Scenic spot passenger flow volume prediction method based on multi-source information fusion

Abstract

The invention provides a scenic spot passenger flow volume prediction method based on multi-source information fusion, and belongs to the field of deep learning. The method comprises the following steps: firstly, collecting multi-source heterogeneous data including weather, holidays and passenger flow volumes of adjacent scenic spots, and carrying out preprocessing, feature extraction and dimension reduction on the multi-source heterogeneous data to obtain more compact and effective feature representation; then introducing a feature-level attention mechanism and a time attention mechanism to highlight the importance of key features and key time slices; and finally, modeling is carried out on the time sequence characteristics in combination with a long-short term memory (LSTM) network, dynamic and accurate prediction of the tourist flow volume of the scenic spot is realized, and powerful support is provided for tourist management and resource scheduling of the scenic spot.

Description

Technical Field

[0001] The present invention belongs to the field of deep learning, and specifically relates to a method for predicting tourist flow in a scenic area based on multi-source information fusion. Background Art

[0002] With the development of the tourism industry, visitor traffic to scenic spots has rapidly increased. However, issues such as oversaturation and irrational resource allocation have become increasingly prominent. These problems not only impact the visitor experience but also pose potential threats to their lives and property. Therefore, effectively managing visitor traffic to scenic spots has become a pressing issue in the public safety field.

[0003] Visitor flow forecasting, a key technology for improving scenic area management, is currently being widely used across major scenic areas. By predicting visitor flow in real time, managers can proactively understand changes in visitor flow over time, enabling the optimized deployment of management measures such as reservation and flow control, resource allocation, personnel scheduling, and security alerts. This not only enhances the visitor experience but also effectively ensures the safe operation of the scenic area. Common visitor flow forecasting methods can be categorized into traditional statistical methods, machine learning methods, and deep learning methods.

[0004] Traditional statistical methods primarily predict visitor flow by analyzing and modeling historical data. While these methods are easy to implement, they struggle to capture the complex nonlinear relationships and dynamic time series characteristics of visitor flow data, resulting in significant limitations in terms of prediction accuracy. In contrast, machine learning methods enhance the nonlinear modeling capabilities of traditional statistical methods, enabling them to uncover underlying patterns and characteristic relationships in visitor flow data, thereby improving prediction results. However, these methods still face bottlenecks in handling long-term dependencies or multi-scale features. In recent years, deep learning methods have been widely used in visitor flow prediction tasks. These methods effectively overcome the limitations of traditional statistical and machine learning methods in terms of nonlinear modeling capabilities and handling time series dependencies, significantly improving visitor flow prediction accuracy. However, existing deep learning-based methods primarily focus on predicting visitor flow over longer timescales (e.g., by year, month, or day), failing to meet the high-resolution prediction requirements (e.g., hourly) required for daily scenic area operations. Furthermore, most studies overlook the impact of factors such as weather and the flow of visitors to neighboring scenic areas on the visitor flow of a target scenic area, limiting the models' prediction accuracy and practicality. Summary of the Invention

[0005] The present invention provides a method for predicting tourist flow in a scenic area based on multi-source information fusion, which specifically includes the following steps: Step 1: Collect passenger flow, weather conditions, and holiday information from the scenic area management system, meteorological service platform, and calendar system to construct a multi-source heterogeneous dataset; Step 2: Preprocess and standardize the collected multi-source heterogeneous data to construct a standardized data set; Step 3: Use the autoencoder to extract features and reduce the dimension of the surrounding scenic spot passenger flow information, remove redundant information, and obtain a more expressive low-dimensional feature representation; Step 4: Introduce the feature-level attention mechanism to dynamically evaluate and calculate the importance weights of various features in the prediction task, and construct a unified feature sequence of weighted fusion; Step 5: The weighted fusion features are combined with the historical passenger flow information of the scenic spot under study and input into the long short-term memory network. The temporal attention mechanism is introduced to highlight the contribution of key time points to the prediction results. Step 6: Connect the fully connected layer to output the final passenger flow prediction result.

[0006] Compared with the prior art, the present invention has the following beneficial effects:

[0007] This application fully considers various influencing factors in the process of scenic spot passenger flow prediction, such as the passenger flow of neighboring scenic spots, weather and holiday information. By comprehensively considering these factors, the prediction accuracy of scenic spot passenger flow is significantly improved.

[0008] This application proposes a scenic spot passenger flow prediction model based on the AE-LSTM-Attention combination. This model fully integrates the advantages of autoencoders (AE), long short-term memory networks (LSTM), and attention mechanisms (Attention), improving the ability to model passenger flow changes in complex scenarios. Among them, the autoencoder (AE) is used to compress and remove redundancy from high-dimensional multi-source features and extract representative low-dimensional features; the long short-term memory network (LSTM) is used to model the temporal evolution of passenger flow data and improve the ability to model long-term dependent information; the attention mechanism (Attention) is used to dynamically assign weights to different features and key moments, further improving the model's responsiveness and prediction accuracy in complex changing scenarios, thereby achieving efficient and accurate prediction of scenic spot passenger flow.

[0009] This application realizes system integration from multi-source data collection, preprocessing, feature extraction to deep model prediction by constructing a complete data processing and modeling process. It has good scalability and versatility and can be widely used in passenger flow prediction scenarios of different types of scenic spots. It provides a scientific basis and technical support for scenic spot tourist management, service scheduling and safety warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 A graph showing the flow of people in each scenic spot.

[0011] Figure 2 Flowchart for multi-source data preprocessing.

[0012] Figure 3 Schematic diagram of the encoding and decoding process of the autoencoder.

[0013] Figure 4 This is a flow chart for predicting tourist flow in scenic spots.

[0014] Figure 5 Schematic diagram of the AE-LSTM-Attention model structure.

[0015] Figure 6 This is a comparison chart of the original flow and predicted flow of the scenic area.

[0016] Figure 7 This is a comparison chart of the prediction results of the model proposed in this invention and the baseline model. DETAILED DESCRIPTION

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings of the present invention. Obviously, the described embodiments are only a part of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0018] See also Figures 1 to 7 The present invention provides a method for predicting the tourist flow of scenic spots based on multi-source information fusion, which aims to achieve accurate and efficient prediction of the tourist flow of scenic spots by fusing multi-source heterogeneous data. The method specifically includes the following steps:

[0019] Step 1: Collect visitor flow, weather conditions, and holiday information from the scenic area management system, meteorological service platform, and calendar system. The specific steps to construct a multi-source heterogeneous dataset include: Step 1.1: Collect hourly passenger flow information from the scenic area management system for the target scenic area and its adjacent scenic areas at the entrances and exits of the gates from July 1, 2022, to July 1, 2024. This data contains 70,176 entries in total. Step 1.2: Collect rainfall and temperature information from the meteorological service platform, which contains 17,544 data items in total; Step 1.3: Call the holiday information query interface to obtain holiday information, which includes 731 pieces of data.

[0020] Step 2: Preprocess and standardize the collected multi-source heterogeneous data. The specific steps for constructing a standardized data set include: Step 2.1: Convert the timestamps of passenger flow, weather, and holiday data into a standard format and align them according to a unified time granularity. Using the hour as the benchmark, upsample and align low-frequency data such as holiday data to the hourly level. Step 2.2: Standardize the units of indicators in different data sources, unify temperature to degrees Celsius and rainfall to millimeters; Step 2.3: Process missing values ​​for the collected real-time passenger flow data, weather information, and holiday information. Passenger flow, temperature, and rainfall are filled using the mean filling method, while holiday information is filled using the forward filling method. Step 2.4: Normalize the multi-source data. For the scenic spot passenger flow data, the Min-Max normalization method is used to map the passenger flow data to the interval [0, 1]. The normalization formula is: ; Among them, x represents the current passenger flow, and Represent the maximum and minimum passenger flow respectively. For temperature, the Z-score standardization method is used for processing. The formula is: ; Where x is the current temperature, is the mean temperature of the region, is the standard deviation of the temperature in the area. The rainfall information is first logarithmically normalized to compress the right-skewed distribution of the data, and then Z-score normalization is performed. The logarithmic transformation formula is recorded as: ; Among them, x is the current rainfall, c is a constant, to avoid In the case of holiday information, binary encoding is used for normalization. The encoding rules are as follows: ; in, Represents the normalized holiday data.

[0021] Step 3: Use the autoencoder to extract features and reduce the dimension of the surrounding scenic spot passenger flow information, remove redundant information, and obtain a more expressive low-dimensional feature representation. The specific steps include: Step 3.1: Arrange the acquired passenger flow data of surrounding scenic spots in chronological order to construct the original input sequence; Step 3.2: Build an autoencoder model. The encoder part consists of a multi-layer fully connected neural network, which is used to compress the input high-dimensional time series data into a low-dimensional latent feature representation. The decoder is responsible for reconstructing the original input from the low-dimensional features to achieve self-supervised learning. In this embodiment, both the encoder and decoder are composed of a 4-layer fully connected network. Step 3.3: Use layer-by-layer unsupervised pre-training to train the autoencoder model. Optimize the parameters of the encoder and decoder to minimize the reconstruction error as the training objective. The loss function is expressed as: ; in, is the original input data sample, is the reconstruction result, T is the total number of samples; Step 3.4: Input the passenger flow data into the trained autoencoder and extract the corresponding low-dimensional feature sequence through the encoder, which is recorded as: ; Among them, T is the length of the time series, that is, the number of time steps of the passenger flow data, and Z is the feature representation of the entire passenger flow time series in low-dimensional space, which can be used as input for subsequent prediction or analysis.

[0022] Step 4: Introduce the feature-level attention mechanism to dynamically evaluate and calculate the importance weights of various features in the prediction task. The specific steps for constructing a weighted fusion unified feature sequence include: Step 4.1, weather information rainfall ,temperature , Holiday Information and passenger flow characteristics of surrounding scenic spots (extracted from encoder) It is projected into a unified latent space through linear transformation to facilitate subsequent weighted fusion operations. Taking rainfall as an example, its embedding expression formula is recorded as: ; in, is the weight matrix, is the bias term, is the embedding representation of weather features; Step 4.2: Introduce the learnable attention query vector v and calculate the attention score through the dot product operation. Taking weather features as an example, the score calculation formula is recorded as: ; in, is the transpose of the attention vector, which is used to score the input features. The larger it is, the more important the feature is. Similarly, we can calculate 、 、 , in order to facilitate the comparison of the relative importance of different features, the above scores are normalized by Softmax to obtain the final weight coefficient, which is recorded as: ; This normalization process ensures that the sum of the weights of various features is 1, forming a probability distribution; Step 4.3: According to the weight coefficient output by the attention mechanism, perform weighted summation on the original three types of features to construct a unified fusion feature sequence, which is recorded as: ; After fusion It not only retains the effective information of various features, but also highlights the feature categories that contribute most to the prediction through the attention mechanism, enhancing the model's ability to perceive environmental changes, holiday impacts and regional linkages.

[0023] Step 5: Combine the weighted fusion features with the historical visitor flow information of the scenic spot under study and input them into the long short-term memory network (LSTM). A temporal attention mechanism is introduced to highlight the contribution of key time points to the prediction results. The specific steps include: Step 5.1: weighted fusion feature sequence Study on the historical passenger flow sequence of scenic spots Splice to form a new input sequence , sent to the long short-term memory network to obtain the hidden state of each time step ; Step 5.2: For each time step Calculate its attention score for the final predicted target , which is achieved through a trainable attention function, and the formula is: ; Among them, W, v, and b are learnable parameters; Step 5.3: Attention scores for all time steps Normalize the attention weights using the Softmax function ; Step 5.4: Perform weighted summation of the hidden states of each time step according to the attention weight to obtain the aggregated context vector c, which can be expressed as: ; in, represents the hidden state vector at the t-th time step, is the attention weight corresponding to the t-th time step.

[0024] Step 6: Connect the fully connected layer to output the final passenger flow prediction result. The specific steps include: Step 6.1: Input the context vector c into a fully connected (FC) neural network layer, which is responsible for mapping the abstract hidden state to the actual prediction value space. The specific mapping process uses a linear transformation method, and the calculation formula is recorded as: ; in, and are the weight matrix and bias term of the fully connected layer respectively; Step 6.2: Since the input data is normalized and preprocessed before model training, the prediction results output by the model are It is actually a dimensionless relative value and needs to be restored to the actual passenger flow range through inverse normalization operation to restore the dimension of the original data. The calculation formula is: ; in, and are the maximum and minimum passenger flow in the training data respectively.

Claims

1. A method for predicting tourist flow in scenic spots based on multi-source information fusion, characterized in that: The following steps are involved: Step 1: Collect passenger flow, weather conditions, and holiday information from the scenic area management system, meteorological service platform, and calendar system to construct a multi-source heterogeneous dataset; Step 2: Preprocess and standardize the collected multi-source heterogeneous data to construct a standardized data set; Step 3: Use the autoencoder to extract features and reduce the dimension of the surrounding scenic spot passenger flow information, remove redundant information, and obtain a more expressive low-dimensional feature representation; Step 4: Introduce the feature-level attention mechanism to dynamically evaluate and calculate the importance weights of various features in the prediction task, and construct a unified feature sequence of weighted fusion; Step 5: The weighted fusion features are combined with the historical passenger flow information of the scenic spot under study and input into the long short-term memory network. The temporal attention mechanism is introduced to highlight the contribution of key time points to the prediction results. Step 6: Connect the fully connected layer to output the final passenger flow prediction result.

2. The method for predicting tourist flow in a scenic spot based on multi-source information fusion according to claim 1, characterized in that: In step 1, passenger flow, weather conditions, and holiday information are collected from the scenic area management system, meteorological service platform, and calendar system. The specific steps to construct a multi-source heterogeneous dataset include: Step 1.1: Collect hourly cumulative passenger flow information at the gates and exits of the target scenic area and adjacent scenic areas from July 1, 2022 to July 1, 2024; Step 1.2: Collect rainfall and temperature information from the meteorological service platform; Step 1.3: Call the holiday information query interface to obtain holiday information.

3. The method for predicting tourist flow in a scenic spot based on multi-source information fusion according to claim 1, characterized in that: In step 2, the collected multi-source heterogeneous data is preprocessed and standardized. The specific steps for constructing a standardized data set include: Step 2.1: Convert the timestamps of passenger flow, weather, and holiday data into a standard format and perform time alignment. Step 2.2: Standardize the indicator units in different data sources; Step 2.3: Process missing values ​​for the collected real-time passenger flow data, weather information, and holiday information. Passenger flow, temperature, and rainfall are filled using the mean filling method, while holiday information is filled using the forward filling method. Step 2.4: Normalize the multi-source data. For the scenic spot passenger flow data, the Min-Max normalization method is used to map the passenger flow data to the interval [0, 1]. The normalization formula is: ; Among them, x represents the current passenger flow, and Represent the maximum and minimum passenger flow respectively. For temperature, the Z-score standardization method is used for processing. The formula is: ； Where x is the current temperature, is the mean temperature of the region, is the standard deviation of the temperature in the area. The rainfall information is first logarithmically normalized to compress the right-skewed distribution of the data, and then Z-score normalization is performed. The logarithmic transformation formula is recorded as: ； Among them, x is the current rainfall, c is a constant, to avoid In the case of holiday information, binary encoding is used for normalization. The encoding rules are as follows: ； in, Represents the normalized holiday data.

4. The method for predicting tourist flow in a scenic spot based on multi-source information fusion according to claim 1, characterized in that: In step 3, the autoencoder is used to extract features and reduce the dimension of the surrounding scenic spot passenger flow information, remove redundant information, and obtain a more expressive low-dimensional feature representation. The specific steps include: Step 3.1: Arrange the acquired passenger flow data of surrounding scenic spots in chronological order to construct the original input sequence; Step 3.2: Build an autoencoder model. The encoder part consists of a multi-layer fully connected neural network, which is used to compress the input high-dimensional time series data into a low-dimensional latent feature representation. The decoder is responsible for reconstructing the original input from the low-dimensional features to achieve self-supervised learning. Step 3.3: Use layer-by-layer unsupervised pre-training to train the autoencoder model. Optimize the parameters of the encoder and decoder to minimize the reconstruction error as the training objective. The loss function is expressed as: ； in, is the original input data sample, is the reconstruction result, T is the total number of samples; Step 3.4: Input the passenger flow data of the surrounding scenic spots into the trained autoencoder, and extract the corresponding low-dimensional feature sequence through the encoder, which is recorded as: ； Among them, T is the length of the time series, that is, the number of time steps of the passenger flow data, and Z is the feature representation of the entire passenger flow time series in low-dimensional space, which can be used as input for subsequent prediction or analysis.

5. The method for predicting tourist flow in a scenic spot based on multi-source information fusion according to claim 1, characterized in that: In step 4, the feature-level attention mechanism is introduced to dynamically evaluate and calculate the importance weights of various features in the prediction task. The specific steps for constructing a weighted fusion unified feature sequence include: Step 4.1, weather information rainfall ,temperature , Holiday Information Passenger flow characteristics of surrounding scenic spots Projected to a unified latent space through linear transformation, it is convenient for subsequent weighted fusion operations. Taking rainfall as an example, the embedding expression formula is recorded as: ； in, is the weight matrix, is the bias term, is the embedding representation of weather features; Step 4.2: Introduce the learnable attention query vector v and calculate the attention score through the dot product operation. Taking weather features as an example, the score calculation formula is recorded as: ； in, is the transpose of the attention vector, which is used to score the input features. The larger it is, the more important the feature is. Similarly, we can calculate 、 、 , in order to facilitate the comparison of the relative importance of different features, the above scores are normalized by Softmax to obtain the final weight coefficient, which is recorded as: ； This normalization process ensures that the sum of the weights of various features is 1, forming a probability distribution; Step 4.3: According to the weight coefficient output by the attention mechanism, perform weighted summation on the original three types of features to construct a unified fusion feature sequence, which is recorded as: ； After fusion It not only retains the effective information of various features, but also highlights the feature categories that contribute most to the prediction through the attention mechanism, enhancing the model's ability to perceive environmental changes, holiday impacts, and regional linkages.

6. The method for predicting tourist flow in a scenic spot based on multi-source information fusion according to claim 1, characterized in that: In step 5, the weighted fusion features are spliced ​​with the historical passenger flow information of the study scenic spot and input into the long short-term memory network. The temporal attention mechanism is introduced to highlight the contribution of key time points to the prediction results. The specific steps include: Step 5.1: weighted fusion feature sequence Study on the historical passenger flow sequence of scenic spots Splice to form a new input sequence , sent to the long short-term memory network to obtain the hidden state of each time step ; Step 5.2: For each time step Calculate its attention score for the final predicted target , which is achieved through a trainable attention function, and the formula is: ； Among them, W, v, and b are learnable parameters; Step 5.3: Attention scores for all time steps Normalize the attention weights using the Softmax function ; Step 5.4: Perform weighted summation of the hidden states of each time step according to the attention weight to obtain the aggregated context vector c, which can be expressed as: ； in, represents the hidden state vector at the t-th time step, is the attention weight corresponding to the t-th time step.

7. The method for predicting tourist flow in a scenic spot based on multi-source information fusion according to claim 1, characterized in that: In step 6, the specific steps of connecting the fully connected layer to output the final passenger flow prediction result include: Step 6.1: Input the context vector c into a fully connected neural network layer, which is responsible for mapping the abstract hidden state to the actual prediction value space. The formula is: ； in, and are the weight matrix and bias term of the fully connected layer respectively; Step 6.2: Since the input data is normalized and preprocessed before model training, the prediction results output by the model are It is actually a dimensionless relative value and needs to be restored to the actual passenger flow range through inverse normalization operation to restore the dimension of the original data. The calculation formula is: ； in, and are the maximum and minimum passenger flow in the training data respectively.

Images