Festival and holiday key business district passenger flow prediction method based on multi-model fusion

Abstract

The invention discloses a holiday key business district passenger flow prediction method based on multi-model fusion, and the method is characterized in that the method comprises the steps: data processing, model construction, multi-model fusion, and data prediction value output. In the data processing part, signaling data is adopted as original data, and feature coding is performed on holiday, workday and dual-holiday data so as to improve data accuracy; in the model building part, a gradient ascending algorithm, a random forest algorithm and a linear regression algorithm are respectively responsible for data prediction tasks; in the multi-model fusion and output data prediction value part, the prediction data output by the multiple models are subjected to normalization error processing and weighting processing, so that the accuracy of the data is improved.

Description

Technical Field

[0001] This invention belongs to the field of passenger flow prediction technology, specifically involving a method for predicting passenger flow in key business districts during holidays based on multi-model fusion. Background Technology

[0002] Commercial districts are crucial carriers of urban commercial activities. The volume of foot traffic directly impacts the business operations and development prospects of merchants within these districts, reflecting the city's consumer vitality. Current methods for predicting foot traffic in commercial districts involve spatial division using pre-defined remote sensing imagery and geoscientific analysis techniques. Based on pre-defined spatial topological relationships, hotspot areas are gridded, and the collected foot traffic data from these gridded areas undergoes spatiotemporal distribution analysis to determine the current foot traffic range within each hotspot area. This allows for the prediction of future foot traffic ranges for these hotspot areas. The methods primarily rely on data from commercial district merchants' operations, building foot traffic, popular tourist attractions, merchant location tables, and merchant rental management. However, this approach involves a large volume of complex data, incurring high costs and a cumbersome analysis process, lacking effective data management tools. Furthermore, each analysis can only predict the flow of a single commercial district, lacking universality. Summary of the Invention

[0003] The purpose of this invention is to provide a method for predicting passenger flow in key business districts during holidays based on multi-model fusion. This method improves the prediction accuracy and efficiency of the model by effectively fusing multiple data prediction methods, and enhances its robustness to noisy data.

[0004] To achieve the above-mentioned technical effects, the technical solution adopted by the present invention is as follows: A method for predicting customer flow in key commercial areas during holidays based on multi-model fusion, characterized by the following steps: S1. Data Processing: Mobile signaling data is used as the raw data. The raw data is cleaned, normalized, and feature-selected to remove noise and outliers and ensure data quality. At the same time, feature coding is performed on data from holidays, weekdays, and weekends to improve data accuracy. S2. Model Construction: Construct gradient boosting algorithm model, random forest algorithm model and linear regression model respectively, train them and output the final prediction model of each model; S3. Multi-model fusion and output data prediction values; normalized error values ​​of each prediction model in the validation set obtained from step S2. , and Various prediction models are used to predict future data and obtain corresponding predicted values. , and The predicted data y after multi-model fusion is:

[0005] In the formula: , and The value range of is [0-1]; 1. [*] indicates an indicator function: it takes the value 1 when the expression inside [*] is true, and 0 otherwise; This is a threshold parameter, set by experience, used to exclude low-confidence predictions in the prediction process; , and These represent the model weights when the gradient boosting algorithm model, random forest algorithm model, and linear regression model are fused.

[0006] Furthermore, the construction of the gradient boosting algorithm model in step S2 includes the following steps: S211. Initialize the model: Set up the initial model for the gradient boosting algorithm to predict the initial values ​​of the data; S212. Constructing a weak learner: Use a decision tree as a weak learner. In each iteration, train a new decision tree based on the residuals of the current model to gradually improve the prediction accuracy of the model. S213, Gradient Calculation: Calculate the gradient of the current model on the objective function so that the model can continuously approach the true value; S214. Model Update: In each iteration, the model's prediction results are updated using the output of the previous decision tree; by adjusting the weight coefficients of each tree, the degree of fit of the model to the training data is controlled to avoid overfitting.

[0007] S215. Regularization: To prevent overfitting, regularization techniques such as L1 and L2 regularization terms are used to control model complexity and to further improve generalization ability through early stopping mechanism. S216. Output Predicted Data: After reaching the preset number of iterations or the model convergence condition, the final prediction model is output, which can be used for future data prediction tasks.

[0008] Furthermore, the construction of the random forest algorithm model in step S2 includes the following steps: S221. Constructing a Random Forest: A random forest consists of multiple independent decision trees. Each decision tree is generated through data sampling and feature selection. Data sampling is performed on the original training dataset, using the Bootstrap method to randomly sample and generate multiple subsets of the dataset. Each subset is used to train a decision tree model. Feature selection is performed during the construction of each decision tree, randomly selecting a subset of features from the feature space for node splitting, thereby improving the diversity of the model and reducing the risk of overfitting. S222, Train multiple decision trees: Build a decision tree for each subset of the dataset; each decision tree constructs multiple nodes by recursively splitting the dataset, and each node divides the data into different subsets based on the best split point of a certain feature; S223, Ensemble Prediction: For classification tasks, a majority voting method is used to vote on the predictions of all decision trees and output the final classification result. For regression tasks, an average method is used to take the average of the predictions of all decision trees as the final prediction result. S224. Model Optimization: The model's performance is optimized by adjusting hyperparameters such as the number of trees in the forest, maximum depth, and minimum number of leaf nodes. In addition, regularization and cross-validation techniques are introduced to prevent overfitting, and the interpretability of the model is improved through feature importance evaluation.

[0009] S225. Output Predicted Data: After all decision trees have been trained, the prediction results of multiple trees are integrated to generate the final prediction model, which can be applied to the prediction task of future data.

[0010] Furthermore, the construction of the linear regression model in step S2 includes the following steps: S231. Establish a linear regression model: Establish a linear regression model; S232. Define the loss function: Define the loss function as mean squared error (MSE). S233. Minimize the loss function: Use the gradient descent algorithm to optimize the model parameters and minimize the loss function to find the optimal regression coefficients; the gradient descent algorithm gradually reduces the error by repeatedly updating the parameters.

[0011] S234. Regularization: To prevent overfitting, regularization terms are added, such as L1 regularization (Lasso) or L2 regularization (Ridge).

[0012] S235. Output Predicted Data: After the model training is completed, save the optimized linear regression model and use it for future data prediction tasks.

[0013] Compared with the prior art, the beneficial effects of the present invention are: This invention effectively solves the problem of fragmented data sources and inaccurate management by collecting mobile phone signaling data as training data for the model. Furthermore, by using machine learning, the model is trained based on key holidays, weekdays, and weekends data within the business district. Through the effective fusion of multiple data prediction methods, the model's prediction accuracy and efficiency are improved, and its robustness to noisy data is enhanced, enabling effective prediction of customer flow in business districts. Moreover, this invention eliminates the need for deploying physical monitoring equipment such as remote sensing devices and can perform corresponding prediction work in various business districts throughout the city, reducing prediction costs and workload. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the overall prediction process of the model in this invention; Figure 2 This is a schematic diagram of the gradient boosting model of the present invention; Figure 3 This is a schematic diagram of the random forest model of the present invention; Figure 4 This is a schematic diagram of the linear regression model of the present invention. Detailed Implementation

[0015] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings. However, these descriptions do not constitute a limitation of the present invention and are merely illustrative. The advantages of the present invention will become clearer and easier to understand through this description.

[0016] like Figure 1 As shown, this invention provides a method for predicting passenger flow in key commercial areas during holidays based on multi-model fusion. The method includes the following steps: S1. Data Processing: Mobile signaling data is used as the raw data. The raw data is cleaned, normalized, and feature-selected to remove noise and outliers, ensuring data quality. Additionally, features such as whether the data falls on a holiday, weekday, weekend, or before / after a holiday are encoded to improve data accuracy.

[0017] S2. Model Construction: Construct gradient boosting algorithm model, random forest algorithm model and linear regression model respectively, and train them to obtain the final prediction model of each model.

[0018] S21. Construction of the gradient boosting algorithm model: The construction of the gradient boosting algorithm model involves the following steps: S211. Initialize the model: Set up the initial model for the gradient boosting algorithm. This model is usually a simple constant model used to predict the initial values ​​of the data.

[0019] S212. Constructing a weak learner: Use a decision tree as a weak learner. In each iteration, train a new decision tree based on the residuals of the current model, thereby gradually improving the prediction accuracy of the model.

[0020] S213. Gradient Calculation: Calculate the gradient of the current model on the objective function. The gradient is used to guide the construction of the next decision tree, so that the model can continuously approach the true value.

[0021] S214. Model Update: In each iteration, the model's predictions are updated using the output of the previous decision tree. By adjusting the weight coefficients of each tree, the model's fit to the training data is controlled, avoiding overfitting.

[0022] S215. Regularization: To prevent overfitting, regularization techniques such as L1 and L2 regularization terms are used to control model complexity, and early stopping mechanisms are used to further improve generalization ability.

[0023] S216. Output Predicted Data: After reaching the preset number of iterations or the model convergence condition, the final prediction model is output. This model can be used for future data prediction tasks, outputting predicted data.

[0024] S22. Construction of the Random Forest Algorithm Model: The construction of the random forest algorithm model includes the following steps: S221. Constructing a Random Forest: A random forest consists of multiple independent decision trees. Each decision tree is generated through data sampling and feature selection. Data sampling is performed on the original training dataset, using the Bootstrap method to randomly sample and generate multiple subsets of the dataset. Each subset is used to train a decision tree model. Feature selection is performed during the construction of each decision tree, randomly selecting a subset of features from the feature space for node splitting, thereby improving the diversity of the model and reducing the risk of overfitting.

[0025] S222, Training Multiple Decision Trees: Construct a decision tree for each subset of the dataset. Each decision tree constructs multiple nodes by recursively splitting the dataset, with each node dividing the data into different subsets based on the optimal split point for a certain feature.

[0026] S223, Ensemble Prediction: For classification tasks, a majority voting method is used to vote on the predictions of all decision trees and output the final classification result. For regression tasks, an average method is used to take the average of the predictions of all decision trees as the final prediction result.

[0027] S224. Model Optimization: The model's performance is optimized by adjusting hyperparameters such as the number of trees in the forest, maximum depth, and minimum number of leaf nodes. Furthermore, regularization and cross-validation techniques are introduced to prevent overfitting, and feature importance evaluation is used to improve the model's interpretability.

[0028] S225. Output Predicted Data: After all decision trees have been trained, the prediction results of multiple trees are integrated to generate the final prediction model. This model can be applied to the prediction task of future data and output predicted data.

[0029] S23. Construction of a Linear Regression Model: The construction of a linear regression model includes the following steps: S231. Establish a linear regression model: Establish a linear regression model.

[0030] S232. Define the loss function: Define the loss function as mean squared error (MSE), which measures the deviation between the predicted value and the true value.

[0031] S233. Minimizing the loss function: The gradient descent algorithm is used to optimize model parameters and minimize the loss function to find the optimal regression coefficients. The gradient descent algorithm gradually reduces the error by repeatedly updating the parameters.

[0032] S234. Regularization: To prevent overfitting, regularization terms are added, such as L1 regularization (Lasso) or L2 regularization (Ridge).

[0033] S235. Output Predicted Data: After the model training is completed, save the optimized linear regression model and use it for future data prediction tasks to output predicted data.

[0034] S3. Construct a multi-model fusion and output the predicted values: The multiple prediction models trained in step S2 and the corresponding normalized error values ​​predicted by each prediction model on the validation set. , and (Values ​​range from [0-1]). Using various prediction models to forecast future data yields corresponding multi-model predictions. , and The predicted data y after multi-model fusion is: (1) Where 1[*] indicates an indicator function: it takes the value 1 when the expression inside [*] is true, and 0 otherwise; This is a threshold parameter, set by experience, used to exclude low-confidence predictions in the prediction process; , and These represent the model weights of the gradient boosting algorithm, random forest algorithm, and linear regression model during fusion, respectively, and are calculated as follows: (2) (3) (4) in, , , , represent the distance coefficients of different models, where the distance is calculated as follows: (5) (6) (7) in, (8) (9) (10) These represent the normalized predicted values ​​from multiple models, This represents the average predicted value.

[0035] After calculating the weights of each model, the model weights need to be normalized so that the sum of the weights of each sub-model is 1, in order to ensure the consistency of the predicted data in terms of scale.

[0036] Through the above steps, this invention effectively improves the accuracy of data prediction and possesses strong robustness and scalability, making it suitable for predicting customer traffic in key commercial areas during holidays. By fusing three data prediction models, it effectively compensates for the overfitting and large errors that may arise from single-model prediction. Based on corresponding machine learning model algorithms, it reduces prediction costs and improves accuracy. This invention employs three data prediction models: Gradient Boosting, Random Forest, and Linear Regression to predict customer flow in commercial areas during key holidays. Of course, other machine learning algorithms exist, such as decision tree algorithms that can replace Random Forest; XGBoost and Adaboost that can replace Gradient Boosting; and Ridge Regression and Multinomial Regression that can replace Linear Regression. However, testing shows that the fusion model algorithm used in this invention offers the best overall performance.

Claims

1. A method for predicting passenger flow in key commercial districts during holidays based on multi-model fusion, characterized in that, The method includes the following steps: S1. Data Processing: Using mobile signaling data as raw data, feature coding is performed on data from holidays, weekdays, and weekends to improve data accuracy; S2. Model Construction: Construct gradient boosting algorithm model, random forest algorithm model and linear regression model respectively, train them and output the final prediction model of each model; S3. Multi-model fusion and output data prediction values; normalized error values ​​of each prediction model in the validation set obtained from step S2. , and Various prediction models are used to predict future data and obtain corresponding predicted values. , and The predicted data y after multi-model fusion is: , In the formula: , and The value range of is [0-1]; 1. [*] indicates an indicator function: it takes the value 1 when the expression inside [*] is true, and 0 otherwise; This is a threshold parameter, set by experience, used to exclude low-confidence predictions in the prediction process; , and These represent the model weights when the gradient boosting algorithm model, random forest algorithm model, and linear regression model are fused.

2. The method for predicting passenger flow in key business districts during holidays based on multi-model fusion as described in claim 1, characterized in that, The mobile signaling data processed in step S1 needs to be cleaned, normalized, and feature-selected to remove noise and outliers from the data.

3. The method for predicting passenger flow in key business districts during holidays based on multi-model fusion as described in claim 1, characterized in that, The gradient boosting algorithm model construction in step S2 includes the following steps: S211. Initialize the model: Set up the initial model for the gradient boosting algorithm to predict the initial values ​​of the data; S212. Constructing a weak learner: Use a decision tree as a weak learner. In each iteration, train a new decision tree based on the residuals of the current model to gradually improve the prediction accuracy of the model. S213, Gradient Calculation: Calculate the gradient of the current model on the objective function so that the model can continuously approach the true value; S214. Model Update: In each iteration, the model's prediction results are updated using the output of the previous decision tree; the model's fit to the training data is controlled by adjusting the weight coefficients of each tree. S215. Regularization: Regularization techniques are used to control model complexity, and early stopping mechanisms are used to improve generalization ability. S216. Output Predicted Data: After reaching the preset number of iterations or the model convergence condition, output the final prediction model for future data prediction tasks.

4. The method for predicting passenger flow in key business districts during holidays based on multi-model fusion as described in claim 1, characterized in that, The construction of the random forest algorithm model in step S2 includes the following steps: S221. Constructing a Random Forest: A random forest consists of multiple independent decision trees. Each decision tree is generated through data sampling and feature selection. Data sampling is performed on the original training dataset, using the Bootstrap method to randomly sample and generate multiple subsets of the dataset. Each subset is used to train a decision tree model. Feature selection is performed during the construction of each decision tree, randomly selecting a subset of features from the feature space for node splitting, thereby improving the diversity of the model and reducing the risk of overfitting. S222, Training multiple decision trees: Construct a decision tree for each subset of the dataset. Each decision tree constructs multiple nodes by recursively splitting the dataset. Each node divides the data into different subsets based on the best split point of a certain feature. S223, Ensemble Prediction: For classification tasks, the majority voting method is used to vote on the prediction results of all decision trees and output the final classification result. For regression tasks, the average method is used to take the average of the prediction results of all decision trees as the final prediction result. S224. Model Optimization: Optimize the model's performance by adjusting the hyperparameters of the number of trees, maximum depth, and minimum number of leaf nodes in the forest; S225. Output Predicted Data: After all decision trees have been trained, the prediction results of multiple trees are integrated to generate the final prediction model for future data prediction tasks.

5. The method for predicting passenger flow in key business districts during holidays based on multi-model fusion according to claim 4, characterized in that, The model optimization step S224 also includes introducing regularization and cross-validation techniques to prevent overfitting, and improving the interpretability of the model through feature importance evaluation.

6. The method for predicting passenger flow in key business districts during holidays based on multi-model fusion according to claim 1, characterized in that, The construction of the linear regression model in step S2 includes the following steps: S231. Establish a linear regression model: Establish a linear regression model; S232. Define the loss function: Define the loss function as mean squared error (MSE). S233. Minimize the loss function: Use the gradient descent algorithm to optimize the model parameters and minimize the loss function to find the optimal regression coefficients; the gradient descent algorithm gradually reduces the error by repeatedly updating the parameters. S234. Regularization: To prevent overfitting of the model, a regularization term is added; S235. Output Predicted Data: After the model training is completed, save the optimized linear regression model for future data prediction tasks.

Images