Storage industrial land development time sequence decision-making method and system considering space perspective

Abstract

The invention discloses a reserve industrial land development time sequence decision-making method and system considering a space perspective, and the method comprises the steps: S1, constructing a development potential index system, and obtaining the data of a research city reserve industrial land; s2, constructing a particle swarm optimization projection pursuit model, wherein the particle swarm optimization projection pursuit model uses all normalized index data as high-dimensional data, uses a particle swarm algorithm to iteratively adjust particle positions and finally projects the particle positions into one-dimensional data; and S3, constructing an SOFM neural network, carrying out spatial clustering on the development potentials of all spatial positions in the research city to obtain N spaces, and sorting the N spaces in sequence according to the development potentials to serve as a time sequence decision result. According to the method, the particle swarm optimization projection pursuit model projects high-dimensional data of a multi-dimensional index into one-dimensional data to represent the development potential of each spatial position, and the SOFM neural network adopts a combination of a spatial domain and an attribute domain to carry out mixed distance clustering processing, so that development potential evaluation and development time sequence decision technical support is provided for reserve industrial land.

Description

Technical Field

[0001] The present invention relates to the field of national land space planning and utilization, and in particular to a method and system for making timing decisions on the development of reserve industrial land taking into account a spatial perspective. Background Art

[0002] With the deepening of national land and space governance concepts and the tightening of land resource constraints, the scientific development and utilization of reserve industrial land has become a key component of urban renewal and industrial transformation. Reserve industrial land generally refers to industrial land that has been included in the reserve system but has not yet been put into use or is in urgent need of revitalization. Improving its utilization efficiency directly impacts the structural optimization of land supply and the rationality of regional industrial layout. The development of this type of land must be scientifically promoted in a dynamic and phased manner, comprehensively considering multiple factors such as economic value, environmental impact, and spatial coordination. The development sequencing of reserve industrial land incorporates the time variable into the land reserve management system. Through quantitative analysis of the development potential of each plot, combined with factors such as spatial characteristics and regional planning positioning, a scientifically formulated short-term, medium-term, and long-term development sequence is established, thus establishing an implementation path that gradually transitions from "near-term remediation" to "long-term optimization." This is essentially a governance strategy of "trading time for space." By rationally scheduling the development of plots, it avoids resource waste and spatial fragmentation caused by blind advancement, thereby improving overall land use efficiency and development benefits.

[0003] Existing research primarily ranks reserved industrial land based on development potential as a basis for development scheduling. While this approach has, to some extent, contributed to improvements in land use efficiency, it also suffers from significant shortcomings: 1. Ignoring spatial characteristics and regional planning positioning: Most studies use plots as the primary unit, lacking systematic consideration of spatial proximity between plots, industrial linkages, and regional planning objectives. This results in fragmented development areas, making it difficult to achieve the optimal goal of concentrated, contiguous, and coordinated development. 2. The evaluation of development potential is limited to a single dimension and lacks data accuracy: Existing industrial land development potential evaluation systems are often limited to economic or physical indicators within the plot, ignoring the impact of the surrounding environment on development outcomes. Furthermore, data collection methods often rely on static statistical calibers, making it difficult to accurately capture the spatial form, architectural style, and environmental quality of the plot. 3. Traditional methods are weakly adaptable and heavily expert-dominated: Industrial land development potential evaluation often employs models such as principal component analysis and the analytic hierarchy process, which suffer from highly subjective weighting and difficulty adapting to diverse regional characteristics, limiting the generalizability and objectivity of the results.

[0004] In summary, existing methods cannot fully respond to key questions such as "where to develop first" and "how to coordinate development". There is an urgent need for a development timing decision-making method that integrates development potential and spatial perspectives to ensure the scientific, systematic and dynamic advancement of reserved industrial land from potential identification to development implementation. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for sequential decision-making on the development of reserve industrial land that takes into account a spatial perspective. The particle swarm optimization projection pursuit model projects the high-dimensional data of multidimensional indicators into one-dimensional data to characterize the development potential of each spatial location. The SOFM neural network uses a combination of spatial domain and attribute domain to perform hybrid distance clustering processing to achieve spatial partitioning. The comprehensive development potential of each space is calculated separately and ranked in sequence as the sequential decision-making result, providing technical support for development potential evaluation and development sequential decision-making for reserve industrial land.

[0006] The purpose of the present invention is achieved through the following technical solutions:

[0007] A method for making time-sequential decisions on the development of reserved industrial land taking into account a spatial perspective includes:

[0008] S1. Construct a development potential index system divided by indicator level, obtain the reserved industrial land data including spatial location information of the research city, and normalize the data according to the corresponding indicators in the development potential index system;

[0009] S2. Construct a particle swarm optimization projection pursuit model. The particle swarm optimization projection pursuit model uses all normalized indicator data as high-dimensional data, uses the particle swarm algorithm to iteratively adjust the particle position and finally projects it into one-dimensional data, and uses the one-dimensional data containing spatial position information as the development potential;

[0010] S3. Construct a SOFM neural network to spatially cluster the development potential of all spatial locations in the study city into N spaces, and sort the N spaces in order of development potential as the temporal decision result.

[0011] In order to better realize the present invention, the development potential index system is divided into layers according to the indicator level. The indicator level includes dimension layer, characteristic layer and indicator layer from top to bottom. The dimension layer includes necessity, maturity and urgency, wherein necessity is a negative indicator and maturity and urgency are positive indicators; the characteristic layer of necessity includes economic benefits and development intensity, the indicator layer of economic benefits includes per capita employment, per mu tax revenue, per mu output value, per mu industrial added value and annual operating income, and the indicator layer of development intensity includes building density and building height; the characteristic layer of maturity includes the surrounding environment, and the indicator layer of the surrounding environment includes traffic convenience, road density, spatial hydrophilicity, industrial agglomeration, sky rate and greening rate; the characteristic layer of urgency includes industry compliance.

[0012] Preferably, the projection direction of the particle swarm optimization projection pursuit model is expressed as a={a(1), a(2), a(3), …a(j) …, a(p)}, where p is the total number of indicators in the development potential index system, and the projection value expression of data sample i is as follows:

[0013] , where i represents the data sample i of all the indicators associated with the reserved industrial land data, x(i, j) is the data of indicator j after the data sample i is normalized, and z(i) represents the projection value of the data sample i, that is, one-dimensional data;

[0014] Construct the objective function Q(a), the expression is as follows: Q(a) = S Z D Z , where S Z represents the standard deviation of the data sample, D Z Represents the local density of the data sample; the particle swarm algorithm is used to iteratively adjust the particle position to obtain the optimal projection direction and the optimal objective function; the particle swarm optimization projection pursuit model uses the optimal projection direction to calculate the one-dimensional data of the data sample i.

[0015] Preferably, in method S1, the method for normalizing the data corresponding to the indicators in the development potential index system is: normalizing the indicators in the indicator layer under the belonging level according to the dimensional layer or feature layer of the development potential index system; in method S2, the particle swarm optimization projection pursuit model statistically calculates all indicator data of the indicator layer or feature layer or dimensional layer as high-dimensional data, and uses the particle swarm algorithm to iteratively adjust the particle position and finally project it into one-dimensional data.

[0016] Preferably, the number of employees per land, per mu tax revenue, per mu output value, per mu industrial added value, annual operating income, building density, building height and road density are data from the reserved industrial land data or data obtained through calculation; the traffic convenience adopts the traffic POI point kernel density, and the kernel density expression is as follows:

[0017] , where r is the calculation radius centered on the research spatial location point, N is the number of spatial location points within the radius r, is the distance between spatial location point n within the radius r and the research spatial location point, and P is the transportation convenience of the research spatial location point.

[0018] Preferably, the method for obtaining the sky rate and greening rate in the reserve industrial land data is as follows:

[0019] Street view images including the sky and the land are collected in the study city, and the sky and vegetation are segmented using the SegFormer model trained based on the Cityscape dataset. Then, the area ratio of the sky is calculated as the sky rate, and the area ratio of the vegetation is calculated as the greening rate. The industrial compliance status is a scoring system based on the industrial compliance assessment system for enterprises using industrial land in the study city.

[0020] Preferably, the standard deviation S of the data samples Z and the local density D ZUsing local window calculation, the standard deviation S of the data sample Z is the standard deviation of the projection value of the local window data sample, and the expression is as follows:

[0021] Where m represents the total number of data samples i, and E(z) is the average value of the projection values of the data samples;

[0022] The local density D of the data sample Z To select the local density of the local density window, the expression is as follows:

[0023] R represents the radius of the local density window, r(i1, i2) represents the distance between the projection values of data samples i1 and i2, and u() represents the unit step function.

[0024] Preferably, the SOFM neural network uses a combination of spatial domain and attribute domain to perform hybrid distance clustering processing, and the hybrid distance expression is as follows:

[0025] ,in is the mixed distance of the projection values of data samples i1 and i2, is the Euclidean distance between data samples i1 and i2 in geographic space, W d It represents the weight of attribute d, which is the attribute in attribute domain D, and attribute domain D is all the indicators of the development potential index system; are the values of data samples i1 and i2 in the attribute d dimension, w s 、w a are the weights of the spatial domain and attribute domain respectively.

[0026] Preferably, in method S3, N spaces are divided into three intervals of low, medium and high according to the size of development potential, and N spatial clustering areas are sorted according to the three intervals of low, medium and high. The sorting decision corresponding to the spatial clustering area in the low interval is a long-term temporal decision, the sorting decision corresponding to the spatial clustering area in the medium interval is a medium-term temporal decision, and the sorting decision corresponding to the spatial clustering area in the high interval is a short-term temporal decision. The temporal decision result, the medium-term temporal decision and the short-term temporal decision as well as the sorting within each temporal decision together constitute the temporal decision result.

[0027] A time-series decision-making system for the development of reserve industrial land that takes into account a spatial perspective comprises a development potential index system, a data acquisition and processing module, a particle swarm optimization projection pursuit model, and a SOFM neural network. The development potential index system is divided into a hierarchy according to an index level. The data acquisition and processing module is used to obtain reserve industrial land data containing spatial location information in a study city and perform data normalization processing according to the corresponding indicators in the development potential index system. The particle swarm optimization projection pursuit model uses all normalized indicator data as high-dimensional data, utilizes a particle swarm algorithm to iteratively adjust particle positions, and ultimately projects the data into one-dimensional data. The one-dimensional data containing spatial location information is used as the development potential. The SOFM neural network spatially clusters the development potential of all spatial locations in the study city into N spaces, sorts the N spaces in sequence according to development potential, and outputs the result of a time-series decision.

[0028] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0029] (1) The development potential index system of the present invention adopts a multi-level and multi-dimensional index structure with hierarchical layers of indicators. The particle swarm optimization projection pursuit model projects the high-dimensional data of the multi-dimensional indicators into one-dimensional data to characterize the development potential of each spatial location. The SOFM neural network uses a combination of spatial domain and attribute domain to perform hybrid distance clustering processing to achieve spatial division. The comprehensive development potential of each space is calculated separately and ranked in sequence as the result of time series decision-making, providing technical support for development potential evaluation and development time series decision-making for reserve industrial land.

[0030] (2) Based on the spatial clustering results and the comprehensive potential level of the region, the present invention scientifically defines the short-term, medium-term and long-term development timelines, promotes the dynamic transformation of reserved industrial land from inefficient idleness to intensive and efficient use, and provides a decision-making basis for regional industrial layout optimization, coordinated allocation of land resources and urban renewal.

[0031] (3) The present invention has achieved a refined and intelligent arrangement of the development sequence of reserved industrial land by constructing a scientific and complete industrial land development potential indicator system and integrating the spatial feature clustering method, and has achieved remarkable results in many aspects.

[0032] (4) The present invention solves the limitations, inaccuracies and incompleteness of the traditional single evaluation dimension, introduces a dual clustering method of spatial domain and attribute domain of spatial coordinates and attribute features, and adopts a self-organizing feature mapping neural network (i.e., SOFM neural network) for spatial clustering processing. With the help of the self-learning and adaptive characteristics of the SOFM neural network, the present invention can flexibly adjust the classification results according to different regions and industrial characteristics, and is suitable for various scenarios such as urban renewal, land consolidation, and industrial optimization. It has good promotion value and practical operability; it ensures that reserve industrial land with similar potential is concentrated and connected in space, avoids fragmentation of development areas, and improves land use efficiency and development synergy; the technical implementation of the present invention is conducive to industrial resource integration and functional agglomeration, promotes the coordinated construction of industrial chains and supply chains, enhances the overall competitiveness of the regional economy, improves the development income of industrial land, and promotes the healthy development of the regional land market.

[0033] (5) The present invention integrates street view images, remote sensing data, POI information and multi-source socioeconomic data, and combines computer vision with spatial analysis methods to truly restore the physical state, spatial pattern and industrial connection of the land parcels, making up for the shortcomings of traditional single-source data analysis and improving the accuracy and reliability of industrial land development potential assessment. The present invention adopts the particle swarm optimization projection pursuit model to effectively overcome the problems of strong subjectivity and difficulty in eliminating regional differences in traditional methods, and realizes objective scoring and weight adaptive optimization under data-driven conditions, ensuring that the evaluation results are fair, reusable and highly adaptable across regions.

[0034] (6) The present invention provides a systematic solution for government departments, land reserve agencies, development companies, etc. from identifying the development potential of industrial land to making development timing decisions. It helps to formulate annual land use implementation plans, improve development efficiency and land resource allocation benefits, and promote industrial transformation and green and healthy development. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 A method flow chart of the method for making decisions on the timing of industrial land reserve development according to the present invention;

[0036] Figure 2 This is an example diagram of semantic segmentation results of street view images of a certain area in a certain city in the embodiment;

[0037] Figure 3 In the embodiment, a schematic diagram is provided to summarize the development potential of a city's reserved industrial land according to the three dimensions of necessity, maturity, and urgency;

[0038] Figure 4 This is a spatial clustering result diagram of various urban areas in a city center as an example in the embodiment;

[0039] Figure 5 This is a time sequence decision diagram for the development of the first central urban area reserve industrial land of a certain city in the embodiment;

[0040] Figure 6 This is a time sequence decision diagram for the development of industrial land reserved in the second central urban area of a certain city in the embodiment;

[0041] Figure 7 This is a time sequence decision diagram for the development of reserved industrial land in the third central urban area of a certain city in the embodiment;

[0042] Figure 8 This is a time sequence decision diagram for the development of reserved industrial land in the fourth central urban area of a certain city in the embodiment;

[0043] Figure 9 This is a time sequence decision diagram for the development of reserved industrial land in the fifth central urban area of a certain city in the embodiment;

[0044] Figure 10 This is a timing decision diagram for the development of reserved industrial land in the sixth central urban area of a city in the embodiment. DETAILED DESCRIPTION

[0045] Below in conjunction with embodiment, the present invention is described in further detail:

[0046] Example

[0047] like Figure 1 As shown, a method for making a time sequence decision on the development of reserve industrial land taking into account a spatial perspective includes:

[0048] S1. Construct a development potential index system divided by indicator levels, obtain the reserve industrial land data containing spatial location information of the research city, and perform data normalization according to the corresponding indicators in the development potential index system. In some embodiments, the development potential index system is divided into layers according to the indicator level. The indicator level includes a dimension layer, a feature layer, and an indicator layer from top to bottom. The dimension layer includes necessity, maturity, and urgency (the present invention evaluates development potential according to multiple dimensions of necessity, maturity, and urgency, which improves the rationality and application value of the evaluation results), where necessity is a negative indicator and maturity and urgency are positive indicators; the feature layer of necessity includes economic benefits (reflecting the contribution of industrial land to the regional economy, which can effectively evaluate whether it has high development potential; generally speaking, industrial land with low economic benefits is often difficult to fully realize the value of the land, so the necessity of development is higher) and development intensity (reflecting the spatial utilization efficiency of industrial land, which can effectively evaluate the development cost and feasibility of land; generally speaking, plots with low development intensity have more available space, greater adjustment flexibility, and higher development potential). The indicator layer includes the number of employees per unit area, per mu tax revenue, per mu output value, per mu industrial added value and annual operating income; the indicator layer of development intensity includes building density and building height; the characteristic layer of maturity (which helps to screen out areas with good development conditions and improve the feasibility and economic benefits of land use; the maturity of industrial land reflects the feasibility and market attractiveness of its transformation; areas with higher maturity usually have complete infrastructure, convenient transportation conditions and a good industrial development environment, which helps to reduce development costs, improve land use efficiency, and attract high-quality industries; on the contrary, industrial land with lower maturity may face problems such as backward infrastructure, poor accessibility, and insufficient supporting facilities, making the implementation of development more difficult) includes the surrounding environment; the indicator layer of the surrounding environment includes transportation convenience, road density, spatial hydrophilicity, industrial agglomeration, sky rate and greening rate; the characteristic layer of urgency includes industrial compliance.

[0049] In this embodiment, necessity is a negative indicator, while maturity and urgency are positive indicators. For negative indicators (the larger the absolute value, the lower the potential, which can be expressed as a negative number), the data normalization expression is as follows:

[0050]

[0051] For positive indicators (the larger the absolute value, the higher the potential, which can be expressed as a positive number), the data normalization expression is as follows:

[0052]

[0053] x jmax and x jminare the maximum and minimum values of the j-th indicator respectively, and x(i, j) is the normalized indicator value.

[0054] In some embodiments, the present invention uses the average number of employees per area (collecting or calculating the average number of employees at the spatial location point), the average tax per mu (collecting or calculating the average tax per spatial location point), the average output value per mu (collecting or calculating the average output value at the spatial location point), the average industrial added value per mu (collecting or calculating the average industrial added value at the spatial location point), the annual operating income (collecting or calculating the annual operating income at the spatial location point, and the annual operating income at the spatial location point can be calculated by allocating the spatial location point using the reserve industrial land data in the area), the building density, the building height, and the road density are data in the reserve industrial land data or data obtained by calculation, and the above-mentioned spatial location points can be spatial location areas. The traffic convenience of the present invention adopts the kernel density of traffic POI points (the building density, building height, and road density of the present invention can all be calculated using the kernel density method), and the kernel density expression is as follows:

[0055] Where r is the calculation radius centered on the research space location point, N is the number of spatial location points within the radius r, and d n is the distance between spatial location point n and the study spatial location point within radius r, and P is the accessibility of the study spatial location point. Each study spatial location point captures street view images in four directions (0°, 90°, 180°, and 270°) (these angles represent the views from the front, right, back, and left of the plot, respectively).

[0056] The building density and building height of the present invention can also be obtained by the following method:

[0057] The average building height and density at a given location are calculated based on multi-source remote sensing features (such as SAR imagery, optical imagery, and terrain data). Alternatively, the GEE platform can be used to extract 175 feature variables from radar, optical, terrain, socioeconomic, and vector data, and use XGBoost regression to estimate building height and density for a given study year. Remote sensing data is primarily based on the study year, with imagery from prior years supplemented for missing areas.

[0058] In some embodiments, the sky rate and green rate in the reserve industrial land data are obtained as follows:

[0059] Collect street view images of the study city containing the sky and the ground, and perform sky and vegetation segmentation using the SegFormer model trained on the Cityscape dataset (generate a mask image of its category division for each street view image, such as Figure 2As shown, this embodiment takes the semantic segmentation results of street view images of a certain street area in a certain city (for example, Ningbo City) as an example, which can clearly segment the two key environmental elements of sky and vegetation. By calculating the area ratio of the two key environmental elements of sky and vegetation, two potential indicators of sky visibility and greening rate are obtained to characterize and evaluate the environmental maturity of the surrounding reserves), and then the area ratio of the sky is calculated as the sky rate, and the area ratio of vegetation is calculated as the greening rate; the industrial compliance status is to study the scoring system of the enterprises using industrial land in the city according to the industrial compliance assessment system. The industrial compliance assessment system includes the locations of prohibited and eliminated industries and locations where safety and environmental protection do not meet the standards. The urgency and potential dimensions of industrial land development are considered from the two aspects of the locations of prohibited and eliminated industries and locations where safety and environmental protection do not meet the standards. It is mainly used to screen industrial land where enterprises with urgent environmental pollution problems or backward technologies are located. With the adjustment of industrial structure and the improvement of environmental protection requirements, some inefficient and heavily polluting industrial land has caused a heavy burden on the environment and public health. The development of these industrial lands has become a top priority. In addition, the technology of enterprises that rely on some industrial lands is relatively backward and their production efficiency is low. They are in urgent need of modernization and upgrading in order to better integrate into the new industrial structure. The following methods can be used for evaluation: According to the "Guiding Catalogue for Industrial Structure Adjustment" issued by the state, all industries are divided into encouragement, restriction and elimination categories. The elimination category mainly includes industries that do not comply with relevant laws and regulations, seriously waste resources, pollute the environment, have serious safety risks, and hinder the realization of carbon peak and carbon neutrality goals. The backward processes, equipment and products that need to be eliminated are of high urgency. The development of industrial land for these industries is of high urgency. The present invention divides and evaluates the industries of the enterprises in the reserve through the "Guiding Catalogue for Industrial Structure Adjustment". The "List of Key Polluting Units" issued by the local government has made a detailed classification of enterprises that discharge pollutants exceeding the standards, divided into a list of key water environment polluting units, a list of key atmospheric environment polluting units, a list of key regulated units for soil environment pollution, and a list of other key polluting units; the present invention uses the "List of Key Polluting Units" to screen enterprises that discharge pollutants exceeding the standards, thereby dividing and evaluating the urgency of reserve development where the enterprises are located.

[0060] S2. Construct a particle swarm optimization projection pursuit model. The particle swarm optimization projection pursuit model uses all normalized indicator data as high-dimensional data, uses the particle swarm algorithm to iteratively adjust the particle position and finally projects it into one-dimensional data, and uses the one-dimensional data containing spatial position information as development potential.

[0061] In some embodiments, the particle swarm optimization projection pursuit model (PSO-PP model for short) projection pursuit (PP) model is a data-driven analysis method that can find the best projection direction from high-dimensional data to objectively reflect the impact of each indicator on the comprehensive evaluation; however, since different projection directions show different data characteristics, determining the optimal projection direction involves a complex multi-dimensional nonlinear optimization problem (traditional methods are difficult to solve effectively), and the particle swarm optimization projection pursuit model can intelligently adjust the particle search path to approach the optimal solution, thereby improving the optimization effect; the global optimization ability of the particle swarm optimization projection pursuit model of the present invention (using the PP model combined with the PSO algorithm) effectively overcomes the challenges of high-dimensional data optimization, and calculates the development potential value of the construction site based on this. The projection direction of the particle swarm optimization projection pursuit model is expressed as a = {a(1), a(2), a(3), ...a(j) ..., a(p)}, p is the total number of indicators in the development potential index system, and the projection value expression of data sample i is as follows:

[0062] Where i represents the data sample i of all the indicators in the reserved industrial land data, x(i, j) is the data of indicator j after the normalization of data sample i, and z(i) represents the projection value of data sample i, that is, one-dimensional data;

[0063] Construct the objective function Q(a), the expression is as follows, Q(a) = S Z D Z , where S Z represents the standard deviation of the data sample, D Z Represents the local density of the data sample; uses the particle swarm algorithm to iteratively adjust the particle position to obtain the optimal projection direction and the optimal objective function (the purpose of the present invention is to iteratively constrain the optimal objective function. The smaller the objective function value is or less than the set threshold, the more it is judged to be the optimal objective function, and the projection direction corresponding to the optimal objective function is the optimal projection direction); the particle swarm optimization projection pursuit model uses the optimal projection direction to calculate the one-dimensional data of the data sample i. This embodiment continuously calculates the projection value z(i) and the objective function Q(a), and obtains the optimal projection direction a* and the optimal projection index function Q*(a). The calculated optimal projection direction a* is substituted into the formula The optimal projection value z*(i) is obtained, which can retain the key information of the original data to the greatest extent while reducing the data dimension and complexity, thereby effectively representing the development potential of industrial land.

[0064] In some embodiments, the standard deviation S of the data samples Z and the local density D Z Using local window calculation, the standard deviation S of the data sample Z is the standard deviation of the projection value of the local window data sample, and the expression is as follows:

[0065] Where m represents the total number of local window data samples i, and E(z) is the average value of the projection values of the local window data samples;

[0066] The local density D of the data sample Z To select the local density of the local density window, the expression is as follows:

[0067] R represents the radius of the local density window, r(i1, i2) represents the distance between the projection values of data samples l1 and l2 in the local window, and u() represents the unit step function.

[0068] S3. Construct a SOFM neural network to spatially cluster the development potential of all spatial locations in the study city into N spaces, and sort the N spaces in order of development potential as the temporal decision result.

[0069] In some embodiments, in method S1, the method for normalizing the data corresponding to the indicators in the development potential index system is: normalizing the indicators in the indicator layer under the attribute layer according to the dimension layer or feature layer of the development potential index system. In method S2, the particle swarm optimization projection pursuit model statistically calculates all indicator data of the indicator layer, feature layer or dimension layer as high-dimensional data, and uses the particle swarm algorithm to iteratively adjust the particle position and finally project it into one-dimensional data. Figure 3 As shown in the figure, this embodiment summarizes the development potential of industrial land reserves in a certain city (for example, Ningbo) according to the three dimensions of necessity, maturity and urgency (all indicator data of the statistical dimension layer of the particle swarm optimization projection pursuit model are used as high-dimensional data, and the particle position is iteratively adjusted by the particle swarm algorithm and finally projected into one-dimensional data). The development potential results are shown as follows: Figure 3 shown; from Figure 3 As can be seen from (a), industrial land with medium development necessity accounts for a significant proportion and is mainly distributed in the central location of the region, which shows that the economic benefits and land development intensity of most reserve industrial land still need further development; Figure 3 (b) It can be seen that there are more industrial land with medium and high development maturity, which means that most industrial land has good surrounding facilities and good development conditions. Figure 3 As can be seen from (c), industrial land with medium and low development urgency is dominant, indicating that most low-efficiency industrial land has not yet reached a state of urgent need for transformation in terms of environmental pollution and technological backwardness; Figure 3(d) It can be seen that the development potential of reserved industrial land shows obvious spatial differentiation characteristics. Industrial land with medium and high development potential is mostly distributed in the urban core area, while industrial land with low development potential is mostly distributed in the urban fringe area. Therefore, before development, it should be developed according to different comprehensive development potentials. Through this phased development strategy, limited resources can be utilized in the most effective way to achieve orderly development of the city.

[0070] In some embodiments, the SOFM neural network uses a combination of spatial domain and attribute domain to perform hybrid distance clustering processing. The hybrid distance calculation expression is as follows:

[0071] in is the mixed distance of the projection values of data samples i1 and i2, is the Euclidean distance between data samples i1 and i2 in geographic space, W d It represents the weight of attribute d, which is the attribute in attribute domain D, and attribute domain D is all the indicators of the development potential index system; are the values of data samples i1 and i2 in the attribute d dimension, w s 、w a are the weights of the spatial domain and attribute domain respectively, w s +w a =1. The present invention is based on hybrid distance clustering processing (based on hybrid distance, the neuron with the smallest distance is selected to determine the winning node), and traverses all samples until the preset maximum number of iterations is reached or the weight change is less than the threshold. The final competition layer outputs the clustering result of the development space of the reserve industrial land (i.e., N spaces), calculates the comprehensive development potential of the N spaces, and sorts the N spaces in order according to the comprehensive development potential as the timing decision result (the development timing of the N spaces). Figure 4 As shown, this embodiment takes the spatial clustering results of the urban areas in the center of a city (for example, Ningbo) as an example. Figure 4 It shows the spatial clustering results of each center and urban area of a city. In this embodiment, the x and y coordinates of the geometric center points of each reserve industrial land are used as geographic input variables, and the three-dimensional evaluation values of development necessity, maturity and urgency calculated by the particle swarm optimization projection pursuit model (PSO-PP model for short) are used as attribute input variables; the design network parameters are: 118×5 input layer matrix, the number of random seeds is 10, the basic learning rate is 0.5, the maximum number of training times is 500, the geographic space weight is 0.4, and the attribute space weight is 0.6; due to different results produced by different numbers of SOFM clusters, after repeated experiments, it was found that when the number of clusters is 3, the centralized development effect of the spatial clustering results is the most obvious, so the output layer, that is, the number of categories, is set to 3, and the reserve industrial land in each district is divided into three cluster areas according to the principle of "spatial adjacency and similar attributes", such as Figure 5As shown; these three clusters will provide spatial constraints for the subsequent short-term, medium-term and long-term development timing arrangements.

[0072] In some embodiments, in method S3, N spaces are divided into three intervals of low, medium and high according to the size of development potential, and N spatial clustering areas are sorted according to the three intervals of low, medium and high. The sorting decision corresponding to the spatial clustering area in the low interval is a long-term temporal decision, the sorting decision corresponding to the spatial clustering area in the medium interval is a medium-term temporal decision, and the sorting decision corresponding to the spatial clustering area in the high interval is a recent temporal decision. The temporal decision result, the medium-term temporal decision and the recent temporal decision as well as the sorting within each temporal decision together constitute the temporal decision result. Figures 5 to 10 As shown, this embodiment takes the six central urban areas of a city (for example, Ningbo) as an example and calculates the development potential of the reserve industrial land according to the reserve industrial land development time sequence decision method of the present invention, and then outputs the development time sequence decision diagrams of the six central urban areas respectively (see Figures 5 to 10 ). This embodiment divides the reserve industrial land to be developed into several areas according to the spatial clustering results, calculates the comprehensive development potential of the reserve industrial land in each area, compares the development potential of these areas, and determines the development sequence of these areas according to the potential size; in order to further determine the development sequence of the cluster area, this embodiment takes the average comprehensive development potential of each cluster area as the basis, and determines the development sequence arrangement of each administrative area according to the calculated comprehensive potential size. According to the actual needs of local development work, the development sequence is defined as three stages: short-term development, medium-term development, and long-term development. Figures 5 to 10 As shown in the figure, the decision results of the six central urban areas of the study city (for example, Ningbo City) are arranged according to the development timing of three stages: short-term development, medium-term development, and long-term development.

[0073] A time-series decision-making system for the development of reserve industrial land that takes into account a spatial perspective comprises a development potential index system, a data set processing module, a particle swarm optimization projection pursuit model, and a SOFM neural network. The development potential index system is divided into a hierarchy according to an index level. The data acquisition and processing module is used to obtain reserve industrial land data containing spatial location information in a study city and perform data normalization processing according to the corresponding indicators in the development potential index system. The particle swarm optimization projection pursuit model uses all normalized indicator data as high-dimensional data, utilizes a particle swarm algorithm to iteratively adjust particle positions, and ultimately projects the data into one-dimensional data. The one-dimensional data containing spatial location information is used as the development potential. The SOFM neural network spatially clusters the development potential of all spatial locations in the study city into N spaces, sorts the N spaces in sequence according to development potential, and outputs the result of a time-series decision.

[0074] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for making decisions on the timing of development of reserved industrial land taking into account spatial perspective, characterized by: The methods include: S1. Construct a development potential index system divided by indicator level, obtain the reserved industrial land data including spatial location information of the research city, and normalize the data according to the corresponding indicators in the development potential index system; S2. Construct a particle swarm optimization projection pursuit model. The particle swarm optimization projection pursuit model uses all normalized indicator data as high-dimensional data, uses the particle swarm algorithm to iteratively adjust the particle position and finally projects it into one-dimensional data, and uses the one-dimensional data containing spatial position information as the development potential; S3. Construct a SOFM neural network to spatially cluster the development potential of all spatial locations in the study city into N spaces, and sort the N spaces in order of development potential as the temporal decision result.

2. The method for making decisions on the development of reserved industrial land taking into account the spatial perspective according to claim 1, characterized in that: The development potential index system is divided into layers according to the indicator hierarchy. The indicator hierarchy includes, from top to bottom, the dimension layer, the characteristic layer and the indicator layer. The dimension layer includes necessity, maturity and urgency, wherein necessity is a negative indicator and maturity and urgency are positive indicators; the characteristic layer of necessity includes economic benefits and development intensity; the indicator layer of economic benefits includes the number of employees per unit area, per mu tax revenue, per mu output value, per mu industrial added value and annual operating income; the indicator layer of development intensity includes building density and building height; the characteristic layer of maturity includes the surrounding environment; the indicator layer of the surrounding environment includes traffic convenience, road density, spatial hydrophilicity, industrial agglomeration, sky rate and greening rate; the characteristic layer of urgency includes industrial compliance.

3. The method for making decisions on the development of reserved industrial land taking into account spatial perspective according to claim 1, characterized in that: The projection direction of the particle swarm optimization projection pursuit model is expressed as a={a(1), a(2), a(3), …, a(j) …, a(p)}, where p is the total number of indicators in the development potential index system. The projection value expression of data sample i is as follows: , where i represents the data sample i of all the indicators associated with the reserved industrial land data, x(i, j) is the data of indicator j after the data sample i is normalized, and z(i) represents the projection value of the data sample i, that is, one-dimensional data; Construct the objective function Q(a), the expression is as follows: Q(a) = S Z D Z , where S Z represents the standard deviation of the data sample, D Z Represents the local density of the data sample; the particle swarm algorithm is used to iteratively adjust the particle position to obtain the optimal projection direction and the optimal objective function; the particle swarm optimization projection pursuit model uses the optimal projection direction to calculate the one-dimensional data of the data sample i.

4. The method for making decisions on the development of reserved industrial land taking into account spatial perspective according to claim 2, characterized in that: In method S1, the data normalization method for the indicators in the development potential index system is as follows: the indicators in the indicator layer under the belonging layer are normalized according to the dimensional layer or feature layer of the development potential index system; in method S2, the particle swarm optimization projection pursuit model statistically calculates all indicator data of the indicator layer, feature layer or dimensional layer as high-dimensional data, uses the particle swarm algorithm to iteratively adjust the particle position, and finally projects it into one-dimensional data.

5. The method for making decisions on the development of reserved industrial land taking into account spatial perspective according to claim 2, characterized in that: The above-mentioned number of employees per land, per-mu tax revenue, per-mu output value, per-mu industrial added value, annual operating income, building density, building height and road density are data from the reserved industrial land data or data obtained through calculation; the above-mentioned traffic convenience adopts the traffic POI point kernel density, and the kernel density expression is as follows: , where r is the calculation radius centered on the research space location point, N is the number of space location points within the radius r, and a n It is the distance between the spatial location point n and the research spatial location point within the radius r, and P is the transportation convenience of the research spatial location point.

6. The method for time-sequential decision-making for development of reserved industrial land taking into account spatial perspective according to claim 2 or 5, characterized in that: The method for obtaining the sky rate and greening rate in the reserve industrial land data is as follows: Street view images including the sky and the land are collected in the study city, and the sky and vegetation are segmented using the SegFormer model trained based on the Cityscape dataset. Then, the area ratio of the sky is calculated as the sky rate, and the area ratio of the vegetation is calculated as the greening rate. The industrial compliance status is a scoring system based on the industrial compliance assessment system for enterprises using industrial land in the study city.

7. The method for making decisions on the development of reserved industrial land taking into account spatial perspective according to claim 3, characterized in that: The standard deviation S of the data sample z and the local density D z Using local window calculation, the standard deviation S of the data sample z is the standard deviation of the projection value of the local window data sample, and the expression is as follows: Where m represents the total number of data samples i, and E(z) is the average value of the projection values of the data samples; The local density D of the data sample Z To select the local density of the local density window, the expression is as follows: R represents the radius of the local density window, r(i1, i2) represents the distance between the projection values of data samples i1 and i2, and u() represents the unit step function.

8. The method for making decisions on the development of reserved industrial land taking into account spatial perspective according to claim 1, characterized in that: The SOFM neural network uses a combination of spatial domain and attribute domain to perform hybrid distance clustering. The hybrid distance expression is as follows: ,in is the mixed distance of the projection values of data samples i1 and i2, is the Euclidean distance between data samples i1 and i2 in geographic space, W d It represents the weight of attribute d, which is the attribute in attribute domain D, and attribute domain D is all the indicators of the development potential index system; are the values of data samples i1 and i2 in the attribute d dimension, w s 、w a are the weights of the spatial domain and attribute domain respectively.

9. The method for making decisions on the development of reserved industrial land taking into account spatial perspective according to claim 1, characterized in that: In method S3, N spaces are divided into three intervals of low, medium and high according to the size of development potential. N spatial clustering areas are sorted according to the three intervals of low, medium and high. The sorting decision corresponding to the spatial clustering areas in the low interval is a long-term temporal decision, the sorting decision corresponding to the spatial clustering areas in the medium interval is a medium-term temporal decision, and the sorting decision corresponding to the spatial clustering areas in the high interval is a short-term temporal decision. The temporal decision result, the medium-term temporal decision, the short-term temporal decision and the sorting within each temporal decision together constitute the temporal decision result.

10. A time-series decision-making system for the development of reserved industrial land that takes into account spatial perspective, characterized by: The method comprises a development potential index system, a data acquisition and processing module, a particle swarm optimization projection pursuit model and a SOFM neural network. The development potential index system is divided into a hierarchy according to the index level. The data acquisition and processing module is used to obtain the reserve industrial land data containing spatial location information of the study city and perform data normalization processing according to the corresponding indicators in the development potential index system. The particle swarm optimization projection pursuit model uses all normalized index data as high-dimensional data, utilizes the particle swarm algorithm to iteratively adjust the particle position and finally projects it into one-dimensional data, and uses the one-dimensional data containing spatial location information as the development potential. The SOFM neural network spatially clusters the development potential of all spatial locations in the study city into N spaces, sorts the N spaces in sequence according to the development potential, and outputs the results of temporal decision-making.

Images