An internet of things network situation deduction method based on knowledge integration

Abstract

The application provides a knowledge integration-based Internet of Things network situation deduction method. The method comprises the following steps: constructing a network time sequence knowledge graph according to entity concepts, resource mapping and data attributes of the Internet of Things; constructing a historical evolution space-time graph sequence based on the network time sequence knowledge graph, and obtaining a space-time evolution factor of a node in the Internet of Things through a space-time graph encoder and a historical information embedder according to the historical evolution space-time graph sequence; designing a historical evolution encoder according to the space-time evolution factor of the node, and solving a network space-time evolution factor through the historical evolution encoder; and perceiving a network situation of the Internet of Things at the next moment through a situation prediction decoder according to the network space-time evolution factor. The method constructs an Internet of Things time sequence knowledge graph, effectively integrates network information of the Internet of Things, further constructs a network space-time graph sequence, and cooperatively designs a historical evolution encoder and a situation prediction decoder to perceive network situation information at a future moment.

Description

Technical Field

[0001] This invention relates to the field of network data analysis technology in Internet of Things (IoT) scenarios, and in particular to a method for IoT network situational estimation based on knowledge integration. Background Technology

[0002] In recent years, the Internet of Things (IoT) and mobile technology have developed rapidly, leading to a dramatic increase in the scale of IoT networks. Massive numbers of devices communicate through these networks, completing tasks in various IoT scenarios independently or collaboratively. With the advancement of intelligent application development, the "Internet of Everything" has been initially achieved, and the focus of future work will shift from "infrastructure construction" to "monitoring and maintenance." Faced with the ever-present threat of network failures, researchers have proposed a series of fault detection methods and operation and maintenance strategies for different application scenarios. However, existing research lacks the integration and utilization of massive, multi-source, heterogeneous information from the IoT, and also lacks the ability to perceive the network status holistically. Given the complex network structure of the IoT, how to integrate dynamic network information based on network topology and configuration, design network situational awareness models, and effectively detect network anomalies is a crucial issue in network operation and maintenance, and a key technical challenge for intelligent operation and maintenance technology in large-scale IoT applications.

[0003] Large-scale Internet of Things (IoT) networks are characterized by: highly variable network environments, complex topologies, and difficulties in network status perception. The emergence of numerous new network services within the IoT, coupled with its continuously expanding scale, complex network topologies, massive traffic data, distributed network device deployments, and dynamic interactions of service requests, results in a rapidly changing network environment, making network maintenance challenging.

[0004] Therefore, how to integrate network information, design anomaly detection models, and perceive network anomalies in real time is an urgent problem to be solved. Summary of the Invention

[0005] The embodiments of the present invention provide a method for extrapolating the network situation of the Internet of Things based on knowledge integration, so as to effectively perceive network situation information at future moments.

[0006] To achieve the above objectives, the present invention adopts the following technical solution.

[0007] A network situational estimation method for the Internet of Things (IoT) based on knowledge integration, comprising:

[0008] Construct a network temporal knowledge graph based on the entity concepts, resource mapping, and data attributes of the Internet of Things;

[0009] Based on the network temporal knowledge graph, a historical evolution spatiotemporal graph sequence is constructed. Based on the historical evolution spatiotemporal graph sequence, the spatiotemporal evolution factors of nodes in the Internet of Things are obtained through a spatiotemporal graph encoder and a historical information embedder.

[0010] A historical evolution encoder is designed based on the spatiotemporal evolution factors of the nodes, and the spatiotemporal evolution factors of the network are solved through the historical evolution encoder.

[0011] Based on the aforementioned network spatiotemporal evolution factors, the situation prediction decoder perceives the network situation of the Internet of Things at the next moment.

[0012] Preferably, the construction of a network temporal knowledge graph based on the entity concepts, resource mappings, and data attributes of the Internet of Things includes:

[0013] Design a network temporal knowledge graph with a three-layer architecture, including: Conceptual layer G basic Resource Mapping Layer G map and data attribute layer G data The concept layer G basic The resource mapping layer G is used to collect network entity concepts and their relationships. Different entities have different semantic relationships. map The data attribute layer G is used to integrate device attribute information, status information, control information, and topology information. data A dimension-attribute data model was designed, which combines network dynamic data flow information and external supporting data from two dimensions: time information and spatial information, to form a unified network dynamic data flow attribute format.

[0014] Preferably, the resource mapping layer G map Network resources are characterized from four aspects: device attribute information, status information, control information, and topology information. Device attribute information includes static information and dynamic information. Static attribute information is used to store fixed information of the device, including device unique identifier, device name, device type, device model, device unified code, manufacturer, device physical size, device purchase time, and device usage mode. The dynamic information includes device energy status, historical sensing data, current sensing data, current sensing time, historical device location, and current device location.

[0015] The equipment status information includes equipment performance status, equipment usage status, and equipment fault status. The equipment usage status attribute values ​​have 7 possible values, including scheduled, idle, half-load, full-load, overload, maintenance, and decommissioning.

[0016] The device's control information includes access control information for the device's Internet of Things (IoT) access and monitoring information. The access control information includes the device's interface information, communication method, communication protocol, return value type, and parameter information. The monitoring information includes the latitude and longitude of the device's monitoring, the monitoring coverage area, and the monitoring object information.

[0017] The topology information of a device includes the physical topology connection information of the device and the dependency information between devices. The dependency information between devices includes concurrency relationships, selection relationships, predecessor / successor relationships, control relationships and support relationships.

[0018] Preferably, the data attribute layer G data The dimensional-attribute data model of the design encapsulates the data flow on network nodes into a two-level information tuple. In the first dimension, each tuple consists of time (Inf). temporal and space Inf spatial Represented in two dimensions, each dimension in the second layer of attributes has multiple attribute information, including time information Inf temporal It includes timestamp information, node status information, and data stream density. Timestamp information is used to indicate whether the current time is a weekday, weekend, or holiday. Node status information includes busy, idle, or interrupted. Data stream density information is used to record the local density, global density, and fluctuation characteristics of the data stream transmitted by the current network node. Data density is used to describe the load of the data stream. Local density represents the data transmitted by a single data stream in a unit interval. Global density represents the density of all data streams. Data stream fluctuation is used to record the rate of change of the data stream per unit time.

[0019] Spatial Information Inf spatial The network is described from four dimensions: relative location information, system location information, physical location information, and environmental information. Relative location information records the network location, i.e., the subnet, system, slice, or module to which the current node belongs. System location indicates whether the node belongs to a collection point or a collection point. Physical location indicates the actual environmental information, i.e., whether it belongs to a computer room or a rack. Environmental attributes indicate the density of equipment, environmental conditions, and climate conditions.

[0020] Preferably, the step of constructing a historical evolutionary spatiotemporal graph sequence based on the network temporal knowledge graph, and obtaining the spatiotemporal evolution factors of nodes in the Internet of Things through a spatiotemporal graph encoder and a historical information embedder based on the historical evolutionary spatiotemporal graph sequence, includes:

[0021] Based on knowledge graphs, a spatiotemporal graph sequence of network historical evolution is constructed. A spatiotemporal graph encoder is designed to embed the spatiotemporal graph, and a historical information embedder is designed to learn the evolutionary rules of the historical spatiotemporal graph.

[0022] Given a network temporal knowledge graph subG, represent subG as {V,A,D,A}. failed}, where V represents devices and network nodes, A represents the adjacency matrix, and D represents the node index matrix. failed The fault matrix representing a node is used as a reference. The knowledge graph information within the corresponding time slice is extracted to form the spatiotemporal graph of the current time slice. All spatiotemporal graphs are combined to obtain the historical evolution spatiotemporal graph sequence corresponding to subG, denoted as g={subG1,subG2,…,subG…}. T}, where the spacetime graph subG at time t t This reflects the overall state of network interaction at time t. Represents node v i and v j The connection between them Represents node v i Data, Represents node v i The operating status is indicated by 1 for normal operation and 0 for fault.

[0023] Based on the aforementioned historical evolution spatiotemporal graph sequence, a spatiotemporal graph encoder is designed. This encoder is then embedded into the spatiotemporal graph at each historical moment, where the spatiotemporal graph at time t is subG. t The adjacency matrix is ​​A t The node feature matrix is ​​X t The encoding structure of a variational graph autoencoder is used to process subG. t , obtain subG t eigenvector Z t :

[0024]

[0025]

[0026]

[0027] Among them, GCN μ and GCN σ These represent two graph convolutional network layers, with feature matrix X. t Represents the spacetime graph subG t Features of all nodes in X t The i-th row represents node v i The feature embeddings, μ and σ, represent the mean vector matrix and variance vector matrix of the low-dimensional vector representation of the nodes. yes The corresponding degree matrix, I is the identity matrix, W μ0 With W σ0 The two graph convolutional networks are identical, meaning they share the weights of the first layer. Represents node v i The embedding features at time t yield the spatial features of each node at time t. Each spatiotemporal graph is input into the spatiotemporal graph encoder to obtain the spatial feature vector of each node at each time step, node v i The spatial feature vectors at each time step are:

[0028] A historical information embedder is designed based on the spatial feature vectors of nodes. The historical information embedder consists of a stacked long short-term memory network (LSTM) and an attention mechanism. Given a node v, ... i Corresponding historical spatial feature sequence Stacked LSTMs are used to capture the long-term dependencies in the historical state evolution of each node, specifically the hidden states of the k-th layer of the LSTM. The output is determined by the k-level hidden states from the previous time step. and the hidden state of layer k-1 at the current moment. Together, we determine that the hidden value of the last LSTM layer is... The hidden value sequence of the LSTM is weighted using an attention mechanism to obtain the node v. i Spatiotemporal evolution factor at time t The spatiotemporal evolution factors of all nodes at time t are obtained using the method described above.

[0029] Preferably, the step of designing a historical evolution encoder based on the spatiotemporal evolution factors of nodes, and solving for the network spatiotemporal evolution factors through the historical evolution encoder, includes:

[0030] A historical evolution encoder is designed based on the spatiotemporal evolution factors of nodes. This encoder includes a spatiotemporal graph encoder and a historical information embedder. The spatiotemporal graph encoder encodes the input historical evolution spatiotemporal graph sequence *g* to obtain the spatial features of nodes at different times. The historical information embedder includes a stacked LSTM network and an attention mechanism. It captures spatiotemporal dynamic factors reflecting network evolution and learns deep encoding mapping relationships based on the spatiotemporal encoder. |V| represents the number of nodes, and d represents the dimension of the vector. The historical evolutionary spatiotemporal graph sequence g is mapped to the hidden layer space vector Φ. <Z1,Z2,…,Z t >, in Z t middle, Representation of graph subG t The spatiotemporal characteristics of the i-th node are captured by a historical information embedder to obtain the spatiotemporal dynamic factor H of the network's dynamic evolution. t .

[0031] Preferably, the step of sensing the network situation of the Internet of Things at the next moment through a situation prediction decoder based on the network spatiotemporal evolution factors includes:

[0032] Based on the spatiotemporal evolution factor H t Design a situation prediction decoder, which includes a network topology prediction decoder f decE (H t ), Data Indicator Prediction Decoder f decD (H t ) and node failure predictor;

[0033] Network topology prediction decoder f decE (H t Based on the spatiotemporal evolution factor, predict the probability of edges existing between nodes, and predict the state of the topological connection graph at the next time step. Achieve dynamic awareness of network topology:

[0034]

[0035]

[0036] Where, f decE (H t The graph is reconstructed by calculating the probability that an edge exists between any two nodes in the graph. node v at time t+1 i and v j The probability that an edge exists. Represents node v i The spatiotemporal evolution factor at time t, Represents node v j The spatiotemporal evolution factor at time t, where T represents the matrix transpose;

[0037] Based on data indicator decoder f decD (H t Design a multilayer rectified linear network, and input the spatiotemporal evolution factor of each node into the rectified linear network. The rectified linear network outputs predicted node metrics for node v. i The input of the first layer of the rectified linear network is The solution for layer l is as follows:

[0038]

[0039] Where ReLU is the rectified linear unit function, W l and b l These are the weight matrix and bias matrix of the l-th ReLU layer, respectively;

[0040] The output of the last layer of the rectified linear network is For the predicted node v i Data index values ​​at time t+1;

[0041] The node fault predictor consists of a convolutional classifier, and the input data of the convolutional classifier is each node v. i Spatiotemporal dynamic factors The output data is the node's running status at time t+1. Identify abnormal network nodes based on their operational status;

[0042] Based on the topology connection state diagram of each node at the next time step Data indicator values and abnormal network nodes Obtain the network state of the Internet of Things in the next moment. This network status includes network topology, network node data, and network node operational status.

[0043] As can be seen from the technical solutions provided by the embodiments of the present invention above, the method of the present invention effectively integrates IoT network information by constructing an IoT temporal knowledge graph, thereby constructing a network spatiotemporal graph sequence, and collaboratively designing a historical evolution encoder and a situation prediction decoder to perceive network situation information at future moments.

[0044] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and will become apparent from the description or may be learned by practice of the invention. Attached Figure Description

[0045] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 A flowchart illustrating a knowledge integration-based IoT network situational estimation method provided in this embodiment of the invention;

[0047] Figure 2 A conceptual layer diagram of a network temporal knowledge graph provided in an embodiment of the present invention;

[0048] Figure 3 This is a schematic diagram of a resource mapping layer for a network temporal knowledge graph, provided as an embodiment of the present invention.

[0049] Figure 4This is a schematic diagram of the data attribute layer of a network temporal knowledge graph provided in an embodiment of the present invention;

[0050] Figure 5 This is a schematic diagram of a historical evolution encoder and situation prediction decoder framework provided in an embodiment of the present invention. Detailed Implementation

[0051] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0052] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or couplings. The term “and / or” as used herein includes any and all combinations of one or more of the associated listed items.

[0053] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as herein.

[0054] To facilitate understanding of the embodiments of the present invention, the following will provide further explanation and description with reference to the accompanying drawings and several specific embodiments. These embodiments do not constitute a limitation on the embodiments of the present invention.

[0055] This invention proposes a method for predicting the situation of an Internet of Things (IoT) network based on knowledge integration. To break down barriers between heterogeneous data sources, integrate fragmented multi-dimensional information from the IoT, and effectively reconstruct the original network environment, this invention constructs a temporal knowledge graph of the IoT network. Based on this, a historical evolutionary spatiotemporal graph sequence is built, and a historical evolution encoder and a situation prediction decoder are jointly designed to model the spatiotemporal dynamic factors affecting network state evolution, enabling the prediction of network topology changes, node data variations, and node operational failures.

[0056] The processing flow of the IoT network situational estimation method based on knowledge integration provided by this invention is as follows: Figure 1 As shown, the processing steps include the following:

[0057] Step S1: Construct a network temporal knowledge graph based on the network's entity concepts, resource mappings, and data attributes.

[0058] First, a network temporal knowledge graph with a three-layer architecture was designed, including: a concept layer G. basic Resource Mapping Layer G map and data attribute layer G data It is used to effectively organize multi-source heterogeneous information in Internet of Things (IoT) networks.

[0059] Figure 2 A conceptual layer diagram of a network temporal knowledge graph provided in an embodiment of the present invention, such as... Figure 2 As shown, the first layer of the network temporal knowledge graph is the concept layer G. basic It is responsible for collecting network entity concepts and their relationships, with different entities having different semantic relationships. Entities include subnets, switches, routers, gateways, devices, and sensors, etc. Figure 3 This is a schematic diagram of a resource mapping layer for a network temporal knowledge graph provided in an embodiment of the present invention, such as... Figure 3 As shown, the second layer of the network temporal knowledge graph is the resource mapping layer G. map It is responsible for integrating the device's attribute information, status information, control information, and topology information. Figure 4 This is a schematic diagram of the data attribute layer of a network temporal knowledge graph provided in an embodiment of the present invention, such as... Figure 4 As shown, the third layer of the network temporal knowledge graph is the data attribute layer G. data A "dimension-attribute" data model was designed, which combines network dynamic data flow information and external supporting data from two dimensions: time information and spatial information, to form a unified network dynamic data flow attribute format.

[0060] For resource mapping layer G mapThis invention characterizes network resources from four aspects: device attribute information, status information, control information, and topology information. (1) Device attribute information includes static information and dynamic information. Static attribute information is used to store fixed information of the device, including the device's unique identifier, device name, device type, device model, device unified code, manufacturer, device physical dimensions (length, width, height), device purchase time, and device usage mode. Dynamic information includes the device's energy status, historical sensing data, current sensing data, current sensing time, historical device location, and current device location. (2) For device status information, it is described from three aspects: device performance status, device usage status, and device fault status. Among them, the attribute values ​​of device usage status are 7 types, including scheduled, idle, half-loaded, full-loaded, overloaded, under maintenance, and decommissioned. (3) For device control information, it is described from two dimensions: access control information for device access to the Internet of Things and device monitoring information. Access control information includes device interface information, communication method, communication protocol, return value type, and parameter information. Some devices have monitoring functions, such as sensors. Their monitoring information includes the latitude and longitude of the device being monitored, the monitoring coverage area, and the information of the monitored objects; (4) For device topology information, it is described from the physical topology connection information of the devices and the dependency information between devices. The dependency information between devices includes concurrency relationship, selection relationship, predecessor / successor relationship, control relationship, and support relationship.

[0061] For the data attribute layer G data This invention combines network dynamic data flow information with external support data to construct a unified network dynamic data flow attribute format and designs a "dimension-attribute" data model, which records the spatiotemporal characteristics of the data flow in detail from the two dimensions of time and space.

[0062] The "dimensional-attribute" data model encapsulates data streams on network nodes into a two-level information tuple. In the first level, the "dimension," each tuple is defined by its time interval (Inf). temporal and space Inf spatial It is represented using two-dimensional information. In the second layer, "attributes," each dimension has multiple attribute information. Specifically, for time information Inf temporal The data flow density is described from three dimensions: timestamp information, node status information, and data flow density. Timestamp information indicates whether the current time is a weekday, weekend, or holiday. Node status information includes whether it is busy, idle, or interrupted. Data flow density information records the local density, global density, and fluctuation characteristics of the data flow transmitted by the current network nodes. Data density describes the load of the data flow; local density represents the data transmitted by a single data flow per unit interval, while global density represents the density of all data flows. Data flow fluctuation records the rate of change of the data flow per unit time; for spatial information Inf... spatialThe network location is described from four dimensions: relative location information, system location information, physical location information, and environmental information. Relative location information records the detailed network location, i.e., the subnet, system, slice, or module to which the current node belongs. System location indicates whether the node belongs to a data collection point or a data aggregation point. Physical location indicates the actual environmental information, i.e., whether it belongs to a data center or a rack, etc. Environmental attributes indicate the density of equipment, environmental conditions, and climate conditions.

[0063] Step S2: Construct a historical evolution spatiotemporal graph sequence based on the network temporal knowledge graph.

[0064] The aforementioned network temporal knowledge graph integrates multi-source heterogeneous information in the IoT network scenario, effectively reconstructing the network environment. This invention focuses on modeling the spatiotemporal factors affecting network evolution and designs a network historical evolution encoder. First, a spatiotemporal graph sequence of network historical evolution is constructed based on the knowledge graph, and a spatiotemporal graph encoder is designed to embed the spatiotemporal graph. Then, a historical information embedder is designed to learn the evolutionary rules of the historical spatiotemporal graph.

[0065] 1) Constructing a historical evolutionary spatiotemporal diagram sequence:

[0066] Given a network temporal knowledge graph subG, represent it as {V,A,D,A}. failed}, where V represents devices and network nodes, A represents the adjacency matrix, and D represents the node index matrix. failed This represents the fault matrix of a node. Using a fixed time slice as a baseline, the knowledge graph information within that time slice is extracted to form the spatiotemporal graph for the current time slice. All spatiotemporal graphs are combined to obtain the historical evolution spatiotemporal graph sequence corresponding to the network temporal knowledge graph subG, denoted as g={subG1,subG2,…,subG…}. T}, where the spacetime graph subG at time t t This reflects the overall state of network interaction at time t. Represents node v i and v j The connection between them Represents node v i Data, Represents node v i The operating status (1 represents normal, 0 represents fault).

[0067] 2) Spatiotemporal graph encoder:

[0068] A spatiotemporal graph encoder is designed based on the historical evolution spatiotemporal graph sequence, and the encoder is embedded into the spatiotemporal graph at each historical moment. For the spatiotemporal graph subG at time t... t Its adjacency matrix is ​​A t Its node feature matrix is ​​X t(This can be obtained using the knowledge graph representation learning tool TransE). The subG encoding structure is processed using a variational graph autoencoder. t , obtain subG t Latent vector (spatial feature) Z t :

[0069]

[0070]

[0071]

[0072] Where μ and σ represent the mean vector matrix and variance vector matrix of the low-dimensional vector representation of the nodes, respectively. yes The corresponding degree matrix, I is the identity matrix, W μ0 With W σ0 The two graph convolutional networks are identical, meaning they share the weights of the first layer. With the mean and variance, a unique multidimensional Gaussian distribution can be determined, from which node embeddings can be obtained through sampling. Represents node v i The embedding features (spatial features) at time t can then be used to obtain the spatial features of each node at time t. Each spatiotemporal graph is input into the spatiotemporal graph encoder to obtain the spatial feature vector of each node at each time step, such as node v. i Spatial feature vectors at various times

[0073] 3) Historical information embedder:

[0074] Based on the spatial feature vectors of nodes, a historical information embedder is designed to learn the evolutionary patterns of spatiotemporal graph sequences, that is, to learn the evolutionary trends of each node. The historical information embedder consists of a stacked long short-term memory network and an attention mechanism. Specifically, given a node v... i Corresponding historical spatial feature sequence Stacked LSTM (Long Short-Term Memory) networks are used to capture the long-term dependencies in the historical state evolution of each node, specifically the hidden states of its k-th layer. The output is determined by the k-level hidden states from the previous time step. and the hidden state of layer k-1 at the current moment. The decision is made jointly. This applies to the hidden values ​​of the last LSTM layer. We use an attention mechanism to weight its latent value sequence in order to obtain v. i Spatiotemporal evolution factor at time t The spatiotemporal evolution factors of all nodes at time t can be obtained using the method described above.

[0075] Step S3: Design a history evolution encoder based on the spatiotemporal evolution factors of nodes, and solve the spatiotemporal evolution factors of the network through the history evolution encoder.

[0076] Figure 5 This is a schematic diagram of a historical evolution encoder and a situation prediction decoder framework provided in an embodiment of the present invention. The historical evolution encoder includes two main modules, such as... Figure 5 As shown on the left, one is a spatiotemporal graph encoder, and the other is a historical information embedder. First, the spatiotemporal graph sequence g of the historical evolution is encoded by the spatiotemporal graph encoder to obtain the spatial features of nodes at different times. Then, a historical information embedder is designed, which includes a stacked LSTM network and an attention mechanism. The historical information embedder captures spatiotemporal dynamic factors reflecting the network evolution. During this process, deep encoding mapping relationships are learned based on the spatiotemporal encoder. |V| represents the number of nodes, and d represents the dimension of the vector, thus mapping g to the hidden space vector Φ = <Z1,Z2,…,Z t >, in Z t middle, Representation of graph subG t The spatiotemporal characteristics of the i-th node are then used to capture the spatiotemporal dynamic factor H of the network's dynamic evolution based on a historical information embedder. t .

[0077] An encoder structure (graph convolutional network) using a variational graph autoencoder is used to process the spatiotemporal graph sequence g to obtain the hidden layer spatial vector Φ. Φ can be understood as the spatial feature matrix of each node at each time step. For any node v in the graph... i Its historical spatiotemporal characteristic sequence is Processing with stacked long short-term memory (LSTM) networks Get node v i Spatiotemporal dynamic factor at time t The time-dynamic factors of all nodes at time t can be obtained using the method described above.

[0078] Step S4: Based on the network spatiotemporal evolution factors, the situation prediction decoder perceives the network situation of the Internet of Things at the next moment.

[0079] This invention designs a situation prediction decoder to infer the network situation of the Internet of Things (IoT) in the next moment. This network situation includes network topology, network node data, and network node operational status. For example... Figure 5(Right) The situation prediction decoder consists of a network topology prediction decoder, a data indicator prediction decoder, and a node fault predictor. The network topology prediction decoder predicts the state of the topology connection diagram at the next moment based on spatiotemporal dynamic factors. The data indicator prediction decoder predicts the data indicator values ​​of each node at the next time step based on spatiotemporal dynamic factors. The node fault predictor predicts abnormal network nodes in the next time step based on spatiotemporal dynamic factors. Making predictions. The joint learning of these three methods can effectively predict the network state at the next time step.

[0080] This invention trains the model by minimizing the loss functions of the network topology prediction decoder, the data metric prediction decoder, and the node failure predictor.

[0081] The spatiotemporal evolution factor H obtained by the aforementioned method t This provides effective information for predicting future trends in networks. Spatiotemporal evolution factor H t This invention captures nonlinear interactions between network nodes both within a single time step and across time steps. It is based on the spatiotemporal evolution factor H. t Design a situation prediction decoder to predict the overall situation of the network. Including the design of network topology prediction decoders Data Indicator Prediction Decoder And a node failure predictor.

[0082] 1) Network Topology Prediction Decoder

[0083] For topology prediction decoder This invention proposes a predictive decoding approach. Specifically, it predicts the probability of edges existing between nodes based on spatiotemporal evolution factors, thereby generating network topology information at time t+1 and achieving dynamic perception of the network topology.

[0084]

[0085]

[0086] in, node v at time t+1 i and v j The probability of an edge existing. The training of the topology prediction decoder uses the error in adjacency matrix reconstruction as the loss function:

[0087]

[0088] Among them, A t+1 Let be the true value of the network topology adjacency matrix at time t+1. Loss is the predicted value of the adjacency matrix.decE The smaller the value, the closer the predicted topological distribution is to the true distribution.

[0089] 2) Data Indicator Prediction Decoder

[0090] For the data index decoder, a multi-layer rectified linear network was designed, with the spatiotemporal evolution factor of each node as input. Output the predicted values ​​of the node metrics. For node v i The input of the first layer is The solution for layer l is as follows:

[0091]

[0092] Where ReLU is the rectified linear unit function, W l and b l These are the weight matrix and bias matrix of the l-th ReLU layer, respectively. The output of the last layer... It is for node v i Estimate the true index value at time t+1. Use the coefficient of determination (R²). 2 Loss as the loss function decD This is used to assess the deviation between the predicted and actual values.

[0093] 3) Node Fault Predictor

[0094] The node fault predictor consists of a convolutional classifier. The input to the convolutional classifier is each node v. i Spatiotemporal dynamic factors The output is the node's running status at time t+1. Its training uses the cross-entropy function as the loss function:

[0095]

[0096] in, Represents node v i The true state of the running state at time t+1, where |V| is the total number of nodes. Loss Anomaly The smaller the value, the closer the predicted distribution of abnormal network nodes is to the true distribution.

[0097] This invention achieves dynamic modeling of network evolution by collaboratively training a topology prediction decoder, a data metric decoder, and a node fault predictor. The overall loss function is designed as follows:

[0098] Loss = Loss decE +Loss decD +Loss Anomaly

[0099] The proposed "historical evolution coding + situation prediction decoding" architecture can learn spatial patterns in the dynamic evolution process of the network. Stacked LSTM is used to capture temporal dependency patterns in the evolution, and fault discrimination is used to learn abnormal patterns in the data. The combined application of the three can fully learn the dynamic evolution law of the network, thereby realizing real-time perception of the IoT network status.

[0100] In summary, the method of this invention effectively integrates IoT network information by constructing an IoT temporal knowledge graph, thereby constructing a network spatiotemporal graph sequence and collaboratively designing a historical evolution encoder and a situation prediction decoder to perceive network situation information at future moments.

[0101] Current network operation and maintenance research lacks the integration and utilization of massive, multi-source, heterogeneous information from the Internet of Things (IoT), and also lacks the ability to perceive the network status as a whole. This invention integrates dynamic network information based on network topology and configuration, and designs a network situational awareness model, enabling holistic perception of network status, early warning of network faults, and reduction of economic losses caused by network failures.

[0102] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of one embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing the present invention.

[0103] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that the present invention can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of the present invention.

[0104] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for apparatus or system embodiments, since they are basically similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. The apparatus and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0105] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A network situational estimation method for the Internet of Things based on knowledge integration, characterized in that, include: Construct a network temporal knowledge graph based on the entity concepts, resource mapping, and data attributes of the Internet of Things; Based on the network temporal knowledge graph, a historical evolution spatiotemporal graph sequence is constructed. Based on the historical evolution spatiotemporal graph sequence, the spatiotemporal evolution factors of nodes in the Internet of Things are obtained through a spatiotemporal graph encoder and a historical information embedder. A historical evolution encoder is designed based on the spatiotemporal evolution factors of the nodes, and the spatiotemporal evolution factors of the network are solved through the historical evolution encoder. Based on the aforementioned network spatiotemporal evolution factors, the situation prediction decoder perceives the network situation of the Internet of Things at the next moment. The construction of a network temporal knowledge graph based on the entity concepts, resource mappings, and data attributes of the Internet of Things includes: Design a network temporal knowledge graph with a three-layer architecture, including: a concept layer. Resource mapping layer and data attribute layer The concept layer This resource mapping layer is used to collect entity concepts and their relationships within the network. Different entities have different semantic relationships. The data attribute layer is used to integrate device attribute information, status information, control information, and topology information. A dimension-attribute data model was designed, which combines network dynamic data flow information and external supporting data from two dimensions: time information and spatial information, to form a unified network dynamic data flow attribute format.

2. The method according to claim 1, characterized in that, The resource mapping layer Network resources are characterized from four aspects: device attribute information, status information, control information, and topology information. Device attribute information includes static information and dynamic information. Static attribute information is used to store fixed information of the device, including device unique identifier, device name, device type, device model, device unified code, manufacturer, device physical size, device purchase time, and device usage mode. The dynamic information includes device energy status, historical sensing data, current sensing data, current sensing time, historical device location, and current device location. The equipment status information includes equipment performance status, equipment usage status, and equipment fault status. The equipment usage status attribute values ​​have 7 possible values, including scheduled, idle, half-load, full-load, overload, maintenance, and decommissioning. The device's control information includes access control information for the device's Internet of Things (IoT) access and monitoring information. The access control information includes the device's interface information, communication method, communication protocol, return value type, and parameter information. The monitoring information includes the latitude and longitude of the device's monitoring, the monitoring coverage area, and the monitoring object information. The topology information of a device includes the physical topology connection information of the device and the dependency information between devices. The dependency information between devices includes concurrency relationships, selection relationships, predecessor / successor relationships, control relationships and support relationships.

3. The method according to claim 2, characterized in that, The data attribute layer The dimensional-attribute data model of the design encapsulates the data flow on network nodes into a two-level information tuple. In the first dimension, each tuple consists of time... and space Represented in two dimensions, each dimension in the second layer of attributes contains multiple attribute information, including time information. It includes timestamp information, node status information, and data stream density. Timestamp information is used to indicate whether the current time is a weekday, weekend, or holiday. Node status information includes busy, idle, or interrupted. Data stream density information is used to record the local density, global density, and fluctuation characteristics of the data stream transmitted by the current network node. Data density is used to describe the load of the data stream. Local density represents the data transmitted by a single data stream in a unit interval. Global density represents the density of all data streams. Data stream fluctuation is used to record the rate of change of the data stream per unit time. Spatial Information The network is described from four dimensions: relative location information, system location information, physical location information, and environmental information. Relative location information records the network location, i.e., the subnet, system, slice, or module to which the current node belongs. System location indicates whether the node belongs to a collection point or a collection point. Physical location indicates the actual environmental information, i.e., whether it belongs to a computer room or a rack. Environmental attributes indicate the density of equipment, environmental conditions, and climate conditions.

4. The method according to any one of claims 1 to 3, characterized in that, The construction of a historical evolution spatiotemporal graph sequence based on the network temporal knowledge graph, and the determination of the spatiotemporal evolution factors of nodes in the Internet of Things through a spatiotemporal graph encoder and a historical information embedder based on the historical evolution spatiotemporal graph sequence, includes: Based on knowledge graphs, a spatiotemporal graph sequence of network historical evolution is constructed. A spatiotemporal graph encoder is designed to embed the spatiotemporal graph, and a historical information embedder is designed to learn the evolutionary rules of the historical spatiotemporal graph. Given a network temporal knowledge graph ,Will Represented as ,in Indicates devices and network nodes. Represents the adjacency matrix. The index matrix representing the nodes, The fault matrix representing a node is used as a reference, and the knowledge graph information within the corresponding time slice is extracted to form the spatiotemporal graph for the current time slice. All spatiotemporal graphs are then combined to obtain... The corresponding historical evolutionary spatiotemporal sequence is denoted as ,in Spacetime diagram of moments Reflects The overall state of network interaction at any given moment, time, Represents a node and The connection between them Represents a node Data, Represents a node The operating status is indicated by 1 for normal operation and 0 for fault. A spatiotemporal graph encoder is designed based on the aforementioned historical evolution spatiotemporal graph sequence, and the spatiotemporal graph encoder is embedded into the spatiotemporal graph of each historical moment. Spacetime diagram of moments The adjacency matrix is The node feature matrix is The encoding structure of the variational graph autoencoder is used for processing. ,get eigenvectors : in, These represent two graph convolutional network layers, and their feature matrices are respectively. Representing a spacetime diagram Features of all nodes in The i-th row represents the node Feature embedding, and The mean vector matrix and variance vector matrix represent the low-dimensional vector representations of the nodes. , yes The corresponding degree matrix, It is the identity matrix. and The two graph convolutional networks are identical, meaning they share the weights of the first layer. Represents a node exist The embedding features at time step are obtained Spatial characteristics of each node at any given time Each spatiotemporal graph is input into the spatiotemporal graph encoder to obtain the spatial feature vector of each node at each time step. The spatial feature vectors at each time step are: ; A historical information embedder is designed based on the spatial feature vectors of nodes. The historical information embedder consists of a stacked long short-term memory network (LSTM) and an attention mechanism. Given a node... Corresponding historical spatial feature sequence Stacked LSTMs are used to capture the long-term dependencies in the historical state evolution of each node. The LSTM of the first generation... Hidden state of a layer The output is from the previous time step. Hidden state and the current moment Hidden state of a layer Together, we determine that the hidden value of the last LSTM layer is... The attention mechanism is used to weight the sequence of hidden values ​​in the LSTM to obtain the node values. exist Spatiotemporal evolution factors at any moment Using the above method, all nodes can be obtained. Spatiotemporal evolution factors at any moment .

5. The method according to claim 4, characterized in that, The aforementioned design of a historical evolution encoder based on the spatiotemporal evolution factors of nodes, and the solution of the network spatiotemporal evolution factors through the historical evolution encoder, includes: A historical evolution encoder is designed based on the spatiotemporal evolution factors of the nodes. The historical evolution encoder includes a spatiotemporal graph encoder and a historical information embedder. The spatiotemporal graph encoder processes the input historical evolution spatiotemporal graph sequence. Encoding is performed to obtain the spatial features of nodes at different times. The historical information embedder includes a stacked LSTM network and an attention mechanism. It captures spatiotemporal dynamic factors that reflect the network evolution and learns deep encoding mapping relationships based on the spatiotemporal encoder. , The number of nodes The dimension of the vector represents the sequence of the historical evolution spatiotemporal graph. Mapped to hidden space vectors ,exist middle, Representation diagram The Middle The spatiotemporal characteristics of each node are captured by a historical information embedder to identify the spatiotemporal dynamic factors of network dynamic evolution. .

6. The method according to claim 5, characterized in that, The method of sensing the network situation of the Internet of Things (IoT) at the next moment through a situation prediction decoder based on the network spatiotemporal evolution factors includes: Based on spatiotemporal evolution factors Design a situation prediction decoder, which includes a network topology prediction decoder. Data indicator prediction decoder and node failure predictor; Network Topology Prediction Decoder Based on the probability of edges existing between nodes predicted by the spatiotemporal evolution factor, the state of the topological connectivity graph at the next time step is predicted. This enables dynamic perception of the network topology. in, The graph is reconstructed by calculating the probability that an edge exists between any two nodes. for Time Node and The probability that an edge exists. node exist The spatiotemporal evolution factors of time, node exist Spatiotemporal evolution factors at any given moment Indicates matrix transpose; Based on data indicator decoder Design a multi-layer rectified linear network, and input the spatiotemporal evolution factor of each node into the rectified linear network. The rectified linear network outputs predicted node performance values ​​for each node. The input of the first layer of the rectified linear network is , No. The solution for the layer is as follows: Where ReLU is the rectified linear unit function. and The first The weight matrix and bias matrix of the ReLU layer; The output of the last layer of the rectified linear network is , For predicted nodes exist Data index values ​​at any given time; The node fault predictor consists of a convolution classifier, and the input data of the convolution classifier is each node. Spatiotemporal dynamic factors The output data is Node running status at any given time Identify abnormal network nodes based on their operational status; Based on the topology connection state diagram of each node at the next time step Data indicator values and abnormal network nodes Obtain the network state of the Internet of Things in the next moment. This network status includes network topology, network node data, and network node operational status.

Images