Intelligent scenic spot multi-dimensional data monitoring and analysis method based on improved deep learning

Abstract

The application discloses a kind of based on improved deep learning's wisdom scenic spot multidimensional data monitoring and analysis method, S1, constructs multidimensional dataset;S2, to the original multidimensional dataset collected is washed;S3, unified data format and unit;S4, construct the analysis model for multidimensional dataset;S5, utilize analysis model to the real-time data in scenic spot is parsed, identifies abnormal condition and promptly alarms;S6, through the analysis of historical data, predict the tourist flow and environmental change trend in future period;S7, provide the real-time monitoring function of various data of scenic spot;S8, when system detects abnormal condition, automatically trigger alarm mechanism;S9, according to data analysis and analysis result, provide decision support for scenic spot manager;S10, based on the analysis of tourist behavior and preference, provide intelligent navigation and real-time information push service.The application realizes the efficient management and accurate monitoring of wisdom scenic spot, with significant technical advantages and application value.

Description

Technical Field

[0001] The present invention relates to the technical field of smart scenic spots, and in particular to a method for monitoring and analyzing multidimensional data of smart scenic spots based on improved deep learning. Background Art

[0002] With the rapid development of the tourism industry, the concept of smart scenic spots has gradually become an important means of improving scenic spot management efficiency and visitor experience. The collection, integration, and analysis of multidimensional data are key technologies in the development of smart scenic spots. In traditional scenic spot management, multidimensional data collection often relies on independent sensor systems. The inconsistent data formats between these systems make data integration difficult and prevent them from accurately reflecting the overall scenic spot situation in real time. Furthermore, traditional data collection systems suffer from low data processing efficiency, failing to meet the needs of real-time monitoring and analysis.

[0003] Existing multidimensional data collection technologies primarily rely on a single or limited number of sensor types, such as visitor flow sensors or environmental monitoring sensors. This approach suffers from the following drawbacks: First, a single sensor type cannot fully reflect the multidimensional state of a scenic area, resulting in incomplete monitoring data; second, data formats and communication protocols for different sensor types are not standardized, increasing the difficulty of data integration; finally, traditional data collection and transmission systems are prone to delays and data loss when dealing with large amounts of data, making them unable to meet the stringent real-time data monitoring requirements of smart scenic areas.

[0004] In terms of data analysis, while existing deep learning models have performed well in areas such as image recognition and natural language processing, they still have many shortcomings when processing complex, multidimensional data. When applied in complex environments like scenic spots, traditional deep learning models often face the following problems: First, model training is slow, especially when dealing with large amounts of multidimensional data, which significantly increases training time. Second, the analysis accuracy is low, making it difficult for traditional models to effectively identify anomalies or make accurate predictions, especially in scenic spots with high data noise and complex environments, where the model's robustness is poor. Third, the model is easily affected by external interference, such as sudden increases in tourist numbers and weather changes, which can lead to unstable analysis results.

[0005] Existing technologies also suffer from insufficient data preprocessing capabilities. Multidimensional data from scenic areas often contains a large amount of noise and invalid data. Traditional data cleaning and standardization methods are unable to completely remove this invalid information, affecting the accuracy of subsequent data analysis. Furthermore, standardizing data from different sensors is a major challenge. The lack of a unified and effective standardization method leads to inconsistent data formats and units, further complicating data integration and analysis. Summary of the Invention

[0006] One purpose of the present invention is to propose a method for multi-dimensional data monitoring and analysis of smart scenic spots based on improved deep learning. The present invention realizes efficient management and precise monitoring of smart scenic spots, and has significant technical advantages and application value.

[0007] According to an embodiment of the present invention, a method for monitoring and analyzing multidimensional data of a smart scenic spot based on improved deep learning includes the following steps:

[0008] S1. Deploy a variety of IoT sensors in the scenic area to collect real-time information on tourist flow, environmental parameters, and facility usage, and construct a multidimensional dataset.

[0009] S2, cleaning the collected original multidimensional data set to remove noise and invalid data;

[0010] S3. Standardize multidimensional datasets from different sources and unify data formats and units;

[0011] S4. Build an analytical model for multidimensional datasets based on improved deep learning algorithms;

[0012] S5. Use analytical models to analyze real-time data within the scenic area, identify abnormal situations and issue timely alarms;

[0013] S6. Predict tourist flow and environmental change trends in the future through analysis of historical data;

[0014] S7, provides real-time monitoring of various data of scenic spots, and managers can view the operation status of scenic spots in real time through a visual interface;

[0015] S8. When the system detects an abnormal situation, it automatically triggers the alarm mechanism and sends a notification to the relevant management personnel;

[0016] S9. Provide decision support to scenic area managers based on data parsing and analysis results;

[0017] S10. Provide intelligent navigation and real-time information push services based on analysis of tourist behavior and preferences.

[0018] Optionally, S1 includes the following sub-steps:

[0019] S11. Deploy a variety of IoT sensors at key locations within the scenic area, including visitor flow sensors, environmental parameter sensors, and facility usage sensors;

[0020] S12. Real-time collection of tourist flow data D t , environmental parameter data E t , facility usage data F t ;

[0021] S13, the tourist flow data D t including the number of tourists entering the scenic spot N in (t) out (t):

[0022] D t = {N in (t), N out (t)};

[0023] S14, the environmental parameter data E t including temperature T(t), humidity H(t), air quality index AQI(t):

[0024] E t = {T(t), H(t), AQI(t)};

[0025] S15, the facility usage data F t including the usage state S i (t) and the failure record R i (t) of the facility:

[0026] F t = {S i (t), R i (t)};

[0027] S16, the collected tourist flow data D t , environmental parameter data E t , and facility usage data F t are transmitted to the central processing system through a wireless network or a wired network;

[0028] S17, the central processing system receives and stores the data, and constructs a multidimensional data set MDS:

[0029] MDS = {D t , E t , F t}.

[0030] Optionally, the S4 comprises the following sub-steps:

[0031] S41, collecting and pre-processing the multidimensional data set MDS, including data cleaning and standardization processing;

[0032] MDScl ean = {Clean(D t ), Clean(E t ), Clean(F t )};

[0033] MDS std={Std(D t ), Std(E t ), Std(F t )};

[0034] Among them, Clean(·) represents the cleaning process to remove noise and invalid data, and Std(·) represents the standardization process to unify the data format and unit;

[0035] S42. Design and improve deep learning algorithms, including optimizing neural network structures, loss functions, and training methods;

[0036] S43, inputting the preprocessed multidimensional dataset into the improved deep learning model for training, and constructing an analytical model AM for parsing the multidimensional dataset;

[0037] S44. Improve the deep learning model by combining graph convolutional networks and dynamic Bayesian networks to optimize the parsing accuracy and data processing capabilities of the analytical model AM.

[0038] H gcn =GCN(MDS std ;W gcn );

[0039] S45. The graph convolutional network captures the spatial and temporal dependencies in the scenic area data, and defines the graph convolution layer parameters as W gcn , the output feature is represented as H gcn :

[0040]

[0041] in, represents the adjacency matrix with self-loops added, yes The degree matrix of is the input feature matrix of the first layer, σ is the activation function; A represents the adjacency matrix, and I represents the identity matrix;

[0042] S46, dynamic Bayesian network modeling data time-varying characteristics and potential structure, define the Bayesian network node and edge weight matrices as W node and W edge , the output potential is represented as z:

[0043]

[0044] Among them, z i The ith node represents the latent variable, Pa(z i ) represents node z i The parent node set of

[0045]

[0046] in, represents the mean and standard deviation of the i-th node;

[0047] S47, through the fusion layer, the output of the graph convolutional network and the dynamic Bayesian network are fused to obtain the comprehensive feature representation H fuse :

[0048] H fuse =γH gcn +δz;

[0049] Among them, γ and δ represent fusion weights;

[0050] S48, optimizing the model parameters through back propagation algorithm and gradient descent to obtain the final analytical model AM;

[0051]

[0052] in, represents the basic loss function, Y represents the actual label, λ represents the regularization parameter, α and β represent the weights of the additional loss terms; Risk(H fuse ,S safety ) represents the analytical model’s assessment of scenic area safety risks, Satisfaction (H fuse , S user ) represents the evaluation of tourist satisfaction by the analytical model.

[0053] Optionally, the S48 includes the following sub-steps:

[0054] S481. Define the basic loss function Among them H fuse is the fusion feature representation, Y is the actual label, and the loss function is used to measure the difference between the predicted value and the actual value:

[0055]

[0056] Where N is the number of samples, Loss(·) is the loss calculation function;

[0057] S482. Define regularization terms to avoid model overfitting, including weight regularization terms for graph convolutional networks and dynamic Bayesian networks:

[0058] Reg(W gcn , W node )=λ(||W gcn || 2 +||W node || 2 );

[0059] wherein, λ is a regularization parameter, ||W gcn || 2 and ||W node || 2 are the two-norm of the weight matrix, respectively;

[0060] S483, defining an additional loss term for evaluating the security risk and user satisfaction of the analytical model:

[0061]

[0062] wherein, α1 and β1 are the weights of the additional loss term, S safety and S user are the evaluation criteria of the security risk and user satisfaction, respectively;

[0063] S484, constructing a comprehensive loss function by fusing the basic loss, the regularization term and the additional loss term;

[0064] S485, calculating the gradient of the loss function to the model parameters by a back propagation algorithm:

[0065]

[0066] S486, updating the model parameters by a gradient descent optimization algorithm:

[0067]

[0068] wherein, η represents the learning rate, and the parameters of the analytical model AM are optimized by repeated iteration until the loss function converges.

[0069] Optionally, the S5 comprises the following sub-steps:

[0070] S51, inputting the real-time data MDS real in the scenic area into the constructed analytical model AM;

[0071] S52, processing the real-time data by the analytical model AM to extract the feature representation H real :

[0072] H real = AM(MDS real );

[0073] S53, comparing the extracted feature representation H real with the pre-set threshold or reference value, defining the threshold as T threshold,i and T threshold , j and if any element h real in H i satisfies the following conditions:

[0074] h i >T threshold,i or h i <T threshold,j ;

[0075] It is considered an abnormal situation;

[0076] S54, classifying abnormal situations and identifying specific abnormal types, including tourist congestion, facility failure, abnormal environmental parameters, etc.;

[0077] S55. Using the preset alarm mechanism, according to the identified abnormality type, an alarm signal is generated and the relevant management personnel are notified in a timely manner;

[0078] S56. The alarm signal includes the abnormality type, occurrence time, specific location and related data information. The alarm signal is defined as:

[0079] A signal ={Type: T type , Time: T time , Location: L location ,Data:D data};

[0080] S57, sending the alarm signal to the scenic area management system via a wireless network or a wired network, and recording it in the system log;

[0081] S58. According to the alarm signal, the status information in the scenic area management system is updated in real time and displayed on the visual interface, so that the management personnel can check and handle abnormal situations in time.

[0082] Optionally, the S52 includes the following sub-steps:

[0083] S521, real-time data MDS real Standardization is done to make it conform to the input requirements of the analytical model AM. The standardized data is represented as MDS. real_std ;

[0084] S522, the standardized real-time data MDS real_std Input to the input layer of the analytical model AM, and define the input layer weight as W in and bias b in , calculate the output of the input layer as:

[0085]

[0086] Among them, σ represents the nonlinear activation function, n is the number of input layer nodes;

[0087] S523, input layer output H inFeature extraction is performed through multiple hidden layers, and the weight and bias of each hidden layer are W l and b l , the output of the lth layer is represented as H l :

[0088]

[0089] Among them, φ represents the nonlinear activation function, m l-1 is the number of nodes in the l-1th layer;

[0090] S524, the output H of the last hidden layer l Input to the output layer, define the output layer weight as W out and bias b out , calculate the feature representation H of the output layer real :

[0091]

[0092] S525. In each layer of calculation, apply the optimized back propagation algorithm to calculate the loss function based on the actual label Y And adjust the weights and biases according to the gradient descent algorithm:

[0093]

[0094] S526, through multiple iterations of training and optimization, the output feature representation H of the analytical model AM is finally obtained real , used for subsequent abnormal situation identification and alarm.

[0095] Optionally, the S53 includes the following sub-steps:

[0096] S531, the extracted features are represented by H real With the pre-set threshold vector T threshold Compare item by item and define each feature representation h i and threshold T threshold,i ;

[0097] S532, if the feature represents h i Exceeds the pre-set upper threshold or below the lower threshold It is considered an abnormal situation if:

[0098] or

[0099] S533, for each feature h i Classify the abnormal situations, identify the specific abnormal types, and define the abnormal type set as E type, including the following categories:

[0100] E type ={tourist crowding, facility failure, abnormal environmental parameters};

[0101] S534. Set the corresponding alarm level A for each abnormal type level , divided into multiple levels according to the severity and impact range of the abnormality, and the alarm level set is defined as A level , including the following levels:

[0102] A level ={Level 1 alarm, Level 2 alarm, Level 3 alarm};

[0103] S535. Generate an alarm signal based on the identified abnormality type and alarm level. The alarm signal is defined as: A signal ={Type:E type , Level: A level , Time: T time , Location: L location ,Data:D data};

[0104] S536: When generating an alarm signal, record the specific time T when the abnormality occurs. time Specific location L location and related data information D data .

[0105] The beneficial effects of the present invention are:

[0106] (1) The present invention adopts an improved deep learning algorithm, combined with graph convolutional networks and dynamic Bayesian networks, to improve the training speed and analysis accuracy of the analytical model. The combination of the two significantly enhances the robustness of the analytical model in complex environments. By analyzing the real-time data in the scenic area through the constructed analytical model, abnormal situations can be identified in a timely manner and an alarm signal can be generated. The alarm signal includes the abnormality type, occurrence time, specific location and related data information to ensure the accuracy and timeliness of the alarm.

[0107] (2) During the model training process, the present invention constructs a comprehensive loss function, including basic loss, regularization term, and additional loss term. The basic loss function is used to measure the difference between the predicted value and the actual value, the regularization term prevents the model from overfitting, and the additional loss term evaluates the safety risk and user satisfaction of the parsing model. Through the optimized backpropagation algorithm and gradient descent algorithm, the loss function is calculated according to the actual label, and the weights and biases are adjusted to ensure that the model is continuously optimized during multiple iterative training. The application of this comprehensive loss function enables the model to not only accurately parse data, but also effectively evaluate the safety risks and user satisfaction of scenic spots, providing more comprehensive decision support. BRIEF DESCRIPTION OF THE DRAWINGS

[0108] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0109] Figure 1 This is a flowchart of a method for monitoring and analyzing multi-dimensional data of a smart scenic area based on improved deep learning proposed by the present invention;

[0110] Figure 2 This is a structural schematic diagram of an improved deep learning model for a smart scenic area multi-dimensional data monitoring and analysis method based on improved deep learning proposed in the present invention. DETAILED DESCRIPTION

[0111] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0112] refer to Figure 1 , a smart scenic spot multi-dimensional data monitoring and analysis method based on improved deep learning, including the following steps:

[0113] S1. Deploy a variety of IoT sensors in the scenic area to collect real-time information on tourist flow, environmental parameters, and facility usage, and construct a multidimensional dataset.

[0114] S2, cleaning the collected original multidimensional data set to remove noise and invalid data;

[0115] S3. Standardize multidimensional datasets from different sources and unify data formats and units;

[0116] S4. Build an analytical model for multidimensional datasets based on improved deep learning algorithms;

[0117] S5. Use analytical models to analyze real-time data within the scenic area, identify abnormal situations and issue timely alarms;

[0118] S6. Predict tourist flow and environmental change trends in the future through analysis of historical data;

[0119] S7, provides real-time monitoring of various data of scenic spots, and managers can view the operation status of scenic spots in real time through a visual interface;

[0120] S8. When the system detects an abnormal situation, it automatically triggers the alarm mechanism and sends a notification to the relevant management personnel;

[0121] S9. Provide decision support to scenic area managers based on data parsing and analysis results;

[0122] S10. Provide intelligent navigation and real-time information push services based on analysis of tourist behavior and preferences.

[0123] In this embodiment, S1 includes the following sub-steps:

[0124] S11. Deploy a variety of IoT sensors at key locations within the scenic area, including visitor flow sensors, environmental parameter sensors, and facility usage sensors;

[0125] S12. Real-time collection of tourist flow data D t , environmental parameter data E t , facility usage data F t ;

[0126] S13. Tourist flow data D t Including the number of tourists entering the scenic area N in (t) and the number of tourists leaving the scenic area N out (t):

[0127] D t ={N in (t), N out (t)};

[0128] S14, Environmental parameter data E t Including temperature T(t), humidity H(t), air quality index AQI(t):

[0129] E t ={T(t), H(t), AQI(t)};

[0130] S15. Facility usage data F t Including the usage status S of the facility i (t) and the facility's fault record R i (t):

[0131] F t ={S i (t), R i (t)};

[0132] S16, collect the tourist flow data D through wireless network or wired network t , environmental parameter data E t , facility usage data F t Transmit to central processing system;

[0133] S17. The central processing system receives and stores data and constructs a multidimensional data set (MDS).

[0134] MDS={D t , E t , F t}.

[0135] In this embodiment, S4 includes the following sub-steps:

[0136] S41. Collect and preprocess multidimensional datasets (MDS), including data cleaning and standardization.

[0137] MDS clean ={Clean(D t ), Clean(E t ), Clean(F t )};

[0138] MDS std ={Std(D t ), Std(E t ), Std(F t )};

[0139] Among them, Clean(·) represents the cleaning process to remove noise and invalid data, and Std(·) represents the standardization process to unify the data format and unit;

[0140] S42. Design and improve deep learning algorithms, including optimizing neural network structures, loss functions, and training methods;

[0141] S43, inputting the preprocessed multidimensional dataset into the improved deep learning model for training, and constructing an analytical model AM for parsing the multidimensional dataset;

[0142] S44. Improve the deep learning model by combining graph convolutional networks and dynamic Bayesian networks to optimize the parsing accuracy and data processing capabilities of the analytical model AM.

[0143] H gcn =GCN(MDS std ;W gcn );

[0144] S45. The graph convolutional network captures the spatial and temporal dependencies in the scenic area data, and defines the graph convolution layer parameters as W gcn , the output feature is represented as H gcn :

[0145]

[0146] in, represents the adjacency matrix with self-loops added, yes The degree matrix of is the input feature matrix of the first layer, and σ is an activation function; A represents an adjacency matrix, and I represents an identity matrix;

[0147] S46, dynamically modeling the time-varying characteristics and potential structure of the data, defining the weight matrix of the Bayesian network nodes and edges as W node and W edge , and the output potential representation is z:

[0148]

[0149] wherein z i represents the i-th node of the latent variable, Pa(z i ) represents the parent node set of the node z i ;

[0150]

[0151] wherein, represents the mean and standard deviation of the i-th node;

[0152] S47, fusing the outputs of the graph convolutional network and the dynamic Bayesian network through a fusion layer to obtain a comprehensive feature representation H fuse :

[0153] H fuse =γH gcn +δz;

[0154] wherein γ and δ represent fusion weights;

[0155] S48, optimizing the model parameters through a back propagation algorithm and gradient descent to obtain a final analytical model AM;

[0156]

[0157] wherein, represents a basic loss function, Y represents an actual label, λ represents a regularization parameter, and α and β represent weights of additional loss terms; Risk(H fuse , S safety ) represents an evaluation of the analytical model on scenic spot safety risk, and Satisfaction(H fuse , S user ) represents an evaluation of the analytical model on tourist satisfaction.

[0158] In the embodiment, S48 includes the following sub-steps:

[0159] S481, defining a basic loss function wherein H fuseis the fusion feature representation, Y is the actual label, and the loss function is used to measure the difference between the predicted value and the actual value:

[0160]

[0161] Where N is the number of samples, Loss(·) is the loss calculation function;

[0162] S482. Define regularization terms to avoid model overfitting, including weight regularization terms for graph convolutional networks and dynamic Bayesian networks:

[0163] Reg(W gcn , W node )=λ(||W gcn || 2 +||W node || 2 );

[0164] Among them, λ is the regularization parameter, ||W gcn || 2 and ||v node || 2 are the two norms of the weight matrix respectively;

[0165] S483. Define additional loss terms to evaluate the security risk and user satisfaction of the analytical model:

[0166]

[0167] Among them, α1 and β1 are the weights of the additional loss terms, S safety and S user These are the evaluation criteria for security risks and user satisfaction, respectively;

[0168] S484. Construct a comprehensive loss function that integrates basic loss, regularization term, and additional loss term;

[0169] S485. Calculate the gradient of the loss function with respect to the model parameters through the back propagation algorithm:

[0170]

[0171] S486. Update model parameters using gradient descent optimization algorithm:

[0172]

[0173] Where η represents the learning rate, and the parameters of the analytical model AM are optimized through repeated iterations until the loss function converges.

[0174] In this embodiment, S5 includes the following sub-steps:

[0175] S51, real-time data MDS in the scenic area real Input into the constructed analytical model AM;

[0176] S52, the analytical model AM processes the real-time data and extracts the feature representation H real :

[0177] H real =AM(MDS real );

[0178] S53, the extracted features are represented by H real Compare with the preset threshold or reference value, and define the threshold as T threshold,i and T threshold ,j, if H real Any element h in i The following conditions are met:

[0179] h i >T threshold,i or h i <T threshold,j ;

[0180] It is considered an abnormal situation;

[0181] S54, classifying abnormal situations and identifying specific abnormal types, including tourist congestion, facility failure, abnormal environmental parameters, etc.;

[0182] S55. Using the preset alarm mechanism, according to the identified abnormality type, an alarm signal is generated and the relevant management personnel are notified in a timely manner;

[0183] S56. The alarm signal includes the abnormality type, occurrence time, specific location and related data information. The alarm signal is defined as:

[0184] A signal ={Type: T type , Time: T time , Location: L location ,Data:D data};

[0185] S57, sending the alarm signal to the scenic area management system via a wireless network or a wired network, and recording it in the system log;

[0186] S58. According to the alarm signal, the status information in the scenic area management system is updated in real time and displayed on the visual interface, so that the management personnel can check and handle abnormal situations in time.

[0187] In this embodiment, S52 includes the following sub-steps:

[0188] S521, real-time data MDS real Standardization is done to make it conform to the input requirements of the analytical model AM. The standardized data is represented as MDS. real_std ;

[0189] S522, the standardized real-time data MDS real_std Input to the input layer of the analytical model AM, and define the input layer weight as W in and bias b in , calculate the output of the input layer as:

[0190]

[0191] Among them, σ represents the nonlinear activation function, n is the number of input layer nodes;

[0192] S523, input layer output H in Feature extraction is performed through multiple hidden layers, and the weight and bias of each hidden layer are W l and b l , the output of the first layer is represented as H l :

[0193]

[0194] Among them, φ represents the nonlinear activation function, m l-1 is the number of nodes in the l-1th layer;

[0195] S524, the output H of the last hidden layer l Input to the output layer, define the output layer weight as W out and bias b out , calculate the feature representation H of the output layer real :

[0196]

[0197] S525. In each layer of calculation, apply the optimized back propagation algorithm to calculate the loss function based on the actual label Y And adjust the weights and biases according to the gradient descent algorithm:

[0198]

[0199]

[0200] S526, through multiple iterations of training and optimization, the output feature representation H of the analytical model AM is finally obtained real , used for subsequent abnormal situation identification and alarm.

[0201] In this embodiment, S53 includes the following sub-steps:

[0202] S531, the extracted features are represented by H real With the pre-set threshold vector T threshold Compare item by item and define each feature representation h i and threshold T threshold,i ;

[0203] S532, if the feature represents h i Exceeds the pre-set upper threshold or below the lower threshold It is considered an abnormal situation if:

[0204] or

[0205] S533, for each feature h i Classify the abnormal situations, identify the specific abnormal types, and define the abnormal type set as E type , including the following categories:

[0206] E type ={tourist crowding, facility failure, abnormal environmental parameters};

[0207] S534. Set the corresponding alarm level A for each abnormal type level , divided into multiple levels according to the severity and impact range of the abnormality, and the alarm level set is defined as A level , including the following levels:

[0208] A level ={Level 1 alarm, Level 2 alarm, Level 3 alarm};

[0209] S535. Generate an alarm signal based on the identified abnormality type and alarm level. The alarm signal is defined as:

[0210] A signal ={Type:E type , Level: A level , Time: T time , Location: L location ,Data:D data};

[0211] S536: When generating an alarm signal, record the specific time T when the abnormality occurs. time Specific location L location and related data information D data .

[0212] Example 1:

[0213] To validate the effectiveness of our improved deep learning-based multidimensional data monitoring and analysis method for smart scenic spots, we selected a large tourist attraction in XX City as an experimental setting. This vast scenic area welcomes millions of visitors annually and features numerous key attractions, service facilities, and rest areas. Due to the large number and widespread distribution of visitors, the scenic area presents significant challenges in management and monitoring, necessitating an efficient and accurate monitoring and analysis method.

[0214] A variety of IoT sensors have been deployed throughout the scenic area, including sensors for visitor flow, environmental parameters, and facility usage. These sensors are installed near the main entrance, key attractions, rest areas, and service facilities to ensure real-time and comprehensive data collection on visitor flow, temperature, humidity, air quality index, and facility usage.

[0215] Over a weekend, the number of visitors to a scenic area increased significantly. Traditional monitoring methods, which rely primarily on manual inspections and data from a single sensor, cannot accurately and timely reflect the overall situation of the scenic area. Especially during peak visitor periods, data processing efficiency is low, making it difficult to detect and address anomalies in a timely manner. The method proposed in this paper efficiently integrates data from multiple sensors, monitors multidimensional data within the scenic area in real time, and accurately analyzes it using an improved deep learning algorithm.

[0216] The improved deep learning algorithm of this invention combines a graph convolutional network (GCN) with a dynamic Bayesian network (DBN). The GCN is used to capture spatial and temporal dependencies in the data, while the DNB is used to model the time-varying characteristics and underlying structure of the data. This combination enables analytical models with greater accuracy and robustness when processing complex, multidimensional data.

[0217] During specific experiments, the method of the present invention successfully identified multiple anomalies. For example, in a densely populated area, the system detected a sudden increase in visitor traffic, exceeding a preset safety threshold. The system immediately generated an alarm signal and notified scenic area management personnel to promptly direct visitors, avoiding a potential stampede. In another example, near a tourist facility, environmental sensors detected a sharp increase in the air quality index. The system identified this anomaly and quickly notified relevant departments for investigation and resolution, ensuring the health and safety of tourists.

[0218] In order to evaluate the effectiveness of the method of the present invention, the application effects of the traditional method and the method of the present invention in the same scenic area were compared. The following Table 1 shows the specific data comparison results:

[0219] Table 1 Comparison data between traditional method and the method of the present invention

[0220] Data Category Traditional methods Method of the present invention Tourist flow monitoring 50％ 95％ Environmental parameter monitoring 60％ 98％ Facility usage monitoring 55％ 97％ Abnormal situation recognition rate 40％ 92％ Data processing efficiency 70％ 98％ Safety accident rate 5 per month 1 per month

[0221] As can be seen from Table 1, traditional methods have significant shortcomings in monitoring visitor flow, environmental parameters, and facility usage, and their recognition rate of abnormal situations is also low. In contrast, the method of the present invention demonstrates significant advantages in all aspects, especially in identifying and handling abnormal situations, achieving a recognition rate of 92%, significantly reducing the incidence of safety accidents in scenic areas.

[0222] Specifically, traditional methods for monitoring tourist flow rely primarily on manual statistics and data from a single sensor, resulting in limited coverage and untimely data updates. However, the method presented here uses real-time data from multiple sensors and analyzes it using an improved deep learning model, enabling accurate monitoring of tourist flow during each time period. Experimental data shows that the accuracy of tourist flow monitoring using this method has increased from 50% with traditional methods to 95%.

[0223] Traditional methods for monitoring environmental parameters primarily rely on a small number of environmental sensors, which limits data collection and fails to fully reflect the environmental conditions within a scenic area. The method proposed in this paper deploys multiple environmental sensors to collect real-time data such as temperature, humidity, and air quality index. Through data preprocessing and deep learning analysis, the accuracy of environmental parameter monitoring has been increased from 60% with traditional methods to 98%.

[0224] Traditional methods for monitoring facility usage rely primarily on inspection records kept by facility managers, resulting in infrequent data updates and a failure to reflect facility usage and fault conditions in real time. The proposed method uses facility usage sensors to collect real-time data and analyzes it using a deep learning model. This improves the accuracy of facility usage monitoring from 55% with traditional methods to 97%.

[0225] In terms of identifying abnormal situations, traditional methods have a low recognition rate for abnormal situations due to their single data source and low processing efficiency, often failing to detect and address abnormal situations in a timely manner. By adopting the method of the present invention, through real-time monitoring of multidimensional data and precise analysis of deep learning models, the recognition rate of abnormal situations has increased from 40% with traditional methods to 92%. For example, in one experiment, the system successfully identified multiple abnormal situations, such as crowded tourists and abnormal air quality, and promptly generated alarm signals to notify relevant departments for processing, effectively avoiding the occurrence of safety accidents.

[0226] The method of the present invention significantly outperforms traditional methods in terms of data processing efficiency. Traditional methods, due to the decentralized nature of data collection and processing, suffer from low processing efficiency and are prone to delays and data loss. However, the method of the present invention, through efficient data collection and integration combined with improved deep learning algorithms, enables real-time monitoring and analysis of multidimensional data, increasing data processing efficiency from 70% to 98% compared to traditional methods.

[0227] Through the above experiments and data comparisons, it can be seen that the multi-dimensional data monitoring and analysis method of smart scenic spots based on improved deep learning of the present invention has significantly improved the management efficiency of scenic spots and the tourist experience, and solved the problems of inaccurate multi-dimensional data collection and integration and poor application effect of deep learning models in complex environments existing in traditional methods. It has broad application prospects and practical value.

[0228] The present invention adopts an improved deep learning algorithm, combined with graph convolutional networks and dynamic Bayesian networks, to improve the training speed and analysis accuracy of the analytical model. The combination of the two significantly enhances the robustness of the analytical model in complex environments. By analyzing the real-time data in the scenic area through the constructed analytical model, abnormal situations can be identified in time and alarm signals can be generated. The alarm signal includes the abnormality type, occurrence time, specific location and related data information to ensure the accuracy and timeliness of the alarm.

[0229] During model training, the present invention constructs a comprehensive loss function, comprising a basic loss, a regularization term, and an additional loss term. The basic loss function measures the difference between predicted and actual values, the regularization term prevents model overfitting, and the additional loss term assesses the safety risk and user satisfaction of the parsing model. Using an optimized backpropagation and gradient descent algorithm, the loss function is calculated based on the actual labels, and weights and biases are adjusted to ensure continuous model optimization over multiple iterations of training. The application of this comprehensive loss function enables the model to not only accurately parse data but also effectively assess the safety risks and user satisfaction of scenic spots, providing more comprehensive decision support.

[0230] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for monitoring and analyzing multidimensional data of smart scenic spots based on improved deep learning, characterized in that: The steps include: S1. Deploy a variety of IoT sensors in the scenic area to collect real-time information on tourist flow, environmental parameters, and facility usage, and construct a multidimensional dataset. S2, cleaning the collected original multidimensional data set to remove noise and invalid data; S3. Standardize multidimensional datasets from different sources and unify data formats and units; S4. Build an analytical model for multidimensional datasets based on improved deep learning algorithms; The S4 includes the following sub-steps: S41. Collect and preprocess multidimensional datasets (MDS), including data cleaning and standardization. MDS clean ={Clean(D t ),Clean(E t ),Clean(F t )}; MDS std ={Std(D t ),Std(E t ),Std(F t )}; Among them, Clean(·) represents the cleaning process to remove noise and invalid data, and Std(·) represents the standardization process to unify the data format and unit; S42. Design and improve deep learning algorithms, including optimizing neural network structures, loss functions, and training methods; S43, inputting the preprocessed multidimensional dataset into the improved deep learning model for training, and constructing an analytical model AM for parsing the multidimensional dataset; S44. Improve the deep learning model by combining graph convolutional networks and dynamic Bayesian networks to optimize the parsing accuracy and data processing capabilities of the analytical model AM. H gcn =GCN(MDS std ;W gcn ); S45. The graph convolutional network captures the spatial and temporal dependencies in the scenic area data, and defines the graph convolution layer parameters as W gcn , the output feature is represented as H gcn : in, represents the adjacency matrix with self-loops added, yes The degree matrix of is the input feature matrix of the lth layer, σ is the activation function; A represents the adjacency matrix, and I represents the identity matrix; S46, dynamic Bayesian network modeling data time-varying characteristics and potential structure, define the Bayesian network node and edge weight matrices as W node and W edge , the output potential is represented as z: Among them, z i The ith node represents the latent variable, Pa(z i ) represents node z i The parent node set of in, represents the mean and standard deviation of the i-th node; S47, through the fusion layer, the output of the graph convolutional network and the dynamic Bayesian network are fused to obtain the comprehensive feature representation H fuse : H fuse =γH gcn +δz; Among them, γ and δ represent fusion weights; S48, optimizing the model parameters through back propagation algorithm and gradient descent to obtain the final analytical model AM; in, represents the basic loss function, Y represents the actual label, λ represents the regularization parameter, α and β represent the weights of the additional loss terms; Risk(H fuse ,S safety ) represents the analytical model’s assessment of scenic area safety risks, Satisfaction (H fuse ,S user ) represents the evaluation of tourist satisfaction by the analytical model; S5. Use analytical models to analyze real-time data within the scenic area, identify abnormal situations and issue timely alarms; The S5 comprises the following sub-steps: S51, real-time data MDS in the scenic area real Input into the constructed analytical model AM; S52, the analytical model AM processes the real-time data and extracts the feature representation H real : H real =AM(MDS real ); S53, the extracted features are represented by H real Compare with the preset threshold or reference value, and define the threshold as T threshold,i and T threshold,j , if H real Any element h in i The following conditions are met: h i >T threshold,i or h i <T threshold,j ; It is considered an abnormal situation; S54, classifying abnormal situations and identifying specific abnormality types, including tourist congestion, facility failure, and abnormal environmental parameters; S55. Using the preset alarm mechanism, according to the identified abnormality type, an alarm signal is generated and the relevant management personnel are notified in a timely manner; S56. The alarm signal includes the abnormality type, occurrence time, specific location and related data information. The alarm signal is defined as: A signal ={Type:T type ,Time:T time ,Location:L location ,Data:D data }; S57, sending the alarm signal to the scenic area management system via a wireless network or a wired network, and recording it in the system log; S58. According to the alarm signal, the status information in the scenic area management system is updated in real time and displayed on the visual interface, so that the management personnel can check and handle abnormal situations in time; S6. Predict tourist flow and environmental change trends in the future through analysis of historical data; S7, provides real-time monitoring of various data of scenic spots, and managers can view the operation status of scenic spots in real time through a visual interface; S8. When the system detects an abnormal situation, it automatically triggers the alarm mechanism and sends a notification to the relevant management personnel; S9. Provide decision support to scenic area managers based on data parsing and analysis results; S10. Provide intelligent navigation and real-time information push services based on analysis of tourist behavior and preferences.

2. The method for multi-dimensional data monitoring and analysis of a smart scenic spot based on improved deep learning according to claim 1 is characterized in that: The S1 includes the following sub-steps: S11. Deploy a variety of IoT sensors at key locations within the scenic area, including visitor flow sensors, environmental parameter sensors, and facility usage sensors; S12. Real-time collection of tourist flow data D t , environmental parameter data E t , facility usage data F t ; S13, the tourist flow data D t Including the number of tourists entering the scenic area N in (t) and the number of tourists leaving the scenic area N out (t): D t ={N in (t),N out (t)}; S14, the environmental parameter data E t Including temperature T(t), humidity H(t), air quality index AQI(t): E t ={T(t),H(t),AQI(t)}; S15, the facility usage data F t Including the usage status S of the facility i (t) and the facility's fault record R i (t): F t ={S i (t),R i (t)}; S16, collect the tourist flow data D through wireless network or wired network t , environmental parameter data E t , facility usage data F t Transmit to central processing system; S17. The central processing system receives and stores the data and constructs a multidimensional dataset MDS: MDS={D t ,E t ,F t }.

3. The method for monitoring and analyzing multi-dimensional data of a smart scenic spot based on improved deep learning according to claim 2 is characterized in that: The S48 includes the following sub-steps: S481. Define the basic loss function Among them H fuse is the fusion feature representation, Y is the actual label, and the loss function is used to measure the difference between the predicted value and the actual value: Where N is the number of samples, Loss(·) is the loss calculation function; S482. Define regularization terms to avoid model overfitting, including weight regularization terms for graph convolutional networks and dynamic Bayesian networks: Reg(W gcn ,IN node )=λ(||W gcn || 2 +||In node || 2 ); Among them, λ is the regularization parameter, ||W gcn || 2 and ||W node || 2 are the two norms of the weight matrix respectively; S483. Define additional loss terms to evaluate the security risk and user satisfaction of the analytical model: Among them, α1 and β1 are the weights of the additional loss terms, S safety and S user These are the evaluation criteria for security risks and user satisfaction, respectively; S484. Construct a comprehensive loss function that integrates basic loss, regularization term, and additional loss term; S485. Calculate the gradient of the loss function with respect to the model parameters through the back propagation algorithm: S486. Update model parameters using gradient descent optimization algorithm: Where η represents the learning rate, and the parameters of the analytical model AM are optimized through repeated iterations until the loss function converges.

4. The method for monitoring and analyzing multidimensional data of a smart scenic spot based on improved deep learning according to claim 1 is characterized in that: The S52 includes the following sub-steps: S521, real-time data MDS real Standardization is done to make it conform to the input requirements of the analytical model AM. The standardized data is represented as MDS. real_std ; S522, the standardized real-time data MDS real_std Input to the input layer of the analytical model AM, and define the input layer weight as W in and bias b in , calculate the output of the input layer as: Among them, σ represents the nonlinear activation function, n is the number of input layer nodes; S523, input layer output H in Feature extraction is performed through multiple hidden layers, and the weight and bias of each hidden layer are W l and b l , the output of the lth layer is represented as H l : Among them, φ represents the nonlinear activation function, m l-1 is the number of nodes in the l-1th layer; S524, the output H of the last hidden layer l Input to the output layer, define the output layer weight as W out and bias b out , calculate the feature representation H of the output layer real : S525. In each layer of calculation, apply the optimized back propagation algorithm to calculate the loss function based on the actual label Y And adjust the weights and biases according to the gradient descent algorithm: S526, through multiple iterations of training and optimization, the output feature representation H of the analytical model AM is finally obtained real , used for subsequent abnormal situation identification and alarm.

5. The method for monitoring and analyzing multi-dimensional data of a smart scenic spot based on improved deep learning according to claim 1 is characterized in that: The S53 includes the following sub-steps: S531, the extracted features are represented by H real With the pre-set threshold vector T threshold Compare item by item and define each feature representation h i and threshold T threshold,i ; S532, if the feature represents h i Exceeds the pre-set upper threshold or below the lower threshold It is considered an abnormal situation if: or S533, for each feature h i Classify the abnormal situations, identify the specific abnormal types, and define the abnormal type set as E type , including the following categories: E type ={tourist crowding, facility failure, abnormal environmental parameters}; S534. Set the corresponding alarm level A for each abnormal type level , divided into multiple levels according to the severity and impact range of the abnormality, and the alarm level set is defined as A level , including the following levels: A level ={Level 1 alarm, Level 2 alarm, Level 3 alarm}; S535. Generate an alarm signal based on the identified abnormality type and alarm level. The alarm signal is defined as: A signal ={Type:E type ,Level:A level ,Time:T time ,Location:L location ,Data:D data }; S536: When generating an alarm signal, record the specific time T when the abnormality occurs. time Specific location L location and related data information D data .

Images