Intelligent water conservancy project management method based on ecological assessment and analysis

Abstract

The invention discloses an intelligent water conservancy project management method based on ecological evaluation and analysis, and the method comprises the steps: a multi-source ecological data intelligent collection step: building a water-land-air three-dimensional sensor monitoring network, employing a time-space synchronous sampling strategy, obtaining multi-dimensional ecological environment data in a water conservancy project influence region in real time, and carrying out the collection of the multi-source ecological data; the monitoring network comprises an underwater biological acoustic array, a floating type water quality sensor cluster, an unmanned aerial vehicle ecological remote sensing system and shore-based hyperspectral imaging equipment.

Description

Technical Field

[0001] This application relates to the field of water conservancy facility safety management technology, and more specifically, to a smart water conservancy project management method based on ecological assessment and analysis. Background Technology

[0002] With the deepening of ecological civilization construction and the widespread application of the concept of sustainable development, traditional water conservancy project management models can no longer meet the requirements of modern ecological protection. Existing water conservancy project management suffers from the following problems: a lack of systematic ecological monitoring systems, making it difficult to fully grasp the impact of water conservancy projects on the ecological environment; traditional manual inspection methods are inefficient and cannot achieve real-time dynamic monitoring; a lack of scientific ecological risk assessment methods makes it difficult to provide early warnings of ecological risks; and management decisions mainly rely on experience-based judgment, lacking data support and intelligent means.

[0003] Therefore, there is an urgent need to develop a smart water conservancy project management method that can realize real-time monitoring, intelligent assessment and automated management of the ecological environment. Summary of the Invention

[0004] 1. Technical problems to be solved

[0005] The purpose of this application is to provide a smart water conservancy project management method based on ecological assessment and analysis, which solves the technical problems mentioned in the background technology and achieves the technical effect of cleaning irrigation ditches in the field.

[0006] 2. Technical Solution

[0007] This application provides a smart water conservancy project management method based on ecological assessment and analysis, including the following steps:

[0008] (1) By constructing a three-dimensional sensor monitoring network covering water, land, and air, and adopting a spatiotemporal synchronous sampling strategy, multi-dimensional ecological and environmental data of the area affected by water conservancy projects can be acquired in real time.

[0009] (2) Establish a data quality evaluation system based on ecological spatiotemporal characteristics, use an improved Kalman filter algorithm to process sensor drift, use an ecological background self-learning algorithm to eliminate environmental noise, and use multi-scale spatiotemporal interpolation technology to complete missing data and achieve standardized fusion of heterogeneous ecological data.

[0010] (3) Construct a feature extraction framework based on ecological principles, use deep convolutional neural networks to extract the structural features of biological communities, use spectral-texture joint analysis technology to identify the distribution patterns of aquatic vegetation, extract the dynamic evolution features of the ecosystem through time series analysis algorithms, and establish a comprehensive ecological health index system.

[0011] (4) Establish a hybrid intelligent assessment model that integrates expert knowledge and data-driven approaches, use an improved long short-term memory network to predict the evolution trend of the ecosystem, use the Monte Carlo method to quantify uncertainty, construct a multi-scenario ecological risk assessment matrix, and realize dynamic hierarchical early warning of ecological risks.

[0012] (5) Establish a hierarchical decision-making mechanism based on the ecological carrying capacity threshold, use a multi-objective optimization algorithm to generate a coordinated scheme for ecological protection and engineering operation, realize intelligent linkage control of ecological protection equipment through Internet of Things technology, and establish a real-time feedback and adaptive adjustment mechanism for ecological management effect.

[0013] Furthermore, the monitoring network includes an underwater bioacoustic array, a floating water quality sensor cluster, an unmanned aerial vehicle (UAV) ecological remote sensing system, and a shore-based hyperspectral imaging device;

[0014] Furthermore, the underwater bioacoustic array adopts a biomimetic design, simulating the working principle of fish sonar systems. It uses multi-band acoustic wave detection technology to identify the distribution density and behavioral patterns of different types of fish. The time difference positioning algorithm between the acoustic array nodes achieves precise positioning in three-dimensional space, with a monitoring accuracy reaching the meter level.

[0015] Furthermore, ecological background self-learning algorithms include:

[0016] (a) Establish an ecological baseline model based on seasonal changes to automatically identify and eliminate data fluctuations caused by natural periodic changes;

[0017] (b) An unsupervised clustering algorithm is used to identify anomalous data patterns, and the authenticity of the anomalous data is verified by ecological mechanism constraints;

[0018] (c) Use spatiotemporal correlation analysis technology to cross-verify data from neighboring monitoring points to improve data reliability;

[0019] (d) Establish a dynamic data quality scoring mechanism to automatically adjust data weights based on the historical performance of sensors and environmental conditions.

[0020] Furthermore, the extraction of biological community structure features includes:

[0021] (a) An aquatic organism target detection model based on the improved YOLO algorithm is established, which can simultaneously identify fish, water birds and benthic animals;

[0022] (b) Graph convolutional neural networks were used to analyze the food web structure of biological communities and quantify the intensity of interactions between species;

[0023] (c) Using recurrent neural networks with attention mechanisms to analyze the temporal patterns of biological behavior and identify abnormal behavioral events;

[0024] (d) Establish a dynamic assessment model for biodiversity based on the Shannon diversity index and the Simpson dominance index.

[0025] Furthermore, the hybrid intelligent evaluation model includes: an expert knowledge module, a data-driven module, a knowledge fusion mechanism, and dynamic weight adjustment.

[0026] Furthermore, the multi-objective optimization algorithm is an improved non-dominated sorting genetic algorithm, the improvements of which include:

[0027] (a) Introduce niche protection mechanisms to avoid the loss of beneficial genes during the optimization process;

[0028] (b) Adaptive crossover and mutation probabilities are used to dynamically adjust genetic operation parameters based on population diversity;

[0029] (c) Establish a feasible remediation operator based on ecological constraints to ensure that the generated management plan meets the requirements of ecological protection;

[0030] (d) Design a multi-objective decision preference learning mechanism to automatically adjust the objective weights based on the manager's historical choices.

[0031] Furthermore, the floating water quality sensor cluster adopts a self-organizing network structure, and the sensor nodes realize data transmission through LoRa wireless communication technology, including a data acquisition layer, a data relay layer and a data aggregation layer, and has automatic fault detection and network reconstruction capabilities.

[0032] Furthermore, the multi-scale spatiotemporal interpolation technique includes:

[0033] (a) Spatial scale interpolation: Spatial interpolation is performed by combining the improved inverse distance weighting method with topographic and hydrological features;

[0034] (b) Time-scale interpolation: Select an appropriate interpolation method based on the time characteristics of ecological processes. Use linear interpolation for rapidly changing parameters and cubic spline interpolation for slowly changing parameters.

[0035] (c) Spatiotemporal coupling interpolation: The spatiotemporal kriging method is used to consider the temporal and spatial correlations simultaneously to establish a spatiotemporal covariance function model.

[0036] 10. The method according to claim 5, wherein the recurrent neural network of the attention mechanism comprises:

[0037] (a) Multi-head self-attention module: Simultaneously focuses on multiple key time points in the biological behavior sequence to extract behavioral features at different scales;

[0038] (b) Location Encoding Module: Adds temporal location information to the behavioral sequence to enhance the model's ability to understand temporal relationships;

[0039] (c) Residual connection mechanism: to prevent the gradient vanishing problem during deep network training and improve the model convergence speed;

[0040] (d) Layer normalization technique: accelerates the training process and improves the model's generalization ability.

[0041] 3. Beneficial effects

[0042] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0043] This application achieves comprehensive and multi-dimensional monitoring of the ecological environment of water conservancy projects by constructing a three-dimensional monitoring network covering water, land, and air.

[0044] This application employs deep learning and artificial intelligence technologies to improve the accuracy of ecological feature identification and risk assessment;

[0045] This application establishes an ecologically adaptive management mechanism, achieving coordination and unity between project operation and ecological protection. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the overall structure and waterway safety management process of the water conservancy facility safety management system disclosed in a preferred embodiment of this application; Detailed Implementation

[0047] The present application will be further described in detail below with reference to the accompanying drawings.

[0048] Reference Figure 1 A smart water conservancy project management method based on ecological assessment and analysis includes the following steps:

[0049] (1) Multi-source ecological data intelligent acquisition steps: By constructing a three-dimensional sensor monitoring network of water, land and air, and adopting a time-space synchronous sampling strategy, multi-dimensional ecological environment data of the water conservancy project area are acquired in real time. The monitoring network includes an underwater biological acoustic array, a floating water quality sensor cluster, an unmanned aerial vehicle ecological remote sensing system and a shore-based hyperspectral imaging device.

[0050] (2) Adaptive preprocessing steps for ecological big data: Establish a data quality evaluation system based on ecological spatiotemporal characteristics, use an improved Kalman filter algorithm to process sensor drift, use an ecological background self-learning algorithm to eliminate environmental noise, and use multi-scale spatiotemporal interpolation technology to complete missing data and achieve standardized fusion of heterogeneous ecological data.

[0051] (3) Intelligent analysis steps of multi-level features of ecosystem: Construct a feature extraction framework based on ecological principles, use deep convolutional neural networks to extract the structural features of biological communities, use spectral-texture joint analysis technology to identify the distribution patterns of aquatic vegetation, extract the dynamic evolution features of the ecosystem through time series analysis algorithms, and establish a comprehensive ecological health index system.

[0052] (4) Steps for intelligent assessment and prediction of ecological risks: Establish a hybrid intelligent assessment model that integrates expert knowledge and data-driven approaches, use an improved long short-term memory network to predict the evolution trend of the ecosystem, use the Monte Carlo method to quantify uncertainty, construct a multi-scenario ecological risk assessment matrix, and realize dynamic hierarchical early warning of ecological risks.

[0053] (5) Ecological Adaptive Smart Management Decision-Making Steps: Establish a hierarchical decision-making mechanism based on the ecological carrying capacity threshold, use a multi-objective optimization algorithm to generate a coordinated scheme for ecological protection and engineering operation, realize intelligent linkage control of ecological protection equipment through Internet of Things technology, and establish a real-time feedback and adaptive adjustment mechanism for ecological management effects.

[0054] As an embodiment of this application, in the multi-source ecological data intelligent acquisition step, the underwater biological acoustic array adopts a biomimetic design to simulate the working principle of fish sonar systems. It identifies the distribution density and behavior patterns of different types of fish through multi-band acoustic wave detection technology. The time difference positioning algorithm between the acoustic array nodes achieves precise positioning in three-dimensional space, and the monitoring accuracy reaches the meter level.

[0055] As an embodiment of this application, the ecological background self-learning algorithm in the ecological big data adaptive preprocessing step includes:

[0056] (a) Establish an ecological baseline model based on seasonal changes to automatically identify and eliminate data fluctuations caused by natural periodic changes;

[0057] (b) An unsupervised clustering algorithm is used to identify anomalous data patterns, and the authenticity of the anomalous data is verified by ecological mechanism constraints;

[0058] (c) Use spatiotemporal correlation analysis technology to cross-verify data from neighboring monitoring points to improve data reliability;

[0059] (d) Establish a dynamic data quality scoring mechanism to automatically adjust data weights based on the historical performance of sensors and environmental conditions.

[0060] As an embodiment of this application, the extraction of biological community structure features in the intelligent analysis step of the multi-level features of the ecosystem includes:

[0061] (a) An aquatic organism target detection model based on the improved YOLO algorithm is established, which can simultaneously identify fish, water birds and benthic animals;

[0062] (b) Graph convolutional neural networks were used to analyze the food web structure of biological communities and quantify the intensity of interactions between species;

[0063] (c) Using recurrent neural networks with attention mechanisms to analyze the temporal patterns of biological behavior and identify abnormal behavioral events;

[0064] (d) Establish a dynamic assessment model for biodiversity based on the Shannon diversity index and the Simpson dominance index.

[0065] As an embodiment of this application, the hybrid intelligent assessment model in the ecological risk intelligent assessment and prediction step includes:

[0066] (a) Expert knowledge module: A rule base is constructed based on ecological theory, including pollutant ecological effect thresholds, species sensitivity levels, and habitat suitability evaluation criteria;

[0067] (b) Data-driven module: Employs ensemble learning algorithms to fuse the prediction results of multiple machine learning models, thereby improving evaluation accuracy;

[0068] (c) Knowledge fusion mechanism: Integrating expert knowledge with data-driven results through Bayesian inference methods to handle knowledge conflicts and uncertainties;

[0069] (d) Dynamic weight adjustment: The weight ratio of expert knowledge and data-driven modules is dynamically adjusted based on historical prediction accuracy.

[0070] As an embodiment of this application, the multi-objective optimization algorithm in the ecological adaptive intelligent management decision-making step is an improved non-dominated sorting genetic algorithm, the improvement of which includes:

[0071] (a) Introduce niche protection mechanisms to avoid the loss of beneficial genes during the optimization process;

[0072] (b) Adaptive crossover and mutation probabilities are used to dynamically adjust genetic operation parameters based on population diversity;

[0073] (c) Establish a feasible remediation operator based on ecological constraints to ensure that the generated management plan meets the requirements of ecological protection;

[0074] (d) Design a multi-objective decision preference learning mechanism to automatically adjust the objective weights based on the manager's historical choices.

[0075] As an embodiment of this application, the floating water quality sensor cluster adopts a self-organizing network structure. Data transmission between sensor nodes is achieved through LoRa wireless communication technology. The communication distance of a single sensor node reaches 5 kilometers. The cluster network adopts a hierarchical topology structure, including a data acquisition layer, a data relay layer, and a data aggregation layer, and has automatic fault detection and network reconstruction capabilities.

[0076] As one embodiment of this application, the multi-scale spatiotemporal interpolation technique includes:

[0077] (a) Spatial scale interpolation: Spatial interpolation is performed by using an improved inverse distance weighting method combined with topographic and hydrological features, and the interpolation accuracy reaches more than 85%;

[0078] (b) Time-scale interpolation: Select an appropriate interpolation method based on the time characteristics of ecological processes. Use linear interpolation for rapidly changing parameters and cubic spline interpolation for slowly changing parameters.

[0079] (c) Spatiotemporal coupling interpolation: The spatiotemporal kriging method is used to consider the temporal and spatial correlations simultaneously to establish a spatiotemporal covariance function model.

[0080] As one embodiment of this application, the recurrent neural network of the attention mechanism includes:

[0081] (a) Multi-head self-attention module: Simultaneously focuses on multiple key time points in the biological behavior sequence to extract behavioral features at different scales;

[0082] (b) Location Encoding Module: Adds temporal location information to the behavioral sequence to enhance the model's ability to understand temporal relationships;

[0083] (c) Residual connection mechanism: to prevent the gradient vanishing problem during deep network training and improve the model convergence speed;

[0084] (d) Layer normalization technique: accelerates the training process and improves the model's generalization ability.

[0085] As an embodiment of this application, the method further includes an intelligent evaluation step of ecological management effectiveness, specifically including:

[0086] (a) Construct a Before-After-Control-Impact analysis framework, set up control and impact areas, and quantify the net effect of management measures;

[0087] (b) Use the difference-in-differences method to identify the causal relationship between management measures and ecological improvement, eliminating the influence of other confounding factors;

[0088] (c) Establish an ecological effect evaluation model based on Bayesian networks, taking into account the interaction of multiple ecological indicators;

[0089] (d) Construct a long-term tracking and evaluation system for management effectiveness and establish an ecological restoration time series database to provide a scientific basis for subsequent management decisions.

[0090] Example 1: Ecological Intelligent Management System for Large Reservoirs

[0091] This embodiment uses a large reservoir as the application scenario. The reservoir has a capacity of 5 billion cubic meters and involves a fish protection area and migratory bird habitat.

[0092] Step 1: Intelligent Collection of Multi-Source Ecological Data

[0093] 300 monitoring nodes will be deployed in the reservoir area, including:

[0094] Fifty sets of underwater bioacoustic arrays, employing multi-band detection technology from 40kHz to 200kHz, are used to monitor fish distribution and migration behavior;

[0095] 100 sets of floating water quality sensors monitor 15 water quality parameters, including water temperature, dissolved oxygen, pH value, and turbidity;

[0096] Ten sets of UAV ecological remote sensing systems, equipped with hyperspectral cameras and multispectral sensors, are used to monitor the distribution of aquatic vegetation.

[0097] Twenty sets of shore-based hyperspectral imaging equipment were used to analyze changes in the optical properties of water bodies.

[0098] The BeiDou satellite timing system is used to synchronize the time of all devices, and the data acquisition frequency is once every 15 minutes.

[0099] Step 2: Adaptive Preprocessing of Ecological Big Data

[0100] An ecological baseline database containing 10 years of historical data was established, and a seasonal decomposition algorithm was used to identify natural cyclical changes. An improved Kalman filter algorithm was applied to handle sensor drift, improving filtering accuracy by 30%. The DBSCAN clustering algorithm was used to identify outlier data, achieving an accuracy of 95%.

[0101] Step 3: Intelligent Analysis of Multi-Level Ecosystem Characteristics

[0102] A ResNet-50-based aquatic organism identification model was constructed, capable of identifying 35 fish species and 20 waterbird species with an accuracy rate of 92%. A graph convolutional neural network was used to analyze the food web structure and identify key species nodes. An LSTM network was used to analyze fish migration time-series patterns, achieving a prediction accuracy of 85%.

[0103] Step 4: Intelligent assessment and prediction of ecological risks

[0104] An assessment system was established that includes three dimensions: pollution risk, habitat degradation risk, and biodiversity loss risk. An ensemble learning algorithm was used to combine three models: random forest, support vector machine, and neural network, resulting in a 15% improvement in prediction accuracy compared to a single model.

[0105] Step 5: Ecologically Adaptive Smart Management Decisions

[0106] The reservoir scheduling scheme is optimized based on the NSGA-III algorithm to maximize ecological benefits while meeting water supply needs. Sixty sets of ecological protection equipment, including aeration devices and fish passage gates, are controlled through an IoT platform.

[0107] After implementing the system, the fish diversity index of the reservoir increased by 18%, the water quality compliance rate increased to 98%, and management efficiency increased by 40%.

[0108] Example 2: Intelligent Management System for River Ecological Corridors

[0109] This embodiment uses a 100-kilometer section of a river in a certain basin as the application scenario. This section includes three cascade hydropower stations and two national wetland protection areas.

[0110] Step 1: Intelligent Collection of Multi-Source Ecological Data

[0111] 200 monitoring nodes will be set up along the river:

[0112] Thirty sets of underwater bioacoustic arrays were deployed to monitor fish migration channels.

[0113] 80 sets of floating sensors are used to monitor water quality and hydrological parameters;

[0114] Six unmanned aerial vehicle (UAV) systems are deployed for regular patrols and monitoring.

[0115] Fifteen shore-based monitoring stations are in place to monitor the ecological status of wetlands.

[0116] Step 2: Adaptive Preprocessing of Ecological Big Data

[0117] A data preprocessing model based on river hydrological characteristics was established, taking into account seasonal variations between flood and dry seasons. Spatiotemporal kriging interpolation was used to process missing data, achieving an interpolation accuracy of 90%.

[0118] Step 3: Intelligent Analysis of Multi-Level Ecosystem Characteristics

[0119] A fish migration behavior identification model was established, capable of identifying three behavioral states: upstream, downstream, and stationary. Remote sensing image analysis technology was used to monitor wetland vegetation changes, achieving a vegetation classification accuracy of 88%.

[0120] Step 4: Intelligent assessment and prediction of ecological risks

[0121] The assessment will focus on the impact of hydropower stations on fish migration and the health of wetland ecosystems. An early warning mechanism based on ecological thresholds will be established, automatically triggering an alert when ecological indicators exceed the thresholds.

[0122] Step 5: Ecologically Adaptive Smart Management Decisions

[0123] The coordinated operation of three hydropower stations ensures ecological flow for fish migration while meeting power generation needs. An automatic control system regulates water levels and flow rates in the fish passages to improve fish passage efficiency.

[0124] After implementing the system, the success rate of fish migration increased by 25%, and the wetland vegetation coverage remained stable, achieving a win-win situation for both power generation and ecological protection.

Claims

1. A smart water conservancy project management method based on ecological evaluation and analysis, characterized in that, The method comprises the following steps: (1) Real-time acquisition of multi-dimensional ecological environment data in the water conservancy project affected area by constructing a water-land-air three-dimensional sensor monitoring network and adopting a space-time synchronous sampling strategy, (2) Standardized fusion of heterogeneous ecological data by establishing a data quality evaluation system based on ecological space-time characteristics, processing sensor drift by using an improved Kalman filtering algorithm, eliminating environmental noise by using an ecological background self-learning algorithm, and completing missing data by using a multi-scale space-time interpolation technology; (3) Extraction of biological community structure features by using a deep convolutional neural network, identification of aquatic vegetation distribution patterns by using a spectral-texture joint analysis technology, and extraction of ecological system dynamic evolution features by using a time series analysis algorithm, and establishment of an ecological health comprehensive index system; (4) Establishment of a hybrid intelligent evaluation model integrating expert knowledge and data driving, prediction of ecological system evolution trend by using an improved long short-term memory network, quantification of uncertainty by using a Monte Carlo method, construction of a multi-scenario ecological risk assessment matrix, and realization of dynamic grading early warning of ecological risk; (5) Establishment of a hierarchical decision mechanism based on ecological carrying capacity threshold, generation of a coordinated scheme of ecological protection and engineering operation by using a multi-objective optimization algorithm, realization of intelligent linkage control of ecological protection equipment by using Internet of Things technology, and establishment of a real-time feedback and self-adaptive adjustment mechanism of ecological management effect.

2. The method of claim 1, wherein: the monitoring network comprises an underwater biological acoustic array, a floating water quality sensor cluster, an unmanned aerial vehicle ecological remote sensing system, and a shore-based hyperspectral imaging device.

3. The method of claim 2, wherein: the underwater biological acoustic array is designed based on bionics, simulates the working principle of fish sonar system, identifies the distribution density and behavior pattern of different types of fish groups by using multi-band sound wave detection technology, and realizes accurate positioning in three-dimensional space by using time difference positioning algorithm between acoustic array nodes, with a monitoring accuracy of meter level.

4. The method of claim 1, wherein: The ecological background self-learning algorithm comprises: (a) Establishing an ecological baseline model based on seasonal changes to automatically identify and eliminate data fluctuations caused by natural periodic changes; (b) Identifying abnormal data patterns by using unsupervised clustering algorithm, and verifying the authenticity of abnormal data by using ecological mechanism constraint; (c) Using space-time correlation analysis technology to verify each other by using data of adjacent monitoring points to improve the reliability of data; (d) Establishing a dynamic data quality scoring mechanism to automatically adjust data weights according to sensor historical performance and environmental conditions.

5. The method of claim 1, wherein: The biological community structure feature extraction comprises: (a) Establishing a water-borne biological target detection model based on an improved YOLO algorithm, which can simultaneously identify fish, water birds, and benthic animals; (b) Using a graph convolutional neural network to analyze the food web structure of biological communities to quantify the interaction strength between species; (c) Using a recurrent neural network with attention mechanism to analyze the time series pattern of biological behavior to identify abnormal behavior events; (d) Establishing a dynamic evaluation model of biological diversity based on Shannon diversity index and Simpson dominance index.

6. The method of claim 1, wherein: The hybrid intelligent evaluation model comprises an expert knowledge module, a data-driven module, a knowledge fusion mechanism and a dynamic weight adjustment.

7. The method of claim 1, wherein: The multi-objective optimization algorithm is an improved non-dominated sorting genetic algorithm, and the improvements include: (a) introducing a niche protection mechanism to avoid the loss of beneficial genes during the optimization process; (b) using adaptive crossover and mutation probabilities to dynamically adjust genetic operation parameters according to population diversity; (c) establishing a feasibility repair operator based on ecological constraints to ensure that the generated management scheme meets the requirements of ecological protection; (d) designing a multi-objective decision preference learning mechanism to automatically adjust the target weight according to the manager's historical selection.

8. The method of claim 2, wherein: The floating water quality sensor cluster adopts a self-organizing network structure, and the sensor nodes realize data transmission through LoRa wireless communication technology, including a data acquisition layer, a data relay layer and a data aggregation layer, and has the capabilities of automatic fault detection and network reconstruction.

9. The method of claim 4, wherein: The multi-scale spatiotemporal interpolation technology comprises: (a) spatial scale interpolation: using an improved inverse distance weighting method combined with terrain and hydrological characteristics for spatial interpolation; (b) time scale interpolation: selecting appropriate interpolation methods based on the time characteristics of ecological processes, using linear interpolation for fast-changing parameters and cubic spline interpolation for slow-changing parameters; (c) spatiotemporal coupling interpolation: using spatiotemporal Kriging method to consider time and space correlation simultaneously, and establishing a spatiotemporal covariance function model.

10. The method of claim 5, wherein: The attention mechanism recurrent neural network comprises: (a) multi-head self-attention module: simultaneously focusing on multiple key time points in the biological behavior sequence to extract behavior features of different scales; (b) position encoding module: adding time position information to the behavior sequence to enhance the model's understanding ability of time sequence relationship; (c) residual connection mechanism: preventing gradient vanishing problem in deep network training process and improving model convergence speed; (d) layer normalization technology: accelerating the training process and improving the generalization ability of the model.

Images