City multi-element space collaborative simulation method based on combined prediction model control

Abstract

The invention discloses a multi-element space collaborative simulation method for controlling a total amount of a city based on a combined prediction model, and relates to the field of macroeconomic prediction, and the multi-element space collaborative simulation method for controlling the total amount of the city based on the combined prediction model mainly comprises the steps: according to economic related data; obtaining a plurality of prediction values of total quantity control by utilizing an economic data driving model and a news text prediction model; according to the plurality of prediction values controlled by the total amount, obtaining a total amount prediction result by using a Kalman filtering combination prediction method; and performing spatial collaborative simulation according to the total prediction result to obtain a collaborative simulation result. By implementing the city multi-element space collaborative simulation method for controlling the total amount based on the combined prediction model, complex distribution of various elements of a city can be simulated more accurately.

Description

Technical Field

[0001] The present invention relates to the field of macroeconomic forecasting, and more specifically, to a method for spatial collaborative simulation of multiple elements of cities based on a combined forecasting model to control total volume. Background Art

[0002] At present, urban development presents the characteristics of co-evolution and coordinated development in land use, population, economy and other factors. However, the existing cellular automaton model has a single cell type and mostly uses static driving factors to simulate the evolution, ignoring the dynamic synergy between multiple factors, resulting in low simulation accuracy. How to achieve spatial cooperative simulation at a large urban scale and explore the dynamic change trends and mutual influence mechanisms between various factors is of great significance for accurately describing the spatial pattern of urban development. Economic forecasting models can be divided into econometric models and machine learning-based forecasting models. In recent years, machine learning methods have been more popular and have higher accuracy, but there is no universal best model. Econometric models are generally used as baselines to evaluate the effects of forecasting factors or added to combined models to improve accuracy. In addition, the data used in regional macroeconomic forecasting research are still mainly economic indicators, and the data types are relatively single. The Cooperative Cellular Automata (CoCA) model takes into account the coordinated development and driving effects of population growth and economic activities on cities to better simulate the land use changes in reality, so that the spatial distribution pattern of urban land is more in line with the actual situation. However, the existing multi-factor spatial collaborative simulation model relies heavily on official news and manual judgment in setting the total amount of future economic indicators, which lacks scientificity and accuracy. How to coordinate multiple factors to simulate urban land use changes and urban element development, achieve future total control, and improve the accuracy of simulation and prediction is an urgent problem to be solved. Summary of the invention

[0003] The purpose of the present invention is to provide a method and computer program product for spatial collaborative simulation of multiple elements of a city based on a combined prediction model to control the total amount, which can more accurately simulate the complex distribution of various elements in the city.

[0004] The present invention provides a method for spatial collaborative simulation of multiple factors of cities for controlling total volume based on a combined prediction model, comprising the following steps: S1: obtaining multiple prediction values ​​of total volume control by using an economic data-driven model and a news text prediction model according to economic-related data; S2: obtaining a total volume prediction result by using a Kalman filter combined prediction method according to the multiple prediction values ​​of total volume control; S3: performing spatial collaborative simulation according to the total volume prediction result to obtain a collaborative simulation result.

[0005] Furthermore, the above-mentioned method for collaborative simulation of urban multi-factor space based on the combined prediction model to control the total amount also includes: obtaining the result accuracy according to the collaborative simulation results.

[0006] Furthermore, the above-mentioned economic-related data include GDP nominal values ​​and macroeconomic-related news; the multiple forecast values ​​of total control include the first nominal GDP forecast value, the second nominal GDP forecast value, the third nominal GDP forecast value, the fourth nominal GDP forecast value, the fifth nominal GDP forecast value, and the sixth nominal GDP forecast value; the economic data-driven models include the ARIMA model, the ordinary least squares regression model, the time series basic model and the XGBoost model; the news text prediction model includes the BiLSTM sentiment scoring model and the Moka large-scale hybrid embedding model; the ARIMA model obtains the first nominal GDP forecast value based on the GDP nominal value; the ordinary least squares regression model obtains the second nominal GDP forecast value based on the GDP nominal value; the time series basic model obtains the third nominal GDP forecast value based on the GDP nominal value; the XGBoost model obtains the fourth nominal GDP forecast value based on the GDP nominal value; the BiLSTM sentiment scoring model obtains the fifth nominal GDP forecast value based on the GDP nominal value and macroeconomic-related news; the Moka large-scale hybrid embedding model obtains the sixth nominal GDP forecast value based on the GDP nominal value and macroeconomic-related news.

[0007] Furthermore, the above-mentioned total volume forecast results include economic density data.

[0008] Further, step S2 specifically includes: according to multiple predicted values ​​of total quantity control, using Kalman filter combined prediction method, obtaining the total quantity prediction result, such as formula: , , , , , , in, represents the GDP forecast for the current period, is the estimate for the previous period, is the state transition matrix, is the matrix of prediction values ​​provided by the model, is the observation matrix, represents the observation noise, is the forecast error covariance, is the forecast error covariance for the previous period, is the process noise covariance, is the Kalman gain, The transposed matrix of the observation matrix, is the observation noise covariance, is the optimal state estimate of GDP in the current period, is the current period forecast error covariance.

[0009] Furthermore, step S3 specifically includes: S31: Construct a density cellular automaton model, such as the formula: , , , in, for The cell state at the moment, for The development density of the moment, For the nearby range, is the cell density change, is the change in density, is the speed of density development, For nearby range The collection of units that have been converted to urban land, For nearby range The development density of inner-city land, is the maximum value of development density, is the radius of the proximity range; S32: Use the random forest model and cellular automaton to build the PLUS model, as shown in the formula: , , , in, For Grid Land use type The conversion probability, is a variable with a value of 0 or 1. is the number of decision trees in the model, and also the number of samples from the original training dataset; For the A decision tree vector The prediction type, It is a vector composed of multiple driving factors; is the indicative function set by the decision tree; Indicates the future land use type The impact of demand is an adaptive driving factor that depends on the iteration Current land quantity and land use type at the time of gap between target needs. and Respectively The iteration and At the iteration, the difference between the current land quantity and the target demand is Represents a grid At the moment Convert to Land Use Type The total probability of For Grid Changes from other types to other land use types The conversion probability, Indicates whether other land use types are converted into land use types , is a random value from 0 to 1. For land use type Generate new land use thresholds, Indicated in The iteration represents the cell Land type Neighborhood effects; S33: Based on the total amount prediction results, the density cellular automaton model and the PLUS model are used to perform spatial collaborative simulation to obtain collaborative simulation results.

[0010] Furthermore, step S33 specifically includes: S331: obtaining the initial overall development probability based on the total amount prediction results and spatial auxiliary data, and obtaining the initial simulation results using the density cellular automaton model and the PLUS model; S332: performing spatial collaborative simulation based on the initial overall development probability and the initial simulation results using the density cellular automaton model and the PLUS model to obtain new collaborative simulation results and a new overall development probability; S333: performing the next collaborative simulation based on the new collaborative simulation results and the new overall development probability and executing it in a loop until the relative error of the comprehensive indicators for multiple consecutive times is within the preset threshold, stopping the calculation, and obtaining the collaborative simulation results.

[0011] Furthermore, the above-mentioned spatial auxiliary data includes natural elements, traffic elements, and location elements.

[0012] Furthermore, the above comprehensive indicators are as follows: , in, As a comprehensive indicator, For land use simulation performance, is the population density simulation error, is the economic density simulation error, , and is the corresponding weight.

[0013] The present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the above-mentioned method for controlling the total amount of urban multi-factor spatial collaborative simulation based on a combined prediction model.

[0014] The implementation of the method and computer program product for simulating the multi-factor spatial coordination of the total urban area based on the combined prediction model control has the following beneficial effects: The present invention combines a variety of econometric forecasting models and a forecasting model based on news text, and utilizes Kalman filter combination optimization to improve the accuracy and stability of economic indicator forecasting results, providing a more reliable total control data support for the multi-factor spatial collaborative simulation model, and effectively improving the accuracy of multi-factor collaborative simulations such as future land use changes and spatial distribution of population and economic density; adopts a dynamic update strategy for driving factors to conduct "land-population-economy" multi-factor spatial collaborative simulation, and explores the mutual influence mechanism among various factors; by considering the urban development boundary, it effectively suppresses over-expansion, and shows a convergence trend during the simulation process, more accurately capturing the interactive mechanism between economic growth and land expansion, so that the distribution pattern of construction land is more in line with reality, and the division between urban and non-urban areas is clearer; on the other hand, the present invention adopts the idea of ​​"hierarchical progression", and uses land, population, and economy as training data in turn through iteration to influence the result of the next simulation, so as to achieve the effect of synchronous collaborative matching and mutual influence; The present invention combines the prediction of the total economic volume by coupling different econometric models, machine learning, deep learning models and Kalman filter model multi-factor spatial collaborative simulation. The results show that the simulation accuracy of the model can be effectively improved in the future. The experimental results show that urban development is the result of the joint action of natural, social and economic factors. On the basis of proving the effectiveness of the collaborative mechanism, further exploring the collaborative mechanism of various factors can more accurately simulate the complex distribution of various elements in the city. Improving the prediction accuracy of the simulated total volume is beneficial to urban development simulation, which can assist decision makers to better predict and plan the spatial distribution of land use and urban elements, assist decision makers to manage future land use, and comprehensively consider various development factors to formulate a reasonable national land space layout plan to promote the sustainable development of the city. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which: Figure 1 It is a flow chart of the method for spatial collaborative simulation of multiple elements of a city based on a combined prediction model to control total volume provided by the present invention; Figure 2 It is the overall flow chart provided by the present invention; Figure 3 It is the ARIMA model flow chart provided by the present invention; Figure 4 It is a flow chart of the BiLSTM sentiment scoring model provided by the present invention; Figure 5 is a flow chart of the collaborative simulation model provided by the present invention; Figure 6 It is a step-by-step "dynamic driver update" flow chart provided by the present invention. DETAILED DESCRIPTION

[0016] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described in detail with reference to the accompanying drawings.

[0017] Figure 1 A schematic diagram of a method for collaboratively simulating the multi-factor space of a city based on a combined prediction model to control the total volume of the present embodiment is shown. In this embodiment, the method for collaboratively simulating the multi-factor space of a city based on a combined prediction model to control the total volume of the present embodiment comprises the following steps: S1: Based on economic data, we use the economic data-driven model and the news text prediction model to obtain multiple forecast values ​​of total volume control; In an exemplary embodiment, economic-related data include GDP nominal values ​​and macroeconomic-related news; multiple forecast values ​​of total control include a first nominal GDP forecast value, a second nominal GDP forecast value, a third nominal GDP forecast value, a fourth nominal GDP forecast value, a fifth nominal GDP forecast value, and a sixth nominal GDP forecast value; the economic data-driven model includes an ARIMA model, an ordinary least squares regression model, a time series basic model, and an XGBoost model; the news text prediction model includes a BiLSTM sentiment scoring model and a Moka large-scale hybrid embedding model; the ARIMA model obtains a first nominal GDP forecast value based on the GDP nominal value; the ordinary least squares regression model obtains a second nominal GDP forecast value based on the GDP nominal value; the time series basic model obtains a third nominal GDP forecast value based on the GDP nominal value; the XGBoost model obtains a fourth nominal GDP forecast value based on the GDP nominal value; the BiLSTM sentiment scoring model obtains a fifth nominal GDP forecast value based on the GDP nominal value and macroeconomic-related news; the Moka large-scale hybrid embedding model obtains a sixth nominal GDP forecast value based on the GDP nominal value and macroeconomic-related news; As an exemplary embodiment, multi-model prediction is performed in step S1. This step mainly uses a variety of data-driven or news text-based economic prediction models to calculate multiple prediction values ​​of total control; specifically, in step S1, the economic prediction model can be divided into an econometric model and a prediction model based on machine learning. In recent years, the machine learning method is more popular and has higher accuracy, but there is no universal best model, and the econometric model is generally used as a baseline to evaluate the effect of the prediction factor or added to the combined model to improve the accuracy; and the data currently used in regional macroeconomic forecasting research are still mainly economic indicators, and the data types are relatively single; therefore, this embodiment introduces data such as news texts, and combines a variety of representative econometric models and prediction models based on machine learning to obtain a variety of prediction results; It should be noted that, according to the data used, the models can be divided into prediction models driven by economic data and prediction models containing news text. Models driven by economic indicator data include ARIMA, OLS, TimesFM, and XGBoost. The data used by these models are the nominal GDP values ​​of each year. The prediction models containing news text also add macroeconomic news crawled from various news websites. The specific situation is shown in Table 1. These news texts are input into the model for sentiment classification or converted into vectors, and then mapped with GDP data to achieve prediction. The output of all prediction models is the nominal GDP prediction value. Table 1: Macroeconomic related news data table

[0018] It should be noted that the AutoRegressive Integrated Moving Average (ARIMA) model is a time series analysis method. The ARIMA model is widely used in the prediction and modeling of time series data. It can predict future trends and changes by analyzing and fitting the historical data of the time series. Its basic principle can be expressed by the following formula: (1) Among them, AR(p) represents the autoregressive model. The ARIMA model is based on autoregression, that is, the value of the current time point is related to the values ​​of several previous time points; the autoregressive order p means that the current time point is related to the values ​​of the previous p time points; I(d) represents the difference model. In order to eliminate the non-stationarity of time series data, the ARIMA model usually needs to perform a difference operation, that is, convert the original data into difference data to eliminate the influence of trends and seasonality; the difference order d represents the number of times the time series data needs to be differentiated; MA(q) represents the moving average model. The ARIMA model is based on moving average, that is, the value of the current time point is related to the errors of several previous time points; the moving average order q means that the current time point is related to the errors of the previous q time points; the ARIMA model can estimate the appropriate model parameters by analyzing and fitting the time series data, so as to perform data prediction and modeling. The model flow chart is as follows: Figure 3 As shown; Estimate the difference order d: If the time series data is not stationary, a difference operation is required. According to the stationarity of the time series data and the ACF and PACF images, the range of the difference order d can be preliminarily estimated; Determine the autoregressive order p and the moving average order q: According to the autocorrelation function (ACF) and partial autocorrelation function (PACF) images of the time series data, the range of the autoregressive order p and the moving average order q can be preliminarily estimated; It should be noted that the OLS least squares model, that is, the ordinary least squares regression model (Ordinary Least Squares Regression Model) uses a univariate linear regression model, which is also the simplest model in econometrics. Its principle is shown in the formula (2) Among them, 𝑥 is the independent variable, which is set to the lagged characteristics of GDP here (if quarterly data is used, it can be set to 1st, 2nd, 3rd, or 4th order lag); The coefficients obtained from the regression; Constant term; It is called the error term or disturbance term, which represents the difference between Other factors that may affect factors; establish a regression model with the lag characteristics of economic indicators as independent variables and the indicator itself Y as the dependent variable, test the accuracy and make forward predictions; It should be noted that the Time Series Foundation Model (TimesFM) is a pre-trained model for time series forecasting developed by Google Research; it can produce accurate predictions in different fields and applications without additional training; the model has trained TimesFM using more than 100 billion data points, all of which come from open source time series data; the dataset covers a wide range of fields, from finance, economy and weather to network traffic, energy and sales; and the time frequency ranges from days, weeks, months to years, with various granularities; the TimesFM architecture takes an input time series of a specific length and decomposes it into multiple input fragments; each patch is then processed into a vector by the residual block defined in the model definition to match the model size of the Transformer layer; the vector is then added to the positional encoding; the vector with the positional encoding is sent to the stacked Transformer layer; It should be noted that the Extreme Gradient Boosting (XGBoost) model is a decision tree algorithm based on gradient boosting; gradient boosting is an ensemble learning method that optimizes the objective function by iteratively adding new weak learners (usually decision trees); the XGBoost method combines multiple simple decision trees to form a powerful prediction model; in each round of iteration, the algorithm trains a new decision tree to fit the residual of the previous model; the objective function consists of two parts: one is the loss function, which measures the prediction error of the model; the other is the regularization term, which includes the complexity of the tree and is used to prevent overfitting, such as the formula: (3) in, is the loss function, It is Complexity penalty for each tree; In each iteration, we try to find a decision tree that minimizes the loss function, thereby continuously improving the performance of the model; It should be noted that the BiLSTM bidirectional long short-term memory network of the BiLSTM sentiment scoring model is composed of multiple forward and backward bidirectional long short-term memory units, which is particularly suitable for processing sequence data such as text. Its core principle is to combine the advantages of long short-term memory network and bidirectional network, and can effectively capture the long-range dependencies and contextual information in the text; in the sentiment scoring task, the BiLSTM model can understand the complexity of context and emotional expression by learning the sequence of words or characters in the text; through the bidirectional structure, the model can learn sequence information from both forward and backward, which helps to more comprehensively capture the emotional changes and context transformations in the text, and improve the accuracy and expression ability of the sentiment scoring. The model flow chart is shown in the figure. Figure 4 As shown; It should be noted that the Moka Massive Mixed Embedding (M3E) model is an open source text embedding model; Moka means that this model is trained, open sourced and evaluated by MokaAI; Massive means that this model is trained on a dataset of tens of millions (22 million+) Chinese sentences; Mixed means that this model supports similarity calculation of homogeneous texts in Chinese and English, heterogeneous text retrieval and other functions, and will support code retrieval in the future; Embedding means that this model is a text embedding model that can convert natural language into dense vectors; with this method, the news text information of each time period is converted into a vector, a mapping is established between the vector and the economic indicator, and grid search is used for parameter tuning to achieve prediction; in this embodiment, the news text information of each month is converted into a 768-dimensional complex vector, and the embedded features are averaged according to the time frequency of the data (month, quarter or half a year) to generate feature vectors under different time slices; next, the SVR (support vector regression) method is used to establish a mapping between the vector and the economic indicator:

[0019] in, is a text vector Feature mapping, mapping data from the input space to a higher-dimensional feature space; is the weight vector, is the bias, which is the intercept term of the regression model; by introducing the kernel function Replace the explicit feature mapping process , which can effectively find support vectors in high-dimensional space; Minimize the objective function while satisfying the constraints:

[0020] in, and is a slack variable, and Insensitive function correlation; is the regularization parameter, which controls the trade-off between model complexity and allowable error; the kernel function of the support vector regression model is searched by grid method. , insensitive function , regularization parameter Set the hyperparameters and select the optimal parameter combination; finally, use the selected parameter configuration to achieve prediction; S2: Based on multiple predicted values ​​of total quantity control, the Kalman filter combined prediction method is used to obtain the total quantity prediction result; In an exemplary embodiment, step S2 specifically includes: according to multiple predicted values ​​of total quantity control, using a Kalman filter combined prediction method, obtaining a total quantity prediction result, such as formula: , , , , , , in, represents the GDP forecast for the current period, is the estimate for the previous period, is the state transition matrix, is the matrix of prediction values ​​provided by the model, is the observation matrix, represents the observation noise, is the forecast error covariance, is the forecast error covariance for the previous period, is the process noise covariance, is the Kalman gain, The transposed matrix of the observation matrix, is the observation noise covariance, is the optimal state estimate of GDP in the current period, is the forecast error covariance for the current period; As an exemplary embodiment, Kalman filter combined prediction is performed in step S2. This step mainly constructs a Kalman filter model, and uses multiple results in step 1 to perform combined prediction to obtain a total amount prediction result with higher accuracy and stability. For the total amount prediction, the prediction results of multiple models obtained in step S1 can be input into the Kalman filter model for optimization. Kalman filtering is a recursive algorithm that estimates the state of the system by combining observation data and the dynamic model of the system. Its basic principle is to use the prediction-update process to continuously correct the state estimate to make it closer to the true value. Specifically, the Kalman filter is used to fuse the outputs of multiple prediction models (GDP prediction numerical results, the output of each model is regarded as observation data), and the Kalman filter generates the optimal GDP estimate after combining these observation data. The Kalman filter algorithm is mainly divided into the following two steps: prediction and update; It should be noted that, in the prediction step, this embodiment predicts the current GDP based on the historical GDP estimation value and the input of the prediction model and obtains the prediction error covariance matrix, that is, the GDP value of the current quarter is predicted using the state transition equation. For GDP prediction, the state is the real GDP value, and the observed value is the GDP prediction value provided by different prediction methods; step S2 specifically includes: First, we define the state transition equation to describe the process of GDP random walk: (4) in, represents the GDP forecast for the current period, is the estimate for the previous period, is the state transfer matrix, which is set to 1 here to simplify the model as an exemplary embodiment; The output (observation) of each model can be described as: (5) in, is the matrix of prediction values ​​provided by the model, is the observation matrix, usually the identity matrix, represents the observation noise; then the covariance of the prediction error is calculated: (6) in, is the forecast error covariance, is the process noise covariance; In the update step, the predicted value is corrected in combination with the actual observation value, that is, the predicted value is corrected using the observation value, the Kalman gain is calculated, and the state estimate and covariance are updated; First calculate the Kalman gain: (7) in, is the Kalman gain, is the observation noise covariance, Kalman gain Weighs the error information between prediction and observation Relationship with observation noise , used to adjust the predicted state estimate to obtain an updated estimate; Next, revise the GDP forecast for the current period , combined with the output results (observations) of each model Get the optimal state estimate of GDP in the current period , used for prediction of the next moment: (8) The forecast error covariance matrix is ​​also updated to reflect the new uncertainty: (9) Through the above steps, Kalman filtering can effectively integrate the results of multiple GDP forecasting models, provide smoother and more accurate GDP estimates, and also provide a more powerful and scientific data support for the collaborative simulation model; S3: Perform spatial collaborative simulation based on the total amount prediction results to obtain collaborative simulation results; As an exemplary embodiment, in step S3, spatial collaborative simulation calculation is performed. This step requires setting the number of iterations and running the cellular automaton model. According to the idea of ​​"hierarchical progression", the result of each run is converted into a spatial auxiliary variable to assist the calculation of the next round of development model, thereby reconstructing the cellular automaton model to achieve a collaborative simulation calculation effect. It should be noted that in step S3, the cellular automaton model in the current study assumes that the cells only contain one type of land use. Due to the complexity of urban space, this single type of simulation method is difficult to accurately simulate and predict the urban development pattern, and is deficient in simulating continuous factors such as population and economy. Therefore, this embodiment comprehensively considers the development and changes of urban land, population, and economy, and constructs a multi-factor spatial collaborative simulation model by coupling patch generation of land use simulation and population and economic density value simulation. The model flow chart is as follows: Figure 5 As shown; In an exemplary embodiment, step S3 specifically includes: S31: Construct a density cellular automaton model, such as the formula: , , , in, for The cell state at the moment, for The development density of the moment, For the nearby range, is the cell density change, is the change in density, is the speed of density development, For nearby range The collection of units that have been converted to urban land, For nearby range The development density of inner-city land, is the maximum value of development density, is the radius of the proximity range; As an exemplary embodiment, in step S31, for the simulation of population and economic density, this embodiment adopts a density cellular automaton model that simultaneously considers the change of the cell unit state and the development density, and uses a continuous "density value" to represent the change of the cell state, thereby constructing a simulation method based on continuous data; (10) In the formula, is the cell state at time t, is the development density at time t, is the adjacent range; in this embodiment, "density" is used to express the change of the cell state. When "density" gradually changes from 0 to 1, it means that the cell is converted into urban land; the change of cell "density" is determined by the development density of adjacent cells and constraints: (11) In the formula, is the change in density, is the speed of density development, is the adjacent range; the higher the economic level in the adjacent range of a unit, the higher the corresponding urban development density, the greater the possibility of conversion into urban land, and the faster the change rate of its "density"; therefore, the change in the "density" value is expressed as follows: (12) In the formula, For nearby range The collection of units that have been converted to urban land, For nearby range The development density of inner-city land, is the maximum value of development density, is the radius of the proximity range; S32: Use the random forest model and cellular automaton to build the PLUS model, as shown in the formula: , , , in, For Grid Land use type The conversion probability, is a variable with a value of 0 or 1. is the number of decision trees in the model, and also the number of samples from the original training dataset; For the A decision tree vector The prediction type, It is a vector composed of multiple driving factors; is the indicative function set by the decision tree; Indicates the future land use type The impact of demand is an adaptive driving factor that depends on the iteration Current land quantity and land use type at the time of gap between target needs. and Respectively The iteration and At the iteration, the difference between the current land quantity and the target demand is Represents a grid At the moment Convert to Land Use Type The total probability of For Grid Changes from other types to other land use types The conversion probability, Indicates whether other land use types are converted into land use types , is a random value from 0 to 1. For land use type Generate new land use thresholds, Indicated in The iteration represents the cell Land type Neighborhood effects; As an exemplary embodiment, in step S32, for land use types, the PLUS model used in this embodiment couples the land extension analysis strategy (LEAS) and the CA model based on multi-type random plot seeds (CARS), and further improves the conversion rule mining mechanism, which can simultaneously predict multiple future development scenarios including urban development expansion; the PLUS model uses the random forest model to establish a new data mining rule to evaluate the driving factors contributing to the growth of each land use type; the raster derived from LEAS Land use type The conversion probability is calculated as follows: (13) In the formula, The value is 0 or 1. In this study, the value 1 indicates a change from other types to other land use types. , while a value of 0 indicates other changes; It is a vector composed of multiple driving factors. is the indicative function set by the decision tree; No. The prediction type of a decision tree for a vector . is the number of decision trees in the model and the number of samples from the original training data set; the CA model is used to iteratively change the transition rules of each grid to simulate land use; the PLUS model is based on traditional cellular automata and adopts an adaptive inertial competition mechanism to manage overall land use demand and control local land use competition: (14) in, Indicates the future land use type The impact of demand is an adaptive driving factor that depends on the iteration Current land quantity and land use type at the time of gap between target needs. and Respectively The iteration and The difference between the current land quantity and the target demand at the iteration; At the moment Convert to Land Use Type The total probability The expression is as follows: (15) in, Indicated in The iteration represents the cell Land type Neighborhood effects; Indicated in Adaptive function at iterations; Indicates whether other land use types are converted into land use types , is a random value from 0 to 1; For land use type Generate new land use thresholds; S33: Based on the total amount prediction results, the density cellular automaton model and the PLUS model are used to perform spatial collaborative simulation to obtain collaborative simulation results; As an exemplary embodiment, in step S3, current research mostly uses historical economic, social and traffic data as static driving factors to simulate future development scenarios, ignoring the dynamic collaborative impact in the development process. Therefore, the collaborative cellular automaton CoCA model constructed in this embodiment adopts a "hierarchical progressive" dynamic driving factor update strategy to achieve dynamic collaborative simulation of multiple elements; Figure 5 The implementation process of the strategy is shown. The strategy follows the cyclical interaction between land, population and economic production. The core idea is to repeatedly train the CA model using the initial simulation result data instead of historical data, and iterate multiple times to reach the required total amount of each type of data in the model to capture the synergistic impact between factors. In an exemplary embodiment, step S33 specifically includes: S331: Based on the total amount prediction results and spatial auxiliary data, the initial overall development probability is obtained, and the initial simulation results are obtained using the density cellular automaton model and the PLUS model; It should be noted that spatial auxiliary data include natural elements, transportation elements, and location elements; It should be noted that the total volume forecast results include economic density data; As an exemplary embodiment, in step S331, when the model is running, the initial simulation is first performed using the PLUS and DCA models, and the initial time and simulation time Land use , population density and economic density The original data is used as the initial input; at the same time, spatial auxiliary data such as natural elements, traffic elements, and location elements are added as driving factors , calculate the overall development probability and generate land use , population density and economic density The initial simulation results of S332: Based on the initial overall development probability and the initial simulation results, a density cellular automaton model and a PLUS model are used to perform spatial collaborative simulation to obtain new collaborative simulation results and a new overall development probability; As an exemplary embodiment, in step S332, after obtaining the initial simulation results, a step-by-step collaborative simulation method is used to input the result data as part of the driving factors into the model to learn the conversion rules between various factors, and output the first collaborative simulation results; taking land use as an example, the initial simulated land use is obtained. After that, the initial simulated population density and economic density As a driving factor, combined with simulation time Land Use , generate the first collaborative land use simulation results , with the population of the first co-simulation and economy Results; Due to the differences in the range and dimension of different elements in the collaborative simulation method, the simulation results of land use and population and economic density need to be further processed before being introduced into another model as driving factors; this embodiment uses the min-max normalization method to process them, and the formula is as follows: (16) in, For the The normalized value of population and economic density of a plot unit, For the The actual value of the population and economic density of a plot unit, and are the minimum and maximum values ​​in the entire study area, respectively; S333: Perform the next collaborative simulation according to the new collaborative simulation result and the new overall development probability and execute it cyclically until the relative error of the comprehensive index for multiple consecutive times is within the preset threshold, stop the calculation, and obtain the collaborative simulation result; In an exemplary embodiment, the comprehensive index is as follows: , in, As a comprehensive indicator, For land use simulation performance, is the population density simulation error, is the economic density simulation error, , and is the corresponding weight; In an exemplary embodiment, the preset threshold is 0.1; As an exemplary embodiment, in step S333, after completing the above processing, the first collaborative simulation result is used as a new driving factor to retrain the CA model, and calculate a new overall development probability to perform the next collaborative simulation; this cycle is continuously executed until the changes in various factors tend to converge; this embodiment adopts a comprehensive indicator The index determines whether the collaborative simulation has converged. The calculation formula is as follows: (17) in, For land use simulation performance, is the population density simulation error, is the economic density simulation error, , and is the corresponding weight, , ,and If continuous Subcomprehensive error If no significant changes occur, the collaborative simulation converges, the calculation stops, and the collaborative simulation results are finally output. ; In an exemplary embodiment, when performing future scenario prediction, the current time The land use, population and economic data and driving factors will be combined in the same form, and the final output will be based on the set forecast year and total demand. The data are the results under the forecast scenario.

[0021] In an exemplary embodiment, the method for collaborative simulation of urban multi-factor space based on the combined prediction model control total volume further includes: obtaining result accuracy according to the collaborative simulation result; As an exemplary embodiment, the above-mentioned method for spatial collaborative simulation of multiple elements of cities based on the combined prediction model to control the total amount also includes result accuracy evaluation. This step uses the determination coefficient, root mean square error, and mean absolute error to evaluate the total amount prediction result; selects the quality factor and the overall classification accuracy to evaluate the simulation accuracy of land use; uses the mean absolute percentage error and root mean square error to evaluate the gray value simulation accuracy of urban elements, such as population and economy; in this step, for the accuracy of the total amount control prediction result, this embodiment uses the determination coefficient RMS Error And the mean absolute error Evaluation; R-squared is used to measure the degree of fit of the model to the data, that is, the proportion of the variation explained by the model to the total variation, and the value range is usually between 0 and 1; the closer the R-squared is to 1, the stronger the explanatory power of the model; (18) in, It is i The actual observed values, It is i Prediction values, is the average of all actual observations, is the number of observations; the root mean square error is the square root of the average of the squares of the deviations between the predicted value and the actual value, which is used to measure the accuracy of the prediction; the smaller its value, the more accurate the prediction; (19) in, It is i The actual observed values, It is i Prediction values, is the number of observations; the mean absolute error is the average of the absolute values ​​of the deviations between the predicted values ​​and the actual values, providing an intuitive measure of the error between the predicted values ​​and the actual values. The smaller the value, the more accurate the prediction; (20) in, It is i The actual observed values, It is i Prediction values, is the number of observations; these three indicators have different focuses when evaluating the accuracy of the macroeconomic indicator prediction model. R-squared focuses on the model explanatory power, RMSE focuses on the size of the error, and MAE focuses on the average absolute size of the error; this embodiment combines these three indicators to comprehensively evaluate the performance of the model.

[0022] It should be noted that the urban development factor results of this embodiment involve discrete data (land use) and continuous data (population density and GDP density). This embodiment uses four indicators to evaluate the simulation results; the quality factor (FoM) is an indicator that pays more attention to the amount of change in the simulation process, which is equivalent to the ratio of the correctly predicted changes to all predicted changes. It is an accuracy evaluation indicator widely used in current vector cellular automaton simulation research. It can provide a relatively comprehensive evaluation of the performance of the model; the overall accuracy (OA) represents the proportion of correctly simulated cellular units to the total number, and is also widely used in model evaluation; therefore, this embodiment selects the FoM indicator and the OA indicator to evaluate the simulation effect of discrete variables, and the formula is as follows: (twenty one) (twenty two) Where: Indicates the number of cells whose actual type changes and whose simulated type does not change; Indicates the actual type change, the number of cells whose simulated type changes correctly; represents the actual type change and the number of cells that simulate the type error change; Indicates the number of cells whose actual type has not changed but whose simulated type has changed; represents the total number of cell units in the study area; Starting from the continuous data dimension, this embodiment uses the mean absolute percentage error (MAPE) and root mean square error (RMSE) to evaluate the performance of urban density simulation; the definitions of these indicators are as follows: (twenty three) (twenty four) In the formula, is the number of all samples, is the true value of population density and GDP density, are the predicted values ​​of population density and GDP density.

[0023] In an exemplary embodiment, the above-mentioned urban multi-factor spatial collaborative simulation method based on the total amount of control of the combined prediction model includes the following steps: like Figure 2 The following is the overall flow chart: First, data collection is carried out, including economic indicator data, news text data, land use data, and geographic driving factor data; the next step is to build a multi-indicator prediction model, which includes data-driven economic prediction models such as ARIMA, TimesFM, OLS, and XGBoost, and news text prediction models such as BiLSTM and M3E. The prediction results of multiple models are input into the Kalman filter optimization to obtain the total result of the combined prediction; then the total demand for each category of urban land use and the total development of the population and economic density values ​​are set, and multi-factor spatial collaborative simulation is carried out to obtain the collaborative simulation results of urban land use and population economy. Finally, there is the accuracy evaluation part, which uses a variety of indicators such as FOM, OA, MAE, R Square, RMSE, and MAPE to evaluate the model simulation effect; Among them, Figure 3The following is the ARIMA model flow chart: First, input a time series data and check whether the data is stationary. If the data is not stationary, differential processing is required until the data becomes stationary. Through this step, we get the parameter d; then we need to determine the autoregressive order p and the moving average order q: based on the autocorrelation function (ACF) and partial autocorrelation function (PACF) images of the time series data, we can preliminarily estimate the range of the autoregressive order p and the moving average order q; specifically, we judge whether they are tailing or truncated. Tailing and truncation are a description of the image state. If the ACF and PACF coefficients are significantly larger than 2 standard deviations at the initial order, and then almost 95% of the coefficients fall within the range of 2 standard deviations, and the process of non-zero coefficients decaying to small fluctuations is very sudden, it is considered to be K-order truncation; if more than 5% of the sample correlation coefficients are larger than 2 standard deviations, or the process of non-zero coefficients decaying to small fluctuations is relatively slow or continuous, it is usually considered to be tailing; after obtaining all the parameters, we can directly fit the model, and then diagnose the fitted model to check whether the residual meets the model assumptions (such as whether it is white noise); if the conditions are met, the predicted results can be obtained, if not, the values ​​of p and q need to be adjusted and refitted; Among them Figure 4 The figure shows the flow chart of the BiLSTM sentiment scoring model: first, the text data is cleaned and input into the model to obtain the sentiment score; at the same time, the economic statistical data is used to obtain the economic autocorrelation index through preliminary calculation and X13-ARIMA seasonal adjustment; the two types of data are superimposed to obtain the fusion feature, and then the regression processing with the true value is performed to achieve the prediction; Among them Figure 5 The figure shows a multi-factor collaborative simulation model: the specific steps are as follows: (1) using the original data and driving factors as input, using the PLUS model and the DCA model to perform independent land use simulation and initial population and economic density simulation, and obtain the simulation results of each type of factor; (2) setting different simulation round number thresholds, each time the number of simulation rounds reaches the threshold, the obtained simulation results are put into another simulation model as driving factors, and the overall development probability is recalculated for simulation to train the model; (3) replacing the original data with the corrected results, performing the above steps multiple times until the changes in each factor converge, and completing the collaborative simulation that comprehensively considers the mutual influence of land use and population and economy; Among them, Figure 6The execution process of the strategy is demonstrated. The "hierarchical progressive" dynamic driving factor update strategy realizes the dynamic collaborative simulation of multiple factors, following the cyclical interaction between land, population and economic production. The core idea is to use land, population and economy as training data in turn during the iteration process to influence the result of the next iteration, so as to achieve the effect of coordinated development and mutual influence. The overall idea is that first, land, population and economy are trained separately using spatial auxiliary variables (such as elevation, road network, etc.) as model training data, and then a set of independent simulation results are obtained. The simulation results of population and economy are normalized and added to the spatial auxiliary variables (second round), and then the vector land use simulation results are converted from vector to raster with the same grid resolution size, and normalized and added to the spatial auxiliary variables. In the second round of simulation, there are not only the original auxiliary data, but also the results of land, population and economy in the first round to assist, so as to achieve the effect of coordinated development and mutual progressive influence.

[0024] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.

Claims

1. A method for spatial collaborative simulation of multiple factors in cities based on a combined prediction model to control total volume, characterized in that: The following steps are involved: S1: Based on economic data, we use the economic data-driven model and the news text prediction model to obtain multiple forecast values ​​of total volume control; S2: according to the multiple predicted values ​​of the total amount control, using the Kalman filter combined prediction method to obtain the total amount prediction result; S3: Perform spatial collaborative simulation according to the total amount prediction result to obtain a collaborative simulation result.

2. A method for spatial collaborative simulation of multiple elements of cities based on a combined prediction model to control the total amount of capital, characterized in that: The method for collaborative simulation of urban multi-factor space based on the combined prediction model to control the total amount also includes: obtaining the result accuracy according to the collaborative simulation result.

3. The method for spatial collaborative simulation of urban multi-factors based on combined prediction model control total volume according to claim 1 is characterized in that: The economic-related data include nominal GDP and macroeconomic-related news; the multiple forecast values ​​of the total amount control include a first nominal GDP forecast value, a second nominal GDP forecast value, a third nominal GDP forecast value, a fourth nominal GDP forecast value, a fifth nominal GDP forecast value, and a sixth nominal GDP forecast value; The economic data driven model includes an ARIMA model, an ordinary least squares regression model, a time series basic model and an XGBoost model; the news text prediction model includes a BiLSTM sentiment scoring model and a Moka large-scale hybrid embedding model; the ARIMA model obtains a first nominal GDP prediction value based on the GDP nominal value; the ordinary least squares regression model obtains a second nominal GDP prediction value based on the GDP nominal value; the time series basic model obtains a third nominal GDP prediction value based on the GDP nominal value; the XGBoost model obtains a fourth nominal GDP prediction value based on the GDP nominal value; the BiLSTM sentiment scoring model obtains a fifth nominal GDP prediction value based on the GDP nominal value and macroeconomic related news; the Moka large-scale hybrid embedding model obtains a sixth nominal GDP prediction value based on the GDP nominal value and macroeconomic related news.

4. The method for spatial collaborative simulation of urban multi-factors based on combined prediction model control total volume according to claim 1 is characterized in that: The total volume forecast results include economic density data.

5. The method for spatial collaborative simulation of urban multi-factors based on combined prediction model control total volume according to claim 1 is characterized in that: Step S2 specifically includes: according to the multiple predicted values ​​of the total amount control, using the Kalman filter combined prediction method, obtaining the total amount prediction result, such as the formula: ， ， ， ， ， ， in, represents the GDP forecast for the current period, is the estimate for the previous period, is the state transition matrix, is the matrix of prediction values ​​provided by the model, is the observation matrix, represents the observation noise, is the forecast error covariance, is the forecast error covariance for the previous period, is the process noise covariance, is the Kalman gain, The transposed matrix of the observation matrix, is the observation noise covariance, is the optimal state estimate of GDP in the current period, is the current period forecast error covariance.

6. The method for spatial collaborative simulation of urban multi-factors based on combined prediction model control total volume according to claim 1 is characterized in that: Step S3 specifically includes: S31: Construct a density cellular automaton model, such as the formula: ， ， ， in, for The cell state at the moment, for The development density of the moment, For the nearby range, is the cell density change, is the change in density, is the speed of density development, For nearby range The collection of units within the area that have been converted to urban land, For nearby range The development density of inner-city land, is the maximum value of development density, is the radius of the proximity range; S32: Use the random forest model and cellular automaton to build the PLUS model, as shown in the formula: ， ， ， in, For Grid Land use type The conversion probability, is a variable with a value of 0 or 1. is the number of decision trees in the model, and also the number of samples from the original training dataset; For the A decision tree vector The prediction type, It is a vector composed of multiple driving factors; is the indicative function set by the decision tree; Indicates the future land use type The impact of demand is an adaptive driving factor that depends on the iteration Current land quantity and land use type at the time of gap between target needs. and Respectively The iteration and At the iteration, the difference between the current land quantity and the target demand is Represents a grid At the moment Convert to Land Use Type The total probability of For Grid Changes from other types to other land use types The conversion probability, Indicates whether other land use types are converted into land use types , is a random value from 0 to 1. For land use type Generate new land use thresholds, Indicated in The iteration represents the cell Land type Neighborhood effects; S33: According to the total amount prediction result, the density cellular automaton model and the PLUS model are used to perform spatial collaborative simulation to obtain a collaborative simulation result.

7. The method for spatial collaborative simulation of urban multi-factors based on combined prediction model control total volume according to claim 6 is characterized in that: Step S33 specifically includes: S331: Obtaining the initial overall development probability according to the total amount prediction result and the spatial auxiliary data, and obtaining the initial simulation result by using the density cellular automaton model and the PLUS model; S332: performing spatial collaborative simulation using the density cellular automaton model and the PLUS model according to the initial overall development probability and the initial simulation result, to obtain a new collaborative simulation result and a new overall development probability; S333: Perform the next collaborative simulation according to the new collaborative simulation result and the new overall development probability and execute it in a loop until the relative error of the comprehensive indicators for multiple consecutive times is within the preset threshold, stop the calculation, and obtain the collaborative simulation result.

8. The method for spatial collaborative simulation of urban multi-factors based on combined prediction model control total volume according to claim 7 is characterized in that: The spatial auxiliary data includes natural elements, traffic elements, and location elements.

9. The method for spatial collaborative simulation of urban multi-factors based on combined prediction model control total volume according to claim 7 is characterized in that: The comprehensive indicator is as follows: ， in, As a comprehensive indicator, For land use simulation performance, is the population density simulation error, is the economic density simulation error, , and is the corresponding weight.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method for spatial collaborative simulation of multiple elements of a city controlling total volume based on a combined prediction model are implemented as described in any one of claims 1 to 9.

Images