Artificial intelligence-based seasonal change monitoring method and system

Abstract

This application discloses an artificial intelligence-based method and system for monitoring seasonal changes. The method acquires regional geographic information of the monitoring area, historical monitoring data sequences within at least one historical monitoring period, and the acquisition time information of the historical monitoring data sequences. It then performs in-depth feature mining on the regional geographic information and the historical monitoring data sequences within the historical monitoring period. Based on reference depth monitoring feature information, depth monitoring feature information, and the acquisition time information of the historical monitoring data sequences from the surrounding reference areas within the historical monitoring period, it analyzes the seasonal characteristics of the monitoring area to obtain the seasonal change characteristics of the monitoring area under different monitoring periods. A machine learning model is then used to predict the seasonal change results of the monitoring area based on the depth monitoring feature information, seasonal change characteristics, and depth geographic feature information. This method can use artificial intelligence technology to predict the seasonal changes of a region, thereby improving the accuracy of seasonal change monitoring.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based method and system for monitoring seasonal changes. Background Technology

[0002] Seasonal variation monitoring is an important subfield at the intersection of environmental science and geography. In this field, accurate monitoring of seasonal variations is crucial for understanding climate change, ecosystem evolution, and agricultural production planning. Traditional methods for monitoring seasonal variations mainly rely on manual observation and simple statistical analysis. For example, by setting up meteorological observation stations, manually recording data such as temperature and precipitation, and then statistically analyzing data over many years to plot temperature change curves, one can roughly determine the patterns of seasonal variation. This method is relatively simple and intuitive in data collection and processing, but it also has many limitations.

[0003] Specifically, traditional monitoring methods suffer from insufficient data accuracy. Manual observation is susceptible to human error, and the limited distribution of observation stations makes it impossible to comprehensively reflect the true situation of a large monitoring area. To address these issues, with technological advancements, satellite remote sensing technology has been widely applied to seasonal change monitoring. Satellites can acquire data on the Earth's surface over large areas and periodically, such as vegetation cover index and surface temperature, significantly increasing the scope and frequency of data acquisition. Simultaneously, Geographic Information System (GIS) technology is used to integrate and analyze this massive amount of data, providing a more comprehensive view of the seasonal variation characteristics within a region.

[0004] However, satellite remote sensing and GIS technologies also have their limitations. While satellite remote sensing data has a wide coverage, its quality is sometimes difficult to guarantee due to factors such as cloud cover and sensor accuracy. Moreover, for some complex geographical environments, such as mountainous areas and densely forested regions, the interpretation and analysis of data are more challenging. This often results in a lack of in-depth data feature mining capabilities when conducting seasonal variation analysis, relying more on surface data analysis and making it difficult to accurately predict seasonal changes. Summary of the Invention

[0005] The main purpose of this application is to provide an artificial intelligence-based method and system for monitoring seasonal changes, which can improve the accuracy of seasonal change monitoring.

[0006] To achieve the above objectives, embodiments of the present invention provide an artificial intelligence-based method for monitoring seasonal changes, the method comprising:

[0007] The process involves: acquiring regional geographic information of the monitoring area, historical monitoring data sequences within at least one historical monitoring period, and the acquisition time information of the historical monitoring data sequences; performing feature depth mining on the regional geographic information and the historical monitoring data sequences within the historical monitoring period to obtain deep geographic feature information of the regional geographic information and deep monitoring feature information of the historical monitoring data sequences within the historical monitoring period; conducting seasonal feature analysis on the monitoring area based on reference deep monitoring feature information of the surrounding reference area within the historical monitoring period, the deep monitoring feature information, and the acquisition time information of the historical monitoring data sequences to obtain the seasonal variation characteristics of the monitoring area under different monitoring periods; and using a machine learning model to predict the seasonal variation results of the monitoring area based on the deep monitoring feature information, the seasonal variation characteristics, and the deep geographic feature information.

[0008] Accordingly, embodiments of this application also provide an artificial intelligence-based seasonal change monitoring system, the system comprising:

[0009] The system comprises the following modules: an acquisition module for acquiring regional geographic information of the monitoring area, historical monitoring data sequences within at least one historical monitoring period, and the collection time information of the historical monitoring data sequences; a feature mining module for performing in-depth feature mining on the regional geographic information and the historical monitoring data sequences within the historical monitoring period to obtain in-depth geographic feature information of the regional geographic information and in-depth monitoring feature information of the historical monitoring data sequences within the historical monitoring period; a feature analysis module for performing seasonal feature analysis on the monitoring area based on reference in-depth monitoring feature information of the surrounding reference area within the historical monitoring period, the in-depth monitoring feature information, and the collection time information of the historical monitoring data sequences to obtain the seasonal variation characteristics of the monitoring area under different monitoring periods; and a prediction module for using a machine learning model to predict the seasonal variation results of the monitoring area based on the in-depth monitoring feature information, the seasonal variation characteristics, and the in-depth geographic feature information.

[0010] In summary, by adopting the technical solution of this application, and by acquiring regional geographic information, historical monitoring data sequences, and their collection time information of the monitoring area, and conducting in-depth feature mining, a comprehensive and in-depth understanding of various characteristic information related to the monitoring area can be obtained. Seasonal feature analysis based on the characteristic information of the surrounding reference areas can take into account the potential impact of the surrounding areas on the monitoring area, making the analysis of seasonal change characteristics more accurate. Furthermore, by using machine learning models to combine in-depth monitoring feature information, seasonal change characteristics, and in-depth geographic feature information to predict seasonal change results, and leveraging the powerful data analysis capabilities of machine learning, complex geographic and temporal data relationships can be handled. This overcomes the shortcomings of traditional methods, such as single data sources, insufficient accuracy, difficulty in handling complex geographic environments, and lack of in-depth data mining capabilities. Therefore, it is possible to more accurately predict the seasonal change results of the monitoring area, providing more reliable data support for climate change research, ecosystem protection, agricultural production planning, and many other aspects. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a schematic diagram of a seasonal change monitoring system based on artificial intelligence, as described in an embodiment of this application.

[0013] Figure 2 A flowchart illustrating the artificial intelligence-based seasonal change monitoring method provided in this application embodiment;

[0014] Figure 3 This is a schematic flowchart of the feature analysis and processing provided in the embodiments of this application;

[0015] Figure 4 This is a flowchart illustrating the short-term feature analysis process provided in an embodiment of this application.

[0016] Figure 5 This is a flowchart illustrating the long-term feature analysis and processing provided in an embodiment of this application.

[0017] Figure 6 A schematic diagram illustrating the process of seasonal variation prediction provided in this application embodiment;

[0018] Figure 7 A schematic diagram of the structure of an artificial intelligence-based seasonal change monitoring system provided in an embodiment of this application;

[0019] Figure 8A schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] This application provides an artificial intelligence-based method and system for monitoring seasonal changes, which will be described in detail below.

[0022] In this application, the "Artificial Intelligence-Based Seasonal Change Monitoring Method" is a comprehensive operational process that uses artificial intelligence technology to monitor, analyze, and predict seasonal changes in a monitoring area. This method integrates multiple data sources and employs artificial intelligence algorithms for in-depth processing to accurately grasp the characteristics and trends of seasonal changes.

[0023] As shown in Figure 1, a seasonal change monitoring system is provided. The system may include at least one meteorological monitoring station deployed in the monitoring area, a soil sensor network, a global positioning system device, and a meteorological monitoring center platform (which may be implemented by at least one server), and communication connections between the meteorological monitoring station, the soil sensor network, and the monitoring center platform.

[0024] The monitoring area can be a large agricultural region encompassing various crop cultivation areas, rivers, hills, and other diverse topographical features, or it can be any other geographical area requiring monitoring of seasonal changes. This application utilizes high-precision Global Positioning System (GPS) equipment, topographic surveying instruments, and satellite image receiving devices to acquire comprehensive regional geographic information. The GPS equipment precisely determines the latitude and longitude coordinates of the monitoring area, the topographic surveying instrument measures detailed topographic data such as the height and slope of hills, and the satellite image receiving device acquires information on the macroscopic topography and land use types of the area. These elements collectively constitute the regional geographic information, which is then transmitted or uploaded to the meteorological monitoring center platform via a wireless network.

[0025] Meteorological monitoring stations collect data on temperature, precipitation, wind speed, and wind direction at regular intervals (e.g., hourly), forming the meteorological portion of the historical monitoring data sequence. The soil sensor network penetrates different depths of the soil, collecting data on soil moisture, fertility (e.g., nitrogen, phosphorus, and potassium content), and temperature. This data is also part of the historical monitoring data sequence, and each data collection moment is precisely recorded as the collection time information. The meteorological monitoring stations and the soil sensor network then transmit the collected historical monitoring data sequence and collection time information to the meteorological monitoring center platform.

[0026] The meteorological monitoring center platform, after obtaining regional geographic information of the monitoring area, historical monitoring data sequences within at least one historical monitoring period, and the acquisition time information of the historical monitoring data sequences, performs feature depth mining on the regional geographic information and the historical monitoring data sequences within the historical monitoring period to obtain deep geographic feature information of the regional geographic information and deep monitoring feature information of the historical monitoring data sequences within the historical monitoring period. Based on the reference deep monitoring feature information of the surrounding reference area of ​​the monitoring area within the historical monitoring period, the deep monitoring feature information, and the acquisition time information of the historical monitoring data sequences, the platform performs seasonal feature analysis on the monitoring area to obtain the seasonal variation characteristics of the monitoring area under different monitoring periods. Finally, a machine learning model is used to predict the seasonal variation results of the monitoring area based on the deep monitoring feature information, the seasonal variation characteristics, and the deep geographic feature information.

[0027] This application's embodiments can employ machine learning models (such as support vector machines) to predict seasonal changes in protected areas based on deep monitoring feature information, seasonal variation characteristics, and deep geographic feature information. For example, it can predict the potential impact of summer water temperature changes in lakes on fish reproduction, and the risk of forest fires during the dry season. Based on these predictions, protected area managers can take corresponding protective measures. For instance, they can strengthen lake monitoring during the fish breeding season to prevent illegal fishing; and conduct advance maintenance and patrols of forest firebreaks during high-risk fire seasons.

[0028] Furthermore, through long-term monitoring and forecasting of seasonal changes, protected areas can better maintain the balance of their ecosystems. For example, ecotourism routes can be rationally planned based on the needs of flora and fauna in different seasons, avoiding disturbance to rare species. When carrying out ecological restoration projects, appropriate plant species and restoration times are selected based on seasonal characteristics. If vegetation degradation in a certain area is predicted in future seasons, replanting and ecological restoration plans are formulated in advance to ensure the long-term stability of the protected area's ecosystem.

[0029] refer to Figure 2 , Figure 2This is a flowchart illustrating an artificial intelligence-based seasonal change monitoring method provided in this application embodiment. The execution entity of this method can be a computer device, which can be a single computer device or a cluster of multiple computer devices. The computer device can be a terminal device or a server, etc. The artificial intelligence-based seasonal change monitoring method provided in this application embodiment specifically includes:

[0030] S100: Obtain regional geographic information of the monitoring area, historical monitoring data sequences within at least one historical monitoring period, and the acquisition time information of the historical monitoring data sequences.

[0031] In this application, the monitoring area refers to a specific geographical area selected for seasonal change monitoring, the scope and boundaries of which can be set according to actual needs. For example, the monitoring area can be a large agricultural area containing various crop planting areas, rivers, hills and other different topographical features, or it can be any other geographical area that needs to monitor seasonal changes.

[0032] The regional geographic information can include numerous geographically related elements of the monitored area. For example, it can include topographic information, specifically the altitude, slope, and orientation of mountains; the area and flatness of plains; the distribution, area, and depth of water bodies (such as rivers and lakes); and the distribution areas of soil types (such as clay, sandy soil, and loam). In one embodiment, geographic location information can also include latitude and longitude ranges. In practical applications, high-precision Global Positioning System (GPS) devices, topographic surveying instruments, and satellite image receiving devices can be deployed to acquire comprehensive regional geographic information.

[0033] The historical monitoring period is a pre-defined time interval used for periodic data collection of the monitoring area. The historical monitoring data sequence is a collection of various data collected within each historical monitoring period, arranged chronologically. The collected monitoring data types can be diverse. For example, meteorological data may include daily maximum, minimum, and average temperatures for each historical monitoring period; precipitation data may cover precipitation amount, frequency, and type (such as rain, snow, and hail); in one embodiment, it may also include wind speed, wind direction, and sunshine duration.

[0034] In one embodiment, historical monitoring data, in addition to meteorological data, may also include vegetation-related data if the monitoring area involves an ecosystem, such as the percentage of vegetation cover, the distribution ratio of different vegetation types (e.g., forests, grasslands, shrublands), and the growth status of vegetation (e.g., vegetation height, density). If water resources are involved, water quality-related data will be included, such as pH value, dissolved oxygen content, and chemical oxygen demand (COD). Historical monitoring data can be selected or set according to actual monitoring needs.

[0035] In practical applications, historical monitoring data can be collected by equipment such as meteorological monitoring stations and soil sensor networks deployed in the monitored area.

[0036] In this embodiment, the acquisition time information refers to the acquisition time of the monitoring data in each historical monitoring data sequence. Since seasonal changes are closely related to time, accurate acquisition time information allows subsequent analysis to determine the data's position on the timeline, thereby revealing the periodic patterns of seasonal changes. For example, by combining years of temperature data with its acquisition time information (accurate to the year, month, day, and even hour and minute), it is possible to analyze the range of temperature changes in the monitored area during different seasons (spring, summer, autumn, and winter), the timing of temperature rises and falls, and other patterns.

[0037] S200: Perform feature depth mining on the regional geographic information and the historical monitoring data sequence within the historical monitoring period to obtain the deep geographic feature information of the regional geographic information and the deep monitoring feature information of the historical monitoring data sequence within the historical monitoring period.

[0038] In this embodiment, feature depth mining is the process of extracting more abstract, representative and distinctive feature information from the original data using specific algorithms and techniques.

[0039] In one embodiment, in-depth mining of regional geographic information can be achieved using techniques such as spatial analysis algorithms. For example, based on topographic data, by analyzing the altitude, slope, and orientation of mountains, in-depth geographic feature information such as the obstruction and guidance effect of mountains on airflow can be extracted. High-altitude mountains may block warm and humid airflow, resulting in significant climate differences on both sides of the mountain; this obstruction effect on airflow is a type of in-depth geographic feature information.

[0040] From the perspective of soil type distribution, the relationship between the distribution areas of different soil types and topography is also a subject for in-depth exploration. For example, sandy soil may be more distributed in alluvial plains near rivers. This correlation between soil type and topography helps to understand ecological processes such as water conservation and vegetation growth in the region, and is also part of in-depth geographical feature information.

[0041] In this embodiment, deep geographic feature information is highly abstract, representative, and reflects intrinsic relationships, obtained through in-depth mining of regional geographic information. Regional geographic information includes numerous basic geographic elements, such as topography (morphology and distribution of mountains, plains, and water bodies), soil types and their distribution, and geographical locations (latitude and longitude). Deep geographic feature information, on the other hand, is the intrinsic connection between geographic elements and their potential impact on seasonal changes, which is further mined from these elements.

[0042] For example, in terms of topography, deep geographic information is not merely the numerical value of a mountain's elevation, but rather the relationship between that elevation and surrounding air currents and precipitation distribution. High-altitude mountains may, to some extent, block warm, moist air currents, resulting in abundant precipitation on the windward slopes and relatively drier slopes. This causal relationship between mountain height and precipitation is part of deep geographic information.

[0043] In one embodiment, in-depth mining of historical monitoring data sequences can be achieved using machine learning algorithms (such as deep neural networks). Taking temperature data as an example, in addition to simple statistical features (such as average, maximum, and minimum values), in-depth mining can discover periodic fluctuation patterns in temperature data over longer time scales. These fluctuation patterns may be related to factors such as solar activity cycles and ocean temperature changes. For precipitation data, not only can seasonal patterns of precipitation be analyzed, but also potential relationships between precipitation and other meteorological factors (such as temperature and air pressure) can be uncovered. For example, in certain seasons, after the temperature rises to a certain level, changes in air pressure can trigger an increase in precipitation; this complex correlation constitutes in-depth monitoring feature information.

[0044] In this embodiment of the application, the deep monitoring feature information is the feature information obtained by deep mining of historical monitoring data sequences, which can reflect the complex relationships and patterns hidden behind the monitoring data.

[0045] In one embodiment, the historical monitoring data sequence contains a rich variety of data types, such as meteorological data (temperature, precipitation, wind speed, wind direction, etc.) and ecological data (vegetation coverage, plant and animal growth status, etc.). In-depth monitoring feature information mining focuses on the deep relationships between these data and between the data and factors such as time and space.

[0046] Taking temperature data as an example, in-depth monitoring features not only include simple statistical characteristics of temperature (such as daily average temperature, monthly average temperature, etc.), but also temperature fluctuation patterns at different time scales. For example, on a longer time scale, there may be temperature fluctuation patterns related to the solar activity cycle. These patterns may be discovered through complex analysis of many years of temperature data (such as wavelet analysis). Moreover, the potential relationships between temperature and other meteorological factors are also part of the in-depth monitoring features. For instance, in a specific season, there is a complex interaction between temperature and air pressure. When the temperature rises to a certain level, changes in air pressure can affect the stability of weather systems, thereby affecting meteorological elements such as precipitation. This deep-seated relationship between temperature and air pressure constitutes in-depth monitoring features.

[0047] In this embodiment, the deep geographic feature information and deep monitoring feature information obtained through deep mining can reveal the hidden patterns and relationships in the original data. This feature information reflects the essential characteristics of the monitored area better than the original data, providing stronger support for subsequent accurate analysis of seasonal variation characteristics. Through deep mining, the interactions between various factors can be better understood, thereby more accurately grasping the intrinsic mechanisms of seasonal changes.

[0048] S300: Based on the reference depth monitoring feature information of the surrounding reference area of ​​the monitoring area within the historical monitoring period, the depth monitoring feature information, and the acquisition time information of the historical monitoring data sequence, seasonal feature analysis is performed on the monitoring area to obtain the seasonal variation characteristics of the monitoring area under different monitoring periods.

[0049] In this embodiment, the reference area surrounding the monitoring area is the area that is geographically adjacent to or ecologically related to the monitoring area.

[0050] In this embodiment, the acquisition method of reference depth monitoring feature information of the surrounding reference area is similar to that of the depth monitoring feature information of the monitoring area, which is also obtained by in-depth mining of various types of data in the surrounding area. For example, if the monitoring area is an inland farmland, and there is a forest nearby as a reference area, the reference depth monitoring feature information of the forest area may include the impact of forest vegetation transpiration on regional humidity, the impact of phenological changes of plants and animals in the forest on regional ecology, etc.

[0051] In this embodiment, when performing seasonal characteristic analysis, the accuracy of seasonal change monitoring can be improved by combining the reference depth monitoring characteristics of surrounding reference areas with the depth monitoring characteristics of the monitoring area itself and the collection time information of historical monitoring data sequences. Taking the seasonal transition period as an example, the collection time information determines the time range of the seasonal transition, the depth monitoring characteristics of the monitoring area itself (such as the changing trends of temperature and precipitation) reflects the internal seasonal change dynamics, while the reference depth monitoring characteristics of surrounding forests (such as the moderating effect of forests on the local climate during seasonal transitions) will affect the seasonal changes of the monitoring area. For example, in spring, forest vegetation begins to revive and grow, and transpiration increases, which will increase the air humidity in the surrounding area. If the monitoring area is close to this forest, this increase in humidity will affect the sowing and growth conditions of crops in the monitoring area, thereby affecting the seasonal change characteristics of the monitoring area.

[0052] The embodiments of this application can broaden the analytical perspective by considering information from surrounding reference areas, avoiding focusing only on the internal aspects of the monitoring area while ignoring the influence of external factors. This comprehensive analysis helps to more accurately capture the seasonal variation characteristics of the monitoring area under different monitoring cycles. Especially when dealing with complex geography and ecosystems, it can better explain some special phenomena and trends in the seasonal variation process.

[0053] S400: A machine learning model is used to predict the seasonal changes in the monitored area based on the deep monitoring feature information, the seasonal change features, and the deep geographic feature information.

[0054] In this embodiment, the machine learning model is a pre-trained algorithm model that can make predictions based on the input feature information.

[0055] Deep monitoring feature information includes in-depth characteristics of various aspects of data within the monitoring area, such as meteorology and ecology; seasonal variation features reflect the seasonal variation patterns exhibited by the monitoring area in different monitoring periods; and deep geographic feature information reflects the impact of regional geographic elements on seasonal variations. This information is used as input to machine learning models.

[0056] Taking the decision tree model as an example, the model has already learned a large amount of sample data with similar characteristics during the training phase (including data on deep monitoring features, seasonal change features, and deep geographic features, as well as the corresponding seasonal change results). When relevant feature information of the monitoring area is input, the model calculates and judges according to its internal decision rules. For example, the nodes of the decision tree may make the first branch judgment based on deep geographic feature information (such as the distribution of mountains), and then make further branches based on deep monitoring feature information (such as the changing trends of temperature and precipitation), and finally combine seasonal change features to predict the future seasonal change results of the monitoring area.

[0057] In one embodiment, models such as convolutional neural networks can also be used to predict seasonal changes.

[0058] In this application, the seasonal variation results of the monitoring area refer to the conclusions drawn from the analysis and prediction of various data of the monitoring area regarding the changes of various elements in the area during the seasonal transition and the future development trend.

[0059] These elements can encompass multiple aspects. Regarding meteorological elements, seasonal variation results include temperature trends, such as the predicted average temperature range for the next season, the magnitude of temperature increases or decreases, and the likelihood of extreme temperatures; precipitation variations, such as the amount of precipitation, its distribution pattern (whether it is evenly distributed or concentrated in certain periods or areas), and precipitation types (the proportion of rain, snow, hail, etc.); and seasonal variations in wind speed and direction, such as whether strong winds are more likely in certain seasons and whether there is a seasonally dominant wind direction.

[0060] In terms of ecological elements, seasonal changes manifest as changes in vegetation. This includes changes in the growth cycle of different vegetation types, such as the timing and duration of tree budding, flowering, fruiting, and leaf fall; trends in vegetation cover increases and decreases, and the impact of these changes on other organisms in the ecosystem (such as animals that depend on vegetation for habitat or foraging); and changes in vegetation community structure, such as whether new plant species invade or the dominant position of existing species changes in certain seasons.

[0061] For water resource-related factors, seasonal changes can also involve changes in the water levels of lakes, rivers, and other water bodies, such as the extent of water level rise during the rainy season and the risk of drying up during the dry season; seasonal fluctuations in water quality, such as the possibility that rising water temperatures in summer may lead to a decrease in dissolved oxygen content in the water body, or the large-scale proliferation of certain algae affecting water quality; and the seasonal succession of biological communities in the water body, such as the relationship between fish migration time and breeding season and seasonal changes.

[0062] In one embodiment, the historical monitoring data sequence within at least one historical monitoring period includes historical monitoring data sequences within a first historical monitoring period and a second historical monitoring period; the second historical monitoring period is earlier than the first historical monitoring period; reference Figure 3 Step S300 may specifically include:

[0063] S301. Based on the reference depth monitoring feature information, the depth monitoring feature information corresponding to the historical monitoring data sequence within the first historical monitoring period, and the collection time information corresponding to the historical monitoring data sequence within the first historical monitoring period, a short-term seasonal variation analysis is performed on the monitoring area to obtain the short-term seasonal variation characteristics of the monitoring area.

[0064] In this application, the historical monitoring data sequence within the first historical monitoring period refers to a series of data sets obtained by collecting data from the monitoring area at a certain monitoring frequency within a specific, relatively recent historical time period. This specific historical time period is the first historical monitoring period, which can be set according to actual needs.

[0065] Taking ecological monitoring in urban parks as an example, the first historical monitoring period can be set as the most recent year. During this year, multifaceted data collection is conducted on the park's ecological environment. Meteorological data collection includes daily temperature (accurate to one decimal place, recording the highest, lowest, and average temperatures of the day), hourly wind speed and direction (wind speed accurate to meters per second), and precipitation (the amount, duration, and type of each precipitation event, such as light rain, moderate rain, and heavy rain). Ecological data collection involves plant growth data within the park. For example, every two weeks, the height and crown width of specific plant samples (such as representative trees and flowers) are measured (accurate to centimeters), and the flowering and fruiting rates of plants are statistically analyzed monthly (expressed as percentages). For animal activity data, monitoring equipment (such as infrared cameras) is used to record the frequency and activity range of common animals (such as squirrels and birds) in specific areas weekly.

[0066] Arranging the collected data chronologically constitutes the historical monitoring data sequence for the first historical monitoring period. Due to its temporal proximity, this sequence reflects the recent ecological and environmental conditions of the monitored area, significantly impacting the identification of short-term trends and emergencies. For example, analyzing annual temperature fluctuations and corresponding changes in plant flowering times reveals the impact of short-term seasonal variations on plant phenology; correlation analysis between animal activity frequency and meteorological data (such as temperature and precipitation) reveals the short-term effects of recent environmental changes on animal behavior. This facilitates timely adjustments to park management strategies, such as strengthening protection measures in areas with frequent animal activity or adjusting irrigation and fertilization plans based on plant growth.

[0067] In this embodiment of the application, the historical monitoring data sequence within the second historical monitoring period is a set of data from the monitoring area collected within a longer historical period than the first historical monitoring period. The second historical monitoring period can be selected based on experience.

[0068] For example, in the long-term ecological monitoring of the aforementioned urban parks, the second historical monitoring period can be set as the past ten years. During these ten years, data will be collected from multiple sources. Regarding meteorological data, in addition to daily basic meteorological elements (such as temperature, precipitation, wind speed, and wind direction), records of special meteorological events will be included, such as the frequency, intensity, and duration of extreme weather events (heavy rain, heavy snow, strong winds, etc.) each year. For soil data, soil fertility indicators (including nitrogen, phosphorus, and potassium content, soil organic matter content, etc., accurate to milligrams per kilogram), soil moisture (expressed as a percentage), and soil pH (accurate to one decimal place) will be collected quarterly. For plant community data, a detailed survey of the plant community will be conducted annually, recording the distribution area of ​​different plant species (accurate to square meters), changes in dominant species, and the succession stage of the community. For animal population data, a comprehensive animal population census will be conducted every two years, statistically analyzing the number, age structure, and distribution range of different animal populations.

[0069] Further arranging these data in ten-year chronological order forms the historical monitoring data sequence for the second historical monitoring period. In this embodiment, the historical monitoring data sequence for the second historical monitoring period reflects the ecological and environmental change trends over a longer timescale in the monitored area. For example, by analyzing the changing trends of soil fertility and plant community succession over ten years, the impact of long-term seasonal variations on the soil-plant ecosystem can be identified; by analyzing the correlation between changes in animal population size and structure over ten years and meteorological data (such as the frequency of extreme weather events), the impact of long-term environmental changes on animal population dynamics can be understood. This helps in developing long-term park planning and ecological protection strategies, such as developing soil improvement plans to address declining soil fertility trends, or adjusting habitat protection strategies based on long-term changes in animal populations.

[0070] Among them, short-term seasonal variation analysis is a technical process that focuses on a relatively short time range to comprehensively evaluate and analyze the factors related to seasonal variation in the monitored area.

[0071] In one embodiment, short-term seasonal variation analysis may include: determining the correlations between different data points, using specific analytical models and algorithms, such as statistical models based on time series analysis, to quantify these relationships and identify patterns, thereby obtaining short-term seasonal variation characteristics. For example, continuing with the aforementioned agricultural ecosystem example, analysis reveals that pond water temperature gradually increased over the past month (referencing depth monitoring characteristic information), while the growth rate of a certain crop in the farmland suddenly slowed down during the same period (depth monitoring characteristic information), and based on the collection time information, it was determined that the two are synchronous in time. Then, specialized analytical models and algorithms, such as statistical models based on time series analysis, are used to quantify these relationships and identify patterns. For example, the model calculates that for every certain degree increase in pond water temperature, the growth rate of the crop will decrease by a certain proportion.

[0072] In this embodiment, the short-term seasonal variation characteristics are obtained through short-term seasonal variation analysis, which can reflect the comprehensive performance and unique attributes of various elements related to seasonal variation in the monitored area within a short period of time (corresponding to the first historical monitoring period).

[0073] Taking coastal fisheries and marine ecological monitoring as an example, short-term seasonal variation characteristics may include multiple aspects. Regarding the marine environment, fluctuations in seawater temperature over the past two months (such as the difference between the highest and lowest temperatures), short-term trends in seawater salinity (whether it gradually increases or decreases), and changes in the speed and direction of ocean currents in the short term are all components of short-term seasonal variation characteristics. For fishery resources, changes in the migration routes of certain fish species over the past month (such as whether they move closer to the coast or away from traditional fishing grounds), fluctuations in catch volume in the short term (such as weekly), and recent changes in the age structure of specific fish populations (such as in the last two months) also fall under the category of short-term seasonal variation characteristics.

[0074] In the embodiments of this application, short-term seasonal variation characteristics are the result of the interaction of multiple factors. For example, short-term fluctuations in seawater temperature may be the result of the combined effects of recent changes in atmospheric circulation (such as the short-term influence of monsoons) and heat exchange processes within the ocean, while changes in fish migration routes may be closely related to short-term changes in seawater temperature, salinity, and the short-term distribution of food sources.

[0075] In one embodiment, reference Figure 4 Step S301 may specifically include:

[0076] S3011: Integrate the first reference depth monitoring feature information of the reference area surrounding the monitoring area within the first historical monitoring period and the depth monitoring feature information corresponding to the historical monitoring data sequence within the first historical monitoring period to obtain integrated feature information.

[0077] In one embodiment, the first reference depth monitoring feature information of the surrounding reference area within the first historical monitoring period is specific feature information obtained after in-depth mining of the surrounding reference area within the first historical monitoring period. For example, if the monitoring area is a small mountain town, the surrounding mountains are used as the reference area. Within the first historical monitoring period (assuming it is the most recent three months), the ecosystem of the mountains is monitored, and the first reference depth monitoring feature information may include changes in chlorophyll content of specific vegetation types (such as coniferous forests) in the mountains, the activity range and behavioral patterns of certain wild animals (such as changes in the foraging radius of a certain small mammal), etc. This feature information is obtained through a series of professional monitoring methods and analysis methods, such as using hyperspectral remote sensing technology to obtain the chlorophyll content of vegetation, and recording the activity range and behavioral patterns of animals through animal trackers and long-term field observation.

[0078] The in-depth monitoring feature information corresponding to the historical monitoring data sequence within the first historical monitoring period is the result of in-depth data mining of the monitoring area itself within this period. Taking a mountain town as an example, for meteorology, the in-depth monitoring feature information may be the relationship between the daily temperature variation and wind direction in the town (such as the temperature rise and fall pattern under a specific wind direction), the spatial distribution of precipitation and its correlation with topography (such as the concentration of precipitation in valley areas), etc.; for ecology, it may be the interaction pattern between crop growth rate and soil moisture in the town in the short term (such as the variation pattern of crop growth rate when soil moisture is within a certain range), etc. This in-depth monitoring feature information reflects the intrinsic relationship between various factors within the monitoring area and is obtained through complex data analysis and model calculations on the original monitoring data, such as using machine learning algorithms to mine the potential relationship between temperature, wind direction, soil moisture and crop growth rate.

[0079] In one embodiment, integrating the first reference depth monitoring feature information and the depth monitoring feature information within the first historical monitoring period is a process of integrating multi-source information. Continuing with the example of a mountain town, the integration process needs to consider the possible connections between changes in chlorophyll content in the surrounding mountain vegetation (first reference depth monitoring feature information) and meteorological conditions within the town (depth monitoring feature information). For example, a decrease in chlorophyll content in the mountain vegetation may indicate changes in light intensity, which may be related to sunshine duration and cloud cover within the town (meteorological depth monitoring feature information). Simultaneously, changes in the activity range of wild animals in the mountains (first reference depth monitoring feature information) may be related to the growth of crops within the town (depth monitoring feature information); for example, an expansion of the wild animal activity range may lead to an increased impact on crops in the town. By integrating these potentially related feature information from different sources, integrated feature information is obtained. Integration can break down information barriers between the monitoring area and the surrounding reference area, providing a richer information foundation for a more comprehensive analysis of short-term seasonal changes, and helping to discover patterns and relationships that cannot be detected by analyzing the monitoring area or the surrounding reference area alone.

[0080] In one embodiment, there are various ways to integrate features or data, such as feature splicing or weight-based fusion.

[0081] In one embodiment, to improve the accuracy of seasonal change monitoring, step S3011 may include:

[0082] An accumulation operation is performed on the first reference depth monitoring feature information and the depth monitoring feature information corresponding to the historical monitoring data sequence within the first historical monitoring period to obtain the accumulation operation result; a multiplication operation is performed on the reference depth monitoring feature information of the reference area surrounding the monitoring area within the first historical monitoring period and the depth monitoring feature information corresponding to the historical monitoring data sequence within the first historical monitoring period to obtain the multiplication operation result; the accumulation operation result, the multiplication operation result, the reference depth monitoring feature information, and the depth monitoring feature information corresponding to the historical monitoring data sequence within the first historical monitoring period are integrated to obtain integrated feature information.

[0083] The first reference depth monitoring feature information is the feature information obtained from the reference area surrounding the monitoring area within the first historical monitoring period. It reflects the specific state and trend of change in the surrounding area. For example, if the monitoring area is a grassland and the surrounding forest is the reference area, the first reference depth monitoring feature information may include the change in the growth rate of trees in the forest (in centimeters per month) within the first historical monitoring period (assuming it is the most recent three months), the fluctuation of the population of specific bird species in the forest (expressed as the increase or decrease in the number of individuals), etc.

[0084] In this embodiment, the summation operation is a way to numerically merge feature information from two different sources. The summation operation can comprehensively consider the characteristics of the surrounding reference area and the monitoring area itself to obtain more comprehensive information. Taking the grassland-forest example, the change in the growth rate of trees in the forest is summed with the change in grassland vegetation cover. Assuming that the forest trees grow at an average rate of 2 cm per month over three months, and the grassland vegetation cover decreases by 5% during the same period, 2 (which can be considered a transformed dimensionless value, such as through standardization or other mapping methods) is summed with -5 (representing the decrease in vegetation cover), resulting in a summation result of -3.

[0085] In this embodiment, performing a multiplication operation integrates these two types of feature information from another perspective. The purpose of the multiplication operation is to capture the non-linear relationships that may exist between different features. For example, multiplying the forest humidity change rate with the grassland herbivore population change rate. Assuming that the forest humidity change rate increases by 1.1 times per week (i.e., 110%, expressed as 1.1 decimals) and the grassland herbivore population change rate decreases by 0.9 times per month (i.e., 90%, expressed as 0.9 decimals), the result of the multiplication operation is 0.99.

[0086] In one embodiment, the reference depth monitoring feature information of the surrounding reference area within the first historical monitoring period, the depth monitoring feature information corresponding to the historical monitoring data sequence within the first historical monitoring period, the accumulation operation result, and the multiplication operation result can be determined as the elements to be integrated; the weights of each element are determined according to their importance to the short-term seasonal variation analysis. Finally, the elements are weighted and summed based on their weights to obtain the integrated feature information.

[0087] S3012: Standardize the integrated feature information and the collection time information corresponding to the historical monitoring data sequence within the first historical monitoring period to obtain the influence weight information corresponding to the historical monitoring data sequence within the first historical monitoring period.

[0088] In this embodiment, standardization is used to enable the quantification and comparison of integrated feature information and acquisition time information of different types and magnitudes on a unified scale. The integrated feature information includes multiple features from the surrounding reference area and the monitoring area itself, which may differ in numerical range, units, etc. For example, the chlorophyll content of vegetation may fall within a certain range, while the numerical range of temperature is completely different, and the acquisition time information is expressed in time scales (such as days or hours). Standardization eliminates the impact of these differences on subsequent analysis.

[0089] In this embodiment, after standardization, the integrated feature information and collection time information can be combined to calculate the influence weight information corresponding to the historical monitoring data sequence within the first historical monitoring period. The influence weight information reflects the relative importance of each feature in short-term seasonal variation analysis. For example, in the standardized integrated feature information, calculations show that the change in chlorophyll content of mountain vegetation (after standardization) has a different influence weight on short-term seasonal variation compared to precipitation in a town over a certain period (after standardization of meteorological feature information). If the change in vegetation chlorophyll content is more closely related to short-term seasonal variation (calculated through a specific algorithm), then its weight value in the influence weight information will be relatively higher. This embodiment uses influence weight information to more specifically focus on factors that have a greater impact on short-term seasonal variation. In subsequent processing, resources and attention can be rationally allocated according to the weights, improving the accuracy and efficiency of short-term seasonal variation analysis.

[0090] Step 3013: Based on the influence weight information, perform comprehensive processing on each historical monitoring data in the historical monitoring data sequence within the first historical monitoring period to obtain the short-term seasonal variation characteristics of the monitoring area.

[0091] In this application, each historical monitoring data point in the first historical monitoring period includes monitoring results from various aspects of the monitored area within that period. Taking a mountain town as an example, these data include meteorological data (such as daily temperature, precipitation, wind speed, and wind direction), ecological data (such as crop growth height, pest and disease conditions, and soil moisture), and other relevant data (such as river levels and water quality within the town). Each historical monitoring data point contains status information of the monitored area at a specific point in time or time period, serving as the foundational data for short-term seasonal variation analysis.

[0092] This application embodiment comprehensively processes historical monitoring data based on influence weight information. Specifically, it involves assigning different weights to various data points and performing a weighted average. Specifically, each historical monitoring data point is weighted according to the calculated influence weight information. For data with higher weights, such as temperature data in meteorological data, a larger coefficient is assigned in the calculation model when analyzing its relationship with other data. Next, data analysis algorithms, such as multiple regression analysis or neural network-based models, are used to uncover the intrinsic relationships between the weighted data. Taking temperature, soil moisture, and crop growth data as examples, key influence paths are identified by analyzing the correlation patterns between them under different weights. Finally, these analytical results are synthesized to form a comprehensive description of the short-term seasonal variation characteristics of the monitoring area, thereby accurately grasping the short-term seasonal variation of the monitoring area within the first historical monitoring period.

[0093] S302. Based on the depth monitoring feature information corresponding to the historical monitoring data sequence within the second historical monitoring period and the collection time information corresponding to the historical monitoring data sequence within the second historical monitoring period, a long-term seasonal variation analysis is performed on the monitoring area to obtain the long-term seasonal variation characteristics of the monitoring area.

[0094] In this application, long-term seasonal variation analysis is a technical operation that delves into relevant data from a monitoring area over a relatively long time span to reveal the patterns and trends of seasonal variations on a long-term scale. In this application, the "relatively long time span" corresponds to the second historical monitoring period.

[0095] In one embodiment, long-term seasonal variation analysis can employ a combination of analytical methods. For example, for meteorological data such as temperature and precipitation, trend analysis methods, such as linear regression analysis, can be used to determine whether temperature has continued to rise or fall, and whether precipitation has tended to increase or decrease over the past decade. For forest ecological data, time series analysis methods, such as autoregressive moving average (ARMA) models, can be used to analyze the long-term correlation between changes in tree population size and meteorological factors. For instance, analysis may reveal that over the past decade, as the annual average temperature has gradually increased (deep monitoring characteristic information), the population size of a certain temperature-sensitive tree species has shown a decreasing trend year by year (deep monitoring characteristic information), and the trend is determined to be temporally consistent based on the data collection time information.

[0096] Long-term seasonal variation analysis helps to understand the evolution of ecosystems and climate systems in the monitored area from a macro and long-term perspective. In forest management, this analysis can provide a basis for formulating long-term forest management strategies. For example, in response to the decline in tree populations due to rising temperatures, afforestation plans can consider introducing tree species that are more adapted to warmer climates.

[0097] In this embodiment of the application, the long-term seasonal variation characteristics are obtained after long-term seasonal variation analysis, which can reflect the comprehensive characteristics and evolution trend information of various elements related to seasonal variation in the monitoring area over a relatively long period of time (corresponding to the second historical monitoring period).

[0098] Taking large lakes and their surrounding wetland ecosystems as an example, long-term seasonal variation characteristics encompass multiple aspects. In terms of hydrological elements, the long-term trend of lake water level changes over the past two decades (e.g., water level decreasing year by year or fluctuating upwards), and the long-term evolution patterns of lake water quality (e.g., pH, dissolved oxygen content, nutrient concentration, etc.), such as whether there is a long-term eutrophication trend, are all manifestations of long-term seasonal variation characteristics. In terms of biological elements, the succession process of wetland plant community structure over the past two decades (e.g., from emergent plants to submerged plants becoming dominant), the long-term changes in the population size of fish in the lake, and the long-term fluctuations in species diversity are all examples of long-term seasonal variation characteristics.

[0099] In one embodiment, the historical monitoring data sequence within the second historical monitoring period includes historical monitoring data sequences in at least one dimension within the second historical monitoring period; (Refer to...) Figure 5 Step S302 may include:

[0100] S3021: For the historical monitoring data sequences under each dimension within the second historical monitoring period, perform weight allocation processing on the deep monitoring feature information corresponding to the historical monitoring data sequences under the dimensions to obtain the initial influence weight information corresponding to the historical monitoring data sequences under the dimensions.

[0101] In this application, the historical monitoring data sequence within the second historical monitoring period includes historical monitoring data sequences under at least one dimension. The dimension in this application refers to the ability to classify monitoring data from different perspectives. For example, in the long-term monitoring of a large wetland ecosystem, one dimension could be an ecological element dimension, whose historical monitoring data sequence includes changes over time (the second historical monitoring period, assumed to be the past ten years) in data such as the coverage area of ​​different plant communities within the wetland, the species diversity of plant communities, and the population size of animal species within the wetland; another dimension could be an environmental element dimension, including changes in wetland water level, water quality parameters (such as pH, dissolved oxygen, nutrient content, etc.), and meteorological data such as temperature and precipitation over the past ten years.

[0102] Each dimension of historical monitoring data sequence has its corresponding in-depth monitoring feature information. Taking plant community cover area in the ecological element dimension as an example, the in-depth monitoring feature information may include the long-term trend of plant community cover area (such as whether it decreases or increases year by year), the relationship with other ecological elements (such as animal population size) (for example, as plant community cover area decreases, the population size of some animal species that depend on plant habitats also decreases accordingly), etc.

[0103] In this embodiment, weighting the depth monitoring feature information across various dimensions is designed to reflect the relative importance of different dimensional features in long-term seasonal variation analysis. Different depth monitoring features have varying degrees of impact on the long-term seasonal variation characteristics of the monitored area. For example, in wetland ecosystems, long-term changes in plant community cover can fundamentally affect the overall structure and function of the wetland ecosystem, while changes in the content of certain trace elements in water quality parameters may have a relatively small impact on the ecosystem.

[0104] S3022: Based on the initial influence weight information and the collection time information corresponding to the historical monitoring data sequence within the second historical monitoring period, the historical monitoring data sequence under the dimension is calculated to obtain the sequence feature components of the historical monitoring data sequence under the dimension.

[0105] In this embodiment, the initial influence weight information reflects the relative importance of monitoring feature information at different depths, while the collection time information clarifies the temporal order of these feature information within the second historical monitoring period.

[0106] Taking a large wetland ecosystem as an example, assuming that temperature data is one of the key environmental monitoring features, with an initial influence weight of 0.15, the data collection time records temperature data for each season over the past ten years. During the analysis, it is necessary to consider the seasonal variation patterns of temperature and its fluctuations across different years. For example, the long-term trend of winter temperature changes may be closely related to ecological processes within the wetland (such as the hibernation behavior of certain animals and the dormancy state of plants), while changes in summer temperature may affect wetland evaporation and water level changes.

[0107] The sequence feature component is a result obtained by performing specific operations on the historical monitoring data sequence under the dimension. It is a new feature representation obtained by processing the data after considering the initial influence weight information and the collection time information. The feature component of this application integrates the importance of the depth monitoring feature information (reflected by the initial influence weight information) and the temporal characteristics of the data (reflected by the collection time information). For example, in a grassland ecosystem containing multiple plant communities, for the historical monitoring data sequence under the dimension of plant community cover, the sequence feature component obtained after the operation may reflect the comprehensive change trend of plant community cover in different seasons and years, and this trend is calculated based on the importance of different depth monitoring feature information (such as the influence weight of plant species diversity on cover) and collection time information (such as the growth status in different seasons).

[0108] In one embodiment, sequence feature components can be obtained based on time series model computation. For example, for historical monitoring data sequences in the dimension of biological population size in marine ecosystems, a time series model, such as the Autoregressive Integral Moving Average (ARIMA) model, can be used. First, initial influence weight information is incorporated into the model parameters. For example, if biological population size is related to factors such as ocean temperature and salinity, weights are determined based on their importance, and these weights are used as coefficients of variables in the ARIMA model. The time information collected is used as the time axis input to the model.

[0109] Then, the ARIMA model is used to fit and predict the historical monitoring data series. The model calculates the sequence feature components for each time period based on the time-series characteristics of the data and the weighting relationships of the variables. This time-series model-based approach better captures the temporal dynamics of the data and the complex relationships between variables.

[0110] In one embodiment, the initial influence weight information can be integrated with the collection time information corresponding to the historical monitoring data sequence within the second historical monitoring period to obtain the target influence weight information corresponding to the historical monitoring data sequence under the dimension; based on the target influence weight information, the historical monitoring data sequence under the dimension is calculated to obtain the sequence feature components of the historical monitoring data sequence under the dimension.

[0111] The target impact weight information is obtained by integrating the initial impact weight information and the data collection time information. This weight comprehensively considers the importance of depth monitoring feature information and the temporal distribution characteristics of the data. For example, in a desert oasis ecosystem, for a certain depth monitoring feature information under the vegetation growth dimension (such as root depth), if collected in spring (data collection time information), the importance of root depth to the stability of the oasis ecosystem and vegetation growth (long-term seasonal variation) will differ from other seasons due to the specific climatic conditions of spring (such as spring wind and sand activity, precipitation distribution, etc.). The target impact weight information can more accurately reflect the weight relationship under such dynamic changes.

[0112] In a mountainous forest ecosystem, for historical monitoring data sequences along the tree growth dimension, such as tree height and diameter at breast height (DBH), a time-feature correlation matrix is ​​constructed. Rows represent different collection times (e.g., the four seasons of each year), and columns represent different depth monitoring features.

[0113] Based on the characteristics of forest ecosystems and research findings, each element in the matrix is ​​assigned a value. For example, in spring, tree height growth has a relatively high importance to the long-term seasonal variation of forests, so it is assumed to be assigned a value of 0.8; diameter at breast height (DBH) growth has a relatively low importance, so it is assigned a value of 0.2. In summer, tree height growth has an importance of 0.6, and DBH growth has an importance of 0.4, etc.

[0114] Then, the initial impact weight information is multiplied by the values ​​in the time-feature correlation matrix and summed. If the initial impact weight of tree height is 0.4 and the initial impact weight of diameter at breast height (DBH) is 0.6, then the target impact weight of tree height in spring is 0.4 × 0.8 = 0.32, and the target impact weight of DBH in spring is 0.6 × 0.2 = 0.12. This process is repeated to calculate the target impact weight information for each season, thus obtaining the target impact weight information corresponding to the historical monitoring data sequence under the tree growth dimension.

[0115] S3023: Integrate the sequence feature components of the historical monitoring data sequences under each dimension within the second historical monitoring period to obtain the long-term seasonal variation characteristics of the monitoring area.

[0116] In one embodiment, integration can be based on the Analytic Hierarchy Process (AHP). First, the relative importance of the sequence feature components of historical monitoring data sequences under each dimension is determined. For example, in a monitoring area covering mountain forests, mountain grasslands, and mountain lakes, the relative importance of each sequence feature component related to tree growth under the forest dimension, the sequence feature component related to vegetation cover under the grassland dimension, and the sequence feature component related to water level change under the lake dimension is quantified by constructing a judgment matrix based on professional knowledge and understanding of the region's ecosystem. Then, the weight of each sequence feature component is calculated. Finally, each sequence feature component is multiplied by its corresponding weight and summed to obtain the long-term seasonal variation characteristics of the monitoring area. This method can systematically consider the relationships between various dimensions, thereby reasonably integrating different sequence feature components to accurately reflect the long-term seasonal variation of the monitoring area.

[0117] In one embodiment, reference Figure 6 Step S400 can be achieved by the following steps:

[0118] S401: Integrate the seasonal variation features and the deep geographic feature information to obtain comprehensive feature information.

[0119] Among them, comprehensive feature information is a new feature representation obtained by integrating seasonal variation features and deep geographic feature information. It integrates the dynamic patterns of seasonal variation and the potential impact of geographic factors on seasonal variation, and is a more comprehensive and integrated feature description. For example, in a grassland-forest interleaved ecosystem, comprehensive feature information may reflect the influence pattern of forest edge on grassland vegetation growth (which is related to seasonal variation features) during seasonal variation due to the geographic features of the forest (such as the shading and wind protection functions of trees, which are part of the deep geographic feature information). It is a comprehensive description that transcends seasonal dynamics and geographic factors.

[0120] In one embodiment, firstly, both seasonal variation features and deep geographic features are converted into vector form. Taking an urban park ecosystem as an example, assume that seasonal variation features can be represented by a vector containing elements such as flowering rates of plants in different seasons, bird activity frequency, and temperature variation range, with each element corresponding to one dimension of the vector. Deep geographic features may include the impact of topographic relief on the microclimate within the park (quantified as a numerical value), the impact of different soil types on vegetation distribution (represented by a numerical value), etc., and are also converted into vector form. Then, these two vectors are directly concatenated to obtain a new vector as the comprehensive feature information. For example, if the seasonal variation feature vector is [0.1, 0.2, 0.3] (this is just a simple example; in reality, there are more dimensions), and the deep geographic feature information vector is [0.4, 0.5], the concatenated comprehensive feature information vector is [0.1, 0.2, 0.3, 0.4, 0.5]. This method is simple and direct, and can retain most of the information of the original features. The weights are determined based on the importance of the seasonal variation features and deep geographic features to the final result. For example, in a farmland ecosystem, if based on past experience and data analysis, it is found that the seasonal pattern of precipitation in seasonal variation characteristics has a weight of 0.6 on the impact of crop growth, and the distribution of soil types in deep geographic feature information has a weight of 0.4 on the impact of crop growth.

[0121] Assuming the seasonal variation pattern of precipitation is quantified as 0.8 in the seasonal variation characteristics and the soil type distribution in the deep geographic features is quantified as 0.7, then the calculated comprehensive feature information is: (0.8 × 0.6 + 0.7 × 0.4) = 0.76. This method can integrate different features according to their importance, highlighting the features that have a greater impact on the results.

[0122] S402: Perform primary feature mining on the regional geographic information and the historical monitoring data sequence to obtain primary geographic feature information of the regional geographic information and primary monitoring feature information of the historical monitoring data sequence.

[0123] Primary geographic feature information refers to the feature information obtained from the preliminary mining of regional geographic information. For example, for a monitoring area in a coastal region, primary geographic feature information might be a simple topographic classification (such as whether the coast is a plain or hilly area) or a general classification of soil type (such as whether it is sandy soil or clay soil). This feature information is relatively basic and not as abstract and in-depth as deep geographic feature information, but it is still an important part of describing the geographic conditions.

[0124] In one embodiment, for regional geographic information, statistical analysis methods can be used to mine primary geographic feature information. Taking a marine ecosystem monitoring area containing multiple islands as an example, basic statistics such as the area and perimeter of each island are calculated as primary geographic feature information. The area range of the islands (e.g., small islands with an area of ​​1-5 square kilometers, and large islands with an area of ​​10-20 square kilometers) and the ratio of perimeter to area can be obtained through measurement and calculation. These statistics can preliminarily describe the geographic characteristics of the islands.

[0125] In one embodiment, for regional geographic information, primary geographic feature information can be mined according to some predefined rules. In a mountainous ecosystem, if areas with an altitude above 1000 meters are defined as high mountain areas and areas below 500 meters as low mountain areas, then classifying the mountains within the region according to this rule, the resulting distribution of high or low mountains constitutes the primary geographic feature information.

[0126] In one embodiment, a rule-based approach can also be used for historical monitoring data sequences. For example, regarding plant growth data, if a plant height increase exceeding 10 cm / year is defined as rapid growth, and less than 5 cm / year as slow growth, the plant growth data can be classified according to this rule. The resulting classification of plant growth rates constitutes the primary monitoring feature information. This rule-based approach can be customized based on specific domain knowledge and needs.

[0127] S 403: Integrate the primary geographic feature information and the primary monitoring feature information to obtain preliminary feature information.

[0128] Preliminary feature information is a new feature representation obtained by integrating primary geographic feature information and primary monitoring feature information. It integrates the preliminary features of geographic and monitoring data, providing a more comprehensive input for subsequent machine learning models. For example, in a grassland ecosystem, the preliminary feature information obtained by integrating soil type (e.g., sandy soil) from primary geographic feature information and vegetation cover (e.g., 30%) from primary monitoring feature information might be represented as a comprehensive feature of 30% vegetation cover in a sandy soil environment. This feature is significant for analyzing the overall condition of the grassland ecosystem.

[0129] In one embodiment, primary geographic feature information and primary monitoring feature information are combined according to a certain structure. Taking a farmland ecosystem as an example, the primary geographic feature information is soil type (e.g., loam), and the primary monitoring feature information is crop yield (e.g., 500 kg per acre). They can be combined into a new feature representation, such as "crop yield is 500 kg per acre in loam environment". This combination method can intuitively link the preliminary features of geographic and monitoring data.

[0130] S404: Using a machine learning model based on the comprehensive feature information and the preliminary feature information, predict the seasonal variation results of the monitored area.

[0131] Comprehensive feature information integrates seasonal variation characteristics and deep geographic features, providing a more comprehensive feature representation. Preliminary feature information, on the other hand, is a comprehensive feature that integrates primary geographic features and primary monitoring features. These two types of feature information describe the relevant characteristics of the monitoring area from different levels and perspectives, providing rich input information for machine learning models.

[0132] In one embodiment, the machine learning model is a pre-trained algorithmic model capable of making predictions based on input feature information. Examples include Support Vector Machine (SVM) models and Random Forest models. These models learn from large amounts of training data (containing feature information and corresponding seasonal variation results) to establish a relationship model between features and results, which is then used to predict new input feature information.

[0133] In one embodiment, a vector machine (SVM) model can be used for prediction, specifically:

[0134] First, the comprehensive feature information and the preliminary feature information are standardized to ensure they fall within the same numerical range. For example, feature values ​​are mapped to the interval [0, 1]. Then, the processed feature information is used as input to the SVM model. The SVM model uses a hyperplane to separate different seasonal change outcomes (such as rising or falling water levels). During the training phase, the SVM model searches for the optimal hyperplane that maximizes the margin between different categories.

[0135] When new comprehensive and preliminary feature information is input, the SVM model uses this hyperplane to determine which seasonal change outcome category the input feature belongs to, thereby achieving prediction. For example, in a lake ecosystem, the comprehensive feature information includes the relationship between the geographical features around the lake and seasonal changes, while the preliminary feature information includes primary monitoring features of the lake's water quality and primary features of the surrounding geography. The SVM model can use this information to predict whether the lake's water level will rise or fall in the future season.

[0136] In one embodiment, to improve prediction accuracy, step S404 may specifically include: using a machine learning model to predict the seasonal change results of the monitored area based on the comprehensive feature information and the preliminary feature information, including: combining the comprehensive feature information and the preliminary feature information in a specific ratio to form a hybrid feature vector;

[0137] The parameters of the selected machine learning model are adjusted based on the hybrid feature vector;

[0138] The seasonal variation results of the monitored area are predicted using a machine learning model with adjusted parameters based on the hybrid feature vector.

[0139] A hybrid feature vector is a vector obtained by combining comprehensive feature information and preliminary feature information in a specific ratio. This vector fuses feature information of two different levels and properties, forming an input format more suitable for machine learning models. For example, in an urban park ecosystem, a hybrid feature vector might contain a feature representation combining the distribution of vegetation types in different areas of the park (part of the comprehensive feature information) and the seasonal variation of visitor traffic (part of the preliminary feature information) in a certain ratio. It can more comprehensively reflect the characteristics of the urban park ecosystem, making it easier for models to analyze.

[0140] In one embodiment, the proportions can be determined by analyzing a large amount of historical data. In a large lake ecosystem, multi-year data on seasonal variations, along with corresponding comprehensive and preliminary characteristic information, are collected. Data mining techniques (such as association rule mining) are then used to analyze the strength of the association between different characteristic information and the final seasonal variation results.

[0141] Suppose that analysis reveals a strong correlation between the vertical distribution of lake water temperature and seasonal variation (part of the comprehensive feature information) and the final seasonal variation results (such as fish migration time), while the preliminary statistical data on lake periphery vegetation cover (part of the preliminary feature information) shows a relatively weak correlation. A proportion is determined based on the correlation strength, for example, 0.6 for comprehensive feature information and 0.4 for preliminary feature information. This proportion is then used to form a hybrid feature vector. This approach, based on actual data, more objectively reflects the importance of feature information to the prediction results.

[0142] Machine learning models are algorithmic models used for prediction, such as neural networks and decision trees. These models have adjustable parameters, which can be adjusted to optimize the model's fit to data and predictive accuracy. For example, weights and biases are important parameters in neural networks, and split thresholds are also adjustable parameters in decision trees.

[0143] Parameter tuning is the process of optimizing the parameters in a machine learning model based on the input blended feature vector. The goal is to enable the model to better adapt to the characteristics of the monitored area represented by the blended feature vector, thereby improving prediction accuracy. For example, in a neural network model, if a certain element in the blended feature vector (such as an element related to seasonal vegetation growth characteristics) has a strong influence on the final seasonal change result (such as the seasonal change in vegetation cover), then during parameter tuning, it is necessary to adjust the connection weights related to that element in the neural network so that the model gives more weight to this feature.

[0144] In one embodiment, for some gradient-based machine learning models (such as neural networks), the gradient descent algorithm can be used for parameter tuning. First, a loss function is defined to measure the difference between the model's predictions and the actual seasonal changes. For example, when predicting seasonal temperature changes in a mountain ecosystem, the loss function could be the mean squared error (MSE) between the predicted and actual temperatures.

[0145] In one embodiment, after parameter tuning, the mixed feature vector can be input into the parameter-adjusted neural network model. The neurons in each layer of the neural network model process the input according to a pre-defined activation function (such as the ReLU function). For example, in a neural network model for predicting seasonal changes in marine ecosystems, the input layer receives elements from the mixed feature vector, and then the hidden layers perform a non-linear transformation of the features.

[0146] The neurons in the hidden layer perform a weighted summation of the input based on weights and biases (these parameters have been adjusted), and then process the summation through an activation function. Finally, the output layer provides a prediction of seasonal changes based on the output of the hidden layers. For example, it can predict the distribution range of a certain fish species in the ocean during different seasons, or the trend of seawater salinity changes during seasonal transitions.

[0147] In one embodiment, the comprehensive feature information includes comprehensive sub-features of at least one dimension; the step of "combining the comprehensive feature information and the preliminary feature information in a specific ratio to form a hybrid feature vector" may include: performing interactive operations on the comprehensive sub-features of each dimension to obtain second-order interactive feature components corresponding to the comprehensive feature information; performing fully connected processing on the comprehensive feature information to obtain fully connected feature components corresponding to the comprehensive feature information; and combining the second-order interactive feature components, the fully connected feature components, the comprehensive feature information, and the preliminary feature information in a specific ratio to obtain a hybrid feature vector.

[0148] The second-order interactive feature component is obtained by performing interactive operations on the comprehensive sub-features of each dimension. This interactive operation aims to capture the second-order relationships between the comprehensive sub-features, that is, the new feature information generated by the interaction between the sub-features. For example, in an urban-suburban ecosystem, if the comprehensive sub-features are the relationship between the intensity of the urban heat island effect and seasonal variation, and the relationship between suburban farmland irrigation patterns and seasonal precipitation, the second-order interactive feature component may reflect complex relationships such as the indirect impact of the urban heat island effect on the irrigation demand of suburban farmland (affected by seasonal precipitation) in different seasons. It is a higher-level feature representation than a single comprehensive sub-feature.

[0149] In one embodiment, the interactive operation can be based on multiplication or on polynomial expansion.

[0150] Taking a forest-grassland ecotone ecosystem as an example, assume that the comprehensive feature information has two dimensions of comprehensive sub-features. In the ecological dimension, comprehensive sub-feature A is the relationship between forest vegetation diversity and seasonal temperature changes, quantified as a vector such as [0.1, 0.2, 0.3] (here it is assumed to be the correlation coefficient of different seasons); in the geographical dimension, comprehensive sub-feature B is the relationship between grassland soil type distribution and seasonal precipitation distribution, quantified as a vector [0.4, 0.5, 0.6].

[0151] Performing an interactive multiplication operation, multiplying corresponding elements of the two vectors, yields the second-order interactive feature component [0.1×0.4, 0.2×0.5, 0.3×0.6] = [0.04, 0.1, 0.18]. This multiplication operation can capture a synergistic relationship between the integrated sub-features. When the element values ​​corresponding to a certain season of two sub-features are larger, the corresponding elements in the resulting second-order interactive feature component are also larger, reflecting a stronger interactive relationship in that season.

[0152] Fully connected processing is an operation that comprehensively connects and integrates the various elements in a composite feature information. It's similar to the concept of a fully connected layer in a neural network, aiming to uncover deeper relationships within the composite feature information. During fully connected processing, each element may interact with other elements, resulting in new feature representations. For example, in a monitoring area containing multiple ecosystem types (forest, wetland, farmland), the element representing the relationship between forest vegetation height and seasonal precipitation in the composite feature information might develop new associations with the element representing the seasonal variation of wetland water levels through fully connected processing. These associations might not be obvious in the original composite sub-features.

[0153] Fully connected feature components are the result of fully connected processing. They are new feature components generated after deep integration of comprehensive feature information, reflecting more complex relationships between elements within the comprehensive feature information. For example, in a large lake ecosystem, the fully connected feature components obtained after fully connected processing of elements such as the relationship between lake area and seasonal evaporation, and the relationship between vegetation type around the lake and seasonal wind direction, may reflect complex relationships such as lake area indirectly affecting evaporation by influencing wind direction. This is a higher-level feature representation.

[0154] In one embodiment, a neural network structure can also be constructed, using comprehensive feature information as the input layer. For example, in a forest ecosystem, comprehensive sub-features such as the relationship between tree species diversity and seasonal precipitation, and the relationship between soil fertility and seasonal temperature, are used as the values ​​of input neurons. A fully connected layer is set in the neural network, and the neurons in this layer are fully connected to all neurons in the input layer. Through a forward propagation algorithm, the comprehensive feature information of the input layer is passed to the fully connected layer. The neurons in the fully connected layer process the input according to a preset activation function (such as the sigmoid function) to obtain fully connected feature components. This approach based on a neural network structure can utilize the powerful computing capabilities of neural networks to explore the relationships within comprehensive feature information, and the effect of fully connected processing can be optimized by adjusting the parameters of the neural network (such as weights and biases).

[0155] In summary, by adopting the technical solution of this application, and by acquiring regional geographic information, historical monitoring data sequences, and their collection time information of the monitoring area, and conducting in-depth feature mining, a comprehensive and in-depth understanding of various characteristic information related to the monitoring area can be obtained. Seasonal feature analysis based on the characteristic information of the surrounding reference areas can take into account the potential impact of the surrounding areas on the monitoring area, making the analysis of seasonal change characteristics more accurate. Furthermore, by using machine learning models to combine in-depth monitoring feature information, seasonal change characteristics, and in-depth geographic feature information to predict seasonal change results, and leveraging the powerful data analysis capabilities of machine learning, complex geographic and temporal data relationships can be handled. This overcomes the shortcomings of traditional methods, such as single data sources, insufficient accuracy, difficulty in handling complex geographic environments, and lack of in-depth data mining capabilities. Therefore, it is possible to more accurately predict the seasonal change results of the monitoring area, providing more reliable data support for climate change research, ecosystem protection, agricultural production planning, and many other aspects.

[0156] Accordingly, to better implement the above methods, this application also provides an artificial intelligence-based seasonal change monitoring system. For example... Figure 7As shown, the seasonal change monitoring system 70 includes an acquisition module 701, a feature mining module 702, a feature analysis module 703, and a prediction module 704, as detailed below:

[0157] The acquisition module 701 is used to acquire regional geographic information of the monitoring area, historical monitoring data sequences within at least one historical monitoring period, and the collection time information of the historical monitoring data sequences; the feature mining module 702 is used to perform feature depth mining on the regional geographic information and the historical monitoring data sequences within the historical monitoring period to obtain deep geographic feature information of the regional geographic information and deep monitoring feature information of the historical monitoring data sequences within the historical monitoring period; the feature analysis module 703 is used to perform seasonal feature analysis on the monitoring area based on reference deep monitoring feature information of the surrounding reference area within the historical monitoring period, the deep monitoring feature information, and the collection time information of the historical monitoring data sequences to obtain the seasonal change features of the monitoring area under different monitoring periods; the prediction module 704 is used to use a machine learning model to predict the seasonal change results of the monitoring area based on the deep monitoring feature information, the seasonal change features, and the deep geographic feature information.

[0158] The implementation details of each module are provided in the preceding method embodiments and will not be repeated here. The technical effects achieved by each module and device are described in the foregoing method embodiments.

[0159] It should be noted that, in practical implementation, the above modules can be arbitrarily combined and integrated into one or more modules, or implemented as independent entities. Furthermore, the above modules can be implemented in hardware or as software functional modules. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The aforementioned storage medium can be a read-only memory, a hard disk, or an optical disk, etc.

[0160] like Figure 8 As shown, this application embodiment also provides a computer device 80, characterized in that it includes a processor 801 and a memory 802, wherein the memory 802 stores a computer program, and when the computer program is executed by the processor 801, the processor 801 performs the steps of any of the methods described above.

[0161] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or modules, and may be electrical, mechanical, or other forms.

[0162] One embodiment of this application provides a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the method provided in one aspect of this application.

[0163] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0164] The above-disclosed embodiments are merely preferred embodiments of this application and should not be construed as limiting the scope of this application. Therefore, any equivalent variations made in accordance with the claims of this application shall still fall within the scope of this application.

Claims

1. A method for monitoring seasonal changes based on artificial intelligence, characterized in that, The method includes: Acquire regional geographic information of the monitoring area, historical monitoring data sequences within at least one historical monitoring period, and the collection time information of the historical monitoring data sequences; The regional geographic information and the historical monitoring data sequences within the historical monitoring period are subjected to in-depth feature mining to obtain in-depth geographic feature information of the regional geographic information and in-depth monitoring feature information of the historical monitoring data sequences within the historical monitoring period. The in-depth geographic feature information is highly abstract, representative, and reflects the inherent relationships of the regional geographic information obtained through in-depth mining. The in-depth monitoring feature information is obtained by in-depth mining of the historical monitoring data sequences, reflecting the complex relationships and patterns hidden behind the monitoring data. Based on the reference depth monitoring feature information of the surrounding reference area within the historical monitoring period, the depth monitoring feature information, and the acquisition time information of the historical monitoring data sequence, seasonal feature analysis is performed on the monitoring area to obtain the seasonal variation characteristics of the monitoring area under different monitoring periods; the historical monitoring data sequence within at least one historical monitoring period includes historical monitoring data sequences within a first historical monitoring period and a second historical monitoring period; the second historical monitoring period is earlier than the first historical monitoring period; the specific methods for obtaining the seasonal variation characteristics include: Based on the reference depth monitoring feature information, the depth monitoring feature information corresponding to the historical monitoring data sequence within the first historical monitoring period, and the collection time information corresponding to the historical monitoring data sequence within the first historical monitoring period, a short-term seasonal variation analysis is performed on the monitoring area to obtain the short-term seasonal variation characteristics of the monitoring area. Based on the depth monitoring feature information corresponding to the historical monitoring data sequence within the second historical monitoring period and the collection time information corresponding to the historical monitoring data sequence within the second historical monitoring period, a long-term seasonal variation analysis is performed on the monitoring area to obtain the long-term seasonal variation characteristics of the monitoring area. A machine learning model is used to predict the seasonal changes in the monitored area based on deep monitoring feature information, seasonal change features, and deep geographic feature information.

2. The method according to claim 1, characterized in that, Based on the reference depth monitoring feature information, the depth monitoring feature information corresponding to the historical monitoring data sequence within the first historical monitoring period, and the acquisition time information corresponding to the historical monitoring data sequence within the first historical monitoring period, a short-term seasonal variation analysis is performed on the monitoring area to obtain the short-term seasonal variation characteristics of the monitoring area, including: The first reference depth monitoring feature information of the reference area surrounding the monitoring area during the first historical monitoring period and the depth monitoring feature information corresponding to the historical monitoring data sequence during the first historical monitoring period are integrated to obtain integrated feature information. The integrated feature information and the collection time information corresponding to the historical monitoring data sequence within the first historical monitoring period are standardized to obtain the influence weight information corresponding to the historical monitoring data sequence within the first historical monitoring period. Based on the influence weight information, the historical monitoring data of each historical monitoring data sequence within the first historical monitoring period are comprehensively processed to obtain the short-term seasonal variation characteristics of the monitoring area.

3. The method according to claim 2, characterized in that, The integrated feature information is obtained by integrating the first reference depth monitoring feature information of the reference area surrounding the monitoring area within the first historical monitoring period and the depth monitoring feature information corresponding to the historical monitoring data sequence within the first historical monitoring period, including: An accumulation operation is performed on the first reference depth monitoring feature information and the depth monitoring feature information corresponding to the historical monitoring data sequence within the first historical monitoring period to obtain the accumulation operation result. A multiplication operation is performed on the reference depth monitoring feature information of the reference area surrounding the monitoring area within the first historical monitoring period and the depth monitoring feature information corresponding to the historical monitoring data sequence within the first historical monitoring period to obtain the multiplication operation result. The accumulated operation result, the multiplication operation result, the reference depth monitoring feature information, and the depth monitoring feature information corresponding to the historical monitoring data sequence within the first historical monitoring period are integrated to obtain integrated feature information.

4. The method according to claim 1, characterized in that, The historical monitoring data sequence within the second historical monitoring period includes historical monitoring data sequences under at least one dimension within the second historical monitoring period; The step of performing long-term seasonal variation analysis on the monitoring area based on the depth monitoring feature information corresponding to the historical monitoring data sequence within the second historical monitoring period and the collection time information corresponding to the historical monitoring data sequence within the second historical monitoring period, to obtain the long-term seasonal variation characteristics of the monitoring area, includes: For the historical monitoring data sequences under each dimension within the second historical monitoring period, the deep monitoring feature information corresponding to the historical monitoring data sequences under the dimensions is weighted and processed to obtain the initial influence weight information corresponding to the historical monitoring data sequences under the dimensions. Based on the initial influence weight information and the collection time information corresponding to the historical monitoring data sequence within the second historical monitoring period, the historical monitoring data sequence under the dimension is calculated to obtain the sequence feature components of the historical monitoring data sequence under the dimension. By integrating the sequence feature components of the historical monitoring data sequences under each dimension within the second historical monitoring period, the long-term seasonal variation characteristics of the monitoring area are obtained.

5. The method according to claim 4, characterized in that, The step of performing calculations on the historical monitoring data sequence under the dimension based on the initial influence weight information and the collection time information corresponding to the historical monitoring data sequence within the second historical monitoring period, to obtain the sequence feature components of the historical monitoring data sequence under the dimension, includes: The initial impact weight information is integrated with the collection time information corresponding to the historical monitoring data sequence within the second historical monitoring period to obtain the target impact weight information corresponding to the historical monitoring data sequence under the dimension. Based on the target influence weight information, the historical monitoring data sequence under the dimension is processed to obtain the sequence feature components of the historical monitoring data sequence under the dimension.

6. The method according to any one of claims 1-5, characterized in that, Using a machine learning model, the seasonal variation results of the monitored area are predicted based on deep monitoring feature information, the seasonal variation features, and the deep geographic feature information, including: By integrating the seasonal variation characteristics and the deep geographic feature information, comprehensive feature information is obtained; Primary feature mining is performed on the regional geographic information and the historical monitoring data sequence to obtain primary geographic feature information of the regional geographic information and primary monitoring feature information of the historical monitoring data sequence; the primary geographic feature information and the primary monitoring feature information are integrated to obtain preliminary feature information; A machine learning model is used to predict the seasonal changes in the monitored area based on the comprehensive feature information and the preliminary feature information.

7. The method according to claim 6, characterized in that, The method of using a machine learning model to predict seasonal changes in the monitored area based on the comprehensive feature information and the preliminary feature information includes: The comprehensive feature information and the preliminary feature information are combined in a specific ratio to form a hybrid feature vector. The specific ratio represents the correlation strength between the comprehensive feature information and the preliminary feature information on the seasonal change results. The parameters of the selected machine learning model are adjusted based on the hybrid feature vector; The seasonal variation results of the monitored area are predicted using a machine learning model with adjusted parameters based on the hybrid feature vector.

8. The method according to claim 7, characterized in that, The comprehensive feature information includes comprehensive sub-features of at least one dimension; The comprehensive feature information and the preliminary feature information are combined in a specific ratio to form a hybrid feature vector, including: Interact with the comprehensive sub-features of each dimension to obtain the second-order interactive feature components corresponding to the comprehensive feature information; The comprehensive feature information is processed by a fully connected layer to obtain the fully connected feature components corresponding to the comprehensive feature information. The second-order interactive feature components, the fully connected feature components, the comprehensive feature information, and the preliminary feature information are combined in a specific ratio to obtain a hybrid feature vector.

9. A seasonal variation monitoring system based on artificial intelligence, characterized in that, The system includes: The acquisition module is used to acquire regional geographic information of the monitoring area, historical monitoring data sequences within at least one historical monitoring period, and the acquisition time information of the historical monitoring data sequences. The feature mining module is used to perform in-depth feature mining on the regional geographic information and the historical monitoring data sequence within the historical monitoring period to obtain in-depth geographic feature information of the regional geographic information and in-depth monitoring feature information of the historical monitoring data sequence within the historical monitoring period. The in-depth geographic feature information is highly abstract, representative, and reflects inherent relationships, obtained through in-depth mining of the regional geographic information. The in-depth monitoring feature information is obtained through in-depth mining of the historical monitoring data sequence, reflecting the complex relationships and patterns hidden behind the monitoring data. The feature analysis module is used to perform seasonal feature analysis on the monitoring area based on reference depth monitoring feature information of the surrounding reference area within the historical monitoring period, the depth monitoring feature information, and the acquisition time information of the historical monitoring data sequence, to obtain the seasonal variation features of the monitoring area under different monitoring periods; the historical monitoring data sequence within at least one historical monitoring period includes historical monitoring data sequences within a first historical monitoring period and a second historical monitoring period; the second historical monitoring period is earlier than the first historical monitoring period; the acquisition method of the seasonal variation features specifically includes: performing short-term seasonal variation analysis on the monitoring area based on the reference depth monitoring feature information, the depth monitoring feature information corresponding to the historical monitoring data sequence within the first historical monitoring period, and the acquisition time information corresponding to the historical monitoring data sequence within the first historical monitoring period, to obtain the short-term seasonal variation features of the monitoring area; and performing long-term seasonal variation analysis on the monitoring area based on the depth monitoring feature information corresponding to the historical monitoring data sequence within the second historical monitoring period and the acquisition time information corresponding to the historical monitoring data sequence within the second historical monitoring period, to obtain the long-term seasonal variation features of the monitoring area. The prediction module is used to predict the seasonal change results of the monitored area based on the deep monitoring feature information, the seasonal change features, and the deep geographic feature information using a machine learning model.

Images