end times of each period in the work schedule to determine the period to which the current time belongs. Based on the work mode identifier of that period, the microcontroller generates device control instructions. When the work mode is on, the control instructions include commands to start the mosquito-attracting module, heating pad, and fan. When the work mode is off, the control instructions include commands to shut down all functional modules. When the work mode is standby, the control instructions include commands to keep the communication module working while shutting down other modules, thus obtaining a time-driven device control instruction sequence.

[0026] S16, the device microcontroller executes the device control command sequence obtained in step S15, and outputs control signals to the power management circuits of each functional module of the device through GPIO pins to realize the switching of each module on and off. When the device enters the on mode, the mosquito-attracting module starts to release mosquito attractant, the heating pad is powered on and heated to the preset working temperature, and the fan starts to generate airflow to diffuse the mosquito attractant scent to the surrounding environment. After being attracted, mosquitoes enter the collection container. When the device enters the off mode, each functional module is powered off and stops working, and the device enters a low-power state. Through precise time synchronization control, the device can automatically turn on during specific periods when mosquitoes are active and automatically turn off during periods when mosquitoes are not active, realizing a precise match between the working cycle and the activity pattern of mosquitoes. This improves capture efficiency while reducing energy consumption, achieving a device working state with precise time control.

[0027] The dual counting fusion module acquires the working status of the equipment, collects container weight change data based on the weighing sensor and calculates the preliminary count value, and uses an image recognition algorithm to visually count mosquitoes to obtain the number of mosquitoes captured. S21, Input the device operating status obtained in step S16. When the device is in the on mode, mosquitoes are lured into the collection container. The weighing sensor installed at the bottom of the collection container senses the weight change of the container in real time. The weighing sensor is a strain gauge force sensor. When the weight of the container increases, the strain gauge inside the sensor deforms, causing a change in resistance. The change in resistance is converted into a voltage signal output through a Wheatstone bridge circuit. After the voltage signal is amplified and filtered by the signal conditioning circuit, it is input to the analog-to-digital converter for digital sampling. The sampling frequency is set to once per second. The device microcontroller records the sampled digital signal in a time series. With five minutes as a statistical period, the arithmetic mean of all the data points sampled in the period is calculated to obtain the average weight value of the period. The average weight values ​​of each period are stored in chronological order to obtain the time-series weight data sequence.

[0028] S22, perform a differential operation on the time-series weight data sequence obtained in step S21 to extract weight increment information. The specific method is as follows: subtract the average weight value of the previous period from the average weight value of the current period to obtain the weight increment of the current period. When the weight increment is negative, it is determined to be an abnormal value caused by sensor noise or container disturbance. The increment value is set to zero, and only positive weight increments are retained. The weight increments of multiple consecutive periods are accumulated to obtain the cumulative weight increase. The cumulative weight increase reflects the total mass of mosquitoes captured over a period of time, and the cumulative weight of mosquitoes captured is obtained.

[0029] S23. Based on the cumulative weight of mosquitoes captured obtained in step S22, the count is converted using a preset average weight parameter per mosquito. The method for obtaining the average weight parameter per mosquito is as follows: collect one hundred local common mosquito samples in a laboratory environment, weigh and record the weight of each mosquito using a precision electronic balance, calculate the arithmetic mean of the weights of all samples, and obtain a reference value for the average weight of a single mosquito. Considering the differences in weight of mosquitoes of different species and developmental stages, a default value of two milligrams is set. Divide the cumulative weight of mosquitoes captured by the average weight of a single mosquito to obtain the initial value of mosquito count based on the weighing method.

[0030] The average weight parameter of a single mosquito can be adjusted according to the mosquito population characteristics of different regions and seasons. The system administrator can modify this parameter value through the cloud platform interface. The modified parameter value is sent to the device through the communication module and stored locally for subsequent counting and conversion.

[0031] S24, Input the device working status and current time obtained in step S16. The built-in miniature camera module automatically triggers image acquisition according to the preset shooting cycle, which is set to once every ten minutes. The microcontroller sends a shooting command to the camera module, and the camera module activates the supplementary LED light to ensure sufficient illumination inside the collection container. Then, image acquisition is performed. The camera uses a five-megapixel CMOS image sensor, and the shooting field of view covers the entire bottom area of ​​the collection container. The image resolution is set to 2,560 x 1,920 pixels. After the shooting is completed, the image data is temporarily stored in the cache memory of the camera module and transmitted to the device microcontroller through the SPI serial interface. The microcontroller compresses the image data in JPEG format to reduce the data volume, thus obtaining a digital image of the mosquitoes inside the collection container.

[0032] S25. The digital image obtained in step S24 is uploaded to the cloud server for intelligent recognition processing. The cloud server deploys a mosquito target detection model based on the improved YOLOv8 architecture. This model is specifically optimized based on the general target detection model. The specific optimization measures include: enhancing the small target detection capability of the network by adding a higher resolution feature layer in the feature pyramid network to address the small size of mosquito targets; improving the threshold strategy of the non-maximum suppression algorithm to reduce the probability of false deletions by addressing the potential overlap and occlusion of individual mosquitoes; expanding the training dataset to address the diversity of different mosquito species and postures by collecting image samples of various common mosquito species under different lighting conditions and background environments. The training set contains 50,000 labeled images, and the validation set contains 10,000 labeled images. After 300 rounds of iterative training, the model achieves an average precision of 93.5 and a recall of 90.2 on the validation set.

[0033] The improved YOLOv8 model's network structure comprises three parts: a backbone feature extraction network, a feature pyramid network, and a detection head. The backbone network uses a CSPDarknet structure to extract multi-level features. The feature pyramid network fuses feature maps of different scales through a top-down path and lateral connections. The detection head performs convolution operations on the fused feature maps to output the bounding box coordinates, class probability, and confidence score of the target. During model inference, the input image is scaled to 640 x 640 pixels. After forward propagation through the network, multiple candidate detection boxes are obtained. Redundant detection boxes are removed through confidence thresholding and non-maximum suppression. Finally, the bounding box position and class label of each mosquito target are output.

[0034] S26, the cloud server uses the mosquito target detection model trained in step S25 to perform inference calculations on the uploaded digital images. The model outputs the bounding box coordinates and confidence scores of all detected mosquito targets in the image. Detection boxes with confidence scores higher than a preset first confidence threshold are counted. This threshold is set to 0.6. Detections below this threshold are considered false detections and discarded. The total number of detection boxes is the number of individual mosquitoes identified in the image. Since the image acquisition time interval is ten minutes, new mosquitoes may enter the collection container between adjacent shots. Therefore, it is necessary to accumulate the recognition results of multiple shots. The specific method is as follows: subtract the recognition number of the previous image from the recognition number of the current image to obtain the number of new mosquitoes in that time interval. Accumulate the new numbers of all time intervals to obtain the cumulative number of mosquitoes captured based on the image recognition method.

[0035] S27, the initial mosquito count obtained in step S23 based on the weighing method and the cumulative mosquito capture count obtained in step S26 based on the image recognition method are cross-validated and fused. First, the relative error between the two count values ​​is calculated. The formula for calculating the relative error is the absolute value of the difference between the two count values ​​divided by the arithmetic mean of the two count values ​​and then multiplied by 100%. When the relative error is less than 15%, it indicates that the results of the two counting methods are in good agreement. A simple averaging fusion strategy is adopted, which adds the two count values ​​and divides them by two to obtain the fused mosquito capture count. When the relative error is greater than or equal to 15%, the fusion is performed. At step 15, it was found that there was a significant difference between the two counting methods. At this point, a weighted fusion strategy was adopted. Considering that the image recognition method can directly observe individual mosquitoes, is less affected by external interference, and has relatively higher accuracy, the image recognition count value was given a higher weight. The weight coefficient of the image recognition count value was set to 0.7, and the weight coefficient of the weighing count value was set to 0.3. The two count values ​​were multiplied by their respective weight coefficients and then added together to obtain the weighted fused mosquito capture count. At the same time, the system recorded an anomaly flag for this count, which was convenient for subsequent manual verification, to obtain the accurate mosquito capture count after double verification and fusion processing.

[0036] The multi-dimensional environmental sensing module obtains the number of mosquitoes captured, collects multi-dimensional environmental parameters based on temperature and humidity sensors, light intensity sensors and infrared sensing modules, performs noise reduction processing on the raw data, and obtains a complete monitoring data package containing environmental feature vectors and equipment operating status. S31, Input the number of mosquitoes captured and the current timestamp obtained in step S27. Multiple environmental sensors configured on the device simultaneously start data acquisition tasks. The temperature and humidity sensor uses a digital integrated temperature and humidity sensor chip, communicating with the device's microcontroller via an I2C bus interface. The sensor performs a measurement every minute, outputting the current ambient temperature and relative humidity values. The temperature measurement range is -20 degrees Celsius to +60 degrees Celsius, with a measurement accuracy of ±0.3 degrees Celsius. The humidity measurement range is 0 to 100% relative humidity, with a measurement accuracy of ±2%. The light intensity sensor uses a silicon photodiode as the photosensitive element; its spectral response characteristics are close to the human eye's visual curve, accurately reflecting ambient illuminance. The sensor outputs an analog voltage signal that is proportional to the light intensity. This signal is sampled by an analog-to-digital converter to obtain a digital light intensity value. The measurement range is from zero to 100,000 lux, and the sampling period is set to once per minute. The infrared sensing module uses a pyroelectric infrared sensor, which can detect the movement of heat sources within the field of view. The sensor's detection distance is three meters, and the field of view is 110 degrees. When a moving heat source is detected, the sensor outputs a high-level trigger signal. The microcontroller captures this trigger signal and records the occurrence time to obtain an infrared trigger event sequence. The microcontroller timestamps and caches the data collected by each sensor to obtain multi-dimensional environmental raw data including temperature, humidity, light intensity, and infrared trigger events.

[0037] S32, the multidimensional environmental raw data obtained in step S31 is preprocessed to eliminate sensor noise and outliers. A sliding window averaging filter method is used to smooth the temperature, humidity, and light intensity data. The length of the sliding window is set to five sampling points. For the data point at the current moment, the data of five consecutive sampling points at the current moment and the previous four moments are extracted, and the arithmetic mean of these five data points is calculated as the filtered data value at the current moment. The sliding window moves forward in chronological order, and the same filtering operation is performed on the data at each moment. Through sliding window averaging filter, the random noise and sudden interference of the sensor can be effectively suppressed, making the data curve smoother, while preserving the overall trend of the data change, and obtaining the denoised temperature time series data, humidity time series data, and light intensity time series data.

[0038] The sliding window averaging filtering method has a good noise reduction effect on signals with stable changes, but it will cause a delay for signals with rapid changes. In mosquito control application scenarios, the rate of change of environmental parameters is relatively slow. This filtering method can achieve a good balance between noise reduction and response speed.

[0039] S33. Based on the filtered time-series data obtained in step S32, statistical features of environmental parameters are extracted to characterize the environmental state. Using each hour as a statistical window, various statistical indicators of temperature, humidity, and light intensity are calculated within that time window, including mean, maximum, minimum, standard deviation, and rate of change. The mean reflects the average environmental level during that period and is calculated as the arithmetic mean of all sampled data points within the time window. The maximum and minimum values ​​reflect the range of environmental fluctuations during that period. The standard deviation reflects the dispersion of the data and is calculated by taking the square root of the sum of the squares of the differences between each sampled data point and the mean, divided by the number of sampled points. The square root; the rate of change reflects the speed of change of environmental parameters, and is calculated by subtracting the data value at the beginning of the time window from the data value at the end of the time window and then dividing by the length of the time window; for infrared trigger event sequences, the total number of times the infrared sensor is triggered within the time window is counted, which reflects the frequency of biological activity in the surrounding environment; the above statistical indicators are arranged and combined in a fixed order to form a multi-dimensional vector, which contains seven feature components: mean temperature, mean humidity, mean light intensity, rate of change of temperature, rate of change of humidity, rate of change of light intensity, and number of infrared triggers, thus obtaining an environmental feature vector characterizing the environmental state.

[0040] S34, acquire the current operating status parameters of the device, including the device power supply voltage, heating pad temperature, fan speed, and mosquito attractant release amount; the device power supply voltage is monitored in real time by a voltage detection circuit, which steps down the power supply voltage through a voltage divider network and inputs it to the analog-to-digital converter for sampling. The microcontroller calculates the actual voltage value based on the sampled value and the voltage division ratio; the heating pad temperature is measured by a temperature sensor integrated on the surface of the heating pad, and the analog signal output by the sensor is processed by a signal conditioning circuit and analog-to-digital conversion to obtain the digital temperature value; the fan speed is detected by a Hall effect sensor, with a magnet installed on the fan rotor, and the Hall sensor outputs a pulse for each revolution. The microcontroller counts the number of pulses per unit time and converts them into rotational speed. The release amount of mosquito attractant is adjusted by controlling the opening degree and opening duration of the release valve. The microcontroller records the control parameters of the valve and calculates the cumulative release amount based on the flow calibration data. The microcontroller collects the real-time values ​​of the above operating status parameters, packages and encapsulates them with the environmental feature vector obtained in step S33, and adds the device's unique identifier, the precise timestamp obtained in step S1, and the number of mosquitoes captured obtained in step S27. The fields are organized according to the predefined data format specifications and serialized and encoded using JSON format to obtain a structured and complete device monitoring data package.

[0041] The data format of the device monitoring data packet is as follows: The data packet contains a device identifier field for uniquely identifying the device, a timestamp field to record the data collection time, an environmental parameter field containing temperature, humidity, light intensity and their statistical characteristics, a device status field containing voltage, heating pad temperature, fan speed and mosquito attractant release amount, and a mosquito data field containing the number of mosquitoes captured and the counting method identifier. Each field is organized in key-value pair form, which is convenient for cloud server parsing and processing.

[0042] The prediction and control module, based on the complete monitoring data package, uses transfer learning and incremental learning to build a mosquito activity prediction model, calculates adaptive operating parameters, obtains equipment control commands with adaptive operating parameters, and dynamically optimizes the equipment operating status after executing the commands. S41, Input the complete device monitoring data package obtained in step S34. The cloud server parses and extracts environmental feature vectors from the data package. To establish a mosquito activity prediction model, it first performs pre-training based on a publicly available pest activity research dataset. This publicly available dataset contains pest monitoring records from multiple regions. Each record includes environmental features such as ambient temperature, humidity, light intensity, time period, and season, as well as the corresponding pest activity score. The cloud server uses the LightGBM gradient boosting tree algorithm to construct the basic prediction model. This algorithm predicts by integrating multiple decision trees and has the advantages of fast training speed, high prediction accuracy, and support for large-scale data. During model training, the publicly available data will be used to predict the mosquito activity. The dataset was divided into a training set and a validation set, with the training set accounting for 80% and the validation set accounting for 20%. The model hyperparameters were set as follows: 500 trees, maximum tree depth of 8 layers, learning rate of 0.05, and minimum number of samples per leaf node of 20. The model input features included five dimensions: mean temperature, mean humidity, mean light intensity, time period encoding, and season encoding. The output target was a standardized pest activity score, ranging from 0 to 100, with higher scores indicating more active pests. After 500 iterations of training, the root mean square error of the model on the validation set converged to 8.3, and the mean absolute error was 6.1, resulting in a basic mosquito activity prediction model pre-trained on a public dataset.

[0043] The LightGBM algorithm employs a histogram-based decision tree learning strategy, discretizing continuous feature values ​​into a finite number of bins, significantly reducing computational complexity. It uses a one-sided gradient sampling technique to undersample samples with small gradients, accelerating the training process while maintaining prediction accuracy. It also uses a mutually exclusive feature bundling method to merge mutually exclusive features, reducing feature dimensionality. These optimizations give LightGBM significant performance advantages when processing large-scale data.

[0044] S42, transfer learning technology is used to perform domain-adaptive fine-tuning on the basic prediction model obtained in step S41, enabling the model to adapt to local environmental characteristics and mosquito population characteristics. The core idea of ​​transfer learning is to utilize the general knowledge learned by the pre-trained model on source domain data and quickly adapt to new tasks through a small amount of training on target domain data. Specifically, historical monitoring data collected by locally deployed mosquito trapping devices is extracted from the cloud database. This data includes local environmental parameters and the actual number of mosquitoes captured. The number of mosquitoes captured is standardized into an activity score as the target label to construct a local training dataset. The structure of the basic prediction model was adjusted while keeping the input feature dimension and output target definition unchanged. The tree structure parameters inside the model were used as initial weights, and only the output values ​​of the leaf nodes of the model were relearned. A small learning rate of 0.01 was used for fine-tuning training to prevent overfitting on small sample data. Even if the number of samples in the local dataset is small, such as only a few hundred records, the prior knowledge of the pre-trained model can be effectively utilized through transfer learning. After fifty rounds of iteration, the prediction accuracy of the model on the local validation set was significantly improved, and the root mean square error was reduced to 5.7, resulting in a mosquito activity prediction model suitable for the local environment.

[0045] S43. An incremental learning mechanism is implemented to achieve continuous model optimization. As the system operates for a long time, the equipment continuously collects new monitoring data, which includes the latest environmental change information and mosquito activity patterns. The cloud server accumulates and statistically analyzes the new data. When the number of accumulated new data records reaches one thousand, an incremental training task is automatically triggered. The goal of incremental learning is to enable the model to learn the patterns in the new data while avoiding forgetting the knowledge already learned. The specific implementation method is as follows: the accumulated new data and historical training data are mixed in a certain proportion, with new data accounting for 30% and historical data accounting for 70%. The mixed dataset is used for model retraining. An experience playback technique is used to randomly sample a representative sample from the historical data and add it to the training set to prevent the model from overfitting to the new data and causing performance degradation. The learning rate of incremental training is set to 0.5 times the initial learning rate, i.e., 0.005. Reducing the learning rate can make the model parameter updates smoother and avoid drastic fluctuations. After thirty rounds of incremental training, the model's prediction performance on the comprehensive validation set containing both new and old data has maintained a stable improvement, resulting in a continuously optimized mosquito activity prediction model.

[0046] S44, the current environmental feature vector obtained in step S34 is input into the mosquito activity prediction model continuously optimized in step S43 for inference calculation. Multiple decision trees within the model judge the input features. Each tree traverses downwards layer by layer according to the branch conditions of the feature value at the tree node, finally reaching the leaf node and outputting the predicted value of that tree. The predicted values ​​of all trees are weighted and summed to obtain the final prediction output of the model. This output value is the mosquito activity score under the current environmental conditions, with a score range of zero to one hundred. Simultaneously, a prediction confidence index is calculated. This index reflects the reliability of the model's current prediction results. The calculation method is: calculate the standard deviation of the predicted values ​​of all decision trees. The smaller the standard deviation, the more consistent the prediction results of each tree, and the higher the confidence. The confidence score is mapped to a range of zero to one using an inverse proportional function. When the standard deviation is less than five, the confidence score is higher than 0.9; when the standard deviation is greater than fifteen, the confidence score is lower than 0.5. The reliability of the prediction result is judged based on the confidence score. When the confidence score is lower than 0.7, the system determines that the current environmental conditions exceed the reliable prediction range of the model, possibly due to special environmental conditions not fully covered in the training data. In this case, a conservative control strategy is adopted, limiting the adjustment range of the device's operating parameters to no more than ±10% of the current value to avoid excessive adjustment leading to device malfunction. When the confidence score is higher than or equal to 0.7, the system considers the prediction result reliable and allows for normal parameter adjustments, resulting in a mosquito activity prediction result that includes both activity score and confidence assessment.

[0047] S45, Based on the mosquito activity prediction results obtained in step S44, fuzzy control logic is used to calculate the adaptive adjustment target values ​​of each working parameter of the device. Fuzzy control can simulate the experience-based decision-making process of human experts and has good control effects for complex systems that are difficult to model precisely. First, fuzzy input variables and fuzzy output variables are defined. The fuzzy input variable is the mosquito activity score, which is divided into five fuzzy subsets, named very low, lower, medium, higher, and very high. A membership function is defined for each fuzzy subset. The membership function describes the degree to which the input value belongs to the fuzzy subset. A triangular or trapezoidal membership function is used, specifically defined as: the very low subset corresponds to an activity score of zero to twenty. The membership function linearly increases to 1 between 0 and 10, and linearly decreases to 0 between 10 and 20; for lower subsets with scores of 15 to 40, the membership function linearly increases to 1 between 15 and 25, and linearly decreases to 0 between 25 and 40; for medium subsets with scores of 30 to 70, the membership function linearly increases to 1 between 30 and 50, and linearly decreases to 0 between 50 and 70; for higher subsets with scores of 60 to 85, the membership function linearly increases to 1 between 60 and 70, and linearly decreases to 0 between 70 and 85; for very high subsets with scores of 80 to 100, the membership function linearly increases to 1 between 80 and 90, and remains at 1 between 90 and 100.

[0048] S46, Define a fuzzy control rule base. The rule base contains multiple control rules in the form of IF-THEN. Each rule describes how the equipment's operating parameters should be adjusted under specific activity levels. The rule contents are as follows: Rule 1: If the activity level is very low, the heating pad temperature adjustment coefficient is set to 0.75, the fan speed adjustment coefficient is set to 0.6, and the mosquito attractant release coefficient is set to 0.5; Rule 2: If the activity level is low, the heating pad temperature adjustment coefficient is set to 0.9, the fan speed adjustment coefficient is set to 0.9, and the mosquito attractant release coefficient is set to 0.9; Rule 3: If the activity level is medium, the heating pad temperature adjustment coefficient is set to 1.0, and the fan speed adjustment coefficient is set to 0.9. Rule 4: If the activity level is high, the heating pad temperature adjustment coefficient is set to 1.2, the fan speed adjustment coefficient is set to 1.2, and the mosquito attractant release rate adjustment coefficient is set to 1.2. Rule 5: If the activity level is very high, the heating pad temperature adjustment coefficient is set to 1.4, the fan speed adjustment coefficient is set to 1.5, and the mosquito attractant release rate adjustment coefficient is set to 1.6. The adjustment coefficients represent the multiples of the operating parameters relative to the reference values. The reference values ​​are the standard operating parameters of the equipment, such as a heating pad reference temperature of 40 degrees Celsius, a fan reference speed of 1000 revolutions per minute, and a mosquito attractant reference release rate of 1 milliliter per hour.

[0049] S47. Execute the fuzzy inference process. First, perform fuzzification by substituting the precise activity score obtained in step S44 into the membership function defined in step S45 to calculate the membership degree of the score to each fuzzy subset, resulting in five membership degrees. Then, perform rule evaluation by taking the membership degree of the input variable to the fuzzy subset of the rule's antecedent as the rule's activation strength for each rule in the rule base. All activated rules contribute to the final output, with the contribution determined by the activation strength. Defuzzification is performed using a weighted average method by multiplying the adjustment coefficient of each rule's output by the corresponding activation strength. The weighted output of all rules is summed and divided by the sum of activation strengths to obtain the final working parameter adjustment coefficients. Multiply the adjustment coefficients by the baseline values ​​of each working parameter to calculate the specific values ​​of the heating pad target temperature, fan target speed, and mosquito attractant target release amount, thus obtaining an adaptive working parameter configuration scheme based on fuzzy control logic.

[0050] S48, the cloud server encapsulates the adaptive working parameter configuration scheme calculated in step S47 into a device control instruction. The control instruction includes an instruction type field, a target device identifier field, a parameter setting field, an instruction priority field, and a validity period field. The instruction type identifier is a parameter adjustment command; the target device identifier is a unique code for the device receiving the instruction; the parameter setting field includes specific values ​​for the heating pad target temperature, the fan target speed, and the mosquito attractant target release amount; the instruction priority is determined based on the prediction confidence level and the magnitude of activity change. When the confidence level is high and the activity change is large, it is set to high priority, and the device should execute it immediately; when the confidence level is moderate... If the activity level changes slightly, it is set to medium priority, and the device can execute it in the next work cycle. The validity period field is set to two hours after the instruction is issued. Instructions that have exceeded the validity period will be discarded by the device and will not be executed to prevent outdated instructions from being executed incorrectly due to network latency. Control instructions are managed through the message queue system of the cloud server. The message queue adopts the first-in, first-out principle to ensure that instructions are issued in chronological order. Priority queueing is also supported, and higher priority instructions can be sent first. The cloud server pushes the control instructions to the target device through the NB-IoT or 4G network to obtain device control instruction messages containing adaptive working parameters.

[0051] S49, the device receives the control command message issued in step S48 through the communication module. The device microcontroller parses and verifies the command. First, it checks whether the target device identifier matches the device; if not, the command is discarded. Then, it checks the command validity period, comparing the issuance time in the command with the current device time; if the validity period has expired, the command is discarded. After verification, it extracts the working parameter settings from the command, compares them with the actual working parameters of the current device, and calculates the parameter adjustment amount. The device microcontroller generates the underlying hardware control signal based on the parameter adjustment amount. For heating pad temperature control, a PID feedback control algorithm is used to calculate the heating power adjustment amount based on the deviation between the target temperature and the actual temperature, and controls the power-on time of the heating pad through a PWM pulse width modulation signal. The microcontroller uses a proportional control system to achieve precise temperature regulation. For fan speed control, it outputs a PWM signal to the motor driver chip, adjusting the drive voltage to regulate the speed. Simultaneously, it uses a Hall sensor to provide feedback on the actual speed for closed-loop control. For mosquito attractant release control, the microcontroller controls a stepper motor to drive the release valve, calculates the valve opening degree and opening time based on the target release amount, and drives the stepper motor to rotate to the target position via a pulse signal. After the device executes the parameter adjustment, the execution result is packaged into a response message and uploaded to the cloud server. The response message includes the execution status, the adjusted actual parameter values, and the execution timestamp. After receiving the response, the cloud server updates the device status record, forming a closed-loop adaptive control process, resulting in a dynamically optimized device operating status based on environmental conditions.

[0052] Health assessment module: Based on dynamically optimized equipment operating status, the module performs local prediction on the equipment and optimizes the data upload strategy to obtain equipment health assessment results and predictive maintenance warnings; S51, Input the device's working status and actual operating parameters obtained in step S49. The edge computing module integrated on the device starts a local data analysis task. This edge computing module uses an embedded processor, has certain computing power, and can run lightweight data analysis algorithms. The edge computing module is pre-installed with a simplified anomaly detection rule engine. This engine contains multiple detection rules based on threshold judgment. The rules include: Rule 1, if the device voltage shows a downward trend for five consecutive sampling cycles and the rate of decline exceeds 0.1 volts per minute, it is judged as an abnormality of rapid power consumption; Rule 2, if the deviation between the actual temperature of the heating pad and the target temperature exceeds 5 degrees Celsius for ten consecutive minutes, it is judged as an abnormality of the temperature control system; Rule 3, if the communication signal strength is measured below -100 decibels for three consecutive times, it is judged as a decline in communication quality; Rule 4, if the number of mosquitoes captured is zero for six consecutive hours and the device is in the on state, it is judged as an abnormality of the trapping function; The edge computing module monitors various operating parameters of the device in real time, substitutes the parameter values ​​into the detection rules for judgment, and when any rule is triggered, generates a local warning sign to obtain the local prediction result of the device's abnormal symptoms.

[0053] S52, based on the local prediction results obtained in step S51, the edge computing module dynamically adjusts the data upload strategy to adapt to different device states. When no abnormal signs are detected, the device maintains the normal data upload frequency, that is, uploads a monitoring data packet once every ten minutes. When abnormal signs are detected, the device immediately increases the data upload frequency to once every two minutes. Increasing the data upload density helps the cloud server to capture the abnormal development process in a timely manner and perform more accurate fault diagnosis. At the same time, the edge computing module adds a warning identifier field to the data packet. This field contains the triggered rule number and the name of the abnormal parameter, enabling the cloud server to quickly identify the warning data and process it first, resulting in a monitoring data packet with a warning identifier and optimized upload frequency.

[0054] S53, the device uploads the monitoring data packet obtained in step S52 via the NB-IoT or 4G communication module. To address the issue of unstable communication signals in outdoor environments, an adaptive data compression and breakpoint resumption mechanism is implemented to improve transmission reliability. The device's microcontroller monitors the communication signal strength index in real time. This index is obtained through the signal quality query interface of the communication module and is represented by the RSSI (Received Signal Strength Index). When the signal strength is higher than -70 decibels, the signal is considered good, and the data packet is transmitted in standard JSON format. When the signal strength is lower than -70 decibels but higher than -90 decibels, the signal is considered weak, and data compression is enabled. The DEFLATE compression algorithm is used to compress JSON format data packets. This algorithm combines LZ77 dictionary encoding and Huffman coding, and can compress text data to 40% to 60% of its original size. The compressed data packets are transmitted over the network, and the cloud server receives them and decompresses them. When the signal strength is below -90 decibels, it is considered a poor signal. In addition to enabling data compression, the data packets are split into multiple small segments for transmission. Each segment is no larger than 500 bytes. The segment header contains a sequence number and the total number of segments. After receiving all the segments, the cloud server reassembles them into a complete data packet according to the sequence number.

[0055] S54 implements a breakpoint resume mechanism to handle data transmission interruptions. After sending each data packet or segment, the device waits for an acknowledgment message from the cloud server. Upon successful data reception, the cloud server returns an acknowledgment message containing an identifier of the successfully received data packet. Upon receiving the acknowledgment message, the device removes the corresponding data packet from its local transmission queue. If the device does not receive an acknowledgment message within a preset timeout period, it determines the transmission has failed, marks the data packet as pending retransmission, and retains it in its local cache. The device microcontroller maintains a circular buffer to store data packets to be uploaded and retransmitted, with a capacity of one hundred data packets. When communication signals return to a usable state, the device prioritizes retransmitting the pending retransmission data packets in the buffer, sending them sequentially according to timestamp order to ensure the integrity of historical data. Through the breakpoint resume mechanism, data is not lost even in the event of communication interruption or temporary power failure. The system can retransmit missing data after normal operation resumes, ensuring a reliable data stream.

[0056] S55: After receiving the device data stream uploaded in step S54, the cloud server performs real-time monitoring and quality assessment of the data stream. The cloud maintains the data reception time series for each device, recording the timestamp of each successful data reception. By analyzing the interval characteristics of the time series, the communication stability of the devices is evaluated. A small standard deviation of the data reception interval indicates stable communication; a large standard deviation or prolonged periods without data uploads indicate communication anomalies. The cloud server prioritizes processing data packets with warning indicators, extracting these packets for in-depth analysis. Using a time series-based trend prediction method, historical data of device operating parameters from the past 24 hours are extracted to construct a time series curve. The time series is then numerically analyzed. Differential operations are performed to calculate the first derivative, which represents the rate of change of the parameter, and the second derivative, which represents the acceleration of the rate of change. The signs and values ​​of the first and second derivatives are analyzed. A continuously negative first derivative indicates a continuous decline in the parameter, while an increase in the absolute value of the second derivative indicates an accelerated rate of decline. The combination of these two factors indicates an accelerating trend of equipment performance deterioration. For example, if the first derivative of the equipment voltage is negative for twelve consecutive hours and the absolute value of the second derivative is greater than 0.01, it is determined that the power consumption is accelerating, and the equipment is expected to shut down due to power depletion within the next twenty-four hours. The cloud server generates predictive maintenance warning messages, which include the warning type, the expected time of failure, and suggested maintenance measures, resulting in a trend-based equipment health assessment.

[0057] S56, Based on the equipment health assessment results obtained in step S55, the cloud server constructs a quantitative equipment health scoring system. This system comprehensively considers multiple dimensions of health indicators, including voltage stability, temperature control accuracy, communication quality, and acquisition efficiency. The voltage stability dimension assesses the health status of the equipment's power supply system. The calculation method is as follows: statistically analyze the equipment voltage data for the past seven days, calculate the voltage mean and standard deviation. When the voltage mean is higher than 90% of the rated voltage and the standard deviation is less than 0.2 volts, this dimension scores 100 points. The score decreases linearly when the voltage mean or standard deviation deviates from the normal range. The temperature control accuracy dimension assesses the performance of the heating pad's temperature control system by calculating the average absolute value of the deviation between the actual temperature and the target temperature. When the deviation is less than one degree Celsius, this dimension scores 100 points. The score decreases by 10 points for every 1 degree Celsius increase in temperature; the communication quality dimension assesses the reliability of data transmission by calculating the data upload success rate over the past seven days. A success rate above 95% earns the score of 100 points, and a decrease of 5 points for every 1% decrease in the success rate; the capture efficiency dimension assesses the device's trapping performance by calculating the average daily number of mosquitoes captured over the past seven days and comparing it with the historical average. A capture rate above 80% of the historical average earns the score of 100 points, and a decrease of 0.8% results in a linear decrease in the score; the scores of the four dimensions are weighted and summed according to preset weights, namely voltage stability 0.3, temperature control accuracy 0.25, communication quality 0.2, and capture efficiency 0.25. The weighted sum is the device's overall health score, ranging from 0 to 100 points.

[0058] S57. Based on the equipment health score calculated in step S56, the cloud server sets tiered early warning thresholds. When the health score is above 80, the equipment is considered to be in good condition and requires no special handling. When the health score is between 60 and 80, the equipment is considered to be in average condition, triggering a maintenance suggestion-level early warning. The warning message includes maintenance suggestions such as regularly checking the equipment, cleaning the collection container, and replenishing mosquito attractants. When the health score is below 60, the equipment is considered to be in poor condition, triggering an emergency warning. The warning message includes specific abnormal dimensions, abnormal parameter values, expected failure time, and detailed maintenance guidance. The warning message is pushed to management personnel through various communication channels, including SMS, WeChat official account push, email, and message notifications on the management platform. After receiving the warning, management personnel can view detailed operating data and historical trend curves of the equipment through the management platform to assist in fault diagnosis. For emergency warnings, the system will also automatically generate maintenance work orders, assign them to the corresponding maintenance personnel, and track the processing progress of the work orders to ensure that the faults are handled in a timely manner, thus achieving a comprehensive predictive maintenance early warning mechanism.

[0059] The regional collaborative optimization module constructs a mosquito density distribution model based on the equipment health assessment results and all equipment data streams using the weighted kriging spatial interpolation method, identifies high-incidence areas and obtains regional collaborative prevention and control results. S61, Input the data streams and health assessment results of all devices obtained in step S57. The cloud server performs a quality assessment of the device data to ensure the accuracy of subsequent spatial analysis. Data quality is affected by both device health status and data integrity. For the assessment of device health status, the device health score calculated in step S56 is directly used as the health status indicator. For the assessment of data integrity, the number of times each device successfully uploaded data in the last 24 hours is counted. The theoretical upload count is once every 10 minutes, totaling 144 times. The ratio of the actual upload count to the theoretical upload count is defined as... Data completeness rate; The data quality weight is calculated by combining the equipment health score and the data completeness rate. Specifically, the health score is divided by 100 to obtain a normalized health coefficient between 0 and 1. The data completeness rate is directly used as the completeness coefficient. The two are then geometrically averaged, i.e., the square root of the product of the health coefficient and the completeness coefficient, to obtain the quality weight value for the equipment data. The quality weight value ranges from 0 to 1; the closer the weight is to 1, the higher the data quality. Corresponding quality weight labels are added to the monitoring data records of all equipment to construct a quality-weighted equipment monitoring dataset, resulting in a quality-weighted mosquito capture data set.

[0060] S62, the spatial distribution of mosquito density in the quality-weighted dataset obtained in step S61 is modeled using weighted kriging spatial interpolation. Kriging interpolation is an optimal linear unbiased estimation method based on spatial statistics, which can infer the value of unknown points using data from known points. Traditional kriging interpolation assumes that the data quality of all observation points is the same, while weighted kriging interpolation introduces quality weights on this basis, making high-quality data contribute more to the interpolation results. First, a spatial coordinate system is constructed, and the geographical coordinates of all devices are extracted. The latitude and longitude coordinates are converted into a plane rectangular coordinate system using Mercator projection. Then, the spatial covariance function is calculated, which describes the spatial correlation of mosquito density with distance. The semivariogram analysis method is used to calculate the semivariogram of the difference between the distance and the mosquito capture amount between any two devices. All devices are grouped according to distance. The mean semivariance of each distance group is calculated, and a theoretical semivariance model is fitted. Commonly used models include the spherical model, exponential model, and Gaussian model. The model parameters are determined by fitting using the least squares method. When constructing the Kriging equations, mass weights are introduced to weight the observed data. Data with high mass weights have larger coefficients in the equations and have a greater impact on the interpolation results. The weight coefficients of each observation point are obtained by solving the Kriging equations. These weight coefficients are used to sum the mosquito capture amount at the observation points to obtain the estimated mosquito density at any spatial location. The entire area is divided into regular grids with a grid resolution of 10 meters by 10 meters. Kriging interpolation is performed on the center point of each grid to obtain the estimated mosquito density at that point. The estimated values ​​of all grids are summarized to form a raster data of the spatial distribution of mosquito density covering the entire area, resulting in a high-precision regional mosquito density spatial distribution model.

[0061] The weighted kriging interpolation method can also introduce environmental covariates to improve interpolation accuracy when calculating spatial covariance. Environmental covariates include environmental factors that affect mosquito breeding and activity, such as vegetation cover, water distribution, and building density. These covariate data can be obtained from remote sensing images and geographic information databases. After rasterizing the covariate data, spatial registration is performed with mosquito density data, which is then used as an auxiliary variable in the kriging model. By using the co-kriging method to interpolate using both mosquito capture data and environmental covariate data, the spatial distribution characteristics of mosquito density can be described more accurately.

[0062] S63, Based on the mosquito density spatial distribution model obtained in step S62, an improved DBSCAN density clustering algorithm is used to identify high-incidence mosquito areas. The DBSCAN algorithm can discover clusters of arbitrary shapes and automatically identify noise points, making it suitable for processing mosquito density data with irregular spatial distribution. The traditional DBSCAN algorithm uses a fixed neighborhood radius parameter. When the distribution of devices is uneven, the fixed radius may lead to over-clustering in densely populated areas and failure to cluster in sparse areas. The improved DBSCAN algorithm introduces an adaptive neighborhood radius mechanism, dynamically adjusting the clustering radius according to the local device density. The specific implementation method is as follows: First, calculate the number of devices within a certain range around each grid point. This range is set as a circular area with a radius of 200 meters. The number of devices within this area is counted as a local device density index. When the local device density is higher than ten devices per square kilometer, it is determined to be a densely populated area, and the clustering radius is set to 50 meters. When the local device density is lower than five devices per square kilometer, it is determined to be a sparsely populated area, and the clustering radius is expanded to 150 meters. When the local device density is between the two, the clustering radius is reduced to 50 meters. The path is determined by linear interpolation; a density threshold parameter is set, and when the estimated mosquito density of a grid point is higher than the preset density threshold, the grid point is marked as a high-density point. The density threshold is set to 1.5 times the median mosquito density of the region; starting from any unvisited high-density point, all high-density points within its neighborhood radius are searched, and these points are grouped into the same cluster. Then, the neighborhood of each point in the cluster is searched repeatedly, continuously expanding the cluster range until no new high-density points can be added; when the number of grid points in a cluster exceeds the preset minimum cluster size, for example, at least sixteen consecutive grid points, i.e., the coverage area is at least 1,600 square meters, the cluster is identified as a high-incidence area of ​​mosquitoes; the above clustering process is performed on all high-density points to obtain multiple high-incidence areas of mosquitoes, each containing a set of consecutive high-density grid points; the geometric center point coordinates and boundary contours of each high-incidence area are calculated. The geometric center point coordinates are the arithmetic mean of the coordinates of all grid points in the area, and the boundary contours are extracted using the convex hull algorithm to extract the minimum convex polygon outside the area, resulting in an adaptively identified list of high-incidence areas of mosquitoes and their spatial location information.

[0063] S64, for each mosquito-prone area identified in step S63, the cloud server performs collaborative optimization configuration of devices within the area. First, it filters all devices located within or near the high-incidence area. The method for determining whether a device is in a high-incidence area is as follows: calculate the shortest distance from the device's location coordinates to the boundary outline of the area. When the distance is less than 100 meters, the device is included in the scope of regional collaborative optimization. The devices included in the scope are sorted according to their distance to the geometric center point of the area. The devices closest to the center point are defined as core control devices. The number of core control devices is determined according to the area area: three devices are selected when the area area is less than 5,000 square meters, five devices are selected when the area area is between 5,000 and 10,000 square meters, and seven devices are selected when the area area is greater than 10,000 square meters. Devices at the edge of the area are defined as auxiliary control devices. The workload of auxiliary devices is lower than that of core devices to save energy.

[0064] S65, the optimal operating parameter configuration of each device in step S64 is calculated using the particle swarm optimization algorithm. Particle swarm optimization is a global optimization algorithm based on swarm intelligence, which searches for the optimal solution by simulating the foraging behavior of bird flocks. The objective function and constraints of the optimization problem are defined. The optimization objective is to maximize the total mosquito capture in the area while minimizing the total energy consumption. A dual-objective optimization function is constructed. The first objective function is the sum of the expected mosquito captures of all devices in the area, which is estimated based on the device operating parameters and environmental conditions using the mosquito activity prediction model from step S4. The second objective function is the total power consumption of all devices in the area, which is determined based on the heating pad temperature, fan speed, and mosquito attractant release. The calculation of the set values ​​is as follows: the power consumption of the heating pad is proportional to the square of the temperature set value, the power consumption of the fan is proportional to the cube of the rotation speed, and the power consumption of the mosquito attractant release is proportional to the release amount. The two objective functions are normalized and weighted and summed. The weight coefficients can be adjusted according to actual needs. When the capture effect is emphasized, the weight of the first objective is set to 0.7 and the weight of the second objective is set to 0.3. When energy saving is emphasized, the weights are set in reverse to obtain the comprehensive objective function. The constraint condition is that the adjustment range of the working parameters of each device does not exceed plus or minus 50% of the standard value, that is, the heating pad temperature is between 20 and 60 degrees Celsius, the fan speed is between 500 and 1500 revolutions per minute, and the mosquito attractant release amount is between 0.5 and 1.5 milliliters per hour.

[0065] S66. Initialize the particle swarm. The swarm contains thirty particles, each representing a set of device operating parameter configurations. The particle dimension is equal to the number of devices included in the optimization scope multiplied by three. The three parameters are heating pad temperature, fan speed, and mosquito attractant release amount, respectively. Randomly initialize the position and velocity of each particle. The position is randomly generated within the parameter range defined by the constraints, and the velocity is initialized to 10% of the parameter range. Calculate the comprehensive objective function value corresponding to each particle, and record the individual optimal position and global optimal position of each particle. The individual optimal position is the parameter configuration corresponding to the best objective function value reached by the particle in history, and the global optimal position is the parameter configuration corresponding to the best objective function value reached by all particles in history. Iteratively update the particle velocity and position. The velocity update formula contains three parts. The first part is the inertia term, which maintains the original direction of particle motion, with a coefficient of 0.7. The second part is the individual cognition term, which guides the particle to move towards its own historical best position, with a coefficient of 1.5. The third part is the social learning term, which guides the particle to move towards the global best position, with a coefficient of 1.5. The position update formula is the current position plus the updated velocity. Boundary processing is performed on the updated position, and parameter values ​​that exceed the constraint range are clipped to the boundary values. The objective function value of each particle is recalculated, and the individual best and global best are updated. The iteration process is repeated for one hundred rounds. The iteration is terminated early when the global best objective function value no longer improves for ten consecutive rounds. After the iteration is completed, the parameter configuration corresponding to the global best position is the optimal working parameter configuration scheme for regional cooperation, and the Pareto optimal equipment working parameter configuration is obtained.

[0066] Pareto optimality refers to the solution in multi-objective optimization where it is impossible to continue improving any objective without compromising other objectives. In mosquito control scenarios, Pareto optimal configuration schemes can achieve the best balance between capture effect and energy consumption. Administrators can choose configuration schemes with different emphases on the Pareto front according to actual needs.

[0067] S67, the cloud server distributes the optimal operating parameter configuration scheme calculated in step S66 to relevant devices in batches. To avoid system instability that may be caused by large-scale simultaneous adjustments, a phased and gradual adjustment strategy is adopted. First, the parameters of the core regional control equipment are adjusted. The operating parameter adjustment instructions of the core equipment are encapsulated into control messages and sent through the communication network. After receiving the instructions, the core equipment executes the parameter adjustment. A ten-minute observation period is waited, during which the cloud server continuously monitors the operating status of the core equipment and changes in mosquito capture data. The number of mosquitoes captured before and after the adjustment is compared, and the rate of change in capture volume is calculated. When the increase in capture volume reaches more than 10%, it is determined that... The parameter adjustment was effective, so the second phase of adjustment continued. The second phase of adjustment targeted the regional auxiliary control equipment, issuing corresponding work parameter adjustment instructions. The parameter adjustment range for auxiliary equipment was relatively small, and the workload was set to 80% of that of the core equipment. After a ten-minute observation period, the overall adjustment effect was evaluated. Every twelve hours, the complete process from steps S61 to S67 was re-executed, and the collaborative configuration scheme was recalculated based on the latest mosquito density distribution and environmental conditions to achieve dynamic optimization. Through continuous cyclical optimization, the system can adaptively track the spatiotemporal changes of mosquito activity, always maintain the optimal control configuration, and achieve a robust and efficient regional collaborative control effect.

[0068] The visualization management module uses the Leaflet map library to build a real-time map display of the device status and visualizes the spatial distribution of mosquito density, based on the results of regional collaborative prevention and control and all device data. S71, Input all device data and regional optimization results obtained in step S67. A web-based visual management platform is built on the cloud server. This platform adopts a front-end / back-end separation architecture. The front-end uses HTML, CSS, and JavaScript technologies to implement the user interface, while the back-end provides a RESTful API interface for the front-end to call and obtain data. The front-end integrates the Leaflet open-source map library for the visualization of geospatial data. Leaflet is a lightweight interactive map JavaScript library that supports various online map services and custom layer overlays. The map display module of the management platform calls Leaflet. The API initializes the map object, setting the initial center point coordinates of the map to the geographic center of the managed area, and the initial zoom level to a suitable level that can fully display the entire managed area. It loads an online tile map as the base map, providing basic geographic information such as streets, buildings, and terrain. It calls the backend API interface to obtain the real-time location coordinates and operating status data of all devices, adding a marker icon to each device on the map, with the icon's position representing the device's latitude and longitude coordinates. Different colored icons are used to distinguish devices based on their operating status: a green circle for normal operation, a yellow triangle for warning status, a red cross for fault status, and a gray circle for offline status. When a user clicks on a device icon on the map, an information window pops up displaying detailed information about the device, including device number, current operating parameters, number of mosquitoes captured, health score, and last reported time. The map supports interactive operations; users can drag and pan the map with the mouse, zoom in and out with the scroll wheel, and click on icons to view details, achieving intuitive and convenient device status monitoring and providing a real-time map visualization interface of device distribution and status.

[0069] S72, based on the mosquito density spatial distribution model obtained in step S62, a mosquito density heatmap layer is overlaid on the map display of the management platform. A heatmap is a visualization method that uses color gradients to represent numerical values, intuitively displaying the distribution characteristics of spatial data. The density estimate and coordinates of each grid point are extracted from the mosquito density spatial distribution raster data, and these data points are passed to the Leaflet heatmap plugin. This plugin, based on the Heatmap.js library, supports efficient rendering of large amounts of data points. The rendering parameters of the heatmap are configured, including the color mapping scheme, transparency, and blur radius. The color mapping scheme uses a gradient from blue to green to yellow to red, with blue representing... The map displays low-density areas in red, high-density areas in red, and medium-density areas in neutral colors. The transparency is set to 60%, making the heatmap semi-transparent while still allowing the geographic information of the base map to be visible. The blur radius is set to 20 pixels to control the smoothness of the heatmap. The heatmap layer is overlaid on the base map and below the device icons, allowing users to simultaneously see the correspondence between mosquito density distribution and device location. The heatmap refreshes automatically every ten minutes, recalculating the density distribution based on the latest device data and updating the display for dynamic density monitoring. Users can switch the heatmap's display on and off via the layer control panel, easily adjusting the map display content as needed to obtain a visual representation of the spatial distribution of mosquito density.

[0070] S73 features a device capture performance comparison and analysis function. It queries the cumulative mosquito capture count of all devices from a cloud database, sorts them by capture count from highest to lowest, and extracts information for the top 20 devices, including device ID, location, and capture count. It calculates the capture efficiency index for each device, defined as the number of mosquitoes captured per unit time, calculated by dividing the cumulative capture count by the device's cumulative working time. This index eliminates the influence of differences in working time between different devices, more accurately reflecting device performance. A front-end charting library such as ECharts or Chart.js is used to create a capture count ranking bar chart. The horizontal axis represents the device ID, and the vertical axis represents the capture count. Each device corresponds to a bar, and the bar height represents the capture count. The chart displays the number of captures; it assigns different shades of color to bars based on capture efficiency, with darker bars for high-efficiency devices and lighter bars for low-efficiency devices, visually distinguishing device performance through color gradients; when the user hovers the mouse over a bar, a tooltip displays detailed data for that device, including the number of captures, capture efficiency, working time, and device location; the chart supports dynamic filtering, allowing users to select a statistical time range, such as the last seven days, the last thirty days, or the last three months, and the system will re-query the data and update the chart based on the selected time range; the chart also supports filtering by region, and when the managed area is divided into multiple sub-regions, users can select to view the leaderboard for a specific sub-region, obtaining comparative analysis charts and leaderboard displays of device capture performance.

[0071] The S74 module features a mosquito activity trend analysis function. It queries historical mosquito capture data from a cloud database and aggregates it statistically by time dimension, with selectable time granularity (hourly, daily, or weekly). For daily aggregation, it calculates the total capture count from all devices each day, resulting in a daily capture time series. A time series line chart displays the capture trend over time, with the horizontal axis representing dates and the vertical axis representing daily capture count. The line connects the capture data points for each date, visually indicating the increase or decrease in mosquito activity. A moving average is overlaid on the line chart, with a seven-day moving average window to calculate the arithmetic mean of the capture counts for each date over the preceding seven days. The moving average provides... This method smooths short-term fluctuations and highlights long-term trends. It performs seasonal decomposition on time-series data, separating trend, seasonal, and random components. The trend component reflects the long-term direction of mosquito activity, while the seasonal component reflects the periodic fluctuations. Seasonal decomposition allows for a deeper understanding of mosquito activity characteristics. The ARIMA time-series forecasting model is used to predict mosquito capture volume for the next seven days. ARIMA model parameters are determined automatically, and extrapolation is performed after fitting the model to historical data. The prediction results are displayed as dashed lines on a line graph, with confidence intervals marked, providing managers with forward-looking decision-making references. This results in a mosquito activity trend analysis and forecast display.

[0072] The S75 features anomaly event statistical analysis function. It queries historical anomaly warning records from the cloud database, extracting information such as anomaly type, occurrence time, and involved devices. Anomalies are categorized and statistically analyzed by type, including insufficient power, temperature control anomalies, communication anomalies, and capture anomalies. The frequency and percentage of each type of anomaly event are calculated, and a pie chart is used to display the distribution of anomaly types. Each sector of the pie chart represents an anomaly type, and the sector area is proportional to the frequency of that type of event. The sector is labeled with the type name and percentage. The pie chart supports click interaction; when a user clicks on a sector, it expands to display a detailed list of anomalies of that type, including the occurrence time, device number, device location, anomaly parameter value, and processing status of each event. Anomalies are analyzed for trends over time, counting the total number of anomalies each week and displaying the weekly anomaly count using a bar chart. This trend analysis allows for assessment of system stability improvements. Spatial distribution analysis of anomalies is also performed, counting the number of anomalies in each region or per device, identifying high-incidence areas or devices, and highlighting them on a map with hotspot markers to alert management personnel. This provides a multi-dimensional statistical analysis and visualization of anomalies.

[0073] S76 features an automatic report generation function, supporting daily, weekly, and monthly reports. Administrators select the report type and statistical time range through the platform interface, and the system automatically extracts data for the corresponding time period from the database for statistical summary. Daily reports include: the device online rate for the day (calculated as the number of online devices divided by the total number of devices multiplied by 100%); the total number of mosquitoes captured for the day (summarizing the capture count of all devices); the total number of abnormal events for the day (counting the number of abnormal alerts); the device health score distribution (counting the number of devices in different score ranges); a list of the top ten devices in terms of capture volume; and a detailed list of abnormal event records. Weekly and monthly reports have similar statistical dimensions, but the time range is expanded to one week or one month, and trend comparison analysis is added. Examples of data include comparisons of capture volume this week versus last week, and comparisons of the number of anomalies this month versus last month. Reports are organized using a combination of tables and charts, with tables displaying specific values ​​and charts showing trends and distribution characteristics. The system uses an HTML template engine to generate report pages, filling statistical data into predefined templates to generate reports with standardized formats and complete content. Reports can be exported to PDF or Excel formats for easy printing and archiving by administrators or for reporting to superiors. The system also supports scheduled automatic report generation; administrators can set fixed times for daily, weekly, or monthly report generation and send them to designated recipients via email, automating the reporting process and providing various types of automated statistical reports and decision support information.

[0074] like Figure 2 As shown, the intelligent mosquito control system includes an intelligent sensing module on the device side, a communication transmission module, a cloud server, and a visual management platform. The intelligent sensing module on the device side is responsible for collecting data on device operating status, environmental parameters, and mosquito capture data; the communication transmission module enables bidirectional data transmission between the device and the cloud; the cloud server is responsible for data storage, intelligent analysis, predictive modeling, and decision calculation; and the visual management platform provides managers with an intuitive data display and operation interface.

[0075] This invention focuses on mosquito control in large residential areas within the field of urban property management. A property management company is responsible for managing a large residential community covering 500,000 square meters. The community includes 20 high-rise residential buildings, five artificial lakes, ten parks and green spaces, and three underground parking garages. The environment is complex and diverse, and mosquito problems seriously affect the quality of life of residents during the summer and autumn seasons.

[0076] The property management company has deployed two hundred mosquito trapping devices in the community. The devices are distributed in different environmental locations such as the lakeside, green space, and around the buildings. All devices have been upgraded to the intelligent system described in this invention, and are equipped with a Beidou GPS dual-mode timing module, a weighing sensor, a miniature camera, a temperature and humidity sensor, a light intensity sensor, an infrared sensing module, and an NB-IoT communication module.

[0077] After deployment, the system began to run normally. The following shows some examples of real-world operational data: Table 1 shows an example of real-time monitoring data from the equipment: Table 1. Real-time monitoring data of the equipment Table 2 shows an example of regional mosquito density distribution and collaborative optimization configuration: Table 2. Regional mosquito density distribution and examples of collaborative optimization configuration Through the application of the intelligent mosquito control system described in this invention, the property management company has achieved intelligent and precise management of mosquito control work. The equipment operation status can be monitored in real time, the mosquito capture data can be accurately counted, the working mode can be adaptively adjusted according to the environment, the equipment failure can be warned in advance, and the regional collaborative optimization configuration can be improved, thus improving management efficiency and control effect and creating a more comfortable living environment for residents.

[0078] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. An intelligent mosquito control system, characterized in that, include: The timing and control module performs weighted fusion processing on dual-channel time signals acquired from Beidou and GPS satellites to obtain a time-driven sequence of device control commands and the device's working status after execution. The dual counting fusion module acquires the working status of the equipment, collects container weight change data based on the weighing sensor and calculates the preliminary count value, and uses an image recognition algorithm to visually count mosquitoes to obtain the number of mosquitoes captured. The multi-dimensional environmental sensing module obtains the number of mosquitoes captured, collects multi-dimensional environmental parameters based on temperature and humidity sensors, light intensity sensors and infrared sensing modules, performs noise reduction processing on the raw data, and obtains a complete monitoring data package containing environmental feature vectors and equipment operating status. The prediction and control module, based on the complete monitoring data package, uses transfer learning and incremental learning to build a mosquito activity prediction model, calculates adaptive operating parameters, obtains equipment control commands with adaptive operating parameters, and dynamically optimizes the equipment operating status after executing the commands. The health assessment module performs local predictions on the device based on dynamically optimized device operating status and optimizes data upload strategies to obtain device health assessment results and predictive maintenance warnings. The regional collaborative optimization module constructs a mosquito density distribution model based on the equipment health assessment results and all equipment data streams using the weighted kriging spatial interpolation method, identifies high-incidence areas and obtains regional collaborative prevention and control results. The visualization management module uses the Leaflet map library to build a real-time map display of the device status and visualize the spatial distribution of mosquito density, based on the results of regional collaborative prevention and control and all device data.

2. The intelligent mosquito control system according to claim 1, characterized in that, The timing and control module includes: It receives navigation signals from BeiDou and GPS satellites, extracts time information, and outputs BeiDou timestamps, GPS timestamps, and corresponding signal quality factors. The validity of BeiDou time data and GPS time data is verified, signal quality factors are extracted and normalized, Kalman filtering algorithm is used to fuse the two time signals, and the observation noise covariance matrix is ​​dynamically adjusted according to the signal quality score to obtain the standard time for accurate synchronization. The precise synchronization standard time is compared with the device's internal real-time clock. When the time deviation exceeds the preset deviation threshold, a clock synchronization calibration operation is triggered. The work schedule is read based on the device's local time, and a sequence of device control commands is generated according to the work mode identifier.

3. The intelligent mosquito control system according to claim 1, characterized in that, The dual-counting fusion module includes: A strain gauge force sensor is used to sense the weight change of the collection container in real time. The resistance change is converted into a voltage signal through a Wheatstone bridge circuit. The sampled data is recorded in time series and the average weight value is calculated. The weight increment information is extracted by differential operation on the time series weight data sequence. The count conversion is performed using the preset average weight parameter of a single mosquito. The miniature camera module automatically triggers image acquisition according to the preset shooting cycle, uploads digital images to the cloud server, and uses an improved YOLOv8 architecture mosquito target detection model for intelligent recognition processing, counting and statistically analyzing detection boxes with confidence scores higher than the preset confidence threshold. The relative error between the weighing count and the image recognition count is calculated. When the relative error is less than a preset error threshold, a simple averaging fusion strategy is adopted. When the relative error is greater than or equal to the preset error threshold, a weighted fusion strategy is adopted.

4. The intelligent mosquito control system according to claim 1, characterized in that, The multi-dimensional environment perception module includes: The temperature and humidity sensor communicates with the device microcontroller via the I2C bus interface and outputs ambient temperature and relative humidity values. The light intensity sensor outputs digital light intensity values. The infrared sensing module detects the movement of heat sources within the field of view and outputs a trigger signal. The sliding window averaging filter method is used to smooth the temperature, humidity and light intensity data, and the arithmetic mean of the continuous sampling point data within the preset time window is calculated. The statistical indicators of environmental parameters are calculated using a preset statistical window to form an environmental feature vector that includes the average temperature, average humidity, average light intensity, temperature change rate, humidity change rate, light intensity change rate, and number of infrared triggers. Collect equipment operating status parameters, package and encapsulate them with environmental feature vectors, and serialize them using JSON format.

5. The intelligent mosquito control system according to claim 1, characterized in that, The prediction and control module includes: Based on publicly available pest activity research datasets, a basic prediction model was constructed using the LightGBM gradient boosting tree algorithm. The model hyperparameters were set and iterative training was performed. Historical monitoring data collected by locally deployed devices is extracted from cloud databases. The number of mosquitoes captured is standardized into an activity score as the target label, and the basic prediction model is then fine-tuned in a domain-adaptive manner. When the number of accumulated new data records reaches a preset threshold, the model incremental training task is automatically triggered, and the new data is mixed with the historical training data in a preset ratio for model retraining. The environmental feature vector is input into the continuously optimized mosquito activity prediction model for inference calculation. Fuzzy control logic is used to calculate the adaptive adjustment target value of the device's operating parameters, which is then encapsulated into device control commands and sent to the target device for execution.

6. The intelligent mosquito control system according to claim 1, characterized in that, The health assessment module includes: The edge computing module integrated on the device comes pre-installed with a simplified anomaly detection rule engine, which monitors the device's operating parameters in real time and substitutes the parameter values ​​into the detection rules for judgment. When any rule is triggered, a local warning sign is generated. The data upload strategy is dynamically adjusted based on local prediction results. When no abnormal signs are detected, the normal data upload frequency is maintained. When abnormal signs are detected, the data upload frequency is increased and a warning flag field is added to the data packet. Real-time monitoring of communication signal strength indicators; different data compression and transmission strategies are adopted according to the signal strength; breakpoint resume mechanism is implemented to cope with data transmission interruption. The cloud server performs real-time monitoring and quality assessment of the data stream, uses time-series-based trend prediction methods for in-depth analysis, constructs a quantitative equipment health scoring system, sets graded early warning thresholds, and pushes early warning messages.

7. The intelligent mosquito control system according to claim 1, characterized in that, The regional collaborative optimization module includes: The data quality weight is calculated by combining the equipment health score and the data integrity rate, and a corresponding quality weight label is attached to the monitoring data records of all equipment. The spatial distribution model of mosquito density is modeled using the weighted kriging spatial interpolation method. A spatial coordinate system is constructed and the spatial covariance function is calculated. Mass weights are introduced to weight the observation data. The entire region is divided into regular grids and kriging interpolation is performed at the center point of each grid. An improved DBSCAN density clustering algorithm is used to identify mosquito-prone areas. An adaptive neighborhood radius mechanism is introduced to dynamically adjust the clustering radius based on the local device density. When the number of grid points in a cluster exceeds the preset minimum cluster size, it is identified as a mosquito-prone area. The optimal operating parameter configuration for each device is calculated using the particle swarm optimization algorithm. A dual-objective optimization function is constructed and constraints are set. The optimal operating parameter configuration scheme is distributed in batches using a phased and gradual adjustment strategy.

8. The intelligent mosquito control system according to claim 1, characterized in that, The visualization management module includes: A cloud server is used to build a web-based visual management platform. The front end integrates the Leaflet open-source map library for the visualization of geospatial data. It calls the backend API interface to obtain the real-time location coordinates and operating status data of all devices, and selects different colored icons to distinguish them according to the operating status of the devices. A mosquito density heatmap layer is overlaid on the map display of the management platform. The density estimate and coordinate position of grid points are extracted from the spatial distribution raster data of mosquito density. The rendering parameters of the heatmap are configured and automatic refresh is supported. Develop functions for comparing and analyzing the capture effect of equipment, analyzing mosquito activity trends, and analyzing abnormal events. Use a front-end chart library to create bar charts, line charts, and pie charts. The system has been developed to automatically generate daily, weekly, and monthly reports. The system automatically extracts data for the corresponding time period from the database for statistical summary, and the reports can be exported as PDF or Excel format.

9. The intelligent mosquito control system according to claim 5, characterized in that, The prediction and control module also includes: Define the fuzzy input variable as the mosquito activity score, divide it into a preset number of fuzzy subsets, define a membership function for each fuzzy subset, and substitute the precise activity score into the membership function to calculate the membership degree of the score to each fuzzy subset; The fuzzy control rule base is defined to contain multiple control rules in the form of IF-THEN. Each rule describes the adjustment coefficient of the equipment operating parameters under a specific activity level. The membership degree of the input variable to the fuzzy subset of the rule's antecedent is taken as the activation strength of the rule. Defuzzing is performed using a weighted average method. The adjustment coefficient of each rule output is multiplied by the corresponding activation intensity. The weighted output of all rules is summed and then divided by the sum of activation intensities to obtain the final working parameter adjustment coefficient. The adjustment coefficient is then multiplied by the baseline value of each working parameter to calculate the specific value.

10. The intelligent mosquito control system according to claim 7, characterized in that, The regional collaborative optimization module also includes: The optimization objective is defined as maximizing the total number of mosquitoes captured in the area while minimizing the total energy consumption in the area. A dual-objective optimization function is constructed. The two objective functions are normalized and weighted to obtain a comprehensive objective function. The constraint condition is set that the adjustment range of the working parameters of each device does not exceed the preset adjustment range of the standard value. The initial particle swarm contains a preset number of particles, each representing a set of equipment operating parameter configuration schemes. The position and velocity of each particle are randomly initialized, the comprehensive objective function value corresponding to each particle is calculated, and the individual optimal position and global optimal position of each particle are recorded. The velocity and position of the particles are updated iteratively. The velocity update formula includes inertial terms, individual cognition terms, and social learning terms. Boundary processing is performed on the updated positions. The objective function values ​​of each particle are recalculated and the individual optimum and global optimum are updated. The iteration is terminated early when the global optimum objective function value no longer improves after a preset number of consecutive rounds.

Images