Park personnel behavior early warning method and system based on big data and artificial intelligence

Abstract

The invention provides a park personnel behavior early warning method and system based on big data and artificial intelligence, and relates to the technical field of intelligent monitoring. The source domain data is used as target domain data, a deep learning network model is established in combination with the historical behavior data set of the park personnel, and the deep learning network model is trained; using the trained deep learning network model to identify and classify the behaviors of the personnel in the park collected in real time; an early-warning model is constructed through a big data technology, the early-warning model is used for analyzing behaviors, recognized in real time, of personnel in the park, abnormal behavior judgment and early warning are carried out, the early-warning behaviors judged manually and the early-warning behaviors judged by a system are labeled and then synchronized to a source domain, the early-warning threshold value and strategy can be dynamically adjusted, and the early-warning effect is improved. The intelligent level of safety management of the park is improved, scientific and accurate decision support is provided for park managers, potential risks are effectively reduced, and safe and stable operation of the park is guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of intelligent monitoring technology, and in particular to a method and system for early warning of park personnel behavior based on big data and artificial intelligence. Background Art

[0002] With the rapid development of society and the acceleration of urbanization, the park, as an important carrier of economic activities, has become increasingly prominent in terms of security management issues. Monitoring and early warning of personnel behavior in the park are crucial to ensuring the safe and stable operation of the park. However, traditional personnel monitoring methods have problems such as low efficiency and poor accuracy, which makes it difficult to meet the needs of modern park security management. The main shortcomings of existing park security management technologies include the following aspects:

[0003] 1. Imperfect safety management system: Multiple evidences point out that the safety management system of the park is imperfect and incomplete. For example, the safety management system of the chemical park is imperfect, the access mechanism is neglected, and the monitoring and management technology is backward. In addition, the safety management organization is not sound, and the safety management responsibility is not proportional to the power.

[0004] 2. Network security threats: The campus network faces a variety of security threats, such as the failure to adopt a secure network architecture and weak user awareness of virus prevention, which leads to network intrusion. In addition, the security risk assessment of the dedicated campus network system shows that there are problems in network security planning, implementation and management.

[0005] 3. Outdated technology and insufficient updates: The park's security management technology is outdated, and advanced monitoring devices and technologies have not been adopted in a timely manner. For example, the traditional large-scale park personnel access and vehicle traffic management model can no longer meet the needs of modern security management. In addition, although the park's intelligent security monitoring system has developed, there are still some problems that need to be solved.

[0006] 4. Lack of effective risk assessment and management: The safety and quality risk assessment system of chemical parks has defects, and it is necessary to deeply analyze its own problems and strengthen the effective control of safety risks. In addition, information security challenges are becoming increasingly prominent in industrial automation systems, and information security management needs to be strengthened.

[0007] 5. Insufficient personnel and training: Insufficient personnel training and education in safety management leads to inefficient safety management. For example, the main drawbacks in safety management include an incomplete safety management organization and a large gap between responsibility and power.

[0008] The main shortcomings of existing park security management technology include imperfect management system, network security threats, backward technology and insufficient updates, lack of effective risk assessment and management, and insufficient personnel and training. These problems need to be solved by strengthening management, adopting advanced technology, improving security awareness and strengthening personnel training. Summary of the invention

[0009] The purpose of the present invention is to propose a method and system for early warning of personnel behavior in a park based on big data and artificial intelligence, aiming to solve the current problem of lack of intelligent monitoring of personnel behavior in the park, so as to meet the needs of monitoring and early warning of personnel behavior in the park.

[0010] The present invention provides a park personnel behavior early warning method based on big data and artificial intelligence, comprising the following steps:

[0011] Collect open source behavioral data of other similar park personnel, build a source domain including source data, models and tasks, use the data, models and tasks in the source domain as the data, models and tasks of the current park target domain, and build a deep learning network model in combination with the historical behavioral data set of the current park personnel. Train the deep learning network model in combination with the data, models and tasks in the source domain and the historical behavioral data set of the current park personnel.

[0012] Based on the real-time collected video or image data of park personnel, the trained deep learning network model is used to identify and classify the behaviors of park personnel;

[0013] Build an early warning model through big data technology, use the early warning model to analyze the behavior of park personnel identified in real time, judge abnormal behavior and issue early warning;

[0014] The real-time collected video or image data in the target domain is synchronized to the source domain in real time, and the warning behaviors determined by humans and the warning behaviors determined by the system are annotated and synchronized to the source domain to realize dynamic adjustment of the warning thresholds and warning strategies in the warning model.

[0015] Preferably, using the data, models and tasks in the source domain as the data, models and tasks in the current park target domain, and combining the historical behavior data set of the current park personnel to establish a deep learning network model includes:

[0016] Use the data, models, and tasks in the source domain as the data, models, and tasks in the current park’s target domain;

[0017] Collect historical video or image data of human behavior in the park;

[0018] Combine the collected historical video or image data of people’s behavior in the park with the data in the source domain to construct a dataset;

[0019] The collected historical video or image data and the data in the source domain are cropped, scaled, and normalized for preprocessing, and a deep learning network model is established.

[0020] Preferably, the deep learning network model adopts a hybrid model of convolutional neural network and recurrent neural network;

[0021] Use convolutional neural networks to extract spatial features from the input dataset;

[0022] Use recurrent neural networks to process time series data and extract time features from the data set;

[0023] Use forward propagation to train the deep learning network model until the preset training end condition is reached;

[0024] Using the trained deep learning network model to identify and classify the behaviors of park personnel includes using the trained deep learning network model to identify and classify the behaviors of park personnel through forward propagation to obtain predicted classification results of the behaviors.

[0025] Preferably, the deep learning network model is optimized and parameters are adjusted by training the deep learning network model, specifically including:

[0026] Define the loss function;

[0027] Calculate the gradient using the back-propagation algorithm;

[0028] Use optimization algorithms to update the parameters of deep learning network models;

[0029] Repeat the above steps until the performance of the deep learning network model reaches the preset standard or the training reaches the maximum number of iterations.

[0030] Preferably, the loss function is defined as a cross entropy loss function, and the parameters of the deep learning network model are updated using the gradient descent method or the Adam optimizer; after the training is completed, the performance of the deep learning network model is evaluated on the test set, and the model parameters of the deep learning network model are adjusted according to the evaluation results.

[0031] Preferably, the analysis of the behaviors of park personnel identified in real time by big data technology specifically includes:

[0032] Collect data from multiple data sources and integrate them into a unified form and store them in a unified database;

[0033] Clean and process the collected data;

[0034] Discover data patterns, trends, and association rules from data based on data characteristics.

[0035] Preferably, cleaning and processing includes removing noisy data, filling missing values ​​and correcting errors, specifically including:

[0036] Identify and remove noisy data by setting thresholds or using statistical methods;

[0037] Fill missing values ​​using mean, median, or mode;

[0038] Check and correct errors based on preset data validation rules and logic.

[0039] Preferably, the mean, variance and correlation coefficient are used to discover patterns, trends and association rules of the data.

[0040] Preferably, data privacy and security protection is performed during the data collection, data storage, and data processing stages, including:

[0041] During the data collection and transmission stage, the type, quantity and source of data to be collected should be clearly defined, and the collected data should be encrypted for transmission;

[0042] During the data storage phase, data is classified and stored according to its sensitivity, ensuring that data with higher sensitivity receives a higher level of protection and that the stored data is encrypted;

[0043] Sensitive data that meets the preset sensitivity level is desensitized and processed using a privacy protection algorithm.

[0044] The present invention provides a park personnel behavior early warning system based on big data and artificial intelligence, including a processor capable of executing the steps of the above-mentioned park personnel behavior early warning method based on big data and artificial intelligence.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] 1. The present invention solves the problem of few park training samples by proposing to use the source domain data of other parks of the same type as the target domain data of the current park. At the same time, the real-time collected video or image data in the target domain is synchronized to the source domain in real time, and the warning behaviors determined by humans and the warning behaviors determined by the system are annotated and synchronized to the source domain, which can realize the dynamic adjustment of the warning threshold and warning strategy in the warning model.

[0047] 2. The park personnel behavior early warning method based on big data and artificial intelligence proposed in this invention realizes comprehensive and accurate monitoring and early warning of the behavior of personnel in the park by integrating deep learning park personnel behavior recognition technology with big data analysis and early warning technology. This innovative solution not only improves the level of intelligence in park safety management, but also provides scientific and accurate decision-making support for park managers, effectively reduces potential risks, and ensures the safe and stable operation of the park. The technical foresight and innovation of this mechanism are of great significance to promoting technological progress and industrial upgrading in the field of park safety management. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying creative work.

[0049] Figure 1 This is a flow chart of a method for early warning of park personnel behavior based on big data and artificial intelligence according to an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] The present invention provides a park personnel behavior early warning method based on big data and artificial intelligence, comprising the following steps:

[0052] By using the collected open source behavioral data of personnel in other similar parks (such as chemical parks, office buildings, shopping malls, etc.), we build a source domain including source data, models, and tasks. We use the data, models, and tasks in the source domain as the data, models, and tasks in the target domain of the current park. We build a deep learning network model based on the historical behavioral data set of the current park personnel. We train the deep learning network model based on the data, models, and tasks in the source domain and the historical behavioral data set of the current park personnel.

[0053] Based on the real-time collected video or image data of park personnel, the trained deep learning network model is used to identify and classify the behaviors of park personnel;

[0054] Build an early warning model through big data technology, use the early warning model to analyze the behavior of park personnel identified in real time, judge abnormal behavior and issue early warning;

[0055] The real-time collected video or image data in the target domain is synchronized to the source domain in real time, and the warning behaviors determined by humans and the warning behaviors determined by the system are annotated and synchronized to the source domain to realize dynamic adjustment of the warning thresholds and warning strategies in the warning model.

[0056] According to a specific embodiment of the present invention, using the data, model and task in the source domain as the data, model and task in the current park target domain, and combining the historical behavior data set of the current park personnel to establish a deep learning network model includes:

[0057] Use the data, models, and tasks in the source domain as the data, models, and tasks in the current park’s target domain;

[0058] Collect historical video or image data of human behavior in the park;

[0059] Combine the collected historical video or image data of people’s behavior in the park with the data in the source domain to construct a dataset;

[0060] The collected historical video or image data and the data in the source domain are cropped, scaled, and normalized for preprocessing, and a deep learning network model is established.

[0061] According to a specific embodiment of the present invention, the deep learning network model adopts a hybrid model of a convolutional neural network and a recurrent neural network;

[0062] Use convolutional neural networks to extract spatial features from the input dataset;

[0063] Use recurrent neural networks to process time series data and extract time features from the data set;

[0064] Use forward propagation to train the deep learning network model until the preset training end condition is reached;

[0065] Using the trained deep learning network model to identify and classify the behaviors of park personnel includes using the trained deep learning network model to identify and classify the behaviors of park personnel through forward propagation to obtain predicted classification results of the behaviors.

[0066] According to a specific embodiment of the present invention, the deep learning network model is optimized and the parameters are adjusted by training the deep learning network model, specifically including:

[0067] Define the loss function;

[0068] Calculate the gradient using the back-propagation algorithm;

[0069] Use optimization algorithms to update the parameters of deep learning network models;

[0070] Repeat the above steps until the performance of the deep learning network model reaches the preset standard or the training reaches the maximum number of iterations.

[0071] According to a specific implementation scheme of the present invention, the loss function is defined as a cross entropy loss function, and the parameters of the deep learning network model are updated using the gradient descent method or the Adam optimizer; after the training is completed, the performance of the deep learning network model is evaluated on the test set, and the model parameters of the deep learning network model are adjusted according to the evaluation results.

[0072] According to a specific embodiment of the present invention, analyzing the behaviors of park personnel identified in real time by big data technology specifically includes:

[0073] Collect data from multiple data sources and integrate them into a unified form and store them in a unified database;

[0074] Clean and process the collected data;

[0075] Discover data patterns, trends, and association rules from data based on data characteristics.

[0076] According to a specific embodiment of the present invention, cleaning and processing include removing noise data, filling missing values ​​and correcting errors, specifically including:

[0077] Identify and remove noisy data by setting thresholds or using statistical methods;

[0078] Fill missing values ​​using mean, median, or mode;

[0079] Check and correct errors based on preset data validation rules and logic.

[0080] According to a specific embodiment of the present invention, the mean, variance and correlation coefficient are used to discover the patterns, trends and association rules of the data.

[0081] According to a specific embodiment of the present invention, privacy and security protection of data is performed during the data collection, data storage, and data processing stages, specifically including:

[0082] During the data collection and transmission stage, the type, quantity and source of data to be collected should be clearly defined, and the collected data should be encrypted for transmission;

[0083] During the data storage phase, data is classified and stored according to its sensitivity, ensuring that data with higher sensitivity receives a higher level of protection and that the stored data is encrypted;

[0084] Sensitive data that meets the preset sensitivity level is desensitized and processed using a privacy protection algorithm.

[0085] According to a specific embodiment of the present invention, the Adtributor algorithm is used for early warning judgment:

[0086] The Adtributor algorithm decomposes the multi-dimensional root cause analysis problem into multiple single-dimensional root cause analysis problems, and locates the abnormal root cause by calculating the explanatory power (EP value) and surprise (S value) of each element in each dimension.

[0087] The Adtributor algorithm consists of the following steps:

[0088] First select the dimension. For example, in a chemical park, smoke sensors are set up to sense the smoke concentration, and the smoke concentration is used as the dimension for early warning. In an office building, when people gather in the video, specific point clouds are used as the dimension for early warning. The alarm events in different parks are different.

[0089] Set an alarm baseline. Set different alarm baselines for different alarm events. The alarm baseline can be set based on common sense or experience. For example, which behaviors are not in compliance with laws, regulations, industry rules, and operating specifications. Set the alarm baseline based on these behaviors. For example, at what degree of body tilt is predicted to be a fall, or how long a person has to stay without getting up after a fall to be considered a particularly dangerous behavior, etc.

[0090] Conduct data collection: Collect multi-dimensional time series data of KPIs, perform preliminary preprocessing, and improve data quality.

[0091] Anomaly detection: The ARMA time series model is used to predict KPI in real time, and the predicted value F is compared with the alarm baseline to determine whether the KPI is abnormal. In the present invention, the KPI is the employee behavior in the park, and different alarm events are selected according to different types of parks.

[0092] Perform early warning root cause analysis: calculate the EP value and S value of each element in each dimension. The EP value is used to measure the element's ability to explain anomalies, and the S value is used to measure the surprise of the element's own fluctuations;

[0093] Through the set thresholds (such as TEP and TEEP), the root causes of abnormalities are screened and located. Generally, elements with EP values ​​greater than the TEEP threshold are considered suspicious elements and added to the root cause set under this dimension. When the sum of the EP values ​​of the root cause set is greater than the TEP threshold, it is believed that the set can largely explain the abnormal fluctuations of the KPI.

[0094] According to the total S value of each dimension, the root cause set is summarized and output. The dimension with a larger S value is more likely to be the root cause of the anomaly.

[0095] Output root cause analysis results, including the dimensions and element sets that cause KPI anomalies.

[0096] According to a specific implementation scheme of the present invention, the behavior of park personnel is identified and classified by forward propagation, which specifically includes the following steps:

[0097] a. Preprocess the real-time or data set park personnel behavior data, such as frame extraction, normalization, etc.;

[0098] Frame extraction involves segmenting video data into a continuous sequence of frames. Assuming the video data is V, frame extraction is expressed as:

[0099] Fi = extract_frame(V,i);

[0100] Among them, Fi represents the i-th frame image, extract_frame() is a frame extraction function, and the input includes video data V and frame number i; normalization processing is to scale the pixel values ​​of the image data to the [0,1] interval, and the formula is as follows:

[0101] Inorm(x,y)=(I(x,y)-Imin) / (Imax-Imin);

[0102] Among them, I(x,y) is the pixel value of the original image at position (x,y), Inorm(x,y) is the pixel value of the normalized image at position (x,y), and Imin and Imax are the minimum and maximum pixel values ​​in the image, respectively.

[0103] b. Input the preprocessed data into the trained model for forward propagation.

[0104] Forward propagation can be expressed as a series of matrix multiplications and the application of activation functions. Assuming X is the input data, W is the weight matrix, b is the bias term, and f is the activation function, then for a simple fully connected layer, the forward propagation is expressed as:

[0105] H = f(WX+b);

[0106] Among them, H is the output of the fully connected layer.

[0107] c. The deep learning network model processes the input data layer by layer, extracts features and performs classification.

[0108] For convolutional neural networks (CNNs), feature extraction usually involves convolution operations. Assuming H is the input feature map, K is the convolution kernel, b is the bias term, and f is the activation function, the convolution operation can be expressed as:

[0109] C = f(K*H+b);

[0110] Among them, C is the output feature map of the convolutional layer.

[0111] For the classification task, assuming that C is the output feature map of the convolutional layer, Wclass is the weight matrix of the classification layer, bclass is the bias term, and the classification is expressed as:

[0112] P=softmax(WclassC+bclass);

[0113] Among them, P is the predicted probability distribution and softmax() is the loss function.

[0114] d. Output recognition and classification results, such as behavior category, timestamp, etc.

[0115] The determination of the behavior category can be achieved by selecting the category with the highest probability. Assuming P is the output probability distribution of the classification layer, the behavior category y is expressed as:

[0116] y = argmax_iPi;

[0117] Where Pi is the probability of the i-th category.

[0118] The timestamp can be obtained by recording the time of the input data. Assuming t is the timestamp of the input data, the output timestamp tout is: tout = t.

[0119] According to a specific embodiment of the present invention, the specific method for calculating the gradient by the back propagation algorithm comprises the following steps:

[0120] Calculate the output of each layer according to the forward propagation until the last layer to get the final prediction value;

[0121] Calculate the loss function: Use the loss function to measure the difference between the predicted value and the true value, such as the mean square error (MSE). In the present invention, the true value is the alarm baseline.

[0122] Starting from the output layer, the error term is calculated backward layer by layer, that is, the partial derivative of the loss function with respect to the output of each layer;

[0123] For the output layer, the error term δ L It can be expressed as:

[0124]

[0125] Among them, ⊙ represents the Hadamard product (i.e., element-by-element product), σ′ is the derivative of the output layer activation function, and a L is the activation value of the output layer, z Lis the weighted input to the output layer.

[0126] Update weights and biases: Use the chain rule to calculate the gradient of the loss function with respect to weights and biases and update them;

[0127] The formula for calculating the gradient of the weight is:

[0128]

[0129] The gradient calculation formula of the bias is:

[0130]

[0131] Among them, L is the difference between the predicted value and the true value, ω j is the weight matrix of the jth layer, b j is the bias vector of the jth layer, a j is the activation value of the jth layer, δ j+1 is the error term of the j+1th layer (note that this is the error term of the next layer, which is used to calculate the gradient of the current layer).

[0132] Iterate the above steps until the gradient drops to a preset acceptable level or reaches a predetermined number of iterations.

[0133] The above steps form the core of the backpropagation algorithm, which is used to calculate the gradients of the weights and biases of each layer in the neural network. Optimization algorithms (such as gradient descent) can also be used to update the weights and biases to minimize the loss function.

[0134] According to a specific embodiment of the present invention, when identifying and removing noise data by setting a threshold, the threshold is determined according to the data type. For example, when performing face recognition, if the degree of overlap between a face image and a face image in a database reaches 70% (which can be set as needed), it is considered to be the same person, and the other non-overlapping pixels are noise.

[0135] According to a specific embodiment of the present invention, the following steps are specifically included according to the preset data validation rules and logic checks:

[0136] Capture device IP address, device status and spatial location information through the embedded DPI program;

[0137] The device IP address, device status and spatial location information are sent to the Beidou satellite navigation and positioning system through the embedded Beidou module;

[0138] The device IP address, device status and spatial location information are then sent to the platform information processing center via short messages from the Beidou satellite navigation and positioning system, and verified according to preset rules.

[0139] According to a specific implementation scheme of the present invention, the patterns, trends and association rules of data are discovered from data according to data characteristics, and used for analyzing the behavioral characteristics of park personnel, which specifically includes the following steps:

[0140] 1. Use the mean to discover the overall characteristics of the behavior pattern:

[0141] The mean can help us understand the average level of a certain behavioral characteristic of park personnel, thereby revealing the overall behavioral pattern.

[0142] Daily activity analysis: For example, by calculating the average number of times park personnel enter and exit the park every day, we can understand the frequency of daily activities of park personnel. If the average value increases significantly over a period of time, it may mean that a large-scale event is held in the park or that the flow of personnel increases during holidays.

[0143] Work habit analysis: By calculating the average working hours and rest time of people in different departments or positions, the differences in work habits and efficiency of different departments or positions can be revealed.

[0144] 2. Use variance analysis to determine the degree of discreteness of behavior:

[0145] Variance can reflect the degree of discreteness of the behavioral characteristics of park personnel and help understand the stability and consistency of behavior.

[0146] Behavioral stability assessment: For example, by calculating the variance of the daily entry and exit times of park personnel, the stability of personnel attendance times can be assessed. A smaller variance means that personnel attendance times are more regular, while a larger variance may indicate that personnel attendance times fluctuate greatly.

[0147] Abnormal behavior detection: Variance can also be used to detect abnormal behavior. If a person's behavior characteristic differs from the mean by a large margin, then this may be an abnormal behavior point that requires further attention and analysis.

[0148] 3. Use correlation coefficients to explore association rules between behaviors:

[0149] The correlation coefficient can help us discover the correlation and changing trends between different behavioral characteristics of park personnel.

[0150] Behavioral correlation analysis: For example, by calculating the correlation coefficient between the number of times people enter and exit the park and the number of times they consume in the park, we can understand the degree of correlation between people's entry and exit and consumption behavior. If the correlation coefficient is high, it may mean that people who enter and exit frequently are also active consumers.

[0151] Behavior pattern mining: By calculating the correlation coefficient matrix between multiple behavior features, we can deeply explore the behavior patterns and association rules of park personnel. For example, by analyzing the correlation between personnel entry and exit time, working hours, consumption habits, etc., we can discover the potential connections and rules between certain behaviors.

[0152] Before conducting behavioral feature analysis, it is necessary to ensure the accuracy and completeness of the data and properly handle missing values, outliers, etc. to ensure the reliability of the analysis results.

[0153] The calculation results of the statistics need to be interpreted and analyzed in combination with the actual situation of the park. For example, the impact of holidays, special events and other factors on personnel behavior should be taken into account.

[0154] Comprehensively use multiple statistics such as mean, variance and correlation coefficient to obtain more comprehensive and in-depth behavioral feature analysis results.

[0155] By using statistics such as mean, variance and correlation coefficient, we can effectively analyze the behavioral characteristics of park personnel and reveal behavioral patterns, trends and association rules between different behaviors. This helps park managers better understand the behavioral characteristics of personnel and optimize park management and services.

[0156] The present invention provides a park personnel behavior early warning system based on big data and artificial intelligence, including a processor capable of executing the steps of the above-mentioned park personnel behavior early warning method based on big data and artificial intelligence.

[0157] Example 1

[0158] The present invention provides a park personnel behavior early warning method based on big data and artificial intelligence, comprising the following steps:

[0159] By using the collected open source behavioral data of other similar park personnel, a source domain including source data, models and tasks is constructed. The data, models and tasks in the source domain are used as the data, models and tasks of the current park target domain. A deep learning network model is established in combination with the historical behavioral data set of the current park personnel. The deep learning network model is trained by combining the data, models and tasks in the source domain and the historical behavioral data set of the current park personnel.

[0160] Based on the real-time collected video or image data of park personnel, the trained deep learning network model is used to identify and classify the behaviors of park personnel;

[0161] Build an early warning model through big data technology, use the early warning model to analyze the behavior of park personnel identified in real time, judge abnormal behavior and issue early warning;

[0162] The real-time collected video or image data in the target domain is synchronized to the source domain in real time, and the warning behaviors determined by humans and the warning behaviors determined by the system are annotated and synchronized to the source domain to realize dynamic adjustment of the warning thresholds and warning strategies in the warning model.

[0163] Example 2

[0164] The present invention provides a park personnel behavior early warning method based on big data and artificial intelligence, comprising the following steps:

[0165] By using the collected open source behavioral data of other similar park personnel, a source domain including source data, models and tasks is constructed. The data, models and tasks in the source domain are used as the data, models and tasks of the current park target domain. A deep learning network model is established in combination with the historical behavioral data set of the current park personnel. The deep learning network model is trained by combining the data, models and tasks in the source domain and the historical behavioral data set of the current park personnel.

[0166] Based on the real-time collected video or image data of park personnel, the trained deep learning network model is used to identify and classify the behaviors of park personnel;

[0167] Build an early warning model through big data technology, use the early warning model to analyze the behavior of park personnel identified in real time, judge abnormal behavior and issue early warning;

[0168] The real-time collected video or image data in the target domain is synchronized to the source domain in real time, and the warning behaviors determined by humans and the warning behaviors determined by the system are annotated and synchronized to the source domain to realize dynamic adjustment of the warning thresholds and warning strategies in the warning model.

[0169] Among them, using the data, models and tasks in the source domain as the data, models and tasks of the current park target domain, combined with the historical behavior data set of the current park personnel to establish a deep learning network model includes:

[0170] Use the data, models, and tasks in the source domain as the data, models, and tasks in the current park’s target domain;

[0171] Collect historical video or image data of human behavior in the park;

[0172] Combine the collected historical video or image data of people’s behavior in the park with the data in the source domain to construct a dataset;

[0173] The collected historical video or image data and the data in the source domain are cropped, scaled, and normalized for preprocessing, and a deep learning network model is established.

[0174] Among them, the deep learning network model adopts a hybrid model of convolutional neural network and recurrent neural network;

[0175] Use convolutional neural networks to extract spatial features from the input dataset;

[0176] Use recurrent neural networks to process time series data and extract time features from the data set;

[0177] Use forward propagation to train the deep learning network model until the preset training end condition is reached;

[0178] Using the trained deep learning network model, the behaviors of park personnel are identified and classified through forward propagation to obtain the predicted classification results of the behaviors.

[0179] Among them, the deep learning network model is optimized and the parameters are adjusted by training the deep learning network model, specifically including:

[0180] Define the loss function;

[0181] Calculate the gradient using the back-propagation algorithm;

[0182] Use optimization algorithms to update the parameters of deep learning network models;

[0183] Repeat the above steps until the performance of the deep learning network model reaches the preset standard or the training reaches the maximum number of iterations.

[0184] Among them, the loss function is defined as the cross entropy loss function, and the gradient descent method or Adam optimizer is used to update the parameters of the deep learning network model; after the training, the performance of the deep learning network model is evaluated on the test set, and the model parameters of the deep learning network model are adjusted according to the evaluation results.

[0185] Among them, the analysis of the behavior of park personnel identified in real time through big data technology specifically includes:

[0186] Collect data from multiple data sources and integrate them into a unified form and store them in a unified database;

[0187] Clean and process the collected data;

[0188] Discover data patterns, trends, and association rules from data based on data characteristics.

[0189] Cleaning and processing include removing noisy data, filling missing values, and correcting errors, including:

[0190] Identify and remove noisy data by setting thresholds or using statistical methods;

[0191] Fill missing values ​​using mean, median, or mode;

[0192] Check and correct errors based on preset data validation rules and logic.

[0193] Among them, the mean, variance and correlation coefficient are used to discover the patterns, trends and association rules of the data.

[0194] Among them, data privacy and security protection are carried out in the data collection, data storage and data processing stages, including:

[0195] During the data collection and transmission stage, the type, quantity and source of data to be collected should be clearly defined, and the collected data should be encrypted for transmission;

[0196] During the data storage phase, data is classified and stored according to its sensitivity, ensuring that data with higher sensitivity receives a higher level of protection and that the stored data is encrypted;

[0197] Sensitive data that meets the preset sensitivity level is desensitized and processed using a privacy protection algorithm.

[0198] In one embodiment of the present invention, in order to utilize big data and artificial intelligence technology to improve the accuracy and real-time performance of monitoring the behavior of park personnel, the following strategies are also adopted:

[0199] 1. Use advanced video surveillance system: The W4 system can monitor people and their activities in outdoor environments in real time, locate people and their body parts through shape analysis and tracking technology, and create appearance models of people for tracking in interactions such as occlusion. In addition, the system can identify objects carried by people, separate objects from outlines, and build appearance models for recognition in subsequent frames.

[0200] 2. Implement AI algorithms for real-time personnel counting: As mentioned above, by installing radars and cameras at the checkpoints in the park and using AI algorithms to process the collected image data in real time, the number of people in the park can be effectively counted. At the same time, an in-vehicle target detection calculation model is designed to identify the number of people in the vehicle, thereby providing real-time personnel safety management guarantees for park managers.

[0201] 3. Develop a real-time abnormal behavior detection system: According to, by combining online tracking algorithms in computer vision, nonlinear pedestrian motion models in crowd simulation, and Bayesian learning techniques, the trajectory-level pedestrian behavior of each agent in the video can be automatically calculated and used to detect abnormal behavior. This method can achieve real-time anomaly detection in low to medium density crowd videos.

[0202] 4. Apply deep learning to intelligently monitor personnel status: As shown, by designing an intelligent monitoring system equipped with deep learning-based target detection and multi-target tracking algorithms, it is possible to achieve intelligent supervision of the working status of on-the-job personnel and the number of visitors in government agencies and corporate units. In particular, the ILF-YOLOv3 algorithm improves the accuracy and real-time performance of detection by improving the loss function and multi-scale feature detection module of the YOLOv3 detection algorithm.

[0203] 5. Use facial recognition technology for personnel management: The application of facial recognition technology in smart parks can include channel entrance and exit management, visitor management, access control management, etc. Facial recognition technology can improve the efficiency and accuracy of park personnel management.

[0204] 6. Application of edge computing technology: As mentioned above, real-time and uninterrupted human object tracking can be achieved by performing video analysis at the edge of the network. This approach reduces dependence on cloud computing centers, reduces latency, and improves the efficiency of real-time monitoring.

[0205] The present invention utilizes advanced information technology and artificial intelligence technology to construct an efficient and intelligent information processing and decision support platform. In one embodiment of the present invention, the specific implementation is as follows:

[0206] 1. Data collection and integration: First, the system needs to collect real-time data in the park through various sensor networks, video surveillance, mobile devices and other technical means, including environmental data, personnel flow data, equipment status data, etc. These data are transmitted to the central processing system through a high-speed data communication network.

[0207] 2. Data analysis and processing: Use big data technology and artificial intelligence algorithms to clean, store and analyze data. This includes using machine learning models for anomaly detection, pattern recognition and predictive analysis, as well as using geographic information systems (GIS) and remote sensing technologies for spatial data analysis.

[0208] 3. Intelligence information fusion: In the context of big data, intelligence information fusion is the key to improving the effectiveness of emergency management. The system needs to integrate information from different sources, including historical data, real-time monitoring data, and external data sources to form a comprehensive intelligence view, such as data from other similar parks.

[0209] 4. Decision support and response: In one embodiment of the present invention, based on the analysis results, the system can also provide real-time decision support to help managers respond quickly. This includes resource allocation, emergency response route planning, and personnel evacuation guidance in emergency situations. At the same time, the system should also have adaptive learning capabilities to continuously optimize the decision model based on historical response data.

[0210] 5. Collaboration and sharing: In one embodiment of the present invention, the smart park safety emergency integrated management system should also support multi-department and multi-level information sharing and collaborative work. By establishing a unified information platform, it ensures that all relevant parties can access the latest information and decision support tools, thereby improving the overall emergency response efficiency.

[0211] The potential risks in park safety management mainly include fire and explosion accidents in chemical parks, network security threats, safety accidents in logistics parks, and environmental risks.

[0212] The above descriptions are only optional embodiments of the present invention, and are not intended to limit the patent scope of the present invention. All equivalent structural changes made using the contents of the present invention's specification and drawings, or directly / indirectly applied in other related technical fields, are included in the patent protection scope of the present invention.

Claims

1. A park personnel behavior early warning method based on big data and artificial intelligence, characterized in that: The method comprises the following steps: Collect open source behavioral data of other similar park personnel, build a source domain including source data, models and tasks, use the data, models and tasks in the source domain as the data, models and tasks of the current park target domain, and build a deep learning network model in combination with the historical behavioral data set of the current park personnel. Train the deep learning network model in combination with the data, models and tasks in the source domain and the historical behavioral data set of the current park personnel. Based on the real-time collected video or image data of park personnel, the trained deep learning network model is used to identify and classify the behaviors of park personnel; Build an early warning model through big data technology, use the early warning model to analyze the behavior of park personnel identified in real time, judge abnormal behavior and issue early warning; The real-time collected video or image data in the target domain is synchronized to the source domain in real time, and the warning behaviors determined by humans and the warning behaviors determined by the system are annotated and synchronized to the source domain to realize dynamic adjustment of the warning thresholds and warning strategies in the warning model.

2. The park personnel behavior early warning method based on big data and artificial intelligence according to claim 1 is characterized in that: Using the data, models, and tasks in the source domain as the data, models, and tasks in the current park's target domain, combined with the historical behavior dataset of the current park's personnel to establish a deep learning network model includes: Use the data, models, and tasks in the source domain as the data, models, and tasks in the current park’s target domain; Collect historical video or image data of human behavior in the park; Combine the collected historical video or image data of people’s behavior in the park with the data in the source domain to construct a dataset; The collected historical video or image data and the data in the source domain are cropped, scaled, and normalized for preprocessing, and a deep learning network model is established.

3. The park personnel behavior early warning method based on big data and artificial intelligence according to claim 2 is characterized in that: The deep learning network model adopts a hybrid model of convolutional neural network and recurrent neural network; Use convolutional neural networks to extract spatial features from the input dataset; Use recurrent neural networks to process time series data and extract time features from the data set; Use forward propagation to train the deep learning network model until the preset training end condition is reached; Using the trained deep learning network model, the behaviors of park personnel are identified and classified through forward propagation to obtain the predicted classification results of the behaviors.

4. The park personnel behavior early warning method based on big data and artificial intelligence according to claim 3 is characterized in that: The deep learning network model is optimized and its parameters are adjusted by training the deep learning network model, including: Define the loss function; Calculate the gradient using the back-propagation algorithm; Use optimization algorithms to update the parameters of deep learning network models; Repeat the above steps until the performance of the deep learning network model reaches the preset standard or the training reaches the maximum number of iterations.

5. The park personnel behavior early warning method based on big data and artificial intelligence according to claim 4 is characterized in that: The loss function is defined as the cross entropy loss function, and the parameters of the deep learning network model are updated using the gradient descent method or the Adam optimizer. After the training, the performance of the deep learning network model is evaluated on the test set, and the model parameters of the deep learning network model are adjusted according to the evaluation results.

6. The park personnel behavior early warning method based on big data and artificial intelligence according to claim 1 is characterized in that: The analysis of the behaviors of park personnel identified in real time through big data technology includes: Collect data from multiple data sources and integrate them into a unified form and store them in a unified database; Clean and process the collected data; Discover data patterns, trends, and association rules from data based on data characteristics.

7. The method for early warning of park personnel behavior based on big data and artificial intelligence according to claim 6 is characterized in that: Cleaning and processing includes removing noisy data, filling missing values, and correcting errors, including: Identify and remove noisy data by setting thresholds or using statistical methods; Fill missing values ​​using mean, median, or mode; Check and correct errors based on preset data validation rules and logic.

8. The method for early warning of park personnel behavior based on big data and artificial intelligence according to claim 7 is characterized in that: Use mean, variance, and correlation coefficient to discover patterns, trends, and association rules in data.

9. The method for early warning of park personnel behavior based on big data and artificial intelligence according to claim 6 is characterized in that: Data privacy and security are protected during the data collection, data storage, and data processing stages, including: During the data collection and transmission stage, the type, quantity and source of data to be collected should be clearly defined, and the collected data should be encrypted for transmission; During the data storage phase, data is classified and stored according to its sensitivity, ensuring that data with higher sensitivity receives a higher level of protection and that the stored data is encrypted; Sensitive data that meets the preset sensitivity level is desensitized and processed using a privacy protection algorithm.

10. A park personnel behavior early warning system based on big data and artificial intelligence, characterized in that: It includes a processor capable of executing the steps of the park personnel behavior early warning method based on big data and artificial intelligence as described in any one of claims 1 to 9.

Images