Model construction method and system for communication interference cognitive system

Abstract

The invention provides a model construction method and system for a communication interference cognitive system. The method comprises the following steps: constructing a communication interference cognitive reference set containing interference scene association features and interference response reference features; then, interference feature correlation modeling is carried out based on the communication interference cognitive reference set to form a correlation map; implementing cognitive model structure adaptation according to the correlation map to generate an initial interference cognitive model; carrying out interference scene adaptive iterative optimization on the initial model to form an optimized target interference cognitive model; and finally, outputting the target model for interference identification and response processing of the communication interference cognitive system. According to the method, the interference characteristics are comprehensively captured, the incidence relation is deeply mined, the model structure is continuously optimized, and the accuracy and adaptability of a communication interference cognitive system in a complex interference environment are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence, and more specifically, to a model construction method and system for cognitive systems dealing with communication interference. Background Technology

[0002] With the rapid development of communication technology, communication interference has become increasingly complex and diverse, posing a serious threat to the stability and reliability of communication systems. Traditional methods for recognizing communication interference mainly rely on manually set rules or simple statistical models to identify interference types and respond accordingly. However, these methods have many limitations. On the one hand, manually set rules are difficult to comprehensively cover various complex interference scenarios, and often cannot be identified accurately and promptly when faced with newly emerging interference forms. On the other hand, simple statistical models have limited ability to process interference characteristics and cannot deeply explore the intrinsic relationship between interference scenarios and responses, resulting in poor accuracy and adaptability in complex interference environments. Therefore, there is an urgent need for a model-building method that can comprehensively, accurately, and adaptively recognize communication interference to improve the performance of communication systems in interference environments. Summary of the Invention

[0003] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a model construction method for a communication interference cognitive system, the method comprising:

[0004] A set of communication interference cognitive benchmarks is constructed, which includes interference scenario-related features and interference response benchmark features.

[0005] Based on the aforementioned communication interference cognitive benchmark set, interference feature correlation modeling is carried out to form an interference feature correlation map;

[0006] Based on the interference feature association map, the cognitive model structure is adapted to generate an initial interference cognitive model;

[0007] The initial interference cognitive model is iteratively optimized to adapt to interference scenarios, resulting in an optimized target interference cognitive model.

[0008] Output the target interference cognitive model.

[0009] In another aspect, embodiments of the present invention also provide a model building system for a cognitive system of communication interference, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the machine-readable storage medium to implement the above-described method.

[0010] Based on the above, this embodiment of the invention constructs a communication interference cognitive benchmark set that includes interference scenario association features and interference response benchmark features. Based on this benchmark set, interference feature association modeling is performed to form an association graph, clearly presenting the complex relationship between interference scenarios and responses, thus enhancing the model's cognitive depth regarding interference features. An initial interference cognitive model generated by adapting the cognitive model structure based on the association graph can initially adapt to interference features. Through iterative optimization and adaptation of the initial model to interference scenarios, the model is continuously adjusted and improved, ultimately resulting in a target interference cognitive model with high accuracy and adaptability. This model can accurately identify various types of communication interference and make reasonable responses, effectively improving the performance and stability of the communication interference cognitive system in complex interference environments. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the execution flow of the model construction method for a cognitive system of communication interference provided in an embodiment of the present invention.

[0012] Figure 2 This is a schematic diagram of exemplary hardware and software components of a model building system for a communication interference cognitive system provided in an embodiment of the present invention. Detailed Implementation

[0013] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a model construction method for a communication interference cognitive system according to an embodiment of the present invention. The following is a detailed description of the model construction method for a communication interference cognitive system.

[0014] Step S110: Construct a communication interference cognitive benchmark set, which includes interference scenario association features and interference response benchmark features.

[0015] This step involves collecting and processing scenario and response data from historical interference events to form the foundational dataset for model building. Interference scenario correlation features reflect the relationships between elements within the interference scenario, while interference response baseline features record standardized parameters of effective response behaviors.

[0016] Step S111: Collect scenario description information and response behavior records related to communication interference.

[0017] The terrestrial mobile communication system network management platform and terminal log system collect scenario description information such as interference occurrence time, regional base station density, interference source type, and signal parameters, as well as response behavior records such as interference suppression algorithm type, parameter configuration, and suppression effect. All information is stored in a structured table, and each record corresponds to a complete interference event.

[0018] Step S112: Extract interference scene elements from the scene description information and separate the core elements of the interference scene.

[0019] Based on preset rules, core elements such as interference source type, signal frequency range, frequency band of interfered base station, and density of regional base stations are extracted from the scene description information. Irrelevant and redundant information is eliminated, and elements that have a direct impact on interference analysis are retained.

[0020] Step S113: Based on the core elements of the interference scenario, conduct correlation analysis to generate the interference scenario correlation features contained in the communication interference cognitive benchmark set.

[0021] The correlation degree between core elements is calculated by correlation analysis algorithm. Elements with correlation degree higher than the threshold are combined into interference scenario correlation features, such as the combination feature of "interference source type-signal frequency range-interfered base station frequency band", which reflects the key correlation pattern of interference scenario.

[0022] Step S114: Extract response patterns from the recorded response behavior information and separate out the core interference response pattern.

[0023] Clustering algorithms are used to cluster response behavior records, and representative response features are extracted from each cluster, such as the combination pattern of "adaptive filtering algorithm - specific parameter configuration - bit error rate reduction range", to form the core pattern of interference response.

[0024] Step S115: Based on the core interference response mode, set the reference parameters and generate the interference response reference features contained in the communication interference cognition reference set.

[0025] The algorithm type, parameter configuration, and suppression effect in the core interference response model are standardized and normalized, and converted into multi-dimensional feature vectors, which are then stored as the baseline features of the interference response.

[0026] Step S116: Integrate the interference scenario association features and the interference response benchmark features to form the communication interference cognitive benchmark set.

[0027] The feature vector corresponding to each interference scenario-related feature is concatenated with the matched interference response benchmark feature vector to form a fused feature vector that simultaneously contains scenario and response information, thus forming a set of communication interference cognitive benchmarks.

[0028] Step S120: Based on the communication interference cognitive benchmark set, perform interference feature association modeling to form an interference feature association map.

[0029] This step uncovers the potential correlations between features associated with the interference scenario and the baseline features of the response, and visualizes the relationships between features in the form of a graph, providing a basis for model structure adaptation.

[0030] Step S121: Extract key attribute information of interference scene association features from the communication interference cognitive benchmark set.

[0031] Key attributes such as interference source type identifier, signal frequency range, and frequency band of the interfered base station are extracted from the interference scene association features and converted into a structured form.

[0032] Step S122: Extract key attribute information of interference response benchmark features from the communication interference cognitive benchmark set.

[0033] Key attributes such as interference suppression algorithm encoding, standardized parameter values, and normalized suppression effect are extracted from the interference response baseline features and converted into a structured form.

[0034] Step S123: Perform association rule mining on the key attribute information of the interference scene association features and the key attribute information of the interference response benchmark features.

[0035] The association analysis module is used to uncover the association rules between the two types of attributes. The specific process is as follows:

[0036] Step S1231: Input the key attribute information of the interference scene association features and the key attribute information of the interference response benchmark features into the association analysis module.

[0037] The two types of attribute information are converted into a format supported by the association analysis module, and input is completed through the data interface.

[0038] Step S1232: Perform co-occurrence frequency statistics on the key attribute information of the interference scene association features and the key attribute information of the interference response benchmark features to form an attribute co-occurrence frequency distribution.

[0039] Count the number of times two types of attributes appear simultaneously in the dataset, and sort them by frequency to form a co-occurrence distribution.

[0040] Step S1233: Set association filtering conditions based on the attribute co-occurrence frequency distribution, and filter co-occurrence attribute combinations that satisfy the association filtering conditions.

[0041] Set a co-occurrence frequency threshold, and filter out attribute combinations with frequencies higher than the threshold, considering them to have potential associations.

[0042] Step S1234: Perform causal relationship verification on the co-occurrence attribute combination to clarify the causal relationship direction between attributes in the co-occurrence attribute combination.

[0043] Analyze the temporal order and logical impact of the attributes to determine the causal direction of "interference scenario attribute → interference response attribute".

[0044] Step S1235: Generate a quantitative result of the association strength by combining the causal association direction and the attribute co-occurrence frequency distribution.

[0045] The causal direction is converted into a directional coefficient, and the co-occurrence frequency is converted into a frequency coefficient. The correlation strength is obtained by multiplying the two.

[0046] Step S1236: Integrate the co-occurrence attribute combination, the causal association direction, and the association strength quantification result to form an association rule set.

[0047] Each set of attribute combinations, causal direction, and correlation strength is integrated into a single correlation rule, such as "If the interference source is spurious emissions from a neighboring base station, then an adaptive filtering algorithm is used, and the correlation strength is X".

[0048] Step S1237: Redundancy removal is performed on the association rule set, retaining the core association rules in the association rule set.

[0049] Remove rules that are repetitive, similar, or have a correlation strength below a threshold, and retain core rules that play an important role in interfering with cognition.

[0050] Step S124: Construct a feature association edge set based on the association rules obtained through mining. Each association edge in the feature association edge set corresponds to a set of association relationships of the key attribute information.

[0051] Each association rule is converted into a feature association edge, with the starting point of the edge being the interference scene attribute node and the ending point being the interference response attribute node, and the edge weight being the association strength.

[0052] Step S125: Using the interference scene association features and interference response benchmark features in the communication interference cognitive benchmark set as nodes, and the feature association edge set as the connection relationship, construct an interference feature association graph.

[0053] By using the interference scene association features and the response baseline features as graph nodes, and the feature association edges as connections between nodes, a visualized association graph is formed, reflecting the association structure between features.

[0054] For example, in step S1251: node encoding is performed on the interference scene association features in the communication interference cognitive benchmark set and the interference response benchmark features in the communication interference cognitive benchmark set to generate a unique node identifier.

[0055] Each interference scenario association feature and interference response baseline feature is assigned a unique string identifier. The identifier includes a feature type prefix and a sequence number. For example, the identifiers for interference scenario association features are “SC-001” and “SC-002”, and the identifiers for interference response baseline features are “RS-001” and “RS-002”, ensuring that each node has a unique identity in the graph.

[0056] Step S1252: Based on the association relationships in the feature-related edge set, establish a connection mapping between the node identifiers.

[0057] Traverse each edge in the feature association edge set, extract the interference scene association feature node identifier and the interference response baseline feature node identifier corresponding to the edge, and establish a triplet mapping relationship of "starting node identifier → ending node identifier → edge weight", such as "SC-001→RS-001→0.85" and "SC-002→RS-002→0.72", to fully record the association direction and strength between nodes.

[0058] Step S1253: Construct the basic framework of the interference feature association map, which includes a node storage structure and edge connection channels.

[0059] A node storage structure is built, using key-value pairs to store the mapping between node identifiers and corresponding features. The key is the node identifier, and the value is the structured information of the feature. An edge connection channel is built, using an adjacency list to store the connection mapping between nodes. Each node identifier corresponds to a list, and the list elements are the identifiers of other nodes pointed to by that node and the edge weights, ensuring that the basic framework can support the efficient storage and retrieval of nodes and edges.

[0060] Step S1254: Import the encoded node identifier into the node storage structure of the basic framework, and establish the node index of the node identifier.

[0061] The generated unique node identifiers are categorized by feature type and imported into the node storage structure. At the same time, an index table is created for each node identifier. The index table contains node identifier, feature type, and storage location information. The storage location corresponding to any node identifier can be quickly located through the index table, improving node query efficiency.

[0062] Step S1255: Based on the connection mapping between the node identifiers, establish the association path between the nodes corresponding to the node identifiers in the edge connection channel of the basic framework.

[0063] Traverse the connection mapping triples of node identifiers, and import the starting node identifier, ending node identifier and edge weight of each triple into the adjacency list of the edge connection channel. For example, add the entry "RS-001:0.85" to the adjacency list of the starting node "SC-001" to complete the physical construction of the association path between nodes and ensure that the association relationship between nodes can transmit signals through the path.

[0064] Step S1256: Perform a signal transmission test on the associated path, and adjust the transmission parameters of the associated path based on the test results to optimize the feature transmission between the nodes.

[0065] Construct a test signal, input the test signal from any starting node, and record the attenuation and delay time of the signal as it travels through the associated path to the ending node. If the attenuation exceeds the preset threshold or the delay time is too long, adjust the transmission parameters of the associated path, such as increasing the signal amplification coefficient of the high-weight side and reducing the signal loss coefficient of the low-weight side, until the signal attenuation and delay of all associated paths meet the preset requirements.

[0066] Step S1257: Integrate the node storage structure, the edge connection channel, and the association path to form a complete interference feature association map.

[0067] The configured node storage structure, edge connection channels, and association paths are logically integrated to generate a complete interference feature association map containing nodes, edges, and association rules. The map can intuitively display the association structure and strength between the interference scene association features and the interference response benchmark features.

[0068] Step S1258: Optimize the node distribution of the interference feature association map to improve the association query efficiency of the interference feature association map.

[0069] By analyzing the distribution density of nodes in the graph, closely related nodes (i.e., nodes with high edge weights and a large number of connections) are clustered into the same logical region, reducing the path length for cross-region queries. At the same time, the sorting method of the node index table is optimized by reordering the index table according to the query frequency of nodes, thereby improving the calling speed of high-frequency query nodes and ultimately achieving an overall improvement in the efficiency of related queries.

[0070] Step S126: Organize the node connection status in the interference feature association graph and strengthen the signal transmission of the core association edges in the feature association edge set.

[0071] Analyze the number of connections and edge weights of nodes in the graph, delete low-weight edges, increase the weight of core related edges (high-weight edges), and optimize the signal transmission efficiency of the graph.

[0072] Step S130: Based on the interference feature association map, perform cognitive model structure adaptation to generate an initial interference cognitive model.

[0073] Adjust the layer structure and parameters of the neural network model based on the structural features of the association graph to generate an initial model that can capture feature associations.

[0074] Step S131: Analyze the node distribution features and edge connection density features of the interference feature association map.

[0075] The distribution of the number of interference scene nodes and response nodes in the statistical graph is used to calculate the edge connection density by calculating the ratio of the number of edges to the number of nodes, which reflects the density of feature associations.

[0076] Step S132: Determine the feature input layer dimension configuration scheme of the model based on the node distribution characteristics.

[0077] The total number of all nodes in the graph is used as the dimension of the feature input layer to ensure that the input layer can cover all scenarios and response features.

[0078] Step S133: Determine the hidden layer network topology of the model based on the edge connection density characteristics.

[0079] Using edge connection density as a reference, the higher the density, the more neurons are in the hidden layer, ensuring that the hidden layer can handle complex feature associations.

[0080] Step S134: Construct a model input layer according to the feature input layer dimension configuration scheme, wherein the model input layer receives the feature vector corresponding to the interference feature association map.

[0081] The number of neurons is determined based on the dimension of the input layer, and the input layer structure is built so that it can receive the feature vectors corresponding to the graph. Each neuron corresponds to a node in the graph.

[0082] Step S1341: Determine the number of input layer neurons of the model input layer based on the feature input layer dimension configuration scheme.

[0083] The dimension value of the feature input layer is directly used as the number of neurons in the input layer.

[0084] Step S1342: Construct the basic architecture of the model input layer according to the number of neurons in the input layer. The basic architecture includes the neuron arrangement and connection ports.

[0085] The neuron architecture is constructed in a linear arrangement, with each neuron corresponding to a connection port used to receive data of the corresponding dimension of the feature vector.

[0086] Step S1343: Set the feature reception protocol of the model input layer, wherein the feature reception protocol is adapted to the transmission format of the feature vector corresponding to the interference feature association map.

[0087] Configure a protocol that matches the feature vector transmission format, including transmission format, rate, and error checking method, to ensure accurate data reception.

[0088] Step S1344: Construct the feature preprocessing submodule of the model input layer, and perform feature alignment on the feature vectors received by the model input layer.

[0089] A preprocessing submodule is built to adjust the dimension and order of the input feature vectors so that they are consistent with the arrangement order of the neurons in the input layer.

[0090] Step S1345: Connect the infrastructure, the feature receiving protocol, and the feature preprocessing submodule to form a complete model input layer.

[0091] The output of the preprocessing submodule is connected to the input of the infrastructure, and the feature receiving protocol serves as the data input interface, forming a complete input layer.

[0092] Step S1346: Test the feature reception efficiency of the model input layer and optimize the response speed of the model input layer.

[0093] By sending test feature vectors to statistically analyze reception time, the protocol rate and preprocessing logic can be adjusted to improve reception efficiency.

[0094] Step S1347: Adjust the neuron activation threshold of the model input layer to improve the accuracy of the model input layer in receiving feature vectors.

[0095] Set an initial activation threshold, and adjust the threshold through testing to ensure that effective features can be accurately identified and avoid noise interference.

[0096] Step S135: Construct a model hidden layer according to the hidden layer network topology, wherein the model hidden layer includes a feature association enhancement sublayer and a feature transformation sublayer.

[0097] Based on the hidden layer topology, a hidden layer containing two sub-layers is constructed. The feature association enhancement sub-layer strengthens the correlation signal between features, and the feature transformation sub-layer realizes the feature dimension transformation.

[0098] Step S1351: Determine the number of neuronal connections in the feature-associated reinforcement sublayer based on the edge connection density features of the interference feature association map.

[0099] Using edge connection density as a reference, the higher the density, the more neuron connections there are, ensuring that the sublayer can process feature associations.

[0100] Step S1352: Construct an internal connection architecture for the feature association reinforcement sublayer based on the number of neuron connections, wherein the internal connection architecture is adapted to the edge connection pattern of the interference feature association graph.

[0101] The neural connection architecture of the sub-layers is constructed according to the edge connection pattern of the graph, so that the connection structure of the sub-layers is consistent with the association pattern of the graph.

[0102] Step S1353: Set the activation function of the feature association enhancement sublayer, wherein the activation function enhances the signal transmission of effective associated features in the interference feature association map.

[0103] By selecting activation functions such as ReLU that can enhance effective signals, configuring function parameters, amplifying signals with high correlation features, and suppressing signals with low correlation features, the overall effect can be achieved.

[0104] Step S1354: Determine the feature dimension conversion ratio of the feature conversion sub-layer based on the node distribution characteristics of the interference feature association map.

[0105] The conversion ratio is determined based on the ratio of the number of scene nodes to the number of response nodes, thereby achieving dimensional adaptation from input features to output features.

[0106] Step S1355: Construct a weight matrix structure for the feature transformation sub-layer based on the feature dimension transformation ratio. The weight matrix structure is matched with the input feature dimension of the feature vector received by the model input layer and the output feature dimension of the interference cognition result output by the model output layer.

[0107] The number of rows and columns of the weight matrix is ​​designed according to the transformation ratio to ensure that the matrix can realize the transformation from the input dimension to the output dimension.

[0108] Step S1356: Set the regularization processing method for the feature transformation sub-layer, wherein the regularization processing method suppresses the overfitting of the initial interference cognitive model.

[0109] Choose methods such as L2 regularization and configure regularization coefficients to prevent the model from overfitting during training.

[0110] Step S1357: Connect the feature association enhancement sublayer and the feature transformation sublayer in a preset order to form a complete model hidden layer.

[0111] The output of the feature association enhancement sublayer is connected to the input of the feature transformation sublayer, forming a processing flow that first enhances the association and then transforms the dimension.

[0112] Step S1358: Debug the sub-layer connection nodes of the hidden layer of the model to optimize the feature transformation effect between the feature association enhancement sub-layer and the feature transformation sub-layer.

[0113] By testing and adjusting the parameters of sub-layer connection nodes, the efficiency and accuracy of feature transfer from the reinforcement sub-layer to the transformation sub-layer can be improved.

[0114] Step S136: Construct the model output layer, which outputs the interference cognition results.

[0115] The number of neurons in the output layer is determined based on the dimensions of the cognitive results of the interference (such as the type of interference, suggested suppression measures, and predicted suppression effect), and the structure of the output layer is constructed.

[0116] Step S137: Connect the model input layer, the model hidden layer, and the model output layer to form an initial interference cognitive model.

[0117] The input layer, hidden layer, and output layer are connected in sequence to form a complete neural network model. The input is the feature vector of the interference scene, and the output is the interference cognitive result.

[0118] Step S138: Optimize the interlayer connections of the initial interference cognitive model to improve the feature transfer efficiency between the model input layer, the model hidden layer and the model output layer.

[0119] Adjusting the weights and bias parameters of inter-layer connections reduces loss during feature propagation and improves the model's processing speed.

[0120] Step S140: Perform interference scenario adaptation and iterative optimization on the initial interference cognitive model to form an optimized target interference cognitive model.

[0121] The initial model was tested using a validation dataset, and the model structure and parameters were iteratively adjusted to adapt it to the target interference scenario.

[0122] Step S141: Obtain the interference scene adaptation verification dataset, which contains feature data and cognitive result labels corresponding to various interference scenes.

[0123] Different types of interference scenarios are selected, and corresponding feature data and labeled cognitive results are collected to form a validation dataset.

[0124] Step S142: Input the interference scene adaptation verification dataset into the initial interference cognitive model to obtain the initial cognitive output result of the initial interference cognitive model.

[0125] The feature data of the validation dataset is input into the initial model to obtain the perturbation cognitive results predicted by the model.

[0126] Step S143: Compare the initial cognitive output of the model with the cognitive result labels in the interference scenario adaptation verification dataset to form cognitive bias quantification data.

[0127] The difference between the model output and the labeled results, such as mean squared error, is used to obtain a quantified cognitive bias.

[0128] Step S144: Based on the cognitive bias quantification data, locate the modules with insufficient adaptability in the initial interference cognitive model structure.

[0129] By decomposing the deviation data hierarchically, the layers or sub-layers with excessive deviations can be located, and module adaptation issues can be analyzed.

[0130] Step S1441: Perform hierarchical decomposition on the cognitive bias quantification data to obtain the input layer bias component, the hidden layer bias component, and the output layer bias component of the initial interference cognitive model, wherein the hidden layer of the initial interference cognitive model includes a feature association enhancement sublayer and a feature transformation sublayer.

[0131] The total bias is decomposed into bias components for the input layer, hidden layer, and output layer. The bias of the hidden layer is further decomposed into bias components for the two sub-layers.

[0132] Step S1442: Compare the input layer deviation component, the hidden layer deviation component, and the output layer deviation component with the deviation threshold of the corresponding layer to determine the target layer with excessive deviation.

[0133] The deviation of each layer is compared with the preset threshold to determine the layer whose deviation exceeds the standard.

[0134] Step S1443: If the target layer is the input layer of the initial interference cognitive model, analyze the matching degree between the input layer dimension configuration scheme of the initial interference cognitive model and the feature dimension of the interference scene adaptation verification dataset, and locate the problem point of insufficient adaptability of the input layer dimension configuration.

[0135] If the input layer deviation exceeds the standard, analyze the matching between the input layer dimension and the feature dimension of the validation data to locate the dimension mismatch problem.

[0136] Step S1444: If the target layer is the hidden layer of the initial interference cognitive model, then the hidden layer bias component is further decomposed into the bias part corresponding to the feature association reinforcement sublayer in the hidden layer and the bias part corresponding to the feature transformation sublayer in the hidden layer.

[0137] If the hidden layer deviation exceeds the limit, it is decomposed into the deviation parts of two sub-layers.

[0138] Step S1445: Compare the deviation portion corresponding to the feature association enhancement sublayer in the hidden layer and the deviation portion corresponding to the feature transformation sublayer in the hidden layer with the deviation threshold of the corresponding sublayer to determine the target sublayer with excessive deviation.

[0139] The sublayer deviation is compared with a preset threshold to identify sublayers with excessive deviation.

[0140] Step S1446: Analyze the degree of fit between the internal connection architecture or weight matrix structure of the target sublayer and the feature association pattern of the interference scenario adaptation verification dataset.

[0141] If the sublayer deviation exceeds the standard, analyze the compatibility between the sublayer's connection architecture or weight matrix and the correlation pattern of the validation data features.

[0142] Step S1447: If the target layer is the output layer of the initial interference cognitive model, then analyze the matching degree between the output dimension of the output layer and the dimension of the cognitive result label in the interference scene adaptation verification dataset, as well as the adaptability of the output activation function of the output layer.

[0143] If the output layer bias exceeds the limit, analyze the matching degree between the output dimension and the label dimension, as well as the suitability of the activation function.

[0144] Step S1448: Integrate the problem analysis results of each target layer or target sub-layer, and locate the modules with insufficient adaptability in the initial interference cognitive model structure.

[0145] Summarize the problem analysis results of each layer to identify the modules with insufficient adaptability that need adjustment.

[0146] Step S1449: Classify the problem types of the modules with insufficient adaptability.

[0147] Module problems are categorized into types such as dimension mismatch, structural incompatibility, and unreasonable parameters.

[0148] Step S145: Perform parameter adjustment and structural optimization on the modules with insufficient adaptability to obtain the adjusted intermediate interference cognitive model.

[0149] For each module's problem type, adjust parameters (such as weights and thresholds) or optimize the structure (such as dimensions and connection methods) to generate an intermediate model.

[0150] Step S146: Input the interference scene adaptation verification dataset into the intermediate interference cognitive model to obtain the intermediate cognitive output result of the intermediate interference cognitive model.

[0151] The validation dataset is input into the intermediate model to obtain new prediction results.

[0152] Step S147: Compare the intermediate cognitive output results of the model with the cognitive result labels in the interference scenario adaptation verification dataset to form new cognitive bias quantification data.

[0153] The deviation between the intermediate model output and the label is calculated to obtain the new quantization deviation.

[0154] Step S148: Based on the new cognitive bias quantification data, repeat the parameter adjustment and structural optimization steps for the module with insufficient adaptability until an intermediate interference cognitive model that meets the adaptability requirements is generated.

[0155] If the new deviation still exceeds the standard, repeat the adjustment and optimization steps; if the deviation meets the standard, stop the iteration.

[0156] Step S149: Determine the intermediate interference cognitive model that meets the adaptation requirements as the optimized target interference cognitive model.

[0157] The intermediate model that meets the deviation target is determined as the target model.

[0158] Step S1410: Integrate the overall structure of the target interference cognitive model to improve its operating efficiency.

[0159] Optimize the layer connections and parameters of the target model to reduce redundant calculations and improve running speed.

[0160] Step S150: Output the target interference cognitive model.

[0161] The target model is converted into a deployable format, and the output model is used in real-world scenarios.

[0162] Figure 2 Schematic diagrams are shown of exemplary hardware and software components of a model building system 100 for a cognitive system of communication interference, which can implement the ideas of this application, according to some embodiments of this application. For example, a processor 120 may be used in the model building system 100 for a cognitive system of communication interference and to perform the functions of this application.

[0163] The model building system 100 for cognitive systems dealing with communication interference can be a general-purpose server or a special-purpose server; both can be used to implement the model building method for cognitive systems dealing with communication interference of this application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the load.

[0164] For example, the model building system 100 for a cognitive system of communication interference may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the model building system 100 for a cognitive system of communication interference may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The methods of this application can be implemented according to these program instructions. The model building system 100 for a cognitive system of communication interference also includes an I / O interface 150 between the computer and other input / output devices.

[0165] For ease of explanation, only one processor is described in the model building system 100 for cognitive systems of communication interference. However, it should be noted that the model building system 100 for cognitive systems of communication interference in this application may also include multiple processors, and therefore the steps performed by one processor as described in this application may also be performed jointly or individually by multiple processors. For example, if the processor of the model building system 100 for cognitive systems of communication interference performs steps A and B, it should be understood that steps A and B may also be performed jointly by two different processors or individually by one processor. For example, the first processor performs step A, the second processor performs step B, or the first processor and the second processor jointly perform steps A and B.

[0166] Furthermore, embodiments of the present invention also provide a readable storage medium, wherein computer-executable instructions are preset in the readable storage medium, and when the processor executes the computer-executable instructions, the above-described model construction method for a cognitive system of communication interference is implemented.

[0167] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A model construction method for a cognitive system of communication interference, characterized in that, The method includes: A set of communication interference cognitive benchmarks is constructed, which includes interference scenario-related features and interference response benchmark features. Based on the aforementioned communication interference cognitive benchmark set, interference feature correlation modeling is carried out to form an interference feature correlation map; Based on the interference feature association map, the cognitive model structure is adapted to generate an initial interference cognitive model; The initial interference cognitive model is iteratively optimized to adapt to interference scenarios, resulting in an optimized target interference cognitive model. Output the target interference cognitive model.

2. The model construction method for the communication interference cognitive system according to claim 1, characterized in that, The construction of a communication interference cognitive benchmark set includes interference scenario-related features and interference response benchmark features, including: Collect scenario descriptions and response behavior records related to communication interference; The scene description information is subjected to interference scene element extraction to separate the core elements of the interference scene; Based on the core elements of the interference scenario, conduct correlation analysis to generate interference scenario correlation features contained in the communication interference cognitive benchmark set; The response behavior recording information is subjected to response pattern extraction to separate the core interference response pattern; Based on the core interference response mode, benchmark parameters are set to generate interference response benchmark features included in the communication interference cognition benchmark set. The interference scenario association features and the interference response benchmark features are integrated to form the communication interference cognitive benchmark set.

3. The model construction method for the communication interference cognitive system according to claim 1, characterized in that, The step of performing interference feature association modeling based on the communication interference cognitive benchmark set to form an interference feature association map includes: Extract key attribute information of interference scene association features from the communication interference cognitive benchmark set; Extract key attribute information of interference response benchmark features from the communication interference cognitive benchmark set; Association rule mining is performed on the key attribute information of the interference scene association features and the key attribute information of the interference response benchmark features; A set of feature association edges is constructed based on the association rules obtained through mining, and each association edge in the set of feature association edges corresponds to a set of association relationships of the key attribute information; Using the interference scenario association features and interference response benchmark features in the communication interference cognition benchmark set as nodes, and the feature association edge set as the connection relationship, an interference feature association graph is constructed. The node connection states in the interference feature association graph are sorted out to strengthen the signal transmission of the core association edges in the feature association edge set.

4. The model construction method for the communication interference cognitive system according to claim 3, characterized in that, The process of performing association rule mining on the key attribute information of the interference scene association features and the key attribute information of the interference response benchmark features includes: Input the key attribute information of the interference scene association features and the key attribute information of the interference response benchmark features into the association analysis module; Co-occurrence frequency statistics are performed on the key attribute information of the interference scene association features and the key attribute information of the interference response benchmark features to form an attribute co-occurrence frequency distribution. Based on the co-occurrence frequency distribution of the attributes, association filtering conditions are set to filter co-occurrence attribute combinations that satisfy the association filtering conditions; Causal relationship verification is performed on the co-occurrence attribute combination to clarify the direction of causal association between attributes in the co-occurrence attribute combination; The correlation strength quantification result is generated by combining the causal association direction and the attribute co-occurrence frequency distribution; The co-occurrence attribute combination, the causal association direction, and the association strength quantification result are integrated to form an association rule set; Redundancy removal is performed on the set of association rules, retaining the core association rules in the set of association rules.

5. The model construction method for the communication interference cognitive system according to claim 1, characterized in that, The step of adapting the cognitive model structure based on the interference feature association map to generate an initial interference cognitive model includes: Analyze the node distribution characteristics and edge connection density characteristics of the interference feature association graph; The feature input layer dimension configuration scheme of the model is determined based on the node distribution characteristics. The hidden layer network topology of the model is determined based on the edge connection density characteristics; The model input layer is constructed according to the feature input layer dimension configuration scheme, and the model input layer receives the feature vector corresponding to the interference feature association map; The model hidden layer is constructed according to the hidden layer network topology, and the model hidden layer includes a feature association enhancement sublayer and a feature transformation sublayer; A model output layer is constructed, which outputs the results of cognitive interference. The model input layer, the model hidden layer, and the model output layer are connected to form an initial interference cognition model; The interlayer connections of the initial interference cognitive model are optimized to improve the feature transfer efficiency between the model input layer, the model hidden layer and the model output layer.

6. The model construction method for the communication interference cognitive system according to claim 5, characterized in that, The model hidden layer is constructed according to the hidden layer network topology, and the model hidden layer includes a feature association enhancement sublayer and a feature transformation sublayer, including: The number of neuron connections in the feature-associated reinforcement sublayer is determined based on the edge connection density characteristics of the interference feature association map. The internal connection architecture of the feature association enhancement sublayer is constructed based on the number of neuron connections, and the internal connection architecture is adapted to the edge connection pattern of the interference feature association graph. An activation function is set for the feature association enhancement sublayer, which enhances the signal transmission of effective associated features in the interference feature association map. The feature dimension transformation ratio of the feature transformation sub-layer is determined based on the node distribution characteristics of the interference feature association map. Based on the aforementioned feature dimension transformation ratio, a weight matrix structure for the feature transformation sub-layer is constructed. The weight matrix structure is matched with the input feature dimension of the feature vector received by the model input layer and the output feature dimension of the interference cognition result output by the model output layer. A regularization processing method is set for the feature transformation sublayer, which suppresses overfitting of the initial interference cognitive model; The feature association enhancement sublayer and the feature transformation sublayer are connected in a preset order to form a complete model hidden layer; The sub-layer connection nodes of the hidden layer of the model are debugged to optimize the feature transformation effect between the feature association enhancement sub-layer and the feature transformation sub-layer.

7. The model construction method for the communication interference cognitive system according to claim 1, characterized in that, The step of performing iterative optimization of the initial interference cognitive model to adapt to interference scenarios, forming an optimized target interference cognitive model, includes: Obtain the interference scene adaptation verification dataset, which contains feature data and cognitive result labels corresponding to various interference scenes; Input the interference scenario adaptation verification dataset into the initial interference cognition model to obtain the initial cognition output result of the initial interference cognition model. Compare the initial cognitive output of the model with the cognitive result labels in the interference scenario adaptation verification dataset to form cognitive bias quantification data; Based on the cognitive bias quantification data, the modules with insufficient adaptability in the initial interference cognitive model structure are located; The modules with insufficient adaptability are subjected to parameter adjustment and structural optimization to obtain the adjusted intermediate interference cognitive model; Input the interference scenario adaptation verification dataset into the intermediate interference cognitive model to obtain the intermediate cognitive output result of the intermediate interference cognitive model. Compare the intermediate cognitive output of the model with the cognitive result labels in the interference scenario adaptation verification dataset to form new cognitive bias quantification data; Based on the new cognitive bias quantification data, the parameter adjustment and structural optimization steps for the poorly adapted modules are repeated until an intermediate interference cognitive model that meets the adaptation requirements is generated. The intermediate interference cognitive model that meets the adaptation requirements is identified as the optimized target interference cognitive model. The overall structure of the target interference cognitive model is integrated to improve its operational efficiency.

8. The model construction method for the communication interference cognitive system according to claim 7, characterized in that, The step of locating modules with insufficient adaptability in the initial interference cognitive model structure based on the cognitive bias quantification data includes: The cognitive bias quantification data is subjected to hierarchical decomposition to obtain the input layer bias component, the hidden layer bias component, and the output layer bias component of the initial interference cognitive model, wherein the hidden layer of the initial interference cognitive model includes a feature association enhancement sublayer and a feature transformation sublayer. By comparing the input layer deviation component, the hidden layer deviation component, and the output layer deviation component with the corresponding layer deviation threshold, the target layer with excessive deviation is determined. If the target layer is the input layer of the initial interference cognition model, then analyze the matching degree between the input layer dimension configuration scheme of the initial interference cognition model and the feature dimension of the interference scene adaptation verification dataset, and locate the problem point of insufficient adaptability of the input layer dimension configuration. If the target layer is the hidden layer of the initial interference cognitive model, then the bias component of the hidden layer is further decomposed into the bias part corresponding to the feature association enhancement sublayer in the hidden layer and the bias part corresponding to the feature transformation sublayer in the hidden layer; By comparing the deviation portion corresponding to the feature association enhancement sublayer in the hidden layer and the deviation portion corresponding to the feature transformation sublayer in the hidden layer with the deviation threshold of the corresponding sublayer, the target sublayer with excessive deviation is determined. Analyze the degree of fit between the internal connection architecture or weight matrix structure of the target sublayer and the feature association patterns of the interference scenario adaptation verification dataset; If the target layer is the output layer of the initial interference cognitive model, then analyze the matching degree between the output dimension of the output layer and the dimension of the cognitive result label in the interference scenario adaptation verification dataset, as well as the adaptability of the output activation function of the output layer. By integrating the problem analysis results of each target layer or target sub-layer, the modules with insufficient adaptability in the initial interference cognitive model structure are located. The problem types of the modules with insufficient adaptability are classified.

9. The model construction method for the communication interference cognitive system according to claim 5, characterized in that, The model input layer is constructed according to the feature input layer dimension configuration scheme. The model input layer receives the feature vector corresponding to the interference feature association map, including: The number of input layer neurons in the model input layer is determined based on the feature input layer dimension configuration scheme. The basic architecture of the model input layer is constructed according to the number of neurons in the input layer, and the basic architecture includes the neuron arrangement and connection ports; A feature reception protocol is set for the model input layer, and the feature reception protocol is adapted to the transmission format of the feature vector corresponding to the interference feature association map; A feature preprocessing submodule for the model input layer is constructed, and the feature preprocessing submodule performs feature alignment on the feature vectors received by the model input layer. The infrastructure, the feature receiving protocol, and the feature preprocessing submodule are connected to form a complete model input layer; The feature reception efficiency of the model input layer was tested, and the response speed of the model input layer was optimized. Adjusting the neuron activation threshold of the model input layer improves the accuracy of the model input layer in receiving feature vectors.

10. A model building system for cognitive systems dealing with communication interference, characterized in that, The model building system for the cognitive system of communication interference includes a processor and a memory, the memory and the processor are connected, the memory is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the memory to implement the model building method for the cognitive system of communication interference as described in any one of claims 1-9.

Images