A method for characterizing key problems of urban sustainable development by fusing machine learning

Abstract

The application relates to a kind of urban sustainable development key problem characterization methods of fusion machine learning, the method includes: constructing urban sustainable development capability evaluation framework, to obtain the evaluation score of each city corresponding each subsystem and the comprehensive evaluation score of each city;Combination identification model based on feature screening module and feature importance quantification module is constructed to identify key indicators, key subsystems and calculate the corresponding relative importance size;Based on the evaluation score of all indicators, all indicator scores are displayed through the score indication panel;Based on the evaluation score of all subsystems, all subsystems are displayed through radar chart;The key subsystems and key indicators under different importance levels and the corresponding relative importance size are displayed through list.The application quantitatively identifies the key indicators and key subsystems in the process of urban sustainable development, and realizes the characterization of the key problems affecting urban development.

Description

Technical Field

[0001] This invention relates to the field of urban management, and in particular to a method for representing key issues in urban sustainable development that integrates machine learning. Background Technology

[0002] Identifying the influencing factors and characterizing key issues in urban development is crucial for decision-making regarding sustainable urban development. Currently, methods for identifying sustainable development factors typically fall into two categories: panel data correlation analysis and spatial data correlation analysis.

[0003] Identifying limiting factors based on panel data correlation analysis typically involves performing correlation analysis or linear regression analysis on the scores of elements related to urban sustainable development capacity, and making judgments based on the degree of correlation and regression trends. Traditional regression analysis generally analyzes the regression results of each indicator element; the sign of the regression coefficient indicates whether the indicator element has a positive or negative impact on sustainable development, and the absolute value of the regression coefficient indicates the magnitude of the indicator element's influence on urban sustainable development capacity. Correlation analysis and linear regression analysis are simple to calculate and easy to implement, and the statistical analysis results are highly interpretable, but they require a large sample size and have weaker ability to identify complex relationships.

[0004] Identifying limiting factors based on spatial data correlation analysis typically involves calculating the Moran's index and identifying hot and cold spots through spatial autocorrelation analysis, and then using autocorrelation statistics to identify influencing factors. In addition, some scholars have attempted to interpret the coefficients of assessment indicators using geographically weighted regression models and multi-scale geographically weighted regression models, thereby further considering geospatial identification of key factors affecting sustainable development. Both geographically weighted regression models and multi-scale geographically weighted regression models can incorporate data regression from the space surrounding the factors of urban sustainable development capacity, effectively addressing spatial heterogeneity. Furthermore, multi-scale geographically weighted regression models can further consider scale differences between different variables, better identifying key limiting factors. However, in the objective research process of urban sustainable development, most analyses of influencing factors are conducted on one or a few cities. The factor data is mostly statistical data at the city scale. Practical problems such as small city sample sizes, weak spatial connections between cities, and low spatial resolution of factor data make it difficult to identify influencing factors from a spatial perspective.

[0005] In summary, currently only linear regression analysis is widely applicable in identifying influencing factors. However, traditional linear regression methods are relatively limited in variety, usually require data correlation to be eliminated before analysis can be performed, and cannot directly quantify the relative importance of factors contributing to urban sustainable development capabilities, resulting in insufficient depth in decision support information mining. Summary of the Invention

[0006] To address the aforementioned issues, this invention proposes a method for representing key issues in urban sustainable development that integrates machine learning.

[0007] The specific plan is as follows:

[0008] A method for representing key issues in urban sustainable development that integrates machine learning includes the following steps:

[0009] S1: Construct a framework for assessing urban sustainable development capacity to obtain assessment scores for each subsystem of each city and a comprehensive assessment score for each city; the process of constructing the assessment framework includes:

[0010] S101: Select assessment indicators for urban sustainable development assessment and assign all assessment indicators to various subsystems;

[0011] S102: Obtain the index values ​​of each evaluation indicator and perform normalization processing;

[0012] S103: Calculate the weight of each evaluation indicator;

[0013] S104: Based on the weights of each indicator and the results of normalization, calculate the evaluation scores of each subsystem and the comprehensive evaluation score of the city.

[0014] S2: Construct a combined recognition model based on the feature selection module and the feature importance quantification module, and perform the following operations:

[0015] The indicator values ​​of each evaluation indicator are used as independent variables and the comprehensive evaluation score is used as the dependent variable. The feature selection module is used to select the key indicators among all evaluation indicators. The key indicators output by the feature selection module are used as independent variables and the comprehensive evaluation score is used as the dependent variable. The key indicators are used as independent variables and the comprehensive evaluation score is used as the dependent variable. The key indicators are used as independent variables and the feature importance quantification module to obtain the relative importance of each key indicator.

[0016] The evaluation scores of each subsystem are used as independent variables and the comprehensive evaluation score is used as the dependent variable. The feature selection module is used to select the key subsystems among all subsystems. The key subsystems output by the feature selection module are used as independent variables and the comprehensive evaluation score is used as the dependent variable. The key subsystems are then used as independent variables and the comprehensive evaluation score is used as the dependent variable. The key subsystems are then used as independent variables and the feature importance quantification module is used to obtain the relative importance of each key subsystem.

[0017] S3: Based on the evaluation scores of all indicators, display all indicators through a score indicator panel; based on the evaluation scores of all subsystems, display all subsystems through a radar chart.

[0018] S4: Based on the relative importance of each key subsystem, each key subsystem is divided into different importance levels, and corresponding key indicators are matched based on the key subsystems. The key indicators of each key subsystem are arranged from high to low relative importance.

[0019] S5: Display key subsystems and key indicators at different importance levels, along with their corresponding relative importance, through a list.

[0020] Furthermore, when selecting assessment indicators for urban sustainable development evaluation, it is necessary to take into account the current policy context.

[0021] Furthermore, the selection of evaluation indicators follows these principles: avoid duplication, emphasize localization, and emphasize compatibility.

[0022] Furthermore, the formula for calculating the weights of the evaluation indicators is as follows:

[0023]

[0024] Among them, W i n represents the weight of each specific evaluation indicator in the i-th subsystem. i This represents the number of specific evaluation indicators involved in the i-th subsystem, and N represents the number of subsystems.

[0025] Furthermore, the formula for calculating the evaluation score of the subsystem is as follows:

[0026]

[0027] Among them, S i I represents the evaluation score of the i-th subsystem. ij W represents the value of the j-th specific evaluation index in the i-th subsystem. i Let represent the weight of each specific evaluation indicator in the i-th subsystem, j represent the index of the evaluation indicator, and n represent the weight of each specific evaluation indicator in the i-th subsystem. i This represents the number of specific evaluation indicators involved in the i-th subsystem.

[0028] Furthermore, the formula for calculating the city's comprehensive evaluation score is as follows:

[0029]

[0030] Where S represents the comprehensive evaluation score, i represents the subsystem number, and N represents the number of subsystems.

[0031] Furthermore, when the subsystem corresponding to a key indicator is not a key subsystem, the importance level of the key indicator and its corresponding key subsystem is set based on the relative importance of the key indicator.

[0032] Furthermore, the feature selection module employs a LASSO regression model.

[0033] Furthermore, the feature importance quantification module employs the random forest algorithm.

[0034] This invention employs the above-mentioned technical solution, leveraging the advantages of machine learning in capturing complex relationships. Based on machine learning, a combined identification model is developed to quantitatively identify key indicators and subsystems in the process of urban sustainable development, while tolerating data correlation. This allows for the characterization of key issues in the process of urban sustainable development. This invention has the following beneficial effects:

[0035] (1) A combined identification method based on LASSO regression and random forest was developed based on machine learning to identify key indicators and key subsystems. Compared with traditional methods, it has stronger ability to identify complex relationships and better data regression fitting effect. It also has unique advantages such as tolerance for data correlation, automated feature selection, and quantification of feature importance.

[0036] (2) It can identify the key factors affecting the process of sustainable urban development more accurately. It also has strong adaptability and accuracy in the identification of specific factors in different cities, which helps decision-makers to identify key links in the urban development process and diagnose development problems more efficiently. Attached Figure Description

[0037] Figure 1 The diagram shown is a flowchart of a method according to an embodiment of the present invention.

[0038] Figure 2 The diagram shown is a schematic representation of the sustainable development capability assessment results of the innovation demonstration zone subsystem in this embodiment.

[0039] Figure 3 The diagram shown is a schematic representation of the comprehensive assessment results of the sustainable development capacity of the innovation demonstration zone in this embodiment.

[0040] Figure 4 The image shown is a panel diagram indicating the score of the innovation demonstration zone in this embodiment. Detailed Implementation

[0041] To further illustrate the various embodiments, the present invention provides accompanying drawings. These drawings are part of the disclosure of the present invention, primarily used to illustrate the embodiments, and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementations and the advantages of the present invention.

[0042] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments.

[0043] Example 1:

[0044] This invention provides a method for characterizing key issues in urban sustainable development that integrates machine learning, such as... Figure 1 As shown, the method includes the following steps:

[0045] S1: Construct a framework for assessing urban sustainable development capabilities to obtain assessment scores for each subsystem of each city and a comprehensive assessment score for each city.

[0046] The purpose of constructing an assessment framework for urban sustainable development capacity is to obtain assessment results of urban sustainable development capacity through evaluation, providing input data for subsequent identification of influencing factors based on machine learning. The construction of the assessment framework mainly includes five specific steps: policy background analysis, selection of assessment indicators, data normalization processing, weight allocation, and construction of a comprehensive index.

[0047] 1) Policy Background Analysis. Policy background analysis is an essential process in constructing an assessment indicator system for urban sustainable development capacity. By analyzing assessment frameworks, construction plans, planning documents, and assessment reports related to sustainable development at the international, national, and city levels, a thorough understanding of the local progress in sustainable development construction is beneficial for summarizing the theoretical basis and construction foundation of the assessment indicator system. Policy background analysis is mainly used for the selection of subsequent subsystems and assessment indicators.

[0048] 2) Selection of Assessment Indicators. Considering the differences in development policies, social backgrounds, and statistical systems across regions, the same assessment indicator may be replaced or disappear in different regions, leading to certain differences in assessment dimensions in practice. When selecting localized sustainable development assessment indicators within the SDGs framework, the principles of scientific and systematic indicator system design must generally be met. In addition, the following key principles should be followed in indicator selection and dimension construction:

[0049] ① Avoid duplication. In reality, the indicators selected in the localization assessment of the SDGs have a certain degree of duplication or similarity. Duplication of indicators will lead to unreliable or unstable assessment results. Therefore, it is necessary to select indicators reasonably while meeting the requirement of low correlation or similarity.

[0050] ② Emphasize localization. While the SDGs, as a UN framework for sustainable development, have strong applicability, different countries or regions may adopt different policies and measures to achieve the same sustainable development goals. Therefore, during the assessment process, indicators should be selected based on the country's or region's statistical foundation or census results. If any indicators in the SDGs are unsuitable for assessment, they should be considered for removal.

[0051] ③ Emphasize compatibility. Localized sustainable development assessment indicators should retain the original indicators to the greatest extent possible. However, since sustainable development involves numerous indicators, a certain degree of flexibility should be maintained when designing the assessment indicator system, allowing for adjustments or expansions at any time based on policy changes or actual data availability.

[0052] 3) Data normalization. Before conducting a comprehensive assessment of sustainable development capabilities, in order to eliminate the impact of discrepancies in the dimensions and magnitudes of the indicator data on the results, it is necessary to normalize the indicator values ​​according to a set threshold.

[0053] When evaluation index I is a positive index, the normalization calculation method is as follows:

[0054]

[0055] When evaluation index I is a negative index, the normalization calculation method is as follows:

[0056]

[0057] When evaluation index I is a moderate index, the normalization calculation method is as follows:

[0058]

[0059] In equations (1), (2), and (3): x represents the value of a specific evaluation indicator. max x represents the maximum value of a specific evaluation indicator. min x represents the minimum value of a specific evaluation indicator. b This indicates the appropriate value for a specific evaluation indicator. It's important to note that when an indicator exceeds the threshold, it should be assigned a value of 0 or 1 based on its inherent characteristics.

[0060] 4) Calculate the weight allocation. This embodiment also includes equal weight allocation for the urban sustainable development subsystem, and secondary equal weight allocation for the specific evaluation indicators under the subsystem. The weight calculation methods for each subsystem and specific evaluation indicator are as follows:

[0061]

[0062] In the formula: W represents the weight of each urban sustainable development subsystem, and N represents the number of urban sustainable development subsystems involved in the evaluation index system.

[0063] The specific evaluation indicators will be further weighted based on the equal weighting of the urban sustainable development subsystem. The weighting calculation method for the specific evaluation indicators is as follows:

[0064]

[0065] In the formula: Wi This represents the weight of each specific evaluation indicator in the i-th urban sustainable development subsystem, where i represents the subsystem number and n represents the subsystem index. i This represents the number of specific evaluation indicators involved in the i-th urban sustainable development subsystem.

[0066] 5) Assessment Score Calculation. Based on the weighting allocation, the final urban sustainable development capacity (USDC) assessment index is calculated using the following formula:

[0067]

[0068] In the formula: S represents the comprehensive evaluation score, S i I represents the evaluation score of the i-th urban sustainable development subsystem. ij This represents the value of the j-th specific evaluation indicator in the i-th urban sustainable development subsystem, where j represents the index of the evaluation indicator.

[0069] Therefore, the construction process of the evaluation framework in this embodiment includes the following steps:

[0070] S101: Select assessment indicators for urban sustainable development assessment and assign all assessment indicators to various subsystems;

[0071] S102: Obtain the index values ​​of each evaluation indicator and perform normalization processing;

[0072] S103: Calculate the weight of each evaluation indicator;

[0073] S104: Based on the weights of each indicator and the results of normalization, calculate the evaluation scores of each subsystem and the comprehensive evaluation score of the city.

[0074] S2: Construct a combined identification model based on the feature screening module and the feature importance quantification module to identify key indicators and key subsystems and calculate their relative importance.

[0075] In this embodiment, the feature selection module preferably uses the LASSO regression model, and the feature importance quantification module preferably uses the random forest algorithm.

[0076] Urban sustainable development systems are complex mega-systems, and their key interrelationships are difficult to capture using general linear regression and correlation analysis. Compared to other general linear regression models, the LASSO regression model has advantages such as automatic key feature selection and better model fit, and its sparsity processing can also effectively handle correlations between indicators. However, LASSO regression is still essentially a linear regression model, with coefficients that can be positive or negative. Since higher scores on all sustainable development indicators lead to better overall evaluation results, further identification of the relative importance of the selected features in improving sustainable development capabilities is still needed. Random forests, as ensemble learning algorithms, can better handle complex relationships between variables, thus effectively identifying the relative importance of key features in the dependent variable's fit and prediction.

[0077] Therefore, this embodiment constructs a LASSO-RF combined identification model based on machine learning to identify key indicators and key subsystems in the process of urban sustainable development. First, specific evaluation indicators (specific urban subsystems) are input into the LASSO regression model as independent variables and the comprehensive evaluation result as the dependent variable. Second, the key indicators (key urban subsystems) obtained from the LASSO regression model are input into the Random Forest (RF) model as independent variables and the comprehensive evaluation result as the dependent variable. The relative importance quantification result of the model (i.e., the relative importance magnitude) is used as the specific impact of the key indicators (key urban subsystems) on the comprehensive evaluation result. Finally, based on the quantification results of the LASSO-RF identification model, the key components for improving urban sustainable development capabilities can be identified.

[0078] 1) LASSO regression model

[0079] Since its inception, the LASSO regression model has been widely used to solve problems related to feature selection and variable estimation. The LASSO regression model is a compression estimation method that constructs a penalty function to compress some regression coefficients, reducing the coefficients of some independent variables that have no significant impact on the sustainable development capability index to zero. This results in a more refined model, thereby improving the problem of multicollinearity among variables.

[0080]

[0081] In the formula: Y represents the dependent variable, X represents a matrix, β represents the parameter vector, λ represents the L1 regularization coefficient, and λ||β||1 represents the penalty function. When λ increases, the penalty of the LASSO regression model becomes stronger, and more estimates of the regression coefficients of the independent variables will approach 0, thus eliminating more feature variables. Conversely, when λ decreases, the penalty of the LASSO regression model decreases, and the model will retain more feature variables.

[0082] 2) Random Forest

[0083] Random forest is a tree model composed of independent regression decision trees. It employs a bagging ensemble strategy, obtaining the optimal regression result by averaging the values ​​of each decision tree. It is one of the most practical algorithms in current research. Random forest does not require normalization of the original data, has no restrictions on the type of input variables, and effectively solves the collinearity problem among multidimensional variables. During model construction, random forest possesses dual randomness in both sample and feature selection, ensuring that the training samples and fitting features of each decision tree's nodes are randomly selected, thus avoiding overfitting. For each feature variable input into the random forest model, its feature importance value is positive, and the sum of the importance values ​​of all feature variables is generally 1.0. The higher the importance value of a feature variable, the greater its contribution to the prediction function, and vice versa.

[0084] This embodiment performs the following operations based on the constructed combined recognition model:

[0085] The values ​​of each evaluation indicator are used as independent variables and the comprehensive evaluation score is used as the dependent variable. The LASSO regression module is used to filter out the key indicators among all evaluation indicators. The key indicators output by the LASSO regression module are used as independent variables and the comprehensive evaluation score is used as the dependent variable. The random forest module is then used to obtain the relative importance of each key indicator.

[0086] The evaluation scores of each subsystem are used as independent variables and the overall evaluation score is used as the dependent variable. The LASSO regression module is used to screen out the key subsystems among all subsystems. The key subsystems output by the LASSO regression module are used as independent variables and the overall evaluation score is used as the dependent variable. The key subsystems are then used as independent variables and the overall evaluation score is used as the dependent variable. The key subsystems are then used as independent variables and the random forest module is used as the dependent variable to obtain the relative importance of each key subsystem.

[0087] It should be noted that, in addition to the LASSO regression module, the feature selection model used in this embodiment can also employ other machine learning models that simultaneously tolerate data correlation (collinearity) and achieve automatic feature selection, including but not limited to LASSO and Elastic Net, etc., without limitation. Similarly, in addition to the random forest algorithm, the feature importance quantization model used in this embodiment can also employ other feature importance quantization algorithms with different ensemble learning strategies, including but not limited to RF, XGBoost, GBDT, etc., without limitation.

[0088] S3: Based on the evaluation scores of all indicators, display all indicator scores through a score indicator panel (e.g., Figure 4Based on the evaluation scores of all subsystems, the evaluation scores of all subsystems are displayed using a radar chart (e.g., Figure 2 ).

[0089] 1) The SDG score indicator panel in the Global Sustainable Development Report is a visualization tool used to assess and display the progress of SDG achievement. This score indicator panel clearly shows the degree of achievement of each goal, facilitating comparison of progress differences between different goals and providing policymakers with clear direction and priorities. Its characteristics include simplicity, intuitiveness, ease of understanding and dissemination, and the ability to quickly identify areas requiring focused attention. This method standardizes each indicator to a score range of 0-100, intuitively reflecting the position of each goal between "worst" and "best." Therefore, this embodiment considers a score of 0-30 to indicate extremely poor sustainability, 30-60 to be relatively poor, 60-80 to be moderate, and 80-100 to be excellent.

[0090] 2) A radar chart is a graphical method for displaying multivariate data in the form of a two-dimensional chart representing three or more quantitative variables on an axis starting from the same point. Radar charts can better show the scores of each sustainable development subsystem.

[0091] S4: Based on the relative importance of each key subsystem, each key subsystem is divided into different importance levels, and corresponding key indicators are matched based on the key subsystems. The key indicators of each key subsystem are arranged from high to low relative importance.

[0092] In addition, when the subsystem corresponding to a key indicator is not a key subsystem, the importance level of the key indicator and its corresponding key subsystem is set based on the relative importance of the key indicator.

[0093] S5: Display the key subsystems and key indicators at different importance levels, along with their corresponding relative importance, through a list (such as Table 3).

[0094] Case Study

[0095] (1) Construction and assessment results of the urban sustainable development capacity assessment framework

[0096] This embodiment uses 11 National Sustainable Development Agenda Innovation Demonstration Zones (hereinafter referred to as "Innovation Demonstration Zones"), namely Taiyuan, Guilin, Shenzhen, Chenzhou, Lincang, Chengde, Ordos, Xuzhou, Huzhou, Zaozhuang, and Hainan Tibetan Autonomous Prefecture (hereinafter referred to as "Hainan Prefecture"), as case study areas. Under the vision of coordinated carbon reduction goals, and referring to international indicator frameworks such as the SDGs global indicator framework and the Global Urban Monitoring Framework (UMF), and combined with various development strategies, it focuses on the relevant indicators involved in the annual sustainable development reports and sustainable development-related construction plans (action plans) of the Innovation Demonstration Zones. Under the premise of available urban data, the city's sustainable development capacity assessment indicator system was finally constructed through direct adoption, local substitution, and local addition. The assessment indicator system mainly covers 45 sustainable development capacity evaluation indicators (see Table 1) across five urban subsystem dimensions: Technological Innovation (TI), Economic Development (ED), People's Wellbeing (PW), Resource Utilization (RU), and Ecological Environment (EE).

[0097] Table 1

[0098]

[0099]

[0100] This embodiment, based on the constructed urban sustainable development capacity assessment index system, calculates the scores of 11 innovation demonstration zones from 2015 to 2022 in five sustainable development subsystems: technological innovation, economic development, people's well-being, resource utilization, and ecological environment through a comprehensive assessment framework. Figure 2 As shown, the changes in the comprehensive assessment results of the sustainable development capabilities of all innovation demonstration zones from 2015 to 2022 were calculated based on the subsystem scores. 2015 was used as the base year, and 2016, 2019, and 2022 were used as the main time nodes for analysis. Figure 3 As shown.

[0101] Depend on Figure 2An analysis of the comprehensive assessment results of the sustainable development capabilities of innovation demonstration zones reveals that Shenzhen's comprehensive score for sustainable development capability is significantly higher than other innovation demonstration zones, indicating a stronger overall sustainable development capability. Hainan Prefecture's comprehensive score for sustainable development capability is lower, which only indicates a certain degree of weakness in its overall sustainable development capability; the main reason for its lower comprehensive assessment score is the presence of numerous missing indicator data. Apart from Shenzhen, Xuzhou, Zaozhuang, and Hainan Tibetan Autonomous Prefecture (hereinafter referred to as "Hainan Prefecture"), the comprehensive scores of the other seven innovation demonstration zones in each year are generally within the 40-50 point range, indicating relatively similar levels of comprehensive sustainable development capability.

[0102] Depend on Figure 3 An analysis of the sustainable development capacity assessment results of the urban subsystems of the innovation demonstration zones reveals that the overall sustainable development capacity of the science and technology innovation subsystem within the innovation demonstration zones is relatively weak, with significant differences observed. Shenzhen and Taiyuan demonstrate relatively stronger comprehensive sustainable development capacity in science and technology innovation. In the economic development subsystem, only Shenzhen scored above 10 points in all years, indicating a superior economic development capacity for the innovation demonstration zone. Compared to 2019, the economic development subsystem scores of the innovation demonstration zones in 2022 generally declined or stagnated. Since 2016, the social welfare subsystem scores of all innovation demonstration zones have gradually increased, indicating a gradual strengthening of the system's sustainable development capacity. Regarding the resource utilization subsystem, except for Chenzhou, whose 2022 score was lower than its 2015 score, the other 10 innovation demonstration zones all scored higher in 2022 than the baseline year. At the ecological environment subsystem level, except for Hainan Prefecture, whose score remained largely unchanged across all stages, other innovation demonstration zones showed an upward trend, particularly Chengde, Taiyuan, Xuzhou, and Zaozhuang, where the ecological environment has significantly improved.

[0103] (2) Accuracy evaluation of machine learning-based influencing factor identification methods

[0104] This embodiment constructs a LASSO-RF combined identification model based on 45 assessment indicators and 5 subsystems for each city year by year. This model uses automatic feature selection technology to intelligently identify key subsystems and main indicators of urban sustainable development capabilities. Experiments show that setting alpha = 0.001 and max_iter = 2000 in the LASSO regression and setting alpha = 0.01 and max_iter = 1000 in the subsystems results in a suitable retention rate for the selected main indicator and subsystem elements. Random forest can achieve good fitting results using default parameters.

[0105] This embodiment uses R 2 The regression modeling accuracy of all models was evaluated using two indicators: RA(higher is better) and RMSE(lower is better). The results are shown in Table 2. 2The evaluation results show that the R-squared of the RF model (0.7707) and the LASSO model (0.7186) based on the subsystem in Hainan Prefecture are significantly higher. 2 Below 0.80, the R-values ​​of all other models are... 2 All were above 0.80. The RMSE evaluation results show that the RF model built based on the evaluation indicators in Hainan Prefecture, and the two models built based on subsystems, have RMSEs significantly higher than 0.10. The RF model built based on the evaluation indicators in Lincang City has an RMSE slightly higher than 0.10, while the RMSEs of all other models are below 0.10. The overall lower accuracy of the recognition model in Hainan Prefecture is likely due to significant data gaps. Overall, the average R-value of all LASSO-RF combined recognition models is... 2 With a score above 0.95 and an RMSE below 0.1, the model has good overall accuracy. Therefore, the identification results obtained based on the model can be used for research on influencing factor identification and key issue characterization.

[0106] Table 2

[0107]

[0108] (3) Quantitative characterization results of key issues in multi-level urban development

[0109] The key subsystems and key indicators of each innovation demonstration zone were identified using the LASSO-RF combined identification model. A comparison table of the identification results is shown in Table 3. This is combined with the subsystem score radar chart (see...). Figure 2 ), indicator score panel chart (see) Figure 4 (Note: 0-30 points indicate extremely poor sustainable development, 30-60 points indicate poor sustainable development, 60-80 points indicate moderate sustainable development, and 80-100 points indicate excellent sustainable development.) Based on the identification results of influencing factors and actual construction progress of each innovation demonstration zone, key issues can be quantitatively characterized, thereby proposing corresponding development suggestions.

[0110] Table 3

[0111]

[0112]

[0113]

[0114] Note: The values ​​in parentheses indicate the relative importance of the shutdown key subsystem or key indicator. "*" indicates a town subsystem that is not identified as a key subsystem, but still has some key influencing factors corresponding to it.

[0115] The future development of Chengde's urban sustainable development capacity should focus on addressing social welfare issues and improving the level of scientific and technological innovation. It should prioritize increasing per capita GDP, striving to increase the urbanization rate, reducing the proportion of people receiving minimum living allowances, increasing forest coverage, and reducing the rate of water resource development and utilization, all of which will help to continuously improve water conservation capacity.

[0116] Taiyuan's future urban sustainable development capacity building should focus on addressing social welfare issues, striving to improve public cultural services and innovation capabilities. Furthermore, Taiyuan needs to pay attention to water sustainability, increasing per capita water resources, reducing water resource development and utilization rates, and lowering per capita daily domestic water consumption. It should also improve land use efficiency and air quality, promoting urban transformation and upgrading through high-quality economic development.

[0117] The future development of Ordos City's sustainable development capacity should focus on addressing social welfare issues and improving technological levels, strengthening infrastructure construction and public cultural services, enhancing import and export trade, strictly controlling carbon emissions and energy efficiency, and achieving green development by accelerating industrial transformation and reducing energy consumption.

[0118] The future development of Xuzhou's urban sustainable development capacity should focus on strengthening ecological and environmental protection, continuously improving scientific and technological innovation capabilities and economic development levels, promoting people's well-being by increasing per capita GDP, wages, and employment rates, continuously improving the efficiency of urban construction land use, and reducing water resource development rates, thereby achieving innovation-led high-quality development.

[0119] The future development of Huzhou's urban sustainable development capacity should focus on strengthening ecological environmental protection and enhancing scientific and technological innovation capabilities, striving to improve urban air quality, increasing the proportion of general public budget expenditure on science and technology, developing higher education, promoting green industrial transformation and reducing energy consumption.

[0120] The future development of sustainable urban capacity in Zaozhuang City should focus on strengthening ecological environmental protection and technological innovation, vigorously protecting land use types that are beneficial to increasing carbon sequestration, improving public cultural services and transportation, increasing the efficiency of water resource and construction land use, and promoting sustainable rural development.

[0121] The future development of sustainable urban development capacity in Chenzhou should focus on enhancing technological innovation capabilities, increasing R&D investment, reducing unemployment by developing the tertiary industry to create employment opportunities, improving the efficiency of water resource and construction land utilization, and achieving high-quality development based on the sustainable utilization of water resources.

[0122] Shenzhen's future urban sustainable development capacity building should continue to maintain the momentum of technological innovation, further strengthen patent protection awareness, promote the transformation of patent achievements, strengthen transportation infrastructure construction, increase the teacher-student ratio in basic education, and at the same time strive to ensure domestic water supply and reduce the rate of water resource development and utilization, so as to lead the high-quality development of a megacity through technological innovation.

[0123] Guilin's future urban sustainable development capacity building should focus on strengthening ecological environmental protection, promoting technological innovation, striving to increase the area of ​​parks and green spaces, protecting and increasing land use types that are beneficial to carbon sequestration, improving the level of basic education and public cultural services, and achieving sustainable development with landscape resource protection as the leading factor.

[0124] The future development of Lincang City's sustainable urban development capacity should focus on enhancing its technological innovation capabilities, striving to reduce the proportion of residents receiving minimum living allowances, improve the salaries and benefits of on-the-job employees, strengthen the construction of municipal water supply facilities, enhance water conservation capacity by increasing forest coverage, and thereby increase per capita water resources, so as to lead the development of underdeveloped areas through innovation.

[0125] The future development of sustainable urban development capacity in Hainan Prefecture should focus on improving resource utilization efficiency and technological innovation, effectively reducing water consumption per 10,000 yuan of GDP, striving to improve food security, reducing the proportion of people receiving minimum living allowances, increasing per capita GDP through high-quality economic development, emphasizing land protection to increase carbon sequestration, and achieving high-quality development while strengthening the protection of river source areas.

[0126] This invention leverages the advantages of machine learning in capturing complex relationships. Based on machine learning, a combined identification model is developed to quantitatively identify key subsystems and key indicators in the process of sustainable urban development while tolerating data correlation, thereby characterizing key issues in the process of sustainable urban development.

[0127] This embodiment has the following beneficial effects:

[0128] (1) This embodiment develops a combined identification method based on LASSO regression and random forest based on machine learning to identify key indicators and key subsystems. Compared with traditional methods, it has a stronger ability to identify complex relationships and a better data regression fitting effect. It also has unique advantages such as tolerating data correlation, automated feature selection, and quantification of feature importance.

[0129] (2) The method in this embodiment can identify the key factors affecting the process of sustainable urban development more accurately. It also has strong adaptability and accuracy in the identification of specific factors in different cities, which helps decision-makers to identify key links in the urban development process and diagnose development problems more efficiently.

[0130] It should be noted that the evaluation indicators in this embodiment can be other sustainable development capability elements, such as specific development dimensions, SDG scores, etc.

[0131] Although the invention has been specifically shown and described in conjunction with preferred embodiments, those skilled in the art should understand that various changes in form and detail may be made to the invention without departing from the spirit and scope of the invention as defined in the appended claims, all of which shall be within the scope of protection of the invention.

Claims

1. A method for representing key issues in urban sustainable development that integrates machine learning, characterized in that, Includes the following steps: S1: Construct a framework for assessing urban sustainable development capacity to obtain assessment scores for each subsystem of each city and a comprehensive assessment score for each city; the process of constructing the assessment framework includes: S101: Select assessment indicators for urban sustainable development assessment and assign all assessment indicators to various subsystems; S102: Obtain the index values ​​of each evaluation indicator and perform normalization processing; S103: Calculate the weight of each evaluation indicator; S104: Based on the weights of each indicator and the results of normalization, calculate the evaluation scores of each subsystem and the comprehensive evaluation score of the city. S2: Construct a combined recognition model based on the feature selection module and the feature importance quantification module, and perform the following operations: The indicator values ​​of each evaluation indicator are used as independent variables and the comprehensive evaluation score is used as the dependent variable. The feature selection module is used to select the key indicators among all evaluation indicators. The key indicators output by the feature selection module are used as independent variables and the comprehensive evaluation score is used as the dependent variable. The key indicators are used as independent variables and the comprehensive evaluation score is used as the dependent variable. The key indicators are used as independent variables and the feature importance quantification module to obtain the relative importance of each key indicator. The evaluation scores of each subsystem are used as independent variables and the comprehensive evaluation score is used as the dependent variable. The feature selection module is used to select the key subsystems among all subsystems. The key subsystems output by the feature selection module are used as independent variables and the comprehensive evaluation score is used as the dependent variable. The key subsystems are then used as independent variables and the comprehensive evaluation score is used as the dependent variable. The key subsystems are then used as independent variables and the feature importance quantification module is used to obtain the relative importance of each key subsystem. S3: Based on the evaluation scores of all indicators, display all indicators through a score indicator panel; based on the evaluation scores of all subsystems, display all subsystems through a radar chart. S4: Based on the relative importance of each key subsystem, each key subsystem is divided into different importance levels, and corresponding key indicators are matched based on the key subsystems. The key indicators of each key subsystem are arranged from high to low relative importance. S5: Display key subsystems and key indicators at different importance levels, along with their corresponding relative importance, through a list.

2. The method for representing key issues of urban sustainable development by integrating machine learning as described in claim 1, characterized in that: When selecting assessment indicators for urban sustainable development evaluation, it is necessary to take into account the current policy context.

3. The method for representing key issues of urban sustainable development by integrating machine learning as described in claim 1, characterized in that: The selection of evaluation indicators follows these principles: avoid duplication, emphasize localization, and emphasize compatibility.

4. The method for representing key issues of urban sustainable development by integrating machine learning as described in claim 1, characterized in that: The formula for calculating the weight of the evaluation indicators is as follows: Among them, W i n represents the weight of each specific evaluation indicator in the i-th subsystem. i This represents the number of specific evaluation indicators involved in the i-th subsystem, and N represents the number of subsystems.

5. The method for representing key issues of urban sustainable development by integrating machine learning according to claim 1, characterized in that: The formula for calculating the evaluation score of a subsystem is: Among them, S i I represents the evaluation score of the i-th subsystem. ij W represents the value of the j-th specific evaluation index in the i-th subsystem. i Let represent the weight of each specific evaluation indicator in the i-th subsystem, j represent the index of the evaluation indicator, and n represent the weight of each specific evaluation indicator in the i-th subsystem. i This represents the number of specific evaluation indicators involved in the i-th subsystem.

6. The method for representing key issues of urban sustainable development by incorporating machine learning as described in claim 5, characterized in that: The formula for calculating the city's comprehensive evaluation score is as follows: Where S represents the comprehensive evaluation score, i represents the subsystem number, and N represents the number of subsystems.

7. The method for representing key issues of urban sustainable development by integrating machine learning according to claim 1, characterized in that: When the subsystem corresponding to a key indicator is not a key subsystem, the importance level of the key indicator and its corresponding key subsystem is set based on the relative importance of the key indicator.

8. The method for representing key issues of urban sustainable development by incorporating machine learning as described in claim 1, characterized in that: The feature selection module uses the LASSO regression model.

9. The method for representing key issues of urban sustainable development by incorporating machine learning as described in claim 1, characterized in that: The feature importance quantification module uses the random forest algorithm.

Images