Ecological function prediction method based on deep learning and ecosystem service

Abstract

The invention relates to an ecological function prediction method based on deep learning and ecological system service, and the method comprises the following steps: data collection and preprocessing: collecting various types of data in a research region, and preprocessing the collected data; determining network nodes and variables: selecting key variables as nodes in the Bayesian network according to the demand of ecological system function prediction; constructing a Bayesian network structure: based on ecological knowledge and existing research, analyzing a causal relationship between nodes, and determining a connection mode and direction between the nodes; model parameter learning: parameter learning can be carried out through methods of maximum likelihood estimation, Bayesian estimation and the like; performing model evaluation and optimization, result interpretation and management suggestion; the ecological system function prediction method established by the method not only can improve the management efficiency of the ecological environment, but also can provide a scientific basis for coping with climate changes, protecting water resources and the like, and promotes the realization of sustainable development.

Description

Technical Field

[0001] The present invention relates to the field of ecosystem function monitoring and management, and specifically to an ecological function prediction method based on deep learning and ecosystem services. Background Art

[0002] An ecosystem is a complex system that includes biological, abiotic and environmental factors. Human activities affect ecosystem functions by changing the structure, composition and environmental conditions of ecosystems. These impacts include a reduction in species diversity, reduced ecosystem stability and productivity, and a decline in water and soil quality. With the increasing human intervention and impact on the natural environment, the functions and stability of ecosystems are seriously threatened. In order to protect and restore the health of ecosystems, accurate predictions of changes in ecosystem functions are needed, which can help people better understand and assess the impact of human activities on ecosystems. Most importantly, measures can be taken in advance for protection and intervention to prevent or reduce damage to the ecosystem. In addition, predicting changes in ecosystem functions can also provide a scientific basis for decision makers to formulate more reasonable and effective protection policies.

[0003] With the rapid development of science and technology, artificial intelligence (AI) technology has been widely used in various fields, including geography and ecology. The introduction of artificial intelligence technology provides new ideas and methods for solving complex geographical problems. By combining artificial intelligence technology, a large amount of geographic data can be automatically processed and analyzed, which greatly improves work efficiency and the accuracy of research results. As an important branch of machine learning, deep learning has made remarkable progress in many fields in recent years. Its powerful feature learning ability and efficient computing power make deep learning perform well in processing complex and large-scale data. Among them, Bayesian network, as a deep learning method based on probabilistic graphical model, has unique advantages in dealing with uncertainty problems and causal reasoning. Applying deep learning to ecosystem service prediction can overcome the limitations of traditional methods and improve the accuracy of prediction. At present, deep learning has achieved some preliminary results in the field of ecosystem service prediction. For example, researchers have used deep learning models to predict and evaluate forest carbon storage, water quality, biodiversity, etc. These research results show that deep learning models perform well in processing complex ecosystem data and can significantly improve the accuracy and efficiency of prediction. In the past few decades, the prediction of changes in watershed ecosystem functions has mainly relied on traditional geographic information systems (GIS) and environmental models. These methods usually require tedious processing and analysis of a large amount of environmental data, and the prediction results are often not accurate. In recent years, with the development of artificial intelligence technology, especially the popularization of deep learning and machine learning algorithms, the prediction model of watershed ecosystem function change based on artificial intelligence has gradually become a research hotspot. Although traditional GIS and environmental models have a certain role in predicting watershed ecosystem function change, they have some limitations. For example, they often require a lot of manpower, time and computing resources, and the prediction results are easily affected by data quality, model parameters and assumptions. In addition, these models are usually unable to handle complex nonlinear relationships and spatiotemporal changes. Therefore, developing an efficient and accurate watershed ecosystem function change prediction model that can handle complex nonlinear relationships and spatiotemporal changes is an urgent problem to be solved. In addition, with the continuous improvement of environmental awareness and the popularization of the concept of sustainable development, the prediction and evaluation of watershed ecosystem function changes will also receive more attention and attention. Summary of the invention

[0004] In response to the above-mentioned problems, the present invention proposes an ecological function prediction method based on deep learning and ecosystem services, which provides a scientific basis for protecting the river basin ecological environment and sustainably utilizing water resources.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A method for predicting ecological functions based on deep learning and ecosystem services, the method comprising the following steps:

[0007] S01. Data collection and preprocessing: Collect various data in the basin and preprocess the collected data, including remote sensing images, meteorological data and soil information;

[0008] S02. Determine network nodes and variables: According to the needs of ecosystem function prediction, select key variables as nodes in the Bayesian network. These nodes can represent biological, environmental factors or functional indicators in the ecosystem. Clarify the specific meaning, type (discrete or continuous) and value range of each node;

[0009] S03. Construct the Bayesian network structure: First, based on ecological theory and expert knowledge, select variables that have an important impact on the prediction target as nodes in the Bayesian network. Then analyze the causal relationship between the nodes, and clarify which nodes are parent nodes (cause) and which nodes are child nodes (result). Finally, draw the structure diagram of the Bayesian network based on the causal relationship obtained by analysis. Each node in the diagram represents a variable, and the directed edges represent the causal relationship between the variables;

[0010] S04, model parameter learning: First, set the prior distribution for each node, which can be based on expert knowledge, historical data or default distribution. Then use the collected data set to train the Bayesian network model through machine learning algorithms (such as maximum likelihood estimation, Bayesian estimation, etc.) to learn the conditional probability table (CPT) of each node. CPT describes the probability of a child node taking each value when a parent node takes a given value;

[0011] S05, Model evaluation and optimization: Use the test set to evaluate the trained model. The evaluation indicators include accuracy, recall, F1 value, etc. In addition to cross-validation, information criteria and sensitivity analysis are also commonly used evaluation methods. According to the evaluation results, the model is optimized and adjusted again to improve the prediction accuracy and stability;

[0012] S06. Interpretation of results and management recommendations: Apply the optimized model to actual data to predict changes in ecosystem functions in the study area; at the same time, interpret the prediction results, understand the model's prediction basis and influencing factors, and provide corresponding management recommendations to provide a scientific basis for protecting the ecological environment.

[0013] As a further technical solution of the present invention, in step S01, the actions of preprocessing the collected data include: image denoising, geometric correction and fusion.

[0014] As a further technical solution of the present invention, in step S02, each node represents a random variable, which can be an observed data point or a potential unknown parameter. A directed edge represents the causal relationship between variables, pointing from the parent node to the child node, reflecting the direct dependency between these variables.

[0015] As a further technical solution of the present invention, in step S03, these "variables that have a significant impact on the prediction target" specifically refer to those variables that can directly or indirectly affect the change trend, state or result of the prediction target. Specifically, these variables may include:

[0016] 1) Directly related variables, i.e. variables that are directly involved in the generation or change process of the forecast target. For example, when predicting the rainfall in a certain area, the local humidity, cloud cover, wind direction, etc. may be directly related variables.

[0017] 2) Indirect influence variables, that is, variables that do not directly participate in the generation or change process of the forecast target, but indirectly affect the forecast target by affecting other variables. For example, when predicting economic growth rate, policy changes, technological innovation, etc. may be indirect influence variables.

[0018] 3) Background variables, which are variables that provide background information or context for the prediction target. These variables may not directly affect the prediction target, but they help to understand and interpret the prediction results more accurately. For example, when predicting the impact of climate change on agricultural yields, soil type, irrigation conditions, etc. may be background variables.

[0019] When selecting these variables, it is necessary to comprehensively consider the domain knowledge of the problem, actual needs, and the availability and quality of data. Through consultation with domain experts, data analysis, or experimental verification, variables that have a significant impact on the prediction target can be screened out and modeled as nodes in the Bayesian network.

[0020] As a further technical solution of the present invention, in step S03, the core idea of ​​the Bayesian network is to describe the conditional probability distribution between variables through a directed acyclic graph based on Bayes' theorem. Each node has a conditional probability table (CPT) that describes the probability distribution of the node state under a given parent node state. This structure enables the Bayesian network to effectively handle complex joint probability distribution problems, because the joint probability grows exponentially with the increase in the number of variables, and direct calculation is very difficult.

[0021] As a further technical solution of the present invention, in step S03, the core idea Bayesian theorem is the theoretical basis of Bayesian network. Bayesian theorem can be expressed by conditional probability formula, which describes the probability of another event occurring under the premise of knowing the probability of occurrence of certain related events. Bayesian theorem can be expressed by the following formula:

[0022]

[0023] Among them, P(A) and P(B) represent the prior probabilities of event A and event B respectively (that is, the estimated probabilities of event A and event B before considering other information); P(B|A) represents the posterior probability of event B occurring after assuming that event A has occurred (also called conditional probability, that is, the new estimate of the probability of event B); P(A|B) is the same as P(B|A).

[0024] As a further technical solution of the present invention, in step S03, the structure of the Bayesian network is a directed acyclic graph (DAG), in which each node represents a random variable and directed edges represent dependencies between variables.

[0025] A directed acyclic graph (DAG) is a graph structure consisting of a set of vertices and a set of directed edges, where the direction of the edge indicates the relationship between the vertices, and there is no path that forms a loop in the graph. If there is a directed edge from node A to node B, then we believe that A is the parent node of B and B is the child node of A. In a Bayesian network, there is no loop, that is, starting from a node and walking along a directed edge, you will not return to the original node. Specifically, a directed acyclic graph satisfies the following two conditions:

[0026] (1) The edges in the graph are directed, that is, they point from one vertex to another;

[0027] (2) There is no path in the graph that can start from a vertex and return to the vertex itself through several edges.

[0028] Nodes are added to the graph in sequence according to the causal relationship between variables. Each time a node is added, its position in the graph and the edges connected to it need to be determined. Based on domain knowledge and statistical results, the network structure is adjusted to ensure that it can accurately reflect the dependencies between variables.

[0029] As a further technical solution of the present invention, in step S04, the prior distribution refers to the initial belief or probability of the occurrence of an event before any data is observed. It is usually derived based on information such as domain knowledge, historical data or personal experience. For example, the prevalence of a disease in a certain area is 1%, that is, if we randomly select a person, the prior probability that the person is sick is 1%. The prior distribution plays a role in constraining parameter estimation, which can make the inference result more accurate and reliable. In addition, there is also a corresponding posterior distribution, which refers to the new probability distribution calculated using Bayes' theorem and prior distribution after obtaining experimental or observed data, called the posterior distribution. The posterior distribution represents our understanding of the probability distribution of parameters after obtaining experimental or observed data.

[0030] As a further technical solution of the present invention, in step S04, the maximum likelihood estimation (MLE) is a commonly used parameter estimation method, which estimates the value of the unknown parameter by finding the parameter value that maximizes the probability of the observed sample based on the observation results of the sample data. Its formula is based on the likelihood function, and the formula is as follows:

[0031]

[0032] Among them, L(θ|X) is the likelihood function, which represents the probability of observing data X under given parameters θ. is the maximum likelihood function estimator, that is, the parameter value that maximizes the likelihood function.

[0033] As a further technical solution of the present invention, in step S04, the Bayesian estimation is another commonly used parameter estimation method, which is based on the Bayesian theorem and estimates parameters by updating the posterior distribution of the parameters based on the prior distribution and the observed data. Among them, the maximum likelihood estimation focuses on the maximum possibility of the data, without considering the prior knowledge; while the Bayesian estimation combines the prior knowledge and the new data to update the estimation of the parameters.

[0034] As a further technical solution of the present invention, in step S04, in the Bayesian network, each node has a conditional probability table (CPT) associated with it, which describes the probability distribution of the node state given the parent node state. This enables the Bayesian network to calculate the joint probability distribution of the entire network through the chain rule.

[0035] Conditional Probability Table (CPT): is a two-dimensional matrix, an important mathematical tool used to describe the probability relationship between random variables under given conditions. CPT is mainly used in Bayesian networks to show the dependency relationship between variables in an intuitive and structured way. It combines the structure of directed acyclic graph (DAG) and together constitutes the core of Bayesian networks. The probability values ​​in CPT can be learned from observed data or set by expert knowledge.

[0036] As a further technical solution of the present invention, in step S05, the evaluation indicators include accuracy, recall rate, F1 value, etc.;

[0037] Among them, accuracy is the proportion of samples correctly predicted by the model, and the formula is as follows:

[0038] (TP+TN) / (TP+FP+TN+FN);

[0039] Recall refers to the ratio of correctly predicted positive samples to all positive samples. The formula is as follows:

[0040] TP / (TP+FN);

[0041] The F1 value (F1score) is the harmonic average of precision and recall, combining the performance of both. The formula is as follows:

[0042] F1 = 2PR / (P+R);

[0043] Among them, TP: the sample is positive and the prediction result is positive; FP: the sample is negative and the prediction result is positive;

[0044] TN: The sample is negative and the predicted result is negative; FN: The sample is positive and the predicted result is negative.

[0045] As a further technical solution of the present invention, in step S05, the information criteria commonly used include Bayesian Information Criterion (BIC) and Akaike Information Criterion (AIC). These criteria can help us find a suitable balance point in the selection of network structure and parameters to avoid the problem of overfitting or underfitting of the model.

[0046] Akaike Information Criterion (AIC) is based on the concept of information theory and is used to measure the goodness of fit of a statistical model. The formula is as follows:

[0047] AIC = 2k-2ln(L);

[0048] Where: L is the maximum likelihood estimate of the model, which reflects the degree of fit of the model to the data set. k is the number of parameters in the model, representing the complexity of the model.

[0049] Bayesian Information Criterion (BIC) is an information criterion proposed by the Bayesian school. It is similar to the AIC criterion and is also used to weigh the complexity and goodness of fit of the model. The formula is as follows:

[0050] BIC = kln(n)-2ln(L);

[0051] Where: k is the number of parameters in the model, L is the maximum likelihood estimate of the model, and n is the number of samples.

[0052] As a further technical solution of the present invention, in step S05, the sensitivity analysis is a method for evaluating the sensitivity of the model to these changes by changing the parameters or input data of the model. In the Bayesian network, we can evaluate the stability of the network structure and parameters, as well as the robustness of the model through sensitivity analysis. This can help us better understand the behavior of the model, so as to have more confidence in the prediction results of the model.

[0053] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention provides an ecological function prediction method based on deep learning and ecosystem services. The present invention integrates multi-source geographic information data, comprehensive analysis of time series data, selection, training and optimization of appropriate prediction models, and improves the accuracy and stability of prediction. The present invention utilizes deep learning technology, combined with remote sensing data and the concept of ecosystem services, to achieve high-precision prediction of changes in ecosystem functions in the study area. The present invention combines spatiotemporal data to comprehensively, dynamically and in real time monitor the changing trends of ecosystem functions in the study area, providing timely reference for decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 Flowchart established for the ecological function prediction method based on deep learning and ecosystem services.

[0055] Figure 2 A simple Bayesian network diagram example.

[0056] Figure 3 This is an example of a more complex Bayesian network diagram.

[0057] Figure 4 Technology roadmap for Bayesian network modeling. DETAILED DESCRIPTION

[0058] The technical solution of this patent is further described in detail below in conjunction with specific implementation methods.

[0059] As an embodiment of the present invention, an ecological function prediction method based on deep learning and ecosystem services, the model is established by the following steps:

[0060] S01. Data collection and preprocessing:

[0061] Collect various data in the watershed, including remote sensing images, meteorological data and soil information;

[0062] The collected data is preprocessed, and the actions of preprocessing the collected data include: image denoising, geometric correction and fusion.

[0063] The processed data set is divided into a training set and a test set for model training and evaluation. The holdout method and K-fold cross-validation method are usually used to randomly divide the data set into training sets and test sets in proportion, and a part of the training set is divided as a validation set. If the data set has obvious time series characteristics, the data set is usually divided into training sets and test sets according to the time sequence, and the latest part of the data is selected as the test set, and the rest is used as the training set and validation set. This step can be implemented using the sort_values ​​function. The training set (trainingset) is used to run the learning algorithm and train the model; the validation set (developmentset) is used to adjust the model hyperparameters; the test set (testset) is used to evaluate the performance of the selected model and predict the results;

[0064] S02. Determine network nodes and variables: According to the needs of ecosystem function prediction, select key variables as nodes in the Bayesian network. These nodes can be environmental data (such as climate, soil, hydrology, etc.), ecological data (such as species distribution, population size, etc.) and socio-economic data (such as human activity intensity, policy impact, etc.). Then clarify the specific meaning, type (discrete or continuous) and value range of each node;

[0065] S03. Construct the Bayesian network structure: First, based on ecological theory and expert knowledge, select variables that have an important impact on the prediction target as nodes in the Bayesian network. Then analyze the causal relationship between the nodes, and clarify which nodes are parent nodes (cause) and which nodes are child nodes (result). Finally, draw the structure diagram of the Bayesian network based on the causal relationship obtained by analysis. Each node in the diagram represents a variable, and the directed edges represent the causal relationship between the variables;

[0066] S04, model parameter learning: First, set the prior distribution for each node, which can be based on expert knowledge, historical data or default distribution. Then use the collected data set to train the Bayesian network model through machine learning algorithms (such as maximum likelihood estimation, Bayesian estimation, etc.) to learn the conditional probability table (CPT) of each node. CPT describes the probability of a child node taking each value when a parent node takes a given value;

[0067] S05, Model evaluation and optimization: Use the test set to evaluate the trained model. The evaluation indicators include accuracy, recall, F1 value, etc. In addition to cross-validation, information criteria and sensitivity analysis are also commonly used evaluation methods. According to the evaluation results, the model is optimized and adjusted again to improve the prediction accuracy and stability;

[0068] S06. Interpretation of results and management recommendations: Apply the optimized model to actual data to predict changes in ecosystem functions in the study area; at the same time, interpret the prediction results, understand the model's prediction basis and influencing factors, and provide corresponding management recommendations to provide a scientific basis for protecting the ecological environment.

[0069] Throughout the entire process, the geographical knowledge and actual needs of the study area need to be fully considered to ensure the reliability and practicality of the model. During the model construction process, multiple experiments and adjustments may be required to find the model architecture and parameter configuration that best suits the current study. The interpretation of the results and management recommendations need to be based on professional knowledge and field research to ensure the scientificity and feasibility of the prediction results.

[0070] In this embodiment, in step S02, each node represents a random variable, which can be an observed data point or a potential unknown parameter. A directed edge represents the causal relationship between variables, pointing from a parent node to a child node, reflecting the direct dependency between these variables.

[0071] See also Figure 2 In this embodiment, in step S03, the structure of the Bayesian network is a directed acyclic graph (DAG), in which each node represents a random variable and the directed edges represent the dependency relationship between the variables.

[0072] Specifically, a directed acyclic graph satisfies the following two conditions:

[0073] (1) The edges in the graph are directed, that is, they point from one vertex to another;

[0074] (2) There is no path in the graph that can start from a vertex and return to the vertex itself through several edges.

[0075] Nodes are added to the graph in sequence according to the causal relationship between variables. Each time a node is added, its position in the graph and the edges connected to it need to be determined. Based on domain knowledge and statistical results, the network structure is adjusted to ensure that it can accurately reflect the dependencies between variables. Figure 2 The figure shows a simple Bayesian network. Since a leads to b, a and b lead to c, and a has no parent node as the prior probability, its joint probability can be expressed as:

[0076] p(a,b,c)=p(c|a,b)p(b|a)p(a);

[0077] Figure 3 is a more complex Bayesian network: Similarly, we can see that x1, x2, and x3 have no parent nodes, so they are all prior probabilities, while x4 has three parent nodes, namely x1, x2, and x3, and so on. Its joint probability distribution is:

[0078] p(x1,...,x7)=p(x1)p(x2)p(x3)p(x4|x1,x2,x3)p(x5|x1,x3)p(x6|x4)p(x7|x4,x5).

[0079] In this embodiment, in step S04, the maximum likelihood estimation (MLE) is a commonly used parameter estimation method, which estimates the value of the unknown parameter by finding the parameter value that maximizes the probability of the observed sample based on the observation results of the sample data. Its formula is based on the likelihood function, and the formula is as follows:

[0080]

[0081] Among them, L(θ|X) is the likelihood function, which represents the probability of observing data X under given parameters θ. is the maximum likelihood function estimator, that is, the parameter value that maximizes the likelihood function.

[0082] In this embodiment, in step S04, the Bayesian estimation is another commonly used parameter estimation method, which is based on the Bayesian theorem and performs parameter estimation by updating the posterior distribution of the parameter based on the prior distribution and the observed data. Among them, the maximum likelihood estimation focuses on the maximum possibility of the data, without considering the prior knowledge; while the Bayesian estimation combines the prior knowledge and the new data to update the estimation of the parameter.

[0083] In this embodiment, in step S05, the evaluation indicators include accuracy, recall, F1 value, etc.;

[0084] Among them, accuracy is the proportion of samples correctly predicted by the model, and the formula is as follows:

[0085] (TP+TN) / (TP+FP+TN+FN);

[0086] Recall refers to the ratio of correctly predicted positive samples to all positive samples. The formula is as follows:

[0087] TP / (TP+FN);

[0088] The F1 value (F1score) is the harmonic average of precision and recall, combining the performance of both. The formula is as follows:

[0089] F1 = 2PR / (P+R);

[0090] Among them, TP: the sample is positive and the prediction result is positive; FP: the sample is negative and the prediction result is positive;

[0091] TN: The sample is negative and the predicted result is negative; FN: The sample is positive and the predicted result is negative.

[0092] In this embodiment, in step S05, the information criteria commonly used include Bayesian Information Criterion (BIC) and Akaike Information Criterion (AIC). These criteria can help us find a suitable balance point in the selection of network structure and parameters to avoid the problem of overfitting or underfitting of the model.

[0093] Akaike Information Criterion (AIC) is based on the concept of information theory and is used to measure the goodness of fit of a statistical model. The formula is as follows:

[0094] AIC = 2k-2ln(L);

[0095] Where: L is the maximum likelihood estimate of the model, which reflects the degree of fit of the model to the data set. k is the number of parameters in the model, representing the complexity of the model.

[0096] Bayesian Information Criterion (BIC) is an information criterion proposed by the Bayesian school. It is similar to the AIC criterion and is also used to weigh the complexity and goodness of fit of the model. The formula is as follows:

[0097] BIC = kln(n)-2ln(L);

[0098] Where: k is the number of parameters in the model, L is the maximum likelihood estimate of the model, and n is the number of samples.

[0099] In this embodiment, in step S05, the sensitivity analysis is a method of evaluating the sensitivity of the model to these changes by changing the parameters or input data of the model. In a Bayesian network, we can evaluate the stability of the network structure and parameters, as well as the robustness of the model through sensitivity analysis. This can help us better understand the behavior of the model and thus have more confidence in the prediction results of the model.

[0100] In order to facilitate those skilled in the art to better understand the technical solution of the present invention, specific embodiments of the present invention are given as follows:

[0101] The following takes the Chaohu Basin as an example, and uses the technical solution of the present invention to predict the trend of changes in the functions of the Chaohu Basin ecosystem, improve the management efficiency of the Chaohu Basin ecological environment, provide a scientific basis for the Chaohu Basin to cope with climate change, water resources protection, etc., and effectively promote the realization of sustainable development. Those skilled in the art should know that the present invention can also be applied to any other basins.

[0102] See also Figure 1 , including a flowchart of ecological function prediction based on deep learning and ecosystem services;

[0103] Step 1: Data collection and preprocessing. Collect remote sensing images, meteorological data, soil information and other data from 2000 to 2020 in the Chaohu Lake Basin, and preprocess the data;

[0104] Step 2: Determine network nodes and variables. According to the needs of ecosystem function prediction, select factors such as environment, meteorology, and soil as key variable nodes in the Bayesian network. These nodes can represent biological, environmental factors, or functional indicators in the ecosystem. Then clarify the specific meaning, type (discrete or continuous), and value range of each selected node;

[0105] Step 3: Construct the Bayesian network structure. According to the nodes and variables determined in step 2, analyze the causal relationship between each node, and clarify which nodes are parent nodes (cause) and which nodes are child nodes (result). Finally, according to the causal relationship obtained by analysis, draw the structure diagram of the Bayesian network, thus obtaining the basic structure of the Bayesian network model;

[0106] like Figure 2 As shown, in a specific embodiment, the structure of the Bayesian network is a directed acyclic graph (DAG), in which each node represents a random variable and directed edges represent dependencies between variables.

[0107] Specifically, a directed acyclic graph satisfies the following two conditions:

[0108] (1) The edges in the graph are directed, that is, they point from one vertex to another;

[0109] (2) There is no path in the graph that can start from a vertex and return to the vertex itself through several edges.

[0110] Nodes are added to the graph in sequence according to the causal relationship between variables. Each time a node is added, its position in the graph and the edges connected to it need to be determined. Based on domain knowledge and statistical results, the network structure is adjusted to ensure that it can accurately reflect the dependencies between variables.

[0111] In a specific embodiment, the Bayesian network modeling process can be found in Figure 4 technology roadmap.

[0112] Step 4: Model parameter learning. First, set the prior distribution for each node, which can be based on expert knowledge, historical data or default distribution. Then use the training set data segmented from the collected data set to train the Bayesian network model through machine learning algorithms (such as maximum likelihood estimation, Bayesian estimation, etc.) to learn the conditional probability table (CPT) of each node. Analyze the changes in the ecosystem function of the Chaohu Basin in the past 20 years, and use the prediction model to predict the ecosystem function of the Chaohu Basin in newer years, and compare it with the actual ecosystem function of the corresponding year. Through continuous iteration, the optimal parameters of the prediction model are obtained;

[0113] In a specific embodiment, the maximum likelihood estimation (MLE) is a commonly used parameter estimation method, which estimates the value of an unknown parameter by finding a parameter value that maximizes the probability of the observed sample based on the observation results of the sample data. Its formula is based on the likelihood function, and the formula is as follows:

[0114]

[0115] Among them, L(θ|X) is the likelihood function, which represents the probability of observing data X under given parameters θ. is the maximum likelihood function estimator, that is, the parameter value that maximizes the likelihood function.

[0116] In a specific embodiment, the Bayesian estimation is another commonly used parameter estimation method, which is based on the Bayesian theorem and estimates parameters by updating the posterior distribution of the parameters based on the prior distribution and the observed data. Among them, the maximum likelihood estimation focuses on the maximum possibility of the data, without considering the prior knowledge; while the Bayesian estimation combines the prior knowledge and the new data to update the estimation of the parameters. The Bayesian theorem can be expressed by the following formula:

[0117]

[0118] Among them, P(A) and P(B) represent the prior probabilities of event A and event B respectively (that is, the estimated probabilities of event A and event B before considering other information); P(B|A) represents the posterior probability of event B occurring after assuming that event A has occurred (also called conditional probability, that is, the new estimate of the probability of event B); P(A|B) is the same as P(B|A).

[0119] Step 5: Model evaluation and optimization. Use the test set to evaluate the trained model. Evaluation indicators include accuracy, recall, F1 value, etc. In addition to cross-validation, information criteria and sensitivity analysis are also commonly used evaluation methods. Based on the evaluation results, the model is optimized and adjusted again to improve prediction accuracy and stability.

[0120] In a specific embodiment, the evaluation indicators include accuracy, recall, F1 value, etc.

[0121] Among them, accuracy is the proportion of samples correctly predicted by the model, and the formula is as follows:

[0122] (TP+TN) / (TP+FP+TN+FN);

[0123] Recall refers to the ratio of correctly predicted positive samples to all positive samples. The formula is as follows:

[0124] TP / (TP+FN);

[0125] The F1 value (F1score) is the harmonic average of precision and recall, combining the performance of both. The formula is as follows:

[0126] F1 = 2PR / (P+R);

[0127] Among them, TP: the sample is positive and the prediction result is positive; FP: the sample is negative and the prediction result is positive;

[0128] TN: The sample is negative and the predicted result is negative; FN: The sample is positive and the predicted result is negative.

[0129] In a specific embodiment, the information criteria commonly used include Bayesian Information Criterion (BIC) and Akaike Information Criterion (AIC). These criteria can help us find a suitable balance between the selection of network structure and parameters to avoid the problem of overfitting or underfitting of the model.

[0130] Akaike Information Criterion (AIC) is based on the concept of information theory and is used to measure the goodness of fit of a statistical model. The formula is as follows:

[0131] AIC = 2k-2ln(L);

[0132] Where: L is the maximum likelihood estimate of the model, which reflects the degree of fit of the model to the data set. k is the number of parameters in the model, representing the complexity of the model.

[0133] Bayesian Information Criterion (BIC) is an information criterion proposed by the Bayesian school. It is similar to the AIC criterion and is also used to weigh the complexity and goodness of fit of the model. The formula is as follows:

[0134] BIC = kln(n)-2ln(L);

[0135] Where: k is the number of parameters in the model, L is the maximum likelihood estimate of the model, and n is the number of samples.

[0136] In a specific embodiment, the sensitivity analysis is a method of evaluating the sensitivity of a model to these changes by changing the parameters or input data of the model. In a Bayesian network, we can evaluate the stability of the network structure and parameters, as well as the robustness of the model through sensitivity analysis. This can help us better understand the behavior of the model and thus have more confidence in the prediction results of the model.

[0137] Step 5: Interpretation of results and management recommendations. Apply the optimized model to the actual data of the Chaohu Basin to predict changes in the basin's ecosystem functions from 2020 to 2030, such as vegetation coverage, soil quality, and water conditions. At the same time, interpret the prediction results, combine the professional knowledge in the corresponding field with the basin's own resource background, understand the model's prediction basis and influencing factors, and provide corresponding management recommendations to comprehensively and dynamically monitor the trend of changes in the ecological functions of the Chaohu Basin, and provide real-time scientific basis for protecting the ecological environment of the Chaohu Basin and sustainable use of water resources.

[0138] In the specific implementation example, it is necessary to fully consider domain knowledge, actual needs, and the basin's own resource base to find the model architecture and parameter configuration that best suits the current research and ensure the reliability, practicality, and accuracy of the model. The interpretation of the results and management recommendations need to be based on professional knowledge and field research to ensure the scientificity and feasibility of the prediction results.

[0139] The functions that can be achieved by the above-mentioned ecological function prediction method based on deep learning and ecosystem services are all completed by a computer device, and the computer device includes one or more processors and one or more memories, and at least one program code is stored in the one or more memories. The program code is loaded and executed by the one or more processors to realize the functions of the ecological function prediction method based on deep learning and ecosystem services.

[0140] The processor takes out instructions from the memory one by one, analyzes the instructions, and then completes the corresponding operations according to the instruction requirements, generating a series of control commands, so that the various parts of the computer can automatically, continuously and coordinately move to become an organic whole, realize the input of programs, the input of data, and the calculation and output of results. The arithmetic operations or logical operations generated in this process are all completed by the operator; the memory includes a read-only memory (ROM), which is used to store computer programs, and a protection device is provided outside the memory.

[0141] Exemplarily, the computer program may be divided into one or more modules, one or more modules are stored in a memory and executed by a processor to implement the present invention. One or more modules may be a series of computer program instruction segments capable of implementing specific functions, and the instruction segments are used to describe the execution process of the computer program in a terminal device.

[0142] Those skilled in the art will understand that the description of the above service equipment is merely an example and does not constitute a limitation on the terminal equipment. It may include more or fewer components than described above, or a combination of certain components, or different components, for example, it may include input and output devices, network access equipment, buses, etc.

[0143] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, and uses various interfaces and lines to connect various parts of the entire user terminal.

[0144] The memory can be used to store computer programs and / or modules. The processor can realize various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as an information collection template display function, a product information release function, etc.); the data storage area can store data created according to the use of the berth status display system (such as product information collection templates corresponding to different product types, product information that different product providers need to release, etc.). In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (SmartMediaCard, SMC), a secure digital (SecureDigital, SD) card, a flash card (FlashCard), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0145] If the module / unit integrated in the terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the modules / units in the above-mentioned embodiment system, and can also be completed by instructing the relevant hardware through a computer program. The above-mentioned computer program can be stored in a computer-readable storage medium, and the computer program can realize the functions of the above-mentioned various system embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. Computer-readable media may include: any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0146] It should be noted that, in this article, the term "comprises" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of more restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0147] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. An ecological function prediction method based on deep learning and ecosystem services, characterized in that: The method comprises the following steps: S01. Data collection and preprocessing: Collect various data in the study area and preprocess the collected data, including remote sensing images, meteorological data and soil information; S02. Determine network nodes and variables: According to the needs of ecosystem function prediction, select key variables as nodes in the Bayesian network; clarify the specific meaning, type and value range of each node; S03. Constructing Bayesian network structure: First, based on ecological theory and expert knowledge, select the prediction target Important Influencing variables as nodes in the Bayesian network; then analyze the causal relationship between the nodes to clarify which nodes are parent nodes and which nodes are child nodes; finally, draw the structure diagram of the Bayesian network based on the causal relationship obtained by the analysis, in which each node represents a variable, and the directed edges represent the causal relationship between the variables; S04, model parameter learning: first, set a prior distribution for each node, which is based on expert knowledge, historical data or default distribution; then use the collected data set to train the Bayesian network model through machine learning algorithms to learn the conditional probability table CPT of each node. CPT describes the probability of a child node taking each value when a parent node takes a given value. Among them, machine learning algorithms include maximum likelihood estimation and Bayesian estimation; S05. Model evaluation and optimization: Use the test set to evaluate the trained model. The evaluation indicators include accuracy, recall and F1 value. In addition to cross-validation, information criteria and sensitivity analysis are also commonly used evaluation methods. According to the evaluation results, the model is optimized and adjusted again to improve the prediction accuracy and stability. S06. Interpretation of results and management recommendations: Apply the optimized model to actual data to predict changes in ecosystem functions in the study area; at the same time, interpret the prediction results, understand the model's prediction basis and influencing factors, and provide corresponding management recommendations to provide a scientific basis for protecting the ecological environment.

2. The method for predicting ecological functions based on deep learning and ecosystem services according to claim 1, characterized in that: In step S01, the actions of preprocessing the collected data include: image denoising, geometric correction and fusion.

3. The method for predicting ecological functions based on deep learning and ecosystem services according to claim 1, characterized in that: In step S02, each node represents a random variable, including observed data points and potential unknown parameters; directed edges represent the causal relationship between variables, pointing from parent nodes to child nodes, reflecting the direct dependency between these variables.

4. The method for predicting ecological functions based on deep learning and ecosystem services according to claim 1, characterized in that: In step S03, the variables in "variables that have a significant impact on the prediction target" include: 1) Directly related variables, i.e. variables that are directly involved in the generation or change process of the predicted target; 2) Indirect influence variables, that is, variables that do not directly participate in the generation or change process of the predicted target, but indirectly affect the predicted target by affecting other variables; 3) Background variables, which are variables that provide background information or context for the prediction target.

5. The method for predicting ecological functions based on deep learning and ecosystem services according to claim 1, characterized in that: In step S03, the structure of the Bayesian network is a directed acyclic graph, in which each node represents a random variable and directed edges represent dependencies between variables; A directed acyclic graph is a graph structure consisting of a set of vertices and a set of directed edges, where the direction of the edge indicates the relationship between the vertices, and there is no path that forms a loop in the graph; if there is a directed edge from node A to node B, then we believe that A is the parent node of B and B is the child node of A; in a Bayesian network, there is no loop, that is, starting from a node and walking along a directed edge, you will not return to the original node; a directed acyclic graph satisfies the following two conditions: 1) The edges in the graph are directed, that is, they point from one vertex to another; 2) There is no path in the graph that starts from a vertex and goes back to the vertex through several edges; According to the causal relationship between variables, nodes are added to the directed acyclic graph in sequence. Each time a node is added, its position in the directed acyclic graph and the edges connected to it need to be determined. Based on domain knowledge and statistical results, the network structure is adjusted to ensure that it accurately reflects the dependency relationship between variables.

6. The method for predicting ecological functions based on deep learning and ecosystem services according to claim 1, characterized in that: In step S04, the prior distribution refers to the initial belief or probability of an event occurring before any data is observed; the prior distribution is derived based on domain knowledge, historical data or personal experience.

7. The method for predicting ecological functions based on deep learning and ecosystem services according to claim 1, characterized in that: In step S04, in the Bayesian network, each node has a conditional probability table CPT associated with it, which describes the probability distribution of the node state under a given parent node state, so that the Bayesian network has the ability to calculate the joint probability distribution of the entire network through the chain rule; Conditional probability table: It is a two-dimensional matrix, an important mathematical tool, used to describe the probability relationship between random variables under given conditions. CPT is mainly used in Bayesian networks to show the dependency relationship between variables in an intuitive and structured way. CPT combines the structure of directed acyclic graph and together constitutes the core of Bayesian network. The probability values ​​in CPT are obtained by learning from observed data and by setting through expert knowledge.

8. The method for predicting ecological functions based on deep learning and ecosystem services according to claim 1, characterized in that: In step S05, the evaluation indicators of the cross-validation include accuracy, recall and F1 value, and the information criterion is the Bayesian information criterion and the Akaike information criterion. The information criterion helps us find a balance between the selection of network structure and parameters to avoid the problem of overfitting or underfitting of the model; Sensitivity analysis is a method to evaluate the sensitivity of a model to these changes by changing the model parameters or input data.

Images