Model for territorial space planning and method of use

Abstract

The application discloses a model and a use method for national space planning, and the method comprises the steps of time-varying feature extraction, mutation point detection, stage-based basic weight distribution, similarity correction and weighted training. The method avoids the interference of early deviation data on the model through the processes of time-varying feature extraction, mutation point detection, stage-based basic weight distribution, similarity correction and weighted training, can then extract and compare the valuable rules in historical data, improves the adaptability of the model to the current and future land use change rules, and meets the needs of users.

Description

Technical Field

[0001] This invention relates to the field of land spatial planning technology, and in particular to models and methods of using land spatial planning. Background Technology

[0002] Land use change simulation models are core tools in the preparation and evaluation of national spatial planning. They are widely used in scenarios such as urban expansion prediction and ecological space identification. These models analyze historical land use data and its driving factors to establish mathematical relationships between land use changes and driving factors, and then simulate and extrapolate land use patterns under different future scenarios. In layman's terms, they learn patterns from historical data and then use these patterns to predict the future.

[0003] Existing models have a problem: they are trained using data from the past twenty years to establish a set of patterns, and then assume that these patterns will apply for the next twenty years or even longer. However, in reality, the rules governing urban development change over time. When these rules change, training the model with all historical data with equal weights will result in the model learning an averaged pattern. Using such a model to predict the future will inevitably lead to biased results. Existing models tend to use a blanket approach by deleting historical data. However, because there is a periodic recurrence of patterns in urban evolution, discarding all old data indiscriminately will result in the loss of this valuable reference information, which has limitations.

[0004] The reason for this problem is that, over time, the relative importance of different driving factors will naturally change. The influence of some factors will gradually increase, while that of others will gradually decrease, and even entirely new dominant factors may emerge. When the structure of driving relationships changes in this way, a difference arises between the patterns reflected by early historical data and the actual situation in the present and future. This early historical data will not only not be helpful to the model, but will also cause the model parameters to shift toward the historical average, thus leading to model bias and limitations. Summary of the Invention

[0005] The purpose of this invention is to provide a model and method for use in land spatial planning, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a model for land spatial planning and a method for using it, including;

[0007] S1. Extracting time-varying features: Obtain historical land data and data of each driving factor for the corresponding year; calculate the influence of each driving factor on land use change in different historical periods; and obtain the data sequence of the influence of each driving factor over time.

[0008] S2. Mutation point detection: Analyze the data sequence of the influence of each driving factor over time to identify the mutation years in which the driving relationship structure changes significantly.

[0009] S3. Assign basic weights according to the time distance. Divide historical data into two stages, before and after the mutation, based on the year of the mutation. Assign basic weights to the data in each period. The more recent the year, the greater the weight, and the more distant the year, the smaller the weight.

[0010] S4. Similarity correction: Using data from a fixed time period after the mutation as a reference, extract the influence features of each driving factor in that period as the current rule template, calculate the similarity between the driving relationship features of each historical year and the current rule template, and adjust the basic weights obtained in S3 upward or downward according to the degree of similarity to obtain the final weights.

[0011] S5. Weighted training: Using the final weights obtained in S4, the land use change simulation model is trained. The greater the weight of the historical data, the greater its role in the training process; the smaller the weight of the historical data, the less its role. This results in a trained adaptive model.

[0012] Preferably, in step S1, the influence of each driving factor on land use change in different historical periods is calculated using a sliding window regression method. With a fixed number of years as the window length, the regression coefficient of the driving factor is calculated window by window as the influence value of that window period.

[0013] Preferably, in step S2, the data sequence of the influence of each driving factor over time is analyzed to identify the abrupt change year in which the driving relationship structure changes significantly. Specifically, a Bayesian online change point detection algorithm is used to calculate the posterior probability of structural change at each time point. When the posterior probability exceeds a preset threshold, the time point is determined to be the abrupt change year.

[0014] Preferably, the expected number of mutations parameter in the Bayesian online change point detection algorithm is automatically selected using multi-scale cross-validation combined with the land and space planning compilation cycle.

[0015] Preferably, in step S3, the historical data is divided into two stages, before and after the mutation, according to the year of the mutation. The historical data in the post-mutation stage are assigned an exponentially decaying basic weight according to the time proximity, and the historical data in the pre-mutation stage are assigned a basic weight close to zero.

[0016] Preferably, in step S4, data from a fixed period within the post-mutation phase is used as a reference, where the fixed period is the most recent complete cycle or a continuous year interval of a preset length after the mutation.

[0017] Preferably, in step S4, the similarity between the driving relationship features of each historical year and the current rule template is calculated. Specifically, the cosine similarity calculation method is used, and the similarity value ranges from 0 to 1. The closer the value is to 1, the more similar the driving relationship features are.

[0018] Preferably, in step S4, the basic weights obtained in S3 are adjusted upwards or downwards according to the degree of similarity. Specifically, the operation is as follows:

[0019] When the similarity exceeds the first preset threshold, the basic weight is positively corrected and the weight value is increased.

[0020] When the similarity is lower than the second preset threshold, the basic weight is negatively corrected and the weight value is reduced.

[0021] When the similarity is between the first preset threshold and the second preset threshold, it is fine-tuned linearly or the basic weight remains unchanged.

[0022] Preferably, the positively corrected weight value does not exceed the upper limit of the weight of the data in the current period, and the negatively corrected weight value is not lower than the preset minimum weight value.

[0023] Preferably, in step S5, the land use change simulation model is any one of a cellular automaton model, an artificial neural network-cellular automaton model, or a random forest-cellular automaton model.

[0024] The technical effects and advantages of this invention are as follows:

[0025] 1. This method, through the process of time-varying feature extraction, mutation point detection, phased basic weight allocation, similarity correction, and weighted training, not only avoids the interference of early biased data on the model, but also extracts and compares valuable patterns from historical data, improving the model's adaptability to current and future land use change patterns and meeting user needs.

[0026] 2. This invention calculates the influence of driving factors in different historical periods through sliding window regression, and identifies the specific time points of abrupt changes by combining a Bayesian online change point detection algorithm, ensuring the rationality of phased weight allocation and meeting user needs.

[0027] 3. This invention allocates basic weights through exponential decay, corrects weights by combining cosine similarity, and sets upper and lower limits for weights. This achieves precise adjustment of historical data weights, ensuring the dominant role of recent data while extracting valuable information from historical data that matches the current rule template through similarity judgment, thus meeting user needs. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the overall structure of the present invention;

[0029] Figure 2 This is a flowchart of the time-varying feature extraction process of the present invention;

[0030] Figure 3 This is a schematic diagram of mutation point detection in this invention;

[0031] Figure 4 This is a schematic diagram of the similarity correction process of the present invention;

[0032] Figure 5 This is a schematic diagram of the weighted training method of the present invention;

[0033] Figure 6 This is a schematic diagram illustrating the weight correction threshold determination of the present invention;

[0034] Figure 7 This is a schematic diagram illustrating the adaptive selection of parameters in this invention. Detailed Implementation

[0035] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0036] This invention provides, for example Figure 1 The model and usage methods shown for land spatial planning include:

[0037] S1. Extracting time-varying features: Obtain historical land data and data of each driving factor for the corresponding year; calculate the influence of each driving factor on land use change in different historical periods; and obtain the data sequence of the influence of each driving factor over time.

[0038] S2. Mutation point detection: Analyze the data sequence of the influence of each driving factor over time to identify the mutation years in which the driving relationship structure changes significantly.

[0039] S3. Assign basic weights according to the time distance. Divide historical data into two stages, before and after the mutation, based on the year of the mutation. Assign basic weights to the data in each period. The more recent the year, the greater the weight, and the more distant the year, the smaller the weight.

[0040] S4. Similarity correction: Using data from a fixed time period after the mutation as a reference, extract the influence features of each driving factor in that period as the current rule template, calculate the similarity between the driving relationship features of each historical year and the current rule template, and adjust the basic weights obtained in S3 upward or downward according to the degree of similarity to obtain the final weights.

[0041] S5. Weighted training: Using the final weights obtained in S4, the land use change simulation model is trained. The greater the weight of the historical data, the greater its role in the training process; the smaller the weight of the historical data, the less its role. This results in a trained adaptive model.

[0042] Example 1: In this example, the principle of this method is to extract the influence of each driving factor on land use change in different historical periods, identify the years of abrupt changes in the driving relationship, assign different weights to historical data according to time proximity and similarity of patterns, and train the model in a weighted manner so that the model can learn effective data that matches the current pattern and is closer in time. In fact, it is similarity correction, avoiding interference from invalid historical data and improving the accuracy of land space planning simulation and prediction.

[0043] Similarity correction has two main advantages. First, it can identify the source of patterns. When a current rule suddenly appears, the system can trace back to the period when a similar pattern first appeared in history, helping to understand how the new rule emerged. Second, it can verify periodic regression. When the model detects that a rule from a certain historical period is highly similar to the current one, it can indicate that the pattern may be periodic rather than a completely new innovation. Therefore, similarity correction can extract useful parts from historical data and utilize them to meet user needs.

[0044] This invention provides, for example Figures 2 to 7 The model and usage methods shown for land spatial planning include:

[0045] S1. Extracting time-varying features: Obtain historical land data and data of each driving factor for the corresponding year; calculate the influence of each driving factor on land use change in different historical periods; and obtain the data sequence of the influence of each driving factor over time.

[0046] S2. Mutation point detection: Analyze the data sequence of the influence of each driving factor over time to identify the mutation years in which the driving relationship structure changes significantly.

[0047] S3. Assign basic weights according to the time distance. Divide historical data into two stages, before and after the mutation, based on the year of the mutation. Assign basic weights to the data in each period. The more recent the year, the greater the weight, and the more distant the year, the smaller the weight.

[0048] S4. Similarity correction: Using data from a fixed time period after the mutation as a reference, extract the influence features of each driving factor in that period as the current rule template, calculate the similarity between the driving relationship features of each historical year and the current rule template, and adjust the basic weights obtained in S3 upward or downward according to the degree of similarity to obtain the final weights.

[0049] S5. Weighted training: Using the final weights obtained in S4, the land use change simulation model is trained. The greater the weight of the historical data, the greater its role in the training process; the smaller the weight of the historical data, the less its role. This results in a trained adaptive model.

[0050] In step S2, the data sequence of the influence of each driving factor over time is analyzed to identify the abrupt change years in which the driving relationship structure changes significantly. Specifically, the Bayesian online change point detection algorithm is used to calculate the posterior probability of structural change at each time point. When the posterior probability exceeds a preset threshold, the time point is determined to be the abrupt change year.

[0051] The expected number of mutations parameter in the Bayesian online change point detection algorithm is automatically selected by combining multi-scale cross-validation with the land and space planning compilation cycle.

[0052] In step S3, the historical data is divided into two stages, before and after the mutation, based on the year of the mutation. The historical data in the post-mutation stage are assigned an exponentially decaying basic weight according to their time proximity, while the historical data in the pre-mutation stage are assigned a basic weight close to zero.

[0053] In step S4, data from a fixed time period within the post-mutation phase is used as a reference. The fixed time period is the most recent complete cycle or a continuous year interval of a preset length after the mutation.

[0054] In step S4, the similarity between the driving relationship features of each historical year and the current rule template is calculated. Specifically, the cosine similarity calculation method is used, and the similarity value ranges from 0 to 1. The closer the value is to 1, the more similar the driving relationship features are.

[0055] In step S4, the basic weights obtained in S3 are adjusted upwards or downwards according to the degree of similarity. The specific operation is as follows:

[0056] When the similarity exceeds the first preset threshold, the basic weight is positively corrected and the weight value is increased.

[0057] When the similarity is lower than the second preset threshold, the basic weight is negatively corrected and the weight value is reduced.

[0058] When the similarity is between the first preset threshold and the second preset threshold, it is fine-tuned linearly or the basic weight remains unchanged.

[0059] The weight value after positive correction shall not exceed the upper limit of the weight of the data in the current period, and the weight value after negative correction shall not be lower than the preset minimum weight value.

[0060] Example 2: In this example, the historical land data in S1 includes, but is not limited to, basic data such as the area and distribution coordinates of various land uses, including cultivated land, construction land, and ecological land, in different years. Simultaneously, data on various driving factors for the corresponding years are acquired. These driving factors include, but are not limited to, population size, total GDP, road network density, policy control indicators, topographic conditions, and climate factors—various influencing factors related to land use change. For each driving factor, a sliding window regression method is used to calculate its influence on land use change in different historical periods. A fixed number of years is set as the window length; in this example, the window length is 5 years, which can be adjusted according to the time scale requirements of actual land spatial planning. Using time as the horizontal axis, the regression coefficient of the driving factor is calculated by sliding the window window by window. This regression coefficient is used as the influence value of the driving factor on land use change within the corresponding window period, ultimately obtaining a data sequence of the influence of each driving factor over time. This data sequence effectively reflects the temporal evolution of the influence of each driving factor.

[0061] S2. Mutation point detection: The data sequence of the influence of each driving factor obtained in step S1 over time is analyzed. The Bayesian online change point detection algorithm is used to identify the mutation years in which the driving relationship structure changes significantly. Specifically, the time sequence of the influence of each driving factor is used as input. The Bayesian online change point detection algorithm calculates the posterior probability of a mutation in the driving relationship structure at each time point. A preset posterior probability threshold is set. When the posterior probability at a certain time point exceeds the preset threshold, the time point is determined to be a mutation year. That is, after the time point, the relationship structure between the driving factor and land use change has changed significantly. In this embodiment, the preset threshold is preferably 0.8, which can be adjusted according to the actual detection accuracy requirements. The expected mutation frequency parameter in the Bayesian online change point detection algorithm is automatically selected using multi-scale cross-validation combined with the land and space planning cycle. The land and space planning cycle is usually 5 years. Therefore, in the multi-scale cross-validation process, the 5-year scale is used as the basic scale, while taking into account both short-term and long-term scales. The expected mutation frequency parameter that is compatible with the current planning area and planning cycle is automatically selected, avoiding detection bias caused by manually setting parameters and improving the accuracy of change point detection.

[0062] S3. Assign basic weights based on time proximity. Based on the abrupt change year identified in step S2, divide all historical land data and corresponding driving factor data into two stages: before and after the abrupt change. Assign basic weights to the data in each stage, following the principle that the more recent the year, the greater the weight, and the more distant the year, the smaller the weight. Specifically, the exponential decay coefficient can be adjusted according to actual needs. In this embodiment, the decay coefficient is preferably 0.9, that is, the closer the year is to the current time, the smaller the weight decay and the larger the weight value. Assign basic weights close to zero to the historical data in the pre-abrupt change stage. In this embodiment, it is preset to 0.01, which weakens the interference of the pre-abrupt change bias data on model training and avoids completely deleting this part of the data, thus avoiding the waste of valuable information in the historical data.

[0063] S4. Similarity correction: Using data from a fixed time period after the mutation as a reference, the fixed time is preferably the most recent complete national spatial planning cycle after the mutation, i.e., 5 years. If a complete planning cycle has not been reached after the mutation, a continuous year interval of a preset length after the mutation is selected. In this embodiment, the preset length is 3 years. The influence characteristics of each driving factor within this reference period are extracted and used as the current rule template. This template can accurately reflect the regular characteristics under the current driving relationship structure. The cosine similarity calculation method is used to calculate the similarity between the driving relationship characteristics of each historical year and the current rule template. The value of cosine similarity ranges from 0 to 1. The closer the value is to 1, the more similar the driving relationship characteristics of the historical year are to the current rule template. The closer the value is to 0, the greater the difference.

[0064] The basic weights obtained in step S3 are adjusted upwards or downwards based on the degree of similarity. Specifically, a first preset threshold and a second preset threshold are set. In this embodiment, the first preset threshold is 0.8 and the second preset threshold is 0.4. When the similarity exceeds the first preset threshold, it indicates that the driving pattern of the historical year is highly matched with the current pattern, and its basic weight is positively corrected by increasing the weight value. However, the weight value after positive correction does not exceed the upper limit of the weight of the current period data. In this embodiment, the upper limit of the weight is 1.0. When the similarity is lower than the second preset threshold, it indicates that the driving pattern of the historical year differs significantly from the current pattern, and its basic weight is negatively corrected by decreasing the weight value. However, the weight value after negative correction is not lower than the preset minimum weight value. In this embodiment, the minimum weight value is 0.005. When the similarity is between the first preset threshold and the second preset threshold, the basic weight is finely adjusted linearly or kept unchanged, and finally the final weight of each historical data is obtained.

[0065] S5. Weighted training: In this embodiment, a random forest-cellular automata model is preferred to balance prediction accuracy and computational efficiency. The final weights obtained in step S4 are used to train the land use change simulation model. During the training process, the historical data with larger final weights play a greater role in the model parameter iteration and pattern learning process, while the historical data with smaller final weights play a smaller role, thus meeting the user's needs.

[0066] It should be noted that the parameters in this embodiment (such as window length, posterior probability threshold, exponential decay coefficient, similarity threshold, etc.) can be adjusted according to the regional characteristics, time scale, and accuracy requirements of the actual land spatial planning. They are not fixed and can be adjusted as long as the logical coherence of each step and the realization of the technical effect are ensured.

[0067] This invention provides, for example Figure 1 The model and usage methods shown for land spatial planning include:

[0068] S1. Extracting time-varying features: Obtain historical land data and data of each driving factor for the corresponding year; calculate the influence of each driving factor on land use change in different historical periods; and obtain the data sequence of the influence of each driving factor over time.

[0069] S2. Mutation point detection: Analyze the data sequence of the influence of each driving factor over time to identify the mutation years in which the driving relationship structure changes significantly.

[0070] S3. Assign basic weights according to the time distance. Divide historical data into two stages, before and after the mutation, based on the year of the mutation. Assign basic weights to the data in each period. The more recent the year, the greater the weight, and the more distant the year, the smaller the weight.

[0071] S4. Similarity correction: Using data from a fixed time period after the mutation as a reference, extract the influence features of each driving factor in that period as the current rule template, calculate the similarity between the driving relationship features of each historical year and the current rule template, and adjust the basic weights obtained in S3 upward or downward according to the degree of similarity to obtain the final weights.

[0072] S5. Weighted training: Using the final weights obtained in S4, the land use change simulation model is trained. The greater the weight of the historical data, the greater its role in the training process; the smaller the weight of the historical data, the less its role. This results in a trained adaptive model.

[0073] In step S1, the influence of each driving factor on land use change in different historical periods is calculated using a sliding window regression method. With a fixed number of years as the window length, the regression coefficient of the driving factor is calculated window by window as the influence value of the window period. In step S5, the land use change simulation model is any one of the following: cellular automata model, artificial neural network-cellular automata model, or random forest-cellular automata model.

[0074] Example 3: In this example, the specific formula for S1 is as follows:

[0075]

[0076] Let represent the influence of the i-th driving factor during the t-year window. The covariance between driving factors and land use change; The variance of the driving factor itself.

[0077] The formula for the time decay base weight in step S3 is as follows:

[0078]

[0079] The time decay weight for the data in year t; It is a natural constant; This represents the decay rate, i.e., a positive number; the larger the number, the faster the decay. The current year; The year to which the data belongs; The number of years since the data was collected;

[0080] The similarity calculation and weight correction in step S4 are as follows:

[0081]

[0082] Similarity score, ranging from 0 to 1; The influence of various driving factors in a certain historical period; The influence of each driving factor in the current rule template; It is the sum of the products of the influence of the corresponding factors;

[0083] Weight correction;

[0084]

[0085] The final weights after correction; The time decay weight obtained from S3.

[0086] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A model and its usage method for land spatial planning, characterized in that, include: S1. Extracting time-varying features: Obtain historical land data and data of each driving factor for the corresponding year; calculate the influence of each driving factor on land use change in different historical periods; and obtain the data sequence of the influence of each driving factor over time. S2. Mutation point detection: Analyze the data sequence of the influence of each driving factor over time to identify the mutation years in which the driving relationship structure changes significantly. S3. Assign basic weights according to the time distance. Divide historical data into two stages, before and after the mutation, based on the year of the mutation. Assign basic weights to the data in each period. The more recent the year, the greater the weight, and the more distant the year, the smaller the weight. S4. Similarity correction: Using data from a fixed time period after the mutation as a reference, extract the influence features of each driving factor in that period as the current rule template, calculate the similarity between the driving relationship features of each historical year and the current rule template, and adjust the basic weights obtained in S3 upward or downward according to the degree of similarity to obtain the final weights. S5. Weighted training: Using the final weights obtained in S4, the land use change simulation model is trained. The greater the weight of the historical data, the greater its role in the training process; the smaller the weight of the historical data, the less its role. This results in a trained adaptive model.

2. The model and method of use for land spatial planning according to claim 1, characterized in that, In step S1, the influence of each driving factor on land use change in different historical periods is calculated using a sliding window regression method. With a fixed number of years as the window length, the regression coefficient of the driving factor is calculated window by window as the influence value of that window period.

3. The model and method of use for land spatial planning according to claim 1, characterized in that, In step S2, the data sequence of the influence of each driving factor over time is analyzed to identify the abrupt change years in which the driving relationship structure changes significantly. Specifically, the Bayesian online change point detection algorithm is used to calculate the posterior probability of structural change at each time point. When the posterior probability exceeds a preset threshold, the time point is determined to be the abrupt change year.

4. The model for territorial spatial planning and method of use according to claim 3, characterized in that, The expected number of mutations parameter in the Bayesian online change point detection algorithm is automatically selected using multi-scale cross-validation combined with the land and space planning compilation cycle.

5. The model for and method of using land space planning of claim 1, wherein, In step S3, the historical data is divided into two stages, before and after the mutation, based on the year of the mutation. The historical data in the post-mutation stage are assigned an exponentially decaying basic weight according to their time proximity, while the historical data in the pre-mutation stage are assigned a basic weight close to zero.

6. The model for and method of using land space planning of claim 1, wherein, In step S4, data from a fixed period within the post-mutation phase is used as a reference. The fixed period is the most recent complete cycle or a continuous year interval of a preset length after the mutation.

7. The model for and method of using land space planning of claim 1, wherein, In step S4, the similarity between the driving relationship features of each historical year and the current rule template is calculated. Specifically, the cosine similarity calculation method is used, and the similarity value ranges from 0 to 1. The closer the value is to 1, the more similar the driving relationship features are.

8. The model for and method of using land space planning of claim 1, wherein, In step S4, the basic weights obtained in S3 are adjusted upwards or downwards according to the degree of similarity. The specific operation is as follows: When the similarity exceeds the first preset threshold, the basic weight is positively corrected and the weight value is increased. When the similarity is lower than the second preset threshold, the basic weight is negatively corrected and the weight value is reduced. When the similarity is between the first preset threshold and the second preset threshold, it is fine-tuned linearly or the basic weight remains unchanged.

9. The model for territorial spatial planning and method of use according to claim 8, characterized in that, The weight value after positive correction does not exceed the upper limit of the weight of the data in the current period, and the weight value after negative correction is not lower than the preset minimum weight value.

10. The model for and method of using land space planning of claim 1, wherein, In step S5, the land use change simulation model is any one of the following: cellular automata model, artificial neural network-cellular automata model, or random forest-cellular automata model.

Images