CN115293231A - A Random Forest Prediction Method for Regional Ecological Harmony - Google Patents

Abstract

The prediction method of the regional ecological harmony random forest comprises the following steps: 1. finely describing lake and grass elements of the mountain and water forest fields from the underground to the surface by integrating different time scales; 2. finely and quantitatively interpreting the natural elements of nearly one hundred years in time intervals by combining long-time satellite remote sensing; 3. collecting human factor characterization data of different years in the research area range; 4. establishing a human activity factor function by a multivariate regression and machine learning method; 5. and analyzing the period and frequency of the curve, predicting the change characteristics of each element in the future by using a mathematical model, and predicting the influence of human activities on other elements. The invention has excellent accuracy; can operate efficiently on large data sets; input samples with high dimensional characteristics can be processed without dimension reduction; the importance of each feature on the classification problem can be evaluated; in the generation process, an unbiased estimation of an internal generation error can be obtained; good results can be obtained also for the default value problem.

Description

Regional Ecological Harmony Random Forest Prediction Method

technical field

The invention belongs to the technical field of ecological environment prediction, and in particular relates to a method for predicting regional ecological harmony random forest.

Background technique

By analyzing the man-land system of a city and estimating its environmental carrying capacity, it can be planned as an ecologically harmonious three-dimensional city. Based on this technology, through the intersection of natural science and social science in the fields of humanities, society, and management, it is necessary to fully analyze the interference of human activities, and use social science evaluation methods to integrate social benefits into secondary evaluation. According to the needs of local economic development, the appeals of different stakeholders, human settlements and other elements, combined with predictable economic and sociological models, the most complete planning combination is selected and provided to planners, and at the same time, annual land use strategies and specific restoration measures are proposed. Therefore, how to improve the prediction efficiency and accuracy of regional ecological harmony is an urgent problem to be solved.

Contents of the invention

In order to solve the deficiencies in the prior art, the present invention provides a regional ecological harmony random forest prediction method; it provides a key baseline for comparison with modern times based on the long-term and rapid environmental change evidence in the history of the earth, and uses intuitive and quantitative methods to explore Mathematical models analyze the relationship between natural systems, and with the help of high-precision dating technology that spans the geological time scale and human time scale, it can solve the problem of connecting human history and geological history, abstract the interference factors and relationships of human elements on the landscape, forest, field, lake and grass system, and use artificial The intelligent algorithm realizes its high-precision restoration and prediction.

In order to solve the above-mentioned technical problems, the present invention adopts the following technical solutions: a regional ecological harmony random forest prediction method, comprising the following steps:

The first step is to comprehensively describe the elements of mountains, rivers, forests, fields, lakes and grasses from the underground to the surface, aiming at multiple elements and synthesizing different time scales;

The second step is to combine the long-term satellite remote sensing to make a fine and quantitative interpretation of the natural elements of the past century. The natural elements include the area of mountains, water, forests, fields, lakes, and grass. The time unit is calibrated in one year, and the comprehensive solution Human factors, human factors include building land area;

The third step is to collect the characterization data of human factors in different years within the research area. The characterization data include population, GDP and industrial development intensity, calibrated with one year as the time unit, and comprehensively interpret human factors;

The fourth step is to collect and judge the environmental carrying capacity of the research area within the research period based on historical data and expert judgment, and use it as a standard for subsequent model training. Through multiple regression and machine learning methods, the function of human activity elements is established; based on systematic thinking, through multiple regression and machine learning methods, a multi-factor and multi-scale fitting relationship between man and nature is established. Multi-scales include time scales to explore nature The spontaneous evolution process of the environment and the influence of human activities on this process are fitted into a time-related mathematical model, and the mathematical model is a curve function;

The fifth step is to set the time as a certain time in the future, analyze the cycle and frequency of the curve, use mathematical models to predict the characteristics of changes in various elements in the future, and predict the impact of human activities on other elements; based on the above work, put forward the environmental carrying capacity The lower limit defines the area or proportion of mountains, water, forests, fields, lakes, and grass in the area, defines the baseline for ecological function protection, the bottom line for environmental quality and safety, and the upper limit for the use of natural resources to guide the three-dimensional planning of an ecologically harmonious city.

The first step is specifically: arrange intensive shallow drilling in key anatomical areas, establish a three-dimensional space model including the elements of mountains, rivers, forests, fields, lakes and grasses through fine and quantitative characterization of underground geological bodies; use multiple boreholes, combined with high-precision dating technology, from In the late Holocene (>5000a), according to the division of the millennium, the time unit of 100 years began to be calibrated by 1000 years ago, the paleogeography and paleoenvironmental pattern were finely restored, and a four-dimensional spatiotemporal geological model with high precision was finally established.

The method of multiple regression and machine learning in the fourth step is the random forest model algorithm, and the geographical area factor indicators of the random forest model algorithm are shown in the following table:

Table 1 Geographical Region Situation Factor Indicators

Assuming that the data collection year is divided into two periods, the first period is from the middle and late Holocene to modern (7000 BC-1950), a total of 50 time points, the second period is 1951-2021, a total of 71 time points, the first period It is mainly used to construct the background of geological evolution. From 1951 to 2020, the evaluation results of this area as an ecologically harmonious three-dimensional city are known (1: can be used; 0: cannot be used), and the evaluation results in 2021 are unknown, which is required Predicted by the trained prediction model; there are 23 indicators in Table 1, so the size of the standardized Z matrix is 23*120, and the corresponding judgment result Y matrix size is 120*1; Known index Z ₂₀₂₁ matrix size is 23*1.

The random forest model algorithm uses the parameters in Table 1 for data cleaning, such as processing missing values, smooth noise, identifying or deleting outliers, and normalizing for data preprocessing; including the following steps:

(1) Random number generation, the growth of each tree in the model is a key step, (2) Calculation of predictive indicators MAE and MAPE, (3) Random forest parameter optimization, (4) According to the principle of the highest accuracy, select the most Optimal model, (5) directly calculate the weight of each feature (non-zero real number) according to the optimal model generated by the random forest, and select a certain number of more important features according to the principle from large to small.

Step (1) consists of the following three main steps:

A. Bootstrap sampling: if the size of the training set is N, for each tree, randomly select n training samples from the training set with replacement as the training set of the tree;

B. Random features: If the feature dimension of each sample is M, specify a constant m<<M, randomly select m feature subsets from M features, and select the most feature from these m features each time the tree is split. Excellent;

C. Each tree grows to its maximum extent and has not been pruned.

为结果的真实值

为结果的估计值。预测指标MAE（Mean Absolute Error）表示平均绝对误差，值域：[0,+∞)；当预测值与真实值完全吻合时等于0，即完美模型；误差越大，MAE 值越大：Step (2) is specifically:

is the true value of the result

Estimated value for the result. The predictive indicator MAE (Mean Absolute Error) represents the mean absolute error, value range: [0,+∞); when the predicted value is completely consistent with the real value, it is equal to 0, that is, the perfect model; the larger the error, the larger the MAE value:

The predictive indicator MAPE (Mean Absolute Percentage Error) represents the mean absolute percentage error, value range: [0,+∞); when the predicted value is completely consistent with the real value, it is equal to 0, that is, the perfect model; the larger the error, the larger the MAE value:

。

.

Step (3) is specifically: using the classic parameter tuning method in machine learning, the number of established trees, the selection method of the largest feature, the maximum depth of the tree, the number of samples required for the minimum split of nodes, the minimum number of samples of leaf nodes, whether Randomly select the most suitable parameter combination, and adjust it with or without Bayesian optimization.

Adopt above-mentioned technical scheme, the present invention has following technical effect:

From the perspective of space, the reflection of the basic situation of various geographical elements that constitute the natural resources and human living environment of a certain area (such as a city) can be regarded as the three different types of geographical information in perception, statistics and analysis according to different needs. Information obtained after in-depth processing. In this method, the core problem to be solved is the establishment of a multi-element multi-scale (time scale) fitting relationship between human and nature.

Referring to (Ma Wanzhong, Du Qingyun. Research on the System Framework of Geographical National Conditions Monitoring[J]. Land and Resources Science and Technology Management, 2011,28(06):104-111), it can be summarized as natural environment elements, social and cultural elements and industrial economy. elements. Referring to the principle of the indicator system proposed in (Liu Kai. Research on the evolution of ecologically fragile human-land systems and the selection of sustainable development models [D]. Shandong Normal University, 2017), a certain area can be planned as an ecologically harmonious three-dimensional city. Target. By using the indicators of the natural environment and economic society, the prediction model is established by using the random forest method, and the weight analysis of the impact of the indicators on the prediction results is further obtained.

For the indicators listed in Table 1, appropriate supplements or deletions can be made. The more features included, the higher the accuracy rate. Try to keep the weighted indicators as much as possible, which can reduce the characteristics and reduce the running time. It is recommended to select some important feature data sets according to the 95% threshold. For the collection time, multiple time points can be collected in the same year, such as one data point per month, which greatly increases the sample size. Increasing the sample size increases the accuracy of the predictive model.

The random forest classification effect (error rate) is related to two factors: the correlation of any two trees in the forest: the greater the correlation, the greater the error rate; the classification ability of each tree in the forest: the higher the classification ability of each tree Stronger, the lower the error rate of the entire forest. Reducing the number of feature selection m will reduce the correlation and classification ability of the tree; increasing m will increase both. So the key question is how to choose the optimal m (or range), which is also the only parameter of random forest.

The present invention uses the random forest model algorithm, which has the following advantages: 1) among all the current algorithms, it has excellent accuracy; 2) it can effectively run on large data sets; 3) it can process input samples with high-dimensional features , and does not require dimensionality reduction; 4) It can evaluate the importance of each feature in the classification problem; 5) During the generation process, an unbiased estimate of the internal generation error can be obtained; 6) For the default value problem, it can also Get great results.

Description of drawings

Figure 1 is a schematic diagram of the random forest model;

Figure 2 is a schematic diagram of the random forest model algorithm flow;

Figure 3 is a schematic diagram of the weight arrangement of predictive indicators;

Figure 4 is a schematic diagram of the comparison between the prediction model results and the actual results.

Detailed ways

As shown in Figures 1-4, the regional ecological harmony random forest prediction method of the present invention comprises the following steps:

The first step is to comprehensively describe the elements of mountains, rivers, forests, fields, lakes and grasses from the underground to the surface, aiming at multiple elements and synthesizing different time scales;

The second step is to combine the long-term satellite remote sensing to make a fine and quantitative interpretation of the natural elements of the past century. The natural elements include the area of mountains, water, forests, fields, lakes, and grass. The time unit is calibrated in one year, and the comprehensive solution Human factors, human factors include building land area;

The third step is to collect the characterization data of human factors in different years within the research area. The characterization data include population, GDP and industrial development intensity, calibrated with one year as the time unit, and comprehensively interpret human factors;

The fourth step is to collect and judge the environmental carrying capacity of the research area within the research period based on historical data and expert judgment, and use it as a standard for subsequent model training. Through multiple regression and machine learning methods, the function of human activity elements is established; based on systematic thinking, through multiple regression and machine learning methods, a multi-factor and multi-scale fitting relationship between man and nature is established. Multi-scales include time scales to explore nature The spontaneous evolution process of the environment and the influence of human activities on this process are fitted into a time-related mathematical model, and the mathematical model is a curve function;

The fifth step is to set the time as a certain time in the future, analyze the cycle and frequency of the curve, use mathematical models to predict the characteristics of changes in various elements in the future, and predict the impact of human activities on other elements; based on the above work, put forward the environmental carrying capacity The lower limit defines the area or proportion of mountains, water, forests, fields, lakes, and grass in the area, defines the baseline for ecological function protection, the bottom line for environmental quality and safety, and the upper limit for the use of natural resources to guide the three-dimensional planning of an ecologically harmonious city.

The first step is specifically: arrange intensive shallow drilling in key anatomical areas, establish a three-dimensional space model including the elements of mountains, rivers, forests, fields, lakes and grasses through fine and quantitative characterization of underground geological bodies; use multiple boreholes, combined with high-precision dating technology, from In the late Holocene (>5000a), according to the division of the millennium, the time unit of 100 years began to be calibrated by 1000 years ago, the paleogeography and paleoenvironmental pattern were finely restored, and a four-dimensional spatiotemporal geological model with high precision was finally established.

The method of multiple regression and machine learning in the fourth step is the random forest model algorithm, and the geographical area factor indicators of the random forest model algorithm are shown in the following table:

Table 1 Geographical Region Situation Factor Indicators

Assuming that the data collection year is divided into two periods, the first period is from the middle and late Holocene to modern (7000 BC-1950), a total of 50 time points, the second period is 1951-2021, a total of 71 time points, the first period It is mainly used to construct the background of geological evolution. From 1951 to 2020, the evaluation results of this area as an ecologically harmonious three-dimensional city are known (1: can be used; 0: cannot be used), and the evaluation results in 2021 are unknown, which is required Predicted by the trained prediction model; there are 23 indicators in Table 1, so the size of the standardized Z matrix is 23*120, and the corresponding judgment result Y matrix size is 120*1; Known index Z ₂₀₂₁ matrix size is 23*1.

The random forest model algorithm uses the parameters in Table 1 for data cleaning, such as processing missing values, smooth noise, identifying or deleting outliers, and normalizing for data preprocessing; including the following steps:

(1) Random number generation, the growth of each tree in the model is a key step, (2) Calculation of predictive indicators MAE and MAPE, (3) Random forest parameter optimization, (4) According to the principle of the highest accuracy, select the most Optimal model, (5) directly calculate the weight of each feature (non-zero real number) according to the optimal model generated by the random forest, and select a certain number of more important features according to the principle from large to small.

Step (1) consists of the following three main steps:

A. Bootstrap sampling: if the size of the training set is N, for each tree, randomly select n training samples from the training set with replacement as the training set of the tree;

B. Random features: If the feature dimension of each sample is M, specify a constant m<<M, randomly select m feature subsets from M features, and select the most feature from these m features each time the tree is split. Excellent;

C. Each tree grows to its maximum extent and has not been pruned.

为结果的真实值

为结果的估计值。预测指标MAE（Mean Absolute Error）表示平均绝对误差，值域：[0,+∞)；当预测值与真实值完全吻合时等于0，即完美模型；误差越大，MAE 值越大：Step (2) is specifically:

is the true value of the result

Estimated value for the result. The predictive indicator MAE (Mean Absolute Error) represents the mean absolute error, value range: [0,+∞); when the predicted value is completely consistent with the real value, it is equal to 0, that is, the perfect model; the larger the error, the larger the MAE value:

The predictive indicator MAPE (Mean Absolute Percentage Error) represents the mean absolute percentage error, value range: [0,+∞); when the predicted value is completely consistent with the real value, it is equal to 0, that is, the perfect model; the larger the error, the larger the MAE value:

。

.

Step (3) is specifically: using the classic parameter tuning method in machine learning, the number of established trees, the selection method of the largest feature, the maximum depth of the tree, the number of samples required for the minimum split of nodes, the minimum number of samples of leaf nodes, whether Randomly select the most suitable parameter combination, and adjust it with or without Bayesian optimization.

This embodiment does not impose any formal restrictions on the shape, material, structure, etc. of the present invention. All simple modifications, equivalent changes and modifications made to the above embodiments according to the technical essence of the present invention belong to the protection of the technical solution of the present invention. scope.

Claims (7)

1. The regional ecological harmony random forest prediction method is characterized in that: comprising the following steps: The first step is to comprehensively describe the elements of mountains, rivers, forests, fields, lakes and grasses from the underground to the surface, aiming at multiple elements and synthesizing different time scales; The second step is to combine the long-term satellite remote sensing to make a fine and quantitative interpretation of the natural elements of the past century. The natural elements include the area of mountains, water, forests, fields, lakes, and grass. The time unit is calibrated in one year, and the comprehensive solution Human factors, human factors include building land area; The third step is to collect the characterization data of human factors in different years within the research area. The characterization data include population, GDP and industrial development intensity, calibrated with one year as the time unit, and comprehensively interpret human factors; The fourth step, according to historical data and expert judgment, collect the environmental carrying capacity of the research area within the research period to make a judgment, and use it as the standard for subsequent model training; through multiple regression and machine learning methods, establish the human activity element function; based on Systematic thinking, through multiple regression and machine learning methods to establish a multi-element and multi-scale fitting relationship between man and nature, multi-scale including time scale, to explore the spontaneous evolution process of the natural environment and the impact of human activities on this process, fitting Form a mathematical model related to time, and the mathematical model is a curve function; The fifth step is to set the time as a certain time in the future, analyze the cycle and frequency of the curve, use mathematical models to predict the characteristics of changes in various elements in the future, and predict the impact of human activities on other elements; based on the above work, put forward the environmental carrying capacity The lower limit defines the area or proportion of mountains, water, forests, fields, lakes, and grass in the area, defines the baseline for ecological function protection, the bottom line for environmental quality and safety, and the upper limit for the use of natural resources to guide the three-dimensional planning of an ecologically harmonious city. 2. The regional ecological harmony random forest prediction method according to claim 1, characterized in that: the first step is specifically: arranging intensive shallow drills in key anatomical areas, and establishing a structure including mountains, rivers, forests, farmland, and lakes through fine and quantitative characterization of underground geological bodies. The three-dimensional space model of grass elements; using multiple boreholes, combined with high-precision dating technology, from the late Holocene (>5000a), according to the millennium-level division, to 1000 years ago, the 100-year time unit was used to calibrate, and the ancient Geography, palaeoenvironmental pattern, and finally establish a four-dimensional space-time geological model with high precision. 3. the regional ecological harmony random forest prediction method according to claim 1, is characterized in that: the multiple regression in the 4th step and the method of machine learning are random forest model algorithm, and the geographic area situation factor index of random forest model algorithm is as follows As shown in the table: Table 1 Geographical Region Situation Factor Indicators

Assuming that the data collection year is divided into two periods, the first period is from the middle and late Holocene to modern (7000 BC-1950), a total of 50 time points, the second period is 1951-2021, a total of 71 time points, the first period It is mainly used to construct the background of geological evolution. From 1951 to 2020, the evaluation results of this area as an ecologically harmonious three-dimensional city are known (1: can be used; 0: cannot be used), and the evaluation results in 2021 are unknown, which is required Predicted by the trained prediction model; there are 23 indicators in Table 1, so the size of the standardized Z matrix is 23*120, and the corresponding judgment result Y matrix size is 120*1; Known index Z ₂₀₂₁ matrix size is 23*1. 4. The regional ecological harmony random forest prediction method according to claim 3, characterized in that: random forest model algorithm adopts parameters in table 1 to carry out data cleaning such as processing missing values, smooth noise, identifying or deleting outliers and normalization Perform data preprocessing; including the following steps: (1) Random number generation, the growth of each tree in the model is a key step, (2) Calculation of predictive indicators MAE and MAPE, (3) Random forest parameter optimization, (4) According to the principle of the highest accuracy, select the most Optimal model, (5) directly calculate the weight of each feature (non-zero real number) according to the optimal model generated by the random forest, and select a certain number of more important features according to the principle from large to small. 5. The regional ecological harmony random forest prediction method according to claim 4, characterized in that: step (1) includes the following three main steps: A. Bootstrap sampling: if the size of the training set is N, for each tree, randomly select n training samples from the training set with replacement as the training set of the tree; B. Random features: If the feature dimension of each sample is M, specify a constant m<<M, randomly select m feature subsets from M features, and select the most feature from these m features each time the tree is split. Excellent; C. Each tree grows to its maximum extent and has not been pruned. 6. The regional ecological harmony random forest prediction method according to claim 5, characterized in that: step (2) is specifically:

is the true value of the result

is the estimated value of the result; The predictive indicator MAE (Mean Absolute Error) represents the mean absolute error, value range: [0,+∞); when the predicted value is completely consistent with the real value, it is equal to 0, that is, the perfect model; the larger the error, the larger the MAE value:

The predictive indicator MAPE (Mean Absolute Percentage Error) represents the mean absolute percentage error, value range: [0,+∞); when the predicted value is completely consistent with the real value, it is equal to 0, that is, the perfect model; the larger the error, the larger the MAE value:

. 7. The regional ecological harmony random forest prediction method according to claim 6, characterized in that: step (3) is specifically: using the classic parameter adjustment method in machine learning, the number of established trees, the selection method of the largest feature, The maximum depth of the tree, the number of samples required for the minimum split of nodes, the minimum number of samples of leaf nodes, whether to randomly select the most suitable parameter combination, and whether to adjust by Bayesian optimization.

Images