Intelligent station security monitoring method and system

Abstract

The embodiment of the present invention provides an intelligent station house security monitoring method and system, which obtains multi-dimensional monitoring data of a reference station house monitoring target and corresponding intrusion activity label data, and processes these data using an illegal intrusion prediction network, thereby achieving effective identification of whether illegal intrusion activities occur in the reference station house monitoring target. The illegal intrusion prediction network can be trained on network parameters based on the intrusion activity label data and the intrusion activity identification data, thereby generating a target illegal intrusion prediction network, thereby improving the accuracy of predicting illegal intrusion activities occurring in the target station house monitoring target. The method and system can be widely used in various station house scenarios that require security monitoring, effectively improving the safety protection level of the station house and reducing the risk of illegal intrusion activities.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of security monitoring technology, and specifically, to a method and system for security monitoring of an intelligent station building. Background Art

[0002] With the continuous development of society and the continuous advancement of science and technology, station security monitoring has become an indispensable part of modern society. In the existing technology, station security monitoring mainly relies on traditional video surveillance systems and manual patrols. Although these methods can provide a certain degree of security, there are many problems. For example, the video surveillance system may affect its monitoring effect due to factors such as complex environment and light conditions; while manual patrols consume a lot of human resources and may cause monitoring loopholes due to human negligence. In addition, traditional security monitoring methods often lack predictability and cannot effectively predict and prevent illegal intrusion activities in a timely manner. Especially in today's context of rapid development of informatization, the demand for security protection of station buildings is increasing, and traditional security monitoring methods can no longer meet the high security requirements of modern station buildings. In other words, the existing technology relies too much on manual monitoring and post-analysis, and cannot achieve real-time, accurate monitoring and early warning of station monitoring targets, and thus cannot timely and effectively discover and respond to illegal intrusion activities. Summary of the invention

[0003] In order to at least overcome the above-mentioned deficiencies in the prior art, an object of the embodiments of the present application is to provide a smart station house security monitoring method and system.

[0004] In a first aspect, an embodiment of the present application provides a method for security monitoring of an intelligent station house, the method comprising:

[0005] Acquire multi-dimensional monitoring data of a reference station house monitoring target and intrusion activity label data of the reference station house monitoring target, wherein the intrusion activity label data is used to reflect whether illegal intrusion activities occur in the reference station house monitoring target;

[0006] Determine the intrusion activity identification data of the reference station house monitoring target based on the multi-dimensional monitoring data of the reference station house monitoring target by using the illegal intrusion prediction network, wherein the intrusion activity identification data is used to reflect the probability value of the illegal intrusion activity occurring in the reference station house monitoring target;

[0007] Based on the intrusion activity label data and the intrusion activity identification data, network parameter learning is performed on the illegal intrusion prediction network to generate a target illegal intrusion prediction network, and the target illegal intrusion prediction network is used to determine the probability value of the illegal intrusion activity occurring in the target station monitoring target.

[0008] In a possible implementation of the first aspect, the illegal intrusion prediction network includes a graph self-attention unit and a fully connected layer;

[0009] The method of using the illegal intrusion prediction network to determine the intrusion activity identification data of the reference station monitoring target based on the multi-dimensional monitoring data of the reference station monitoring target includes:

[0010] Encoding the multi-dimensional monitoring data of the reference station house monitoring target by using the graph self-attention unit to generate a first embedding representation vector distribution of the reference station house monitoring target;

[0011] Determine a reference embedded representation vector distribution of the reference station house monitoring target according to the first embedded representation vector distribution of the reference station house monitoring target;

[0012] The fully connected layer is used to classify and output the reference embedding representation vector distribution to generate the intrusion activity identification data.

[0013] In a possible implementation of the first aspect, the graph self-attention unit includes a preprocessing unit and a feature focusing unit;

[0014] The method of encoding the multi-dimensional monitoring data of the reference station house monitoring target by using the graph self-attention unit to generate a first embedded representation vector distribution of the reference station house monitoring target includes:

[0015] Using the preprocessing unit to perform regularized mapping conversion on the multi-dimensional monitoring data of the reference station house monitoring target to generate a reference vector distribution of the reference station house monitoring target;

[0016] Using the feature focusing unit to perform feature focusing processing on the reference vector distribution to generate a reference focusing vector distribution of the reference station monitoring target;

[0017] A first embedded representation vector distribution of the reference station house monitoring target is determined based on the reference focusing vector distribution.

[0018] In a possible implementation manner of the first aspect, determining the reference embedded representation vector distribution of the reference station house monitoring target according to the first embedded representation vector distribution of the reference station house monitoring target includes:

[0019] Acquire a reference knowledge network, the reference knowledge network comprising a plurality of reference network members and a plurality of reference member links, any one of the reference network members being used to reflect an interactive monitoring target or a station house interactive logical space or a station house operation activity in a reference station house interactive scene where the reference station house monitoring target is located, any one of the reference member links being used to reflect a reference interactive cost parameter or a reference interactive relationship or a reference affiliation, the reference interactive cost parameter being an interactive cost parameter of an interactive monitoring target represented by a reference network member connected to any one of the reference member links, the reference interactive relationship being used to reflect an interactive monitoring target represented by a reference network member connected to any one of the reference member links executing or being executed by another connected reference network member represented by another connected reference network member, the reference affiliation being used to reflect an interactive monitoring target represented by a reference network member connected to any one of the reference member links belonging to a connected station house interactive logical space represented by another connected reference network member;

[0020] For any reference network member, determine the member embedding representation vector distribution of the any reference network member through a knowledge network embedding representation unit according to the any reference network member, a reference member link connected to the any reference network member, and reference network members connected to the any reference network member through the reference member link;

[0021] If any one of the reference network members is used to reflect the reference station house monitoring target, using the member embedding representation vector distribution of any one of the reference network members as the second embedding representation vector distribution of the reference station house monitoring target;

[0022] The first embedding representation vector distribution and the second embedding representation vector distribution of the reference station house monitoring target are fused to generate the reference embedding representation vector distribution.

[0023] In a possible implementation manner of the first aspect, before the step of using the member embedding representation vector distribution of any one of the reference network members as the second embedding representation vector distribution of the reference station house monitoring target, the method further includes:

[0024] Acquire a sample knowledge network, wherein the sample knowledge network includes a plurality of sample network members and a plurality of sample member links, wherein any one of the sample network members is used to reflect a sample interaction monitoring target or a sample station room interaction logical space or a sample station room operation activity in a sample station room interaction scene, and any one of the sample member links is used to reflect a sample interaction cost parameter or a sample interaction relationship or a sample affiliation relationship, wherein the sample interaction cost parameter is an interaction cost parameter of a sample interaction monitoring target represented by a sample network member connected to the any one of the sample member links, and the sample interaction relationship is used to reflect that a sample interaction monitoring target represented by a sample network member connected to the any one of the sample member links executes or is executed by another sample station room operation activity represented by another sample network member connected to the sample interaction monitoring target represented by a sample network member connected to the any one of the sample member links belongs to a sample station room interaction logical space represented by another sample network member connected to the sample interaction monitoring target represented by a sample network member connected to the any one of the sample member links;

[0025] Embedding the sample knowledge network using an initial embedding representation unit to generate a member embedding representation vector distribution of each sample network member;

[0026] According to the member embedding representation vector distribution of each sample network member, network parameter learning is performed on the initial embedding representation unit to generate the knowledge network embedding representation unit.

[0027] In a possible implementation of the first aspect, performing network parameter learning on the initial embedding representation unit according to the member embedding representation vector distribution of each sample network member to generate the knowledge network embedding representation unit includes:

[0028] Restoring the member embedding representation vector distribution of each sample network member to generate a prediction knowledge network;

[0029] Determining a knowledge network error parameter based on the sample knowledge network and the predicted knowledge network;

[0030] Network parameter learning is performed on the initial embedding representation unit according to the knowledge network error parameter to generate the knowledge network embedding representation unit.

[0031] In a possible implementation of the first aspect, performing network parameter learning on the initial embedding representation unit according to the member embedding representation vector distribution of each sample network member to generate the knowledge network embedding representation unit includes:

[0032] Extracting at least one first network member and at least one second network member from the plurality of sample network members, wherein any one of the first network members is used to reflect a sample interactive monitoring target, and any one of the second network members is used to reflect a sample station house operation activity;

[0033] For any one of the first network members and any one of the second network members, a predicted interaction relationship is determined based on the member embedding representation vector distribution of any one of the first network members and any one of the second network members, wherein the predicted interaction relationship is used to reflect the probability value of the sample interaction monitoring target represented by any one of the first network members executing the sample station room operation activity represented by any one of the second network members;

[0034] Acquire a marked interaction relationship, where the marked interaction relationship is used to reflect whether the sample interaction monitoring target represented by any one of the first network members performs the sample station room operation activity represented by any one of the second network members;

[0035] Determine an interaction relationship error parameter based on the predicted interaction relationship and the labeled interaction relationship; perform network parameter learning on the initial embedding representation unit based on the interaction relationship error parameter to generate the knowledge network embedding representation unit.

[0036] In a possible implementation manner of the first aspect, the illegal intrusion activity includes at least one of an intrusion activity of damaging equipment and an intrusion activity of illegally entering a sensitive area;

[0037] The intrusion activity label data includes at least one of first label data and second label data, wherein the first label data is used to reflect whether the reference station monitoring target has any intrusion activity that damages equipment, and the second label data is used to reflect whether the reference station monitoring target has any intrusion activity that illegally enters a sensitive area;

[0038] The intrusion activity identification data includes at least one of a first identification result and a second identification result, wherein the first identification result is used to reflect the probability value of an intrusion activity that damages equipment in the reference station monitoring target, and the second identification result is used to reflect the probability value of an intrusion activity that illegally enters a sensitive area in the reference station monitoring target;

[0039] The step of performing network parameter learning on the illegal intrusion prediction network based on the intrusion activity label data and the intrusion activity identification data to generate a target illegal intrusion prediction network includes:

[0040] Determining a first error parameter according to the first tag data and the first recognition result;

[0041] determining a second error parameter according to the second tag data and the second recognition result;

[0042] Based on the first error parameter and the second error parameter, network parameter learning is performed on the illegal intrusion prediction network to generate a target illegal intrusion prediction network.

[0043] In a possible implementation of the first aspect, the method further includes:

[0044] Obtain multi-dimensional monitoring data of the target station monitoring target;

[0045] Determine the illegal intrusion prediction result of the target station house monitoring target according to the multi-dimensional monitoring data of the target station house monitoring target by using the target illegal intrusion prediction network, wherein the illegal intrusion prediction result is used to reflect the probability value of illegal intrusion activity occurring in the target station house monitoring target;

[0046] When the illegal intrusion prediction result meets the preset requirements, determining that the illegal intrusion activity occurs in the target station monitoring target;

[0047] Wherein, the use of the target illegal intrusion prediction network to determine the illegal intrusion prediction result of the target station monitoring target according to the multi-dimensional monitoring data of the target station monitoring target includes:

[0048] Encoding the multi-dimensional monitoring data of the target station house monitoring target by using the target illegal intrusion prediction network to generate a first embedded representation vector distribution of the target station house monitoring target;

[0049] Determine the reference vector distribution of the target station house monitoring target according to the first embedded representation vector distribution of the target station house monitoring target;

[0050] The target illegal intrusion prediction network is used to classify and output the reference vector distribution to generate the illegal intrusion prediction result.

[0051] For example, in a possible implementation of the first aspect, after determining that the target station house monitoring target has the illegal intrusion activity, the method further includes:

[0052] Determine data of a first gathering activity according to the multi-dimensional monitoring data of the target station room interaction scene and the multi-dimensional monitoring data of the target station room monitoring target, wherein the first gathering activity is a gathering activity in which the target station room monitoring target intervenes;

[0053] Determine data of a second gathering activity based on the multi-dimensional monitoring data of the target station house interaction scene and the multi-dimensional monitoring data of the candidate station house monitoring target, wherein the candidate station house monitoring target is a station house monitoring target observed to be involved in a second gathering activity in the target station house interaction scene, and the second gathering activity is an existing gathering activity;

[0054] Data of a target aggregation activity is determined based on the data of the first aggregation activity and the data of the second aggregation activity.

[0055] For example, in a possible implementation of the first aspect, after determining that the target station house monitoring target has the illegal intrusion activity, the method further includes:

[0056] Determine, based on the multi-dimensional monitoring data of the target station room interaction scene and the multi-dimensional monitoring data of the target station room monitoring target, collaborative activity intervention information of the target station room monitoring target, wherein the collaborative activity intervention information is used to reflect the probability value of the target station room monitoring target intervening in the collaborative activity, wherein the collaborative activity is the intrusion activity to be coordinated;

[0057] Determine an intrusion weight coefficient of the target monitoring target of the target station house according to the collaborative activity intervention information and the illegal intrusion prediction result, wherein the intrusion weight coefficient is used to reflect the intrusion attention degree of the target monitoring target of the target station house;

[0058] When the intrusion weight coefficient is greater than a preset coefficient, the target station house monitoring target is determined to be a collaborative station house monitoring target, and the collaborative station house monitoring target is an observed station house monitoring target to be coordinated.

[0059] In the second aspect, the embodiments of the present application also provide a security monitoring system, which includes a processor and a machine-readable storage medium, wherein the machine-readable storage medium stores machine-executable instructions, and the machine-executable instructions are loaded and executed by the processor to implement the intelligent station security monitoring method in any possible implementation manner in the first aspect.

[0060] In accordance with any of the above aspects, the present application obtains multi-dimensional monitoring data of the reference station house monitoring target and the corresponding intrusion activity label data, and processes these data using an illegal intrusion prediction network, thereby achieving effective identification of whether illegal intrusion activities occur in the reference station house monitoring target. The application can learn network parameters of the illegal intrusion prediction network based on the intrusion activity label data and the intrusion activity identification data, thereby generating a target illegal intrusion prediction network, thereby improving the accuracy of predicting illegal intrusion activities occurring in the target station house monitoring target. The application can be widely used in various station house scenarios that require security monitoring, effectively improving the safety protection level of the station house and reducing the risk of illegal intrusion activities.

[0061] That is to say, the present application obtains the multi-dimensional monitoring data and intrusion activity label data of the reference station house monitoring target, and then uses the illegal intrusion prediction network to determine the intrusion activity identification data of the reference station house monitoring target. These intrusion activity identification data reflect the probability value of illegal intrusion activities occurring in the reference station house monitoring target. Based on the intrusion activity label data and the intrusion activity identification data, network parameter learning is performed to generate a target illegal intrusion prediction network, which is used to determine the probability value of illegal intrusion activities occurring in the target station house monitoring target. As a result, real-time monitoring of the security status of the station house is effectively achieved, and possible illegal intrusion activities can be accurately predicted, so that protective measures can be taken in time, thereby improving the safety protection level of the station house. At the same time, the network parameters are automatically optimized through the machine learning algorithm, which improves the prediction accuracy and efficiency and saves a lot of human resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required to be used in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be extracted in combination with these drawings without creative work.

[0063] Figure 1 A schematic diagram of a process flow of a smart station house security monitoring method provided by an embodiment of the present invention;

[0064] Figure 2 A schematic block diagram of the structure of a security monitoring system for implementing the above-mentioned intelligent station security monitoring method provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0065] The following description is intended to enable one of ordinary skill in the art to implement and incorporate the present invention, and the description is provided in the context of a specific application scenario and its requirements. It will be apparent to one of ordinary skill in the art that various changes may be made to the disclosed embodiments, and that the general principles defined in the present invention may be applied to other embodiments and application scenarios without departing from the principles and scope of the present invention. Therefore, the present invention is not limited to the described embodiments, but should be given the broadest scope consistent with the claims.

[0066] Figure 1 It is a flow chart of a smart station house security monitoring method provided by an embodiment of the present invention. The smart station house security monitoring method is introduced in detail below.

[0067] Step S110, obtaining multi-dimensional monitoring data of the reference station house monitoring target and intrusion activity label data of the reference station house monitoring target, wherein the intrusion activity label data is used to reflect whether illegal intrusion activity occurs to the reference station house monitoring target.

[0068] In this embodiment, the security monitoring system is first connected to the monitoring system of the reference station house. This monitoring system may include a variety of devices such as video cameras, temperature sensors, humidity sensors, access control systems, etc., which provide different types of multi-dimensional monitoring data. In other words, the multi-dimensional monitoring data refers to data collected from multiple dimensions or angles about the monitoring target of the reference station house. These data may include video images, sound recordings, environmental parameters (such as temperature, humidity), access control status, etc. For example, multiple cameras are installed in a station house, and each camera can provide video stream data from different angles; at the same time, temperature and humidity sensors are also deployed in the station house for real-time monitoring of the environmental conditions in the station house; in addition, the access control system can also provide data about the door status (open / closed) and entry and exit records.

[0069] At the same time, the security monitoring system also obtains intrusion activity label data from the monitoring system. These intrusion activity label data are used to mark whether the multi-dimensional monitoring data corresponds to illegal intrusion activities. They are usually generated by manual annotation or automatic detection algorithms, and are used to train or verify the accuracy of illegal intrusion prediction networks, and to clearly mark whether illegal intrusion activities have occurred at a specific time point. For example, if someone is found to have entered the station room without authorization during a certain period of time through video surveillance, then the video data of this period will be marked as "intrusion occurred", forming a positive sample; and other periods where no intrusion occurred will be marked as "no intrusion", forming a negative sample.

[0070] Step S120, using the illegal intrusion prediction network to determine the intrusion activity identification data of the reference station house monitoring target based on the multi-dimensional monitoring data of the reference station house monitoring target, and the intrusion activity identification data is used to reflect the probability value of the illegal intrusion activity occurring in the reference station house monitoring target.

[0071] In this embodiment, the security monitoring system will then use a pre-built illegal intrusion prediction network to process these multi-dimensional monitoring data. This illegal intrusion prediction network is a machine learning model that predicts whether illegal intrusion activities will occur by analyzing multi-dimensional monitoring data. This illegal intrusion prediction network can be a deep learning model, such as a convolutional neural network (CNN) or a recurrent neural network (RNN), or other types of machine learning models. For example, a CNN-based illegal intrusion prediction network can accept video frames as input, extract image features through the convolution layer, and then output a probability value through the fully connected layer, indicating the possibility that the current video frame corresponds to an illegal intrusion activity.

[0072] The intrusion activity identification data is the data output by the illegal intrusion prediction network, which reflects the probability value of the multi-dimensional monitoring data corresponding to the illegal intrusion activity. This probability value is usually a number between 0 and 1, and the higher the value, the greater the possibility of illegal intrusion activity. For example, assuming that the output of the illegal intrusion prediction network for a certain video frame is 0.8, this means that the probability of the video frame corresponding to the illegal intrusion activity is 80%, that is, there is a high probability that an illegal intrusion has occurred.

[0073] Step S130, based on the intrusion activity label data and the intrusion activity identification data, network parameter learning is performed on the illegal intrusion prediction network to generate a target illegal intrusion prediction network, and the target illegal intrusion prediction network is used to determine the probability value of the illegal intrusion activity occurring in the target station monitoring target.

[0074] After obtaining the intrusion activity label data and intrusion activity identification data, the security monitoring system combines the two to train the illegal intrusion prediction network. Specifically, the intrusion activity identification data (probability value) output by the illegal intrusion prediction network can be compared with the actual intrusion activity label data (real situation). If the probability value output by the illegal intrusion prediction network is significantly different from the actual situation, the security monitoring system will adjust the parameters of the illegal intrusion prediction network to reduce this difference. This process may involve machine learning algorithms such as gradient descent and back propagation. After multiple iterations of training, the parameters of the illegal intrusion prediction network will be optimized and its prediction accuracy will be improved. Finally, the security monitoring system will generate a target illegal intrusion prediction network, which can more accurately predict the probability value of illegal intrusion activities occurring in the target station monitoring target. In this process, cross-validation, regularization and other techniques can also be used to prevent overfitting and improve the generalization ability of the target illegal intrusion prediction network.

[0075] For example, during the training process, if the illegal intrusion prediction network outputs a very low probability value for a positive sample (actual intrusion), such as only 0.2, then the gradient of the loss function for the sample will be calculated through the back-propagation algorithm, and the weights and bias parameters of the illegal intrusion prediction network will be updated accordingly, so that the network can output a higher probability value the next time it encounters a similar sample. Therefore, after multiple rounds of training and optimization, the target illegal intrusion prediction network may have better prediction performance for new, unseen multi-dimensional surveillance data, such as being able to more accurately identify intruders hidden in videos or respond more sensitively to environmental changes.

[0076] Based on the above steps, the present application obtains multi-dimensional monitoring data of the reference station house monitoring target and the corresponding intrusion activity label data, and uses the illegal intrusion prediction network to process these data, thereby achieving effective identification of whether illegal intrusion activities occur in the reference station house monitoring target. It can learn the network parameters of the illegal intrusion prediction network based on the intrusion activity label data and the intrusion activity identification data, thereby generating a target illegal intrusion prediction network, improving the accuracy of predicting illegal intrusion activities in the target station house monitoring target, and can be widely used in various station house scenarios that require security monitoring, effectively improving the safety protection level of the station house and reducing the risk of illegal intrusion activities.

[0077] That is to say, the present application obtains the multi-dimensional monitoring data and intrusion activity label data of the reference station house monitoring target, and then uses the illegal intrusion prediction network to determine the intrusion activity identification data of the reference station house monitoring target. These intrusion activity identification data reflect the probability value of illegal intrusion activities occurring in the reference station house monitoring target. Based on the intrusion activity label data and the intrusion activity identification data, network parameter learning is performed to generate a target illegal intrusion prediction network, which is used to determine the probability value of illegal intrusion activities occurring in the target station house monitoring target. As a result, real-time monitoring of the security status of the station house is effectively achieved, and possible illegal intrusion activities can be accurately predicted, so that protective measures can be taken in time, thereby improving the safety protection level of the station house. At the same time, the network parameters are automatically optimized through machine learning algorithms, which improves the prediction accuracy and efficiency and saves a lot of human resources.

[0078] In a possible implementation, the illegal intrusion prediction network includes a graph self-attention unit and a fully connected layer. In this embodiment, the graph self-attention unit is used to process complex relationships in multi-dimensional monitoring data, and the fully connected layer is used for the final classification output.

[0079] Step S120 may include:

[0080] Step S121, using the graph self-attention unit to encode the multi-dimensional monitoring data of the reference station house monitoring target, and generate a first embedded representation vector distribution of the reference station house monitoring target.

[0081] In this embodiment, the multi-dimensional monitoring data of the reference station house may include video streams, sensor readings (such as temperature and humidity), access control status, etc. These multi-dimensional monitoring data have complex correlations in time and space.

[0082] The security monitoring system first encodes these multi-dimensional monitoring data using graph self-attention units. Graph self-attention units can capture the dependencies between data points and assign different attention weights to each data point. For example, at a certain moment, an abnormal action captured in a video stream may be more important than a normal temperature reading. Through the processing of the graph self-attention unit, the security monitoring system generates a first embedding representation vector distribution of the reference station room monitoring target. This first embedding representation vector distribution captures the key features in the multi-dimensional monitoring data and the relationships between them, providing a compact and effective representation for the next step of classification.

[0083] Step S122, determining the reference embedded representation vector distribution of the reference station house monitoring target based on the first embedded representation vector distribution of the reference station house monitoring target.

[0084] After obtaining the first embedding representation vector distribution, the security monitoring system further processes the first embedding representation vector distribution to determine a reference embedding representation vector distribution. This step may involve averaging, pooling, or selecting the most representative embedding vector for the first embedding representation vector distribution. The reference embedding representation vector distribution is a more abstract and refined representation of the multi-dimensional monitoring data, which highlights the most important information for illegal intrusion prediction.

[0085] Step S123: using the fully connected layer to classify and output the reference embedding representation vector distribution to generate the intrusion activity identification data.

[0086] Finally, the security monitoring system inputs the reference embedding representation vector distribution into the fully connected layer. The fully connected layer is a standard neural network layer that maps the input to the output space through weights and biases. In this scenario, the output of the fully connected layer is the intrusion activity recognition data, that is, a value representing the probability of illegal intrusion activity occurring in the reference station monitoring target. This value is obtained based on multi-dimensional monitoring data and processed by the graph self-attention unit and the fully connected layer.

[0087] In one possible implementation, the graph self-attention unit includes a preprocessing unit and a feature focusing unit. In this embodiment, the structure of the graph self-attention unit includes a preprocessing unit and a feature focusing unit. The preprocessing unit is responsible for performing preliminary regularized mapping transformation on the multi-dimensional monitoring data, while the feature focusing unit focuses on extracting key features from the preprocessed data.

[0088] Step S121 may include:

[0089] Step S1211, using the preprocessing unit to perform regularized mapping conversion on the multi-dimensional monitoring data of the reference station house monitoring target, to generate a reference vector distribution of the reference station house monitoring target.

[0090] In this embodiment, the security monitoring system receives the original multi-dimensional monitoring data of the reference station house, which may have different formats, dimensions and distribution characteristics. The preprocessing unit first cleans, standardizes and normalizes these data to eliminate the dimensional differences and outlier effects between the data. Next, the preprocessing unit maps the cleaned data into a unified feature space to generate a reference vector distribution of the monitoring target of the reference station house. This mapping process may involve technologies such as linear transformation, kernel function mapping, or embedding layers in deep learning, with the aim of making the data easier to analyze and compare in the new feature space.

[0091] For example, assume that multiple sensors are installed in the reference station room to monitor temperature, humidity, wind speed, and access control status. The multidimensional monitoring data received by the security monitoring system are the readings of these sensors at different time points, forming a multidimensional time series data. The preprocessing unit first performs regular mapping conversion on these multidimensional time series data. For example, for temperature and humidity data, due to their different dimensions (degrees Celsius and percentage), the preprocessing unit will normalize them to the same numerical range, such as [0, 1]. For wind speed data, if the original data is in meters per second, the preprocessing unit may convert it to a more intuitive wind force level. For access control status data, the preprocessing unit may convert it from the original switch signal to binary data indicating whether the access control is open or closed. After the regular mapping conversion, the multidimensional monitoring data is converted into a unified reference vector distribution. Each dimension in this reference vector distribution corresponds to a specific sensor reading, and all readings are converted to the same numerical range, which is convenient for subsequent feature extraction and classification processing.

[0092] Step S1212: Use the feature focusing unit to perform feature focusing processing on the reference vector distribution to generate a reference focusing vector distribution of the reference station monitoring target.

[0093] In this embodiment, the security monitoring system inputs the reference vector distribution obtained after preprocessing into the feature focusing unit. The task of the feature focusing unit is to extract the features most relevant to illegal intrusion activities from the reference vector distribution. In order to achieve this goal, the feature focusing unit may adopt structures such as attention mechanism, autoencoder or convolutional neural network. For example, the attention mechanism can assign different weights to different features, so that more attention can be paid to important features when processing data; the autoencoder can learn the compact representation of data through the encoding-decoding process, thereby removing redundant information and highlighting key features. After processing by the feature focusing unit, the security monitoring system obtains the reference focus vector distribution of the monitoring target of the reference station room. Compared with the original reference vector distribution, this reference focus vector distribution is more compact and targeted, highlighting the features most relevant to illegal intrusion activities.

[0094] Exemplarily, continuing with the above scenario, after the security monitoring system obtains the reference vector distribution after the regularized mapping conversion, it will be input into the feature focusing unit for feature focusing processing. The feature focusing unit may use an attention mechanism to automatically learn the weight relationship between different sensor readings. For example, in some cases, abnormal changes in temperature and humidity may be important indicators of illegal intrusion activities, so the feature focusing unit will give these two dimensions a higher weight. In other cases, if the wind speed suddenly increases, it may mean that the window or door is opened, which may also be a sign of intrusion, so the wind speed dimension will also be given a corresponding weight. Through feature focusing processing, the security monitoring system obtains a reference focus vector distribution. Each dimension in this reference focus vector distribution corresponds to a specific sensor reading, and each reading is assigned a weight adjusted according to its importance. In this way, in subsequent classification tasks, more attention can be paid to the features most relevant to illegal intrusion activities to improve the accuracy of classification.

[0095] Step S1213, determining a first embedded representation vector distribution of the reference station house monitoring target based on the reference focusing vector distribution.

[0096] For example, the security monitoring system determines the first embedding representation vector distribution of the reference station monitoring target based on the reference focusing vector distribution. This step may involve further transformation or selection operations on the reference focusing vector distribution to obtain the final embedding representation for classification. The first embedding representation vector distribution is the result of encoding the multidimensional monitoring data. It captures the key information and features in the data and provides effective input for subsequent classification tasks. Through the preprocessing and feature focusing processing of the graph self-attention unit, the security monitoring system successfully converts the original multidimensional monitoring data into the first embedding representation vector distribution that is more suitable for the classification task.

[0097] In a possible implementation, step S122 may include:

[0098] Step S1221, obtaining a reference knowledge network, wherein the reference knowledge network includes multiple reference network members and multiple reference member links, wherein any reference network member is used to reflect an interactive monitoring target or a station house interactive logical space or a station house operation activity in the reference station house interactive scene where the reference station house monitoring target is located, and any reference member link is used to reflect a reference interactive cost parameter or a reference interactive relationship or a reference affiliation, wherein the reference interactive cost parameter is an interactive cost parameter of the interactive monitoring target represented by the reference network member connected to any one of the reference member links, and the reference interactive relationship is used to reflect the station house operation activity represented by another reference network member connected to which an interactive monitoring target represented by a reference network member connected to any one of the reference member links executes or is executed, and the reference affiliation is used to reflect the station house interactive logical space represented by another reference network member connected to which an interactive monitoring target represented by a reference network member connected to any one of the reference member links belongs.

[0099] In this embodiment, the security monitoring system starts to build or access a pre-existing reference knowledge network. This reference knowledge network is a highly structured graph model, in which nodes represent reference network members and edges represent reference member links. Each node (reference network member) may contain rich attribute information, such as type (camera, sensor, door, room, etc.), location (coordinates, floor, area, etc.), state (on, off, active, inactive, etc.) and other type-specific parameters (such as camera resolution, focal length, etc.).

[0100] Edges (reference member links) are used to describe the relationships between nodes. These relationships can be physical (such as a camera monitoring a room), logical (such as an action that needs to be performed after a sensor is triggered), or auxiliary (such as a camera belonging to a monitoring system). Each edge can also be attached with reference interaction cost parameters, which can quantify the cost or effect of interaction between nodes, such as data transmission rate, latency, signal strength, etc.

[0101] By analyzing the network structure, node attributes and edge relationships, the security monitoring system can fully understand the various components in the reference station monitoring environment and the interactions between them.

[0102] Step S1222, for any reference network member, based on the any reference network member, a reference member link connected to the any reference network member, and the reference network members connected to the any reference network member using the reference member link, determine the member embedding representation vector distribution of the any reference network member through the knowledge network embedding representation unit.

[0103] When processing the reference knowledge network, the security monitoring system uses a technology called knowledge network embedding, which can map the nodes (reference network members) and edges (reference member links) in the reference knowledge network into a continuous vector space, thereby obtaining the member embedding representation vector distribution of each node.

[0104] To determine the member embedding representation of a node, the security monitoring system considers the properties of the node, the properties of the edges directly connected to the node, and the properties of the neighboring nodes. The security monitoring system uses an embedding algorithm (such as GraphEmbedding, Node2Vec, etc.) to generate a vector representation for each node that captures its structural and semantic features in the network through a training and optimization process.

[0105] This member embedding representation vector distribution can capture the local and global structure of the node in the network, the role of the node (such as central node, edge node, etc.), and the similarity and relationship between the node and other nodes. By comparing the embedding representations of different nodes, the security monitoring system can infer their similarity or association in the network.

[0106] Step S1223: If any one of the reference network members is used to reflect the reference station house monitoring target, the member embedding representation vector distribution of any one of the reference network members is used as the second embedding representation vector distribution of the reference station house monitoring target.

[0107] After processing the entire reference knowledge network, the security monitoring system will pay special attention to the nodes that directly represent the monitoring targets of the reference station. These nodes may be cameras, sensors and other devices responsible for collecting monitoring data.

[0108] The security monitoring system will select the embedding representations of these specific nodes from the previously generated member embedding representations as the second embedding representation vector distribution of the reference station monitoring target. This second embedding representation vector distribution captures the contextual information of the monitoring target in the network, including other entities directly related to it (such as monitored rooms, triggered sensors, etc.) and the relationship between them.

[0109] In this way, the security monitoring system can utilize the rich information in the knowledge network to enhance the understanding of the monitoring target, rather than relying solely on the limited features extracted from the monitoring data.

[0110] Step S1224, fusing the first embedded representation vector distribution and the second embedded representation vector distribution of the reference station house monitoring target to generate the reference embedded representation vector distribution.

[0111] After obtaining the first embedding representation (encoded from the multidimensional monitoring data) and the second embedding representation (extracted from the knowledge network) of the reference station monitoring target, the security monitoring system fuses the two representations to generate a more comprehensive reference embedding representation vector distribution. The fusion process can be carried out in a variety of ways, such as simple concatenation, weighted averaging, and nonlinear transformation using neural networks. The security monitoring system can choose the most appropriate fusion method based on specific application requirements and experimental results.

[0112] The generated reference embedding representation vector distribution combines the features extracted from the actual monitoring data and the contextual information obtained from the knowledge network, so it has stronger expressiveness and generalization ability. The security monitoring system can use this distribution to perform subsequent illegal intrusion detection, abnormal behavior recognition and other tasks, thereby improving the intelligence level and performance of the monitoring system.

[0113] In a possible implementation, before step S1223, the method further includes:

[0114] Step A110, obtaining a sample knowledge network, wherein the sample knowledge network includes multiple sample network members and multiple sample member links, wherein any sample network member is used to reflect a sample interaction monitoring target or a sample station room interaction logical space or a sample station room operation activity in a sample station room interaction scene, and any sample member link is used to reflect a sample interaction cost parameter or a sample interaction relationship or a sample affiliation relationship, wherein the sample interaction cost parameter is an interaction cost parameter of a sample interaction monitoring target represented by a sample network member connected to any one of the sample member links, and the sample interaction relationship is used to reflect that a sample interaction monitoring target represented by a sample network member connected to any one of the sample member links executes or is executed by a sample station room operation activity represented by another sample network member connected to the sample interaction monitoring target, and the sample affiliation relationship is used to reflect that a sample interaction monitoring target represented by a sample network member connected to any one of the sample member links belongs to a sample station room interaction logical space represented by another sample network member connected to the sample station room interaction target.

[0115] Step A120, using the initial embedding representation unit to embed the sample knowledge network, and generate a member embedding representation vector distribution of each sample network member.

[0116] Step A130, performing network parameter learning on the initial embedding representation unit according to the member embedding representation vector distribution of each sample network member to generate the knowledge network embedding representation unit.

[0117] In this embodiment, the security monitoring system first accesses or constructs a sample knowledge network. This sample knowledge network is similar to the reference knowledge network, but it is built based on historical data or simulation data and is used for the process of training and learning embedded representations. The sample knowledge network is also composed of multiple sample network members and multiple sample member links. Each sample network member can represent an entity in the sample station room interaction scene, such as a sample camera, a sample sensor, or a sample room. The sample member link is used to describe the relationship between these sample network members, such as a sample camera monitoring a sample room. The definitions of sample interaction cost parameters, sample interaction relationships, and sample affiliation relationships are similar to the corresponding concepts in the reference knowledge network, but are applied to sample network members and sample member links.

[0118] Then, the security monitoring system deploys an initial embedding representation unit, which may be an untrained machine learning model, such as a neural network. Its task is to map the sample network members and sample member links in the sample knowledge network into the vector space. The security monitoring system provides the sample knowledge network as input to the initial embedding representation unit. Through a series of calculations and transformations, the initial embedding representation unit generates a member embedding representation vector distribution for each sample network member. This process may involve matrix operations, optimization algorithms, etc. to ensure that the generated embedding representation can capture the structural and semantic characteristics of the sample network members in the network.

[0119] After obtaining the member embedding representation vector distribution of each sample network member, the security monitoring system uses these distributions as supervisory signals to train and optimize the initial embedding representation unit. During this process, the security monitoring system compares the generated embedding representation with the desired output (which may be predefined or obtained by other means) and calculates the loss function. The security monitoring system then uses gradient descent or other optimization algorithms to adjust the network parameters of the initial embedding representation unit to minimize the loss function. This process is iterated until a certain stopping criterion is met (such as loss function convergence, reaching the maximum number of iterations, etc.). Through this process, the initial embedding representation unit gradually learns how to map the sample network members in the sample knowledge network to a meaningful vector space. This trained unit is called a knowledge network embedding representation unit.

[0120] After the above training process, the security monitoring system obtains a well-trained knowledge network embedding representation unit. This knowledge network embedding representation unit has learned to extract useful structural and semantic information from the sample knowledge network and encode this information into a vector representation. Now, the security monitoring system can use this knowledge network embedding representation unit to process new, unseen reference knowledge networks and generate high-quality embedding representations for them. These representations can be used for various downstream tasks such as classification, clustering, anomaly detection, etc., thereby improving the security monitoring system's understanding and response capabilities to the station monitoring environment.

[0121] In a possible implementation, step A130 may include:

[0122] Step A131, restoring the member embedding representation vector distribution of each sample network member to generate a prediction knowledge network.

[0123] In this embodiment, the security monitoring system first obtains the member embedding representation vector distribution of each sample network member after being processed by the initial embedding representation unit. These member embedding representation vector distributions are representations of the sample network members in the vector space, capturing their structural and semantic information in the sample knowledge network.

[0124] In order to verify the quality of the distribution of these member embedding representation vectors, the security monitoring system performs a restoration process. The purpose of this process is to try to reconstruct the original sample knowledge network from the embedding representation. The security monitoring system may use a decoder or similar mechanism that can predict the link relationship, attribute information, etc. between the sample network members based on the embedding representation.

[0125] Through this process, the security monitoring system generates a predictive knowledge network, which is reconstructed based on the embedded representation and restores the structure and semantic information of the original sample knowledge network as much as possible.

[0126] Step A132: determining a knowledge network error parameter based on the sample knowledge network and the predicted knowledge network.

[0127] The security monitoring system now has two knowledge networks: the original sample knowledge network and the predicted knowledge network reconstructed based on the embedded representation. In order to evaluate the quality of the embedded representation, the security monitoring system needs to compare the differences between the two networks. The security monitoring system defines a knowledge network error parameter, which is used to quantify the difference between the sample knowledge network and the predicted knowledge network. This knowledge network error parameter can be calculated based on a variety of indicators, such as differences in network structure, differences in node attributes, differences in link relationships, etc. The security monitoring system calculates the knowledge network error parameter by comparing the attributes and link relationships of corresponding nodes in the two networks. This knowledge network error parameter provides a quantitative indicator for the security monitoring system to evaluate the extent to which the embedded representation captures the information of the original sample knowledge network.

[0128] Step A133, performing network parameter learning on the initial embedding representation unit according to the knowledge network error parameter to generate the knowledge network embedding representation unit.

[0129] After calculating the knowledge network error parameter, the security monitoring system uses this knowledge network error parameter as a feedback signal to adjust the network parameters of the initial embedding representation unit. This process is usually implemented through gradient descent or other optimization algorithms. For example, the security monitoring system calculates the gradient of the knowledge network error parameter with respect to the network parameters of the embedding representation unit, and updates the network parameters based on this gradient to minimize the knowledge network error parameter. This update process may involve multiple iterative steps until a certain stopping criterion is met. Through this learning process, the initial embedding representation unit gradually improves the quality of its embedding representation, so that the predicted knowledge network reconstructed from the embedding representation is closer to the original sample knowledge network.

[0130] After the above network parameter learning process, the security monitoring system obtains an improved embedding representation unit, which can now better capture the structure and semantic information in the sample knowledge network. The security monitoring system named this trained embedding representation unit the knowledge network embedding representation unit. This unit can now be used to process new, unseen reference knowledge networks and generate high-quality embedding representations for them. These embedding representations will be used in subsequent monitoring tasks to improve the security monitoring system's ability to understand and respond to the station environment.

[0131] In a possible implementation, step A130 may further include:

[0132] Step A134, extracting at least one first network member and at least one second network member from the multiple sample network members, any one of the first network members is used to reflect a sample interactive monitoring target, and any one of the second network members is used to reflect a sample station room operation activity.

[0133] In this embodiment, the security monitoring system first accesses the constructed sample knowledge network, which contains multiple sample network members, each of which represents an entity in the station room interaction scene (such as a camera, sensor, door, etc.). The goal of the security monitoring system is to extract two special types of members from these sample network members: first network members and second network members.

[0134] The first network member is selected to reflect the sample interactive monitoring target, such as a specific camera or sensor that is responsible for monitoring a certain important area or activity in the station room. The second network member is selected to reflect the sample station room operation activity, such as the opening or closing of a door, the switching of a light, etc.

[0135] The security monitoring system identifies and extracts at least one first network member and at least one second network member according to the metadata and context information of the sample knowledge network, and prepares for subsequent interactive relationship prediction learning.

[0136] Step A135, for any one of the first network members and any one of the second network members, determine a predicted interaction relationship based on the member embedding representation vector distribution of any one of the first network members and any one of the second network members, wherein the predicted interaction relationship is used to reflect the probability value of the sample interaction monitoring target represented by any one of the first network members executing the sample station room operation activity represented by any one of the second network members.

[0137] The security monitoring system now focuses on a specific first network member (e.g., camera A) and a specific second network member (e.g., the opening operation of door B). The security monitoring system has obtained the member embedding representation vector distributions of these two network members through the initial embedding representation unit. Using these member embedding representation vector distributions, the security monitoring system performs a prediction process to determine the probability value of camera A (the first network member) performing the opening of door B (the operation activity represented by the second network member). This prediction process may involve calculating the similarity between the embedding representations, applying machine learning models, etc.

[0138] The predicted interaction relationship output by the security monitoring system is a probability value that reflects the possibility of the first network member (camera A) performing the operation activity represented by the second network member (door B opened).

[0139] Step A136, obtaining a marked interaction relationship, wherein the marked interaction relationship is used to reflect whether the sample interaction monitoring target represented by any one of the first network members performs the sample station room operation activity represented by any one of the second network members.

[0140] In order to evaluate the accuracy of the predicted interaction relationship, the security monitoring system needs to obtain the annotated interaction relationship. These annotated relationships usually come from human expert annotations or reliable historical data records. In this case, the annotated interaction relationship clearly indicates whether camera A actually monitors the opening operation of door B. This is a binary annotation, either "yes" (indicating that camera A does monitor the opening of door B) or "no" (indicating that there is no monitoring). The security monitoring system retrieves the annotated interaction relationship corresponding to the current first network member and the second network member from the annotated dataset.

[0141] Step A137, determining an interaction relationship error parameter based on the predicted interaction relationship and the labeled interaction relationship.

[0142] The security monitoring system now has two sets of data: one set is the predicted interaction probability value, and the other set is the annotated actual interaction relationship. In order to measure the accuracy of the prediction, the security monitoring system needs to calculate the difference between the two sets of data. For example, the security monitoring system defines an interaction error parameter, which quantifies the difference between the predicted interaction relationship and the annotated interaction relationship. This interaction error parameter may be calculated based on cross entropy loss, mean square error, etc. By comparing the predicted probability value and the actual annotated value, the security monitoring system calculates the interaction error parameter, which will be used in the subsequent network parameter learning process.

[0143] Step A138, performing network parameter learning on the initial embedding representation unit according to the interaction relationship error parameter to generate the knowledge network embedding representation unit.

[0144] In this embodiment, the security monitoring system uses the calculated interaction relationship error parameter as a feedback signal to adjust the network parameters of the initial embedding representation unit. This process is usually implemented by back propagation algorithm and gradient descent optimization. The security monitoring system calculates the gradient of the interaction relationship error parameter with respect to the network parameters of the embedding representation unit, and updates the network parameters according to this gradient to minimize the interaction relationship error parameter. This update process may involve multiple iterative steps until a certain stopping criterion (such as error convergence, reaching the maximum number of iterations, etc.) is met. Through this learning process, the initial embedding representation unit gradually improves the quality of its embedding representation, so that the predicted interaction relationship is closer to the actual labeled interaction relationship. After the above-mentioned network parameter learning process, the security monitoring system obtains an improved embedding representation unit, which can now better capture the interaction relationship information in the sample knowledge network. The security monitoring system names this trained embedding representation unit as a knowledge network embedding representation unit. This unit can now be used to process new and unseen knowledge networks and generate high-quality embedding representations for them. These embedding representations will be used for subsequent monitoring tasks to improve the security monitoring system's understanding and response capabilities to the station environment.

[0145] In a possible implementation manner, the illegal intrusion activity includes at least one of an intrusion activity of damaging equipment and an intrusion activity of illegally entering a sensitive area.

[0146] The intrusion activity label data includes at least one of a first label data and a second label data, wherein the first label data is used to reflect whether an intrusion activity causing equipment damage occurs to the reference station house monitoring target, and the second label data is used to reflect whether an intrusion activity causing illegal entry into a sensitive area occurs to the reference station house monitoring target.

[0147] The intrusion activity identification data includes at least one of a first identification result and a second identification result, the first identification result is used to reflect the probability value of an intrusion activity causing equipment damage to the reference station house monitoring target, and the second identification result is used to reflect the probability value of an intrusion activity illegally entering a sensitive area of ​​the reference station house monitoring target.

[0148] In a large station facility, the security monitoring system is responsible for monitoring various possible illegal intrusion activities. These illegal intrusion activities may include intrusion activities that damage equipment (such as deliberate destruction of cameras, sensors, etc.) and intrusion activities that illegally enter sensitive areas (such as unauthorized entry into the control room, data storage area, etc.). In order to effectively identify these illegal intrusion activities, the security monitoring system uses the previously constructed knowledge network embedding representation unit for processing and analysis.

[0149] The intrusion activity label data includes:

[0150] First label data: This is a binary label that is used to clearly mark whether an intrusion activity that damages the equipment has occurred in the reference station monitoring target. For example, if a camera is deliberately damaged, the first label data will be marked as "yes" or "1".

[0151] Second tag data: Also a binary tag, used to mark whether an unauthorized person has entered a sensitive area. For example, if someone illegally enters the control room, the second tag data will be marked accordingly.

[0152] The intrusion activity identification data includes:

[0153] First recognition result: This is a probability value that indicates the possibility of equipment damage intrusion activity occurring at the reference station monitoring target. For example, based on video analysis and machine learning algorithms, the security monitoring system may output a probability value of 0.8, indicating an 80% probability that equipment damage has occurred.

[0154] Second recognition result: It is also a probability value, indicating the possibility of illegal entry into sensitive areas. The security monitoring system calculates this probability based on multiple sensor data and surveillance videos.

[0155] Step S130 may include:

[0156] Step S131: determining a first error parameter according to the first tag data and the first recognition result.

[0157] Step S132: determining a second error parameter according to the second tag data and the second recognition result.

[0158] Step S133: performing network parameter learning on the illegal intrusion prediction network according to the first error parameter and the second error parameter to generate a target illegal intrusion prediction network.

[0159] In this embodiment, when processing the station monitoring data, the security monitoring system first focuses on the intrusion activity of equipment damage. It has the first label data, which is real and labeled data, and clearly indicates whether there is an intrusion activity of equipment damage at a certain point in time. At the same time, the security monitoring system also generates the first recognition result through the illegal intrusion prediction network, which is a probability value indicating the possibility of the predicted equipment damage intrusion activity.

[0160] To evaluate the accuracy of the prediction, the security monitoring system needs to compare the difference between the first label data and the first recognition result. The security monitoring system defines a first error parameter that quantifies the inconsistency between the predicted result and the true label. This is usually achieved by calculating some distance or difference between the predicted probability and the true label (such as cross entropy loss).

[0161] Similar to intrusions that damage equipment, security monitoring systems also focus on intrusions that illegally enter sensitive areas. It has second label data, which is real, labeled data that clearly indicates whether there is an intrusion that illegally enters sensitive areas at a certain point in time. At the same time, the illegal intrusion prediction network also generates a second recognition result, which indicates the predicted possibility of intrusions that illegally enter sensitive areas. The security monitoring system also needs to compare the difference between the second label data and the second recognition result, and define a second error parameter to quantify this difference. The calculation method of this error parameter is similar to the first error parameter, both of which are to measure the accuracy of the prediction.

[0162] After obtaining the first error parameter and the second error parameter, the security monitoring system now has feedback on the accuracy of the prediction of equipment damage intrusion activities and illegal entry into sensitive areas intrusion activities. The security monitoring system uses these error parameters as guidance to adjust the network parameters of the illegal intrusion prediction network.

[0163] This learning process may involve a back-propagation algorithm, where the security monitoring system calculates the gradient of the error parameter with respect to the network parameters and updates the network parameters based on these gradients to minimize the error parameter. This process may involve multiple iterations until a certain stopping criterion is met (such as error convergence, reaching the maximum number of iterations, etc.).

[0164] After the above network parameter learning process, the illegal intrusion prediction network gradually improves its prediction performance, making the prediction of equipment damage intrusion activities and illegal entry into sensitive areas more accurate. The security monitoring system named this trained and improved illegal intrusion prediction network the target illegal intrusion prediction network. The target illegal intrusion prediction network can now more accurately identify various illegal intrusion activities in the station house, and issue alarms or take other security measures in time to ensure the safety and normal operation of the station house.

[0165] In a possible implementation, the method further includes:

[0166] Step S140, obtaining multi-dimensional monitoring data of the target station house monitoring target.

[0167] Step S150, using the target illegal intrusion prediction network to determine the illegal intrusion prediction result of the target station house monitoring target based on the multi-dimensional monitoring data of the target station house monitoring target, and the illegal intrusion prediction result is used to reflect the probability value of illegal intrusion activities occurring in the target station house monitoring target.

[0168] Step S160, when the illegal intrusion prediction result meets the preset requirements, it is determined that the illegal intrusion activity occurs in the target station house monitoring target.

[0169] For example, the security monitoring system first establishes connections with various monitoring devices in the station room, including but not limited to video cameras, infrared sensors, access control systems, etc. The goal of the security monitoring system is to obtain the multi-dimensional monitoring data generated by these devices to fully understand the real-time situation in the station room. Multi-dimensional monitoring data may include video streams, sensor readings, the switch status of the access control system, etc. For example, the video stream can provide real-time visual information in the station room, the sensor readings can reflect the environmental parameters such as temperature, humidity, pressure, etc. in the station room, and the status of the access control system can show which areas are currently open or closed.

[0170] The security monitoring system receives this data through its network interface and performs necessary preprocessing, such as data cleaning and format conversion, for subsequent analysis and processing.

[0171] After acquiring the multi-dimensional surveillance data, the security monitoring system then processes it using a previously generated target intrusion prediction network that has been trained to recognize various patterns and features associated with intrusion activities.

[0172] The security monitoring system inputs multi-dimensional monitoring data into the target illegal intrusion prediction network, and the network analyzes and processes the data through its internal complex structure. This processing may involve steps such as feature extraction, pattern matching, and probability calculation.

[0173] Finally, the target illegal intrusion prediction network outputs an illegal intrusion prediction result, which is a probability value indicating the possibility of illegal intrusion activity occurring at the target monitoring target. For example, if the output probability value is 0.85, it means that there is an 85% probability that illegal intrusion activity is occurring or is about to occur at the target monitoring target.

[0174] After receiving the intrusion prediction result, the security monitoring system will compare it with a preset threshold. This threshold is set in advance to determine whether the prediction result is reliable enough to trigger further action.

[0175] If the illegal intrusion prediction result is higher than or equal to the preset threshold, the security monitoring system will determine that illegal intrusion has indeed occurred in the target station monitoring target. For example, if the preset threshold is 0.8, and the illegal intrusion prediction result is 0.85, the security monitoring system will determine that illegal intrusion has occurred. Once illegal intrusion is confirmed, the security monitoring system may trigger a series of security measures such as the alarm system, notify security personnel, and record event logs to deal with potential threats.

[0176] Wherein, step S150 may include:

[0177] Step S151, using the target illegal intrusion prediction network to encode the multi-dimensional monitoring data of the target station house monitoring target, and generate a first embedded representation vector distribution of the target station house monitoring target.

[0178] Step S152: determining a reference vector distribution of the target station house monitoring target according to the first embedded representation vector distribution of the target station house monitoring target.

[0179] Step S153: using the target illegal intrusion prediction network to classify and output the reference vector distribution to generate the illegal intrusion prediction result.

[0180] After receiving the multi-dimensional monitoring data of the target station monitoring target, the security monitoring system must first process these raw multi-dimensional monitoring data so that they can be understood and analyzed by the target illegal intrusion prediction network. Multi-dimensional monitoring data may include video streams, sensor readings, access control status and other types of data. Then, the security monitoring system inputs these multi-dimensional monitoring data into the target illegal intrusion prediction network. The first layer of the target illegal intrusion prediction network is usually the encoding layer, which is responsible for encoding these input data. The purpose of encoding is to extract the features of the data and convert these features into a format that can be understood by the network. During the encoding process, the target illegal intrusion prediction network may use various techniques, such as convolutional neural networks (for image data), recurrent neural networks (for sequence data), etc., to capture the spatial and temporal characteristics of the data. After encoding, the multi-dimensional monitoring data is converted into a first embedding representation vector distribution, which captures the key features of the data and is ready for subsequent processing.

[0181] After obtaining the first embedding representation vector distribution, the security monitoring system next needs to determine a reference vector distribution, and then use the classification layer of the target illegal intrusion prediction network to classify and output the differences. The purpose of the classification layer is to determine whether these differences are sufficient to indicate the occurrence of illegal intrusion activities. The classification layer may use various machine learning algorithms, such as support vector machines, decision trees, or neural networks, to make classification decisions based on the input differences. In this process, the classification layer will output a probability value or category label, indicating the possibility of illegal intrusion activities occurring in the target station monitoring target. Ultimately, this probability value or category label is output as an illegal intrusion prediction result. Based on this prediction result, the security monitoring system can decide whether to trigger an alarm, notify security personnel, or take other security measures to deal with potential illegal intrusion activities.

[0182] For example, in a possible implementation, after determining that the target station house monitoring target has the illegal intrusion activity, the method further includes:

[0183] Step S170, determining data of a first gathering activity based on the multi-dimensional monitoring data of the target station room interaction scene and the multi-dimensional monitoring data of the target station room monitoring target, wherein the first gathering activity is a gathering activity in which the target station room monitoring target is involved.

[0184] Step S180, determining the data of the second gathering activity based on the multi-dimensional monitoring data of the target station room interaction scene and the multi-dimensional monitoring data of the candidate station room monitoring target, wherein the candidate station room monitoring target is the station room monitoring target observed to be involved in the second gathering activity in the target station room interaction scene, and the second gathering activity is an existing gathering activity.

[0185] Step S190: determining data of a target gathering activity based on the data of the first gathering activity and the data of the second gathering activity.

[0186] After the security monitoring system determines that the target station monitoring target has an illegal intrusion activity, it needs to further analyze the interaction scene in the station to understand the specific content and context of the illegal intrusion activity. The security monitoring system will access the multi-dimensional monitoring data of the target station interaction scene, which may include video streams, audio recordings, sensor readings, etc., which together depict the real-time situation in the station. At the same time, the security monitoring system will also refer to the multi-dimensional monitoring data of the target station monitoring target collected previously. These data provide direct information about the illegal intrusion activities, such as the movement trajectory and behavior patterns of the intruders. By combining these two sets of data, the security monitoring system can identify the gathering activities (i.e., the first gathering activities) in which the target station monitoring target intervenes. For example, by analyzing the video stream, the security monitoring system may observe that the intruder is communicating with a group or participating in some collective activities. The security monitoring system extracts this information about the first gathering activity to form the data of the first gathering activity.

[0187] In addition to focusing on the gathering activities that the target station monitors, the security monitoring system will also focus on other gathering activities that already exist in the station. To this end, the security monitoring system will analyze the multi-dimensional monitoring data of the target station's interactive scenes to find ongoing gathering activities.

[0188] During this process, the security monitoring system identifies candidate station room monitoring targets - these are station room monitoring targets that are discovered during the observation process and are intervening in existing gathering activities (i.e., second gathering activities). The security monitoring system collects multi-dimensional monitoring data of these candidate station room monitoring targets in order to understand their behavior and context. By analyzing this data, the security monitoring system is able to extract information about the second gathering activity, such as the number of participants, the nature of the activity, and the location where it occurs. This information constitutes the data of the second gathering activity, providing the security monitoring system with a comprehensive view of the existing gathering activities in the station room.

[0189] After collecting data about the first gathering activity and the second gathering activity, the security monitoring system now needs to integrate and analyze this information to determine the final target gathering activity data. For example, the security monitoring system will compare the data of the first gathering activity and the second gathering activity to find similarities and differences between them. This comparison may involve multiple aspects such as the type of activity, the identity of the participants, and the duration of the activity. By comprehensively analyzing these data, the security monitoring system can come up with a comprehensive understanding of the target gathering activity. This target gathering activity may be a combination of the first gathering activity and the second gathering activity, or it may be an extension or evolution of one of the activities. In any case, the security monitoring system will integrate all relevant information about the target gathering activity to form the data of the target gathering activity. These data provide the security monitoring system with in-depth insights into the gathering activities taking place in the station room, which helps the security monitoring system make more accurate decisions and take more effective actions to deal with illegal intrusion activities.

[0190] For example, in a possible implementation manner, after determining that the target station house monitoring target has the illegal intrusion activity, the method further includes:

[0191] Step B110, based on the multi-dimensional monitoring data of the target station room interaction scene and the multi-dimensional monitoring data of the target station room monitoring target, determine the collaborative activity intervention information of the target station room monitoring target, wherein the collaborative activity intervention information is used to reflect the probability value of the target station room monitoring target intervening in the collaborative activity, and the collaborative activity is the intrusion activity to be coordinated.

[0192] Step B120, determining the intrusion weight coefficient of the target monitoring target of the target station house according to the collaborative activity intervention information and the illegal intrusion prediction result, wherein the intrusion weight coefficient is used to reflect the intrusion concern of the target monitoring target of the target station house.

[0193] Step B130, when the intrusion weight coefficient is greater than a preset coefficient, determining that the target station house monitoring target is a collaborative station house monitoring target, and the collaborative station house monitoring target is an observed station house monitoring target to be coordinated.

[0194] In this embodiment, after determining that the target station monitoring target has illegal intrusion activities, the security monitoring system will further analyze whether the target is likely to be involved in a wider range of collaborative activities. This collaborative activity may be a complex intrusion behavior jointly planned or executed by multiple intruders. In order to determine this, the security monitoring system will combine the multi-dimensional monitoring data of the target station interaction scene and the multi-dimensional monitoring data of the target station monitoring target itself. These data include video, audio, sensor readings, etc., which can reveal the real-time situation in the target station and the specific behavior of the target monitoring object. The security monitoring system will analyze these multi-dimensional monitoring data to find signs that the target station monitoring target is engaging in collaborative activities with other individuals or groups. For example, through video analysis, the security monitoring system may observe the target monitoring object communicating with others, passing items, or operating equipment together. These observations will be integrated into collaborative activity intervention information, reflecting the probability value of the target station monitoring target intervening in collaborative activities.

[0195] After obtaining the collaborative activity intervention information and illegal intrusion prediction results, the security monitoring system will further calculate the intrusion weight coefficient of the target station monitoring target. This intrusion weight coefficient is a quantitative indicator used to reflect the security monitoring system's intrusion attention to the target. When calculating the intrusion weight coefficient, the security monitoring system will comprehensively consider the collaborative activity intervention information and illegal intrusion prediction results. If the probability of the target station monitoring target intervening in the collaborative activity is high, and the illegal intrusion prediction results also show a high intrusion risk, then the intrusion weight coefficient of the target will increase accordingly. This calculation process may involve multiple algorithms and models to ensure that the intrusion weight coefficient can accurately reflect the actual threat level of the target station monitoring target.

[0196] Finally, the security monitoring system will compare the calculated intrusion weight coefficient with a preset threshold. This preset coefficient is set based on historical data and security policies to determine whether a station monitoring target is important enough to be given special attention. If the intrusion weight coefficient of the target station monitoring target is greater than the preset coefficient, the security monitoring system will mark it as a collaborative station monitoring target. This means that the target not only has the risk of illegal intrusion itself, but also may collaborate with other intruders to commit crimes, posing a greater threat to the safety of the station. Once the collaborative station monitoring target is determined, the security monitoring system can take more targeted security measures, such as strengthening monitoring, raising the alarm level, or notifying security personnel to intervene.

[0197] Figure 2 The hardware structure of the security monitoring system 100 provided in the embodiment of the present invention for implementing the above-mentioned intelligent station security monitoring method is shown. Figure 2As shown, the security monitoring system 100 may include a processor 110 , a machine-readable storage medium 120 , a bus 130 , and a communication unit 140 .

[0198] In an exemplary design concept, the security monitoring system 100 can be a single security monitoring system or a server group. The server group can be centralized or distributed (for example, the security monitoring system 100 can be a distributed system). In an exemplary design concept, the security monitoring system 100 can be local or remote. For example, the security monitoring system 100 can access data and / or data stored in the machine-readable storage medium 120 via a network. For another example, the security monitoring system 100 can be directly connected to the machine-readable storage medium 120 to access the stored data and / or data. In an exemplary design concept, the security monitoring system 100 can be implemented on a cloud platform. As an example only, the cloud platform can include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, etc. or any combination thereof.

[0199] The machine-readable storage medium 120 can store data and / or instructions. In an exemplary design concept, the machine-readable storage medium 120 can store multidimensional monitoring data obtained from an external terminal. In an exemplary design concept, the machine-readable storage medium 120 can store multidimensional monitoring data and / or instructions used by the security monitoring system 100 to execute or use to complete the exemplary method described in the present invention. In an exemplary design concept, the machine-readable storage medium 120 may include a large-capacity memory, a removable memory, a volatile read-write memory, a read-only memory (ROM), etc. or any combination thereof. Exemplary large-capacity memory may include a disk, an optical disk, a solid-state disk, etc. Exemplary removable memory may include a flash drive, a floppy disk, an optical disk, a memory card, a compressed disk, a tape, etc. Exemplary volatile read-write memory may include a random access memory (RAM). Exemplary RAM may include active random access memory (DRAM), double data rate synchronous active random access memory (DDR SDRAM), passive random access memory (SRAM), thyristor random access memory (T-RAM) and zero capacitance random access memory (Z-RAM), etc. Exemplary read-only memories may include mask read-only memories (MROMs), programmable read-only memories (PROMs), erasable programmable read-only memories (PEROMs), electrically erasable programmable read-only memories (EEPROMs), compact disk read-only memories (CD-ROMs), and digital versatile disk read-only memories, etc. In an exemplary design concept, the machine-readable storage medium 120 may be implemented on a cloud platform. As an example only, a cloud platform may include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, etc., or any combination thereof.

[0200] During the specific implementation process, at least one processor 110 executes computer executable instructions stored in the machine-readable storage medium 120, so that the processor 110 can execute the intelligent station security monitoring method of the above method embodiment. The processor 110, the machine-readable storage medium 120 and the communication unit 140 are connected through the bus 130, and the processor 110 can be used to control the sending and receiving actions of the communication unit 140.

[0201] The specific implementation process of the processor 110 can refer to the various method embodiments executed by the above-mentioned security monitoring system 100. The implementation principles and technical effects are similar, and this embodiment will not be repeated here.

[0202] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer executable instructions are preset. When a processor executes the computer executable instructions, the above-mentioned intelligent station security monitoring method is implemented.

[0203] It should be understood that the above description is for illustrative purposes only and is not intended to limit the scope of the present invention. For those of ordinary skill in the art, various modifications and variations can be made in conjunction with the description of the present invention. However, these modifications and variations will not deviate from the scope of the present invention.

[0204] The basic concepts have been described above. Obviously, for those of ordinary skill in the art who have read this application, the above invention disclosure is only for example and does not constitute a limitation of the present invention. Although not explicitly stated herein, those skilled in the art may make various modifications, improvements and corrections to the present invention. Such modifications, improvements and corrections are suggested in the present invention, so such modifications, improvements and corrections still belong to the spirit and scope of the exemplary embodiments of the present invention.

[0205] At the same time, the present invention uses specific words to describe the embodiments of the present invention. For example, "one embodiment", "an embodiment", and / or "some embodiments" refer to a certain feature, structure or characteristic related to at least one embodiment of the present invention. Therefore, it should be emphasized and noted that "one embodiment" or "an embodiment" or "an alternative embodiment" mentioned twice or more in different positions in this specification does not necessarily refer to the same embodiment. In addition, some features, structures or characteristics in one or more embodiments of the present invention can be appropriately combined.

[0206] In addition, it will be appreciated by those skilled in the art that various aspects of the present invention may be illustrated and described in conjunction with a plurality of patentable categories or situations, including any new and useful process, machine, product, or combination of substances, or any new and useful improvements thereto. Accordingly, various aspects of the present invention may be performed entirely by hardware, entirely by software (including firmware, resident software, microcode, etc.), or by a combination of hardware and software. The above hardware or software may all be referred to as "units," "modules," or "systems." In addition, various aspects disclosed in the present invention may take the form of a computer program product embodied in one or more computer-readable media, wherein computer-readable program code is contained therein.

[0207] A computer readable signal medium may include a propagated data signal containing computer program code, such as in baseband or as part of a carrier wave. Such propagated signals may be in a variety of forms, including electromagnetic, optical, etc., or any suitable combination. A computer readable signal medium may be any computer readable medium other than a computer readable storage medium, which may be coupled to an instruction execution system, device, or apparatus to communicate, propagate, or transmit a program for use. The program code on the computer readable signal medium may be transmitted in conjunction with any suitable medium, including radio, cable, fiber optic cable, RF, or the like, or any combination thereof.

[0208] The computer program code required for the operation of each part of the present invention can be written in any one or more programming languages, including subject-oriented programming languages ​​such as Java, Scala, Smalltalk, Eiffel, JADE, Emerald, C++, C#, VB.NET, Python, etc., conventional procedural programming languages ​​such as C language, Visual Basic, Fortran 2003, Perl, COBOL 2002, PHP, ABAP, active programming languages ​​such as Python, Ruby and Groovy, or other programming languages, etc. The program code can be run entirely on the computer, or run on the computer as an independent software package, or run partially on the computer and partially on the remote computer, or run entirely on the remote computer or server. In the latter case, the remote computer can be connected to the computer in combination with any network form, such as a local area network (LAN) or a wide area network (WAN), or connected to an external computer (e.g., in combination with the Internet), or in a cloud computing environment, or used as a service such as software as a service (SaaS).

[0209] In addition, unless explicitly stated in the claims, the order of the processing elements and sequences described in the present invention, the use of alphanumeric characters, or the use of other names are not intended to limit the order of the processes and methods of the present invention. Although the above disclosure discusses some invention embodiments currently considered useful in conjunction with various examples, it should be understood that such details are only for illustrative purposes, and the attached claims are not limited to the disclosed embodiments. On the contrary, the claims are intended to cover all modifications and equivalent combinations that match the essence and scope of the embodiments of the present invention. For example, although the system components described above can be implemented in conjunction with hardware devices, they can also be implemented in conjunction with software solutions alone, such as installing the described system on an existing server or mobile device.

[0210] Similarly, it should be noted that in order to simplify the description of the present invention disclosure and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the invention, sometimes multiple features are grouped into one embodiment, drawings, or descriptions thereof. Similarly, it should be noted that in order to simplify the description of the present invention disclosure and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the invention, sometimes multiple features are grouped into one embodiment, drawings, or descriptions thereof.

Claims

1. A method for intelligent station security monitoring, characterized in that: The method comprises: Acquire multi-dimensional monitoring data of a reference station house monitoring target and intrusion activity label data of the reference station house monitoring target, wherein the intrusion activity label data is used to reflect whether illegal intrusion activities occur in the reference station house monitoring target; Determine the intrusion activity identification data of the reference station house monitoring target based on the multi-dimensional monitoring data of the reference station house monitoring target by using the illegal intrusion prediction network, wherein the intrusion activity identification data is used to reflect the probability value of the illegal intrusion activity occurring in the reference station house monitoring target; Based on the intrusion activity label data and the intrusion activity identification data, network parameter learning is performed on the illegal intrusion prediction network to generate a target illegal intrusion prediction network, where the target illegal intrusion prediction network is used to determine the probability value of the illegal intrusion activity occurring in the target monitoring target of the target station house; The illegal intrusion prediction network includes a graph self-attention unit and a fully connected layer; The method of using the illegal intrusion prediction network to determine the intrusion activity identification data of the reference station monitoring target based on the multi-dimensional monitoring data of the reference station monitoring target includes: Encoding the multi-dimensional monitoring data of the reference station house monitoring target by using the graph self-attention unit to generate a first embedding representation vector distribution of the reference station house monitoring target; Determining a reference embedded representation vector distribution of the reference station house monitoring target according to the first embedded representation vector distribution of the reference station house monitoring target; Using the fully connected layer to classify and output the reference embedding representation vector distribution to generate the intrusion activity identification data; The step of determining the reference embedded representation vector distribution of the reference station house monitoring target based on the first embedded representation vector distribution of the reference station house monitoring target comprises: Acquire a reference knowledge network, the reference knowledge network comprising a plurality of reference network members and a plurality of reference member links, any one of the reference network members being used to reflect an interactive monitoring target or a station house interactive logical space or a station house operation activity in a reference station house interactive scene where the reference station house monitoring target is located, any one of the reference member links being used to reflect a reference interactive cost parameter or a reference interactive relationship or a reference affiliation, the reference interactive cost parameter being an interactive cost parameter of an interactive monitoring target represented by a reference network member connected to any one of the reference member links, the reference interactive relationship being used to reflect an interactive monitoring target represented by a reference network member connected to any one of the reference member links executing or being executed by another connected reference network member represented by another connected reference network member, the reference affiliation being used to reflect an interactive monitoring target represented by a reference network member connected to any one of the reference member links belonging to a connected station house interactive logical space represented by another connected reference network member; For any reference network member, determining a member embedding representation vector distribution of the any reference network member through a knowledge network embedding representation unit according to the any reference network member, a reference member link connected to the any reference network member, and reference network members connected to the any reference network member through the reference member link; If any one of the reference network members is used to reflect the reference station house monitoring target, using the member embedding representation vector distribution of any one of the reference network members as the second embedding representation vector distribution of the reference station house monitoring target; The first embedding representation vector distribution and the second embedding representation vector distribution of the reference station house monitoring target are fused to generate the reference embedding representation vector distribution.

2. The intelligent station house security monitoring method according to claim 1 is characterized in that: The graph self-attention unit includes a preprocessing unit and a feature focusing unit; The method of encoding the multi-dimensional monitoring data of the reference station house monitoring target by using the graph self-attention unit to generate a first embedded representation vector distribution of the reference station house monitoring target includes: Using the preprocessing unit to perform regularized mapping conversion on the multi-dimensional monitoring data of the reference station house monitoring target to generate a reference vector distribution of the reference station house monitoring target; Using the feature focusing unit to perform feature focusing processing on the reference vector distribution to generate a reference focusing vector distribution of the reference station monitoring target; A first embedded representation vector distribution of the reference station house monitoring target is determined based on the reference focusing vector distribution.

3. The intelligent station house security monitoring method according to claim 1 is characterized in that: Before the step of using the member embedding representation vector distribution of any one of the reference network members as the second embedding representation vector distribution of the reference station house monitoring target, the method further includes: Acquire a sample knowledge network, wherein the sample knowledge network includes a plurality of sample network members and a plurality of sample member links, wherein any one of the sample network members is used to reflect a sample interaction monitoring target or a sample station room interaction logical space or a sample station room operation activity in a sample station room interaction scene, and any one of the sample member links is used to reflect a sample interaction cost parameter or a sample interaction relationship or a sample affiliation relationship, wherein the sample interaction cost parameter is an interaction cost parameter of a sample interaction monitoring target represented by a sample network member connected to the any one of the sample member links, and the sample interaction relationship is used to reflect that a sample interaction monitoring target represented by a sample network member connected to the any one of the sample member links executes or is executed by another sample station room operation activity represented by another sample network member connected to the sample interaction monitoring target represented by a sample network member connected to the any one of the sample member links belongs to a sample station room interaction logical space represented by another sample network member connected to the sample interaction monitoring target represented by a sample network member connected to the any one of the sample member links; Embedding the sample knowledge network using an initial embedding representation unit to generate a member embedding representation vector distribution of each sample network member; According to the member embedding representation vector distribution of each sample network member, network parameter learning is performed on the initial embedding representation unit to generate the knowledge network embedding representation unit.

4. The intelligent station house security monitoring method according to claim 3 is characterized in that: The method of performing network parameter learning on the initial embedding representation unit according to the member embedding representation vector distribution of each sample network member to generate the knowledge network embedding representation unit includes: Restoring the member embedding representation vector distribution of each sample network member to generate a prediction knowledge network; Determining a knowledge network error parameter based on the sample knowledge network and the predicted knowledge network; Network parameter learning is performed on the initial embedding representation unit according to the knowledge network error parameter to generate the knowledge network embedding representation unit.

5. The intelligent station house security monitoring method according to claim 3 is characterized in that: The method of performing network parameter learning on the initial embedding representation unit according to the member embedding representation vector distribution of each sample network member to generate the knowledge network embedding representation unit includes: Extracting at least one first network member and at least one second network member from the plurality of sample network members, wherein any one of the first network members is used to reflect a sample interactive monitoring target, and any one of the second network members is used to reflect a sample station house operation activity; For any one of the first network members and any one of the second network members, a predicted interaction relationship is determined based on the member embedding representation vector distribution of any one of the first network members and any one of the second network members, wherein the predicted interaction relationship is used to reflect the probability value of the sample interaction monitoring target represented by any one of the first network members executing the sample station room operation activity represented by any one of the second network members; Acquire a marked interaction relationship, where the marked interaction relationship is used to reflect whether the sample interaction monitoring target represented by any one of the first network members performs the sample station room operation activity represented by any one of the second network members; Determine an interaction relationship error parameter based on the predicted interaction relationship and the labeled interaction relationship; perform network parameter learning on the initial embedding representation unit based on the interaction relationship error parameter to generate the knowledge network embedding representation unit.

6. The intelligent station house security monitoring method according to claim 1, characterized in that: The illegal intrusion activity includes at least one of an intrusion activity of damaging equipment and an intrusion activity of illegally entering a sensitive area; The intrusion activity label data includes at least one of first label data and second label data, wherein the first label data is used to reflect whether the reference station monitoring target has any intrusion activity that damages equipment, and the second label data is used to reflect whether the reference station monitoring target has any intrusion activity that illegally enters a sensitive area; The intrusion activity identification data includes at least one of a first identification result and a second identification result, wherein the first identification result is used to reflect the probability value of an intrusion activity that damages equipment in the reference station monitoring target, and the second identification result is used to reflect the probability value of an intrusion activity that illegally enters a sensitive area in the reference station monitoring target; The step of performing network parameter learning on the illegal intrusion prediction network based on the intrusion activity label data and the intrusion activity identification data to generate a target illegal intrusion prediction network includes: Determining a first error parameter according to the first tag data and the first recognition result; determining a second error parameter according to the second tag data and the second recognition result; Based on the first error parameter and the second error parameter, network parameter learning is performed on the illegal intrusion prediction network to generate a target illegal intrusion prediction network.

7. The intelligent station house security monitoring method according to claim 1, characterized in that: The method further comprises: Obtain multi-dimensional monitoring data of the target station monitoring target; Determine the illegal intrusion prediction result of the target station house monitoring target according to the multi-dimensional monitoring data of the target station house monitoring target by using the target illegal intrusion prediction network, wherein the illegal intrusion prediction result is used to reflect the probability value of illegal intrusion activity occurring in the target station house monitoring target; When the illegal intrusion prediction result meets the preset requirements, determining that the illegal intrusion activity occurs in the target station monitoring target; Wherein, the use of the target illegal intrusion prediction network to determine the illegal intrusion prediction result of the target station monitoring target according to the multi-dimensional monitoring data of the target station monitoring target includes: Encoding the multi-dimensional monitoring data of the target station house monitoring target by using the target illegal intrusion prediction network to generate a first embedded representation vector distribution of the target station house monitoring target; Determine the reference vector distribution of the target station house monitoring target according to the first embedded representation vector distribution of the target station house monitoring target; The target illegal intrusion prediction network is used to classify and output the reference vector distribution to generate the illegal intrusion prediction result.

8. A security monitoring system, characterized in that: The security monitoring system includes a processor and a machine-readable storage medium, wherein the machine-readable storage medium stores machine-executable instructions, and the machine-executable instructions are loaded and executed by the processor to implement the intelligent station security monitoring method of any one of claims 1-7.

Images