Semantic understanding-based network planning method and device, storage medium and equipment

Abstract

The invention relates to a semantic understanding-based network planning method and device, a storage medium and equipment, and relates to the technical field of natural language processing, the method comprises the following steps: receiving a to-be-queried question sentence corresponding to a to-be-planned network and network context information, and generating a text semantic representation according to the to-be-queried question sentence; generating a context feature representation according to the network context information, and generating a domain knowledge representation according to network planning domain knowledge corresponding to the to-be-planned network; fusing the text semantic representation, the context feature representation and the domain knowledge representation to obtain an overall semantic feature vector, and performing logic dimension reasoning on the overall semantic feature vector to obtain a first reasoning result; and performing semantic dimension reasoning on the overall semantic feature vector to obtain a second reasoning result, and determining a target network planning result corresponding to the to-be-planned network according to the first reasoning result and the second reasoning result. According to the invention, the accuracy of the network planning result is improved.

Description

Technical Field

[0001] This disclosure relates to the field of natural language processing technology, and more specifically, to a network planning method based on semantic understanding, a network planning device based on semantic understanding, a computer-readable storage medium, and an electronic device. Background Technology

[0002] Existing network planning techniques lack the ability to deeply integrate knowledge from multimodal domains, resulting in low accuracy of the network planning results.

[0003] It should be noted that the information in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0004] The purpose of this disclosure is to provide a network planning method, a network planning device, a computer-readable storage medium, and an electronic device based on semantic understanding, thereby overcoming, to at least some extent, the problem of low accuracy of network planning results caused by the limitations and defects of related technologies.

[0005] According to one aspect of this disclosure, a network planning method based on semantic understanding is provided, comprising: receiving a query question sentence corresponding to a network to be planned and network context information, and generating a text semantic representation based on the query question sentence; generating a context feature representation based on the network context information, and generating a domain knowledge representation based on network planning domain knowledge corresponding to the network to be planned; fusing the text semantic representation, context feature representation and domain knowledge representation to obtain an overall semantic feature vector, and performing logical dimension reasoning on the overall semantic feature vector to obtain a first reasoning result; performing semantic dimension reasoning on the overall semantic feature vector to obtain a second reasoning result, and determining a target network planning result corresponding to the network to be planned based on the first reasoning result and the second reasoning result.

[0006] In one example embodiment of this disclosure, generating a text semantic representation based on the query question sentence includes: performing deep semantic parsing on the query question sentence based on a preset element extraction model to obtain key semantic elements included in the query question; analyzing the semantic relationships between the key semantic elements based on a preset semantic analysis model to obtain the association relationships between the key semantic elements; and generating the text semantic representation based on the key semantic elements and the association relationships between the key semantic elements.

[0007] In one example embodiment of this disclosure, generating a context feature representation based on network context information includes: extracting network type information, network planning stage information, and network planning technical parameter information corresponding to the network to be planned from the network context information; wherein, the network planning technical parameter information includes at least one of the operating frequency band, network coverage area, and expected user density of the network to be planned; constructing a network planning background model corresponding to the network to be planned based on the network type information, network planning stage information, and network planning technical parameter information, and generating the context feature representation based on the network planning background model.

[0008] In one example embodiment of this disclosure, generating a domain knowledge representation based on network planning domain knowledge corresponding to the network to be planned includes: querying network planning domain knowledge corresponding to the network to be planned from a preset dynamic knowledge graph based on key semantic elements in the query question sentence and network context information; generating a domain knowledge representation based on the network planning domain knowledge; wherein the network planning domain knowledge includes at least one of propagation characteristics, influence factor calculation models, network deployment specifications, and climate influence factors.

[0009] In one example embodiment of this disclosure, the fusion of text semantic representation, context feature representation, and domain knowledge representation to obtain an overall semantic feature vector includes: performing embedding mapping processing on the text semantic representation based on a preset first embedding mapping model to obtain a first representation vector, and performing embedding mapping processing on the context feature representation based on a preset second embedding mapping model to obtain a second representation vector; performing embedding mapping processing on the domain knowledge representation based on a preset third embedding mapping model to obtain a third representation vector, and projecting the first representation vector, the second representation vector, and the third representation vector onto a trained projection matrix for semantic alignment to obtain the overall semantic feature vector.

[0010] In one example embodiment of this disclosure, a first reasoning result is obtained by logically reasoning the overall semantic feature vector, including: matching the standard network planning rule corresponding to the network to be planned from a preset network planning rule base based on the overall semantic feature vector; and performing logical reasoning on the network planning details information corresponding to the network to be planned in the standard network planning rule based on a preset symbol model to obtain the first reasoning result.

[0011] In one example embodiment of this disclosure, the second reasoning result is obtained by performing semantic dimension reasoning on the overall semantic feature vector, including: matching planned network cases associated with the network to be planned from a preset network planning database based on the overall semantic feature vector, and extracting historical network planning details from the planned network cases; inputting the overall semantic feature vector, historical network planning details, and preset first parameter prompt information into a preset content generation big model to obtain the second reasoning result.

[0012] In one example embodiment of this disclosure, determining the target network planning result corresponding to the network to be planned based on the first inference result and the second inference result includes: inputting the first inference result, the second inference result, and preset second parameter prompt information into a preset weight value prediction model to obtain the first weight value and the second weight value of the first inference result and the second inference result in the rule coverage dimension, the third weight value and the fourth weight value in the logical consistency dimension, and the fifth weight value and the sixth weight value in the confidence dimension; determining the first target weight value corresponding to the first inference result based on the first weight value, the third weight value, and the fifth weight value, and determining the second target weight value corresponding to the second inference result based on the second weight value, the fourth weight value, and the sixth weight value; and determining the target network planning result corresponding to the network to be planned based on the first inference result, the first target weight value, the second inference result, and the second target weight value.

[0013] In one example embodiment of this disclosure, the semantic understanding-based network planning method further includes: dynamically evolving a preset dynamic knowledge graph based on the target network planning result; wherein the dynamic evolution is achieved by: determining the current network planning technology concept included in the target network planning result, and matching historical technology concepts corresponding to the current network planning technology concept in the preset dynamic knowledge graph; determining new technology concepts from the current network planning technology concepts based on the matched historical technology concepts, and extracting technology concepts to be added from the new technology concepts; evaluating the reliability and consistency of the technology concepts to be added to obtain a concept evaluation result, and when it is determined that the concept evaluation result indicates that the technology concepts to be added have reliability and consistency, integrating the technology concepts to be added into the preset dynamic knowledge graph to achieve the dynamic evolution.

[0014] In one example embodiment of this disclosure, extracting a technical concept to be added from the newly added technical concept includes: evaluating the importance of the newly added technical concept in a network planning scenario based on a preset importance assessment model to obtain an importance assessment result; extracting a technical concept to be added from the newly added technical concept according to the importance assessment result; wherein, if the importance assessment result is greater than or equal to a preset importance threshold, then the newly added technical concept corresponding to the importance assessment result is determined as a technical concept to be added.

[0015] In one example embodiment of this disclosure, the reliability and consistency of the technical concept to be added are evaluated to obtain a concept evaluation result, including: inputting the technical concept to be added into a generator in a preset adversarial generative network to obtain a concept generation result corresponding to the technical concept to be added; judging the reliability and consistency of the technical concept to be added and the concept generation result corresponding to the technical concept to be added based on the discriminator in the preset adversarial generative network to obtain the concept evaluation result.

[0016] In one example embodiment of this disclosure, the technical concept to be added is integrated into the preset dynamic knowledge graph to achieve dynamic evolution, including: determining the network planning entity included in the technical concept to be added and the entity attribute information of the network planning entity, and determining the mounting position of the technical concept to be added in the preset dynamic knowledge graph based on the network planning entity and the entity attribute information; and integrating the technical concept to be added into the preset dynamic knowledge graph based on the mounting position of the technical concept to be added in the preset dynamic knowledge graph to achieve dynamic evolution.

[0017] According to one aspect of this disclosure, a network planning apparatus based on semantic understanding is provided, comprising: a text semantic representation generation module, configured to receive a query question sentence corresponding to a network to be planned and network context information, and generate a text semantic representation based on the query question sentence; a domain knowledge representation generation module, configured to generate a context feature representation based on the network context information, and generate a domain knowledge representation based on network planning domain knowledge corresponding to the network to be planned; an overall semantic feature vector generation module, configured to fuse the text semantic representation, context feature representation and domain knowledge representation to obtain an overall semantic feature vector, and perform logical dimension reasoning on the overall semantic feature vector to obtain a first reasoning result; and a target network planning result determination module, configured to perform semantic dimension reasoning on the overall semantic feature vector to obtain a second reasoning result, and determine a target network planning result corresponding to the network to be planned based on the first reasoning result and the second reasoning result.

[0018] According to one aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the semantic understanding-based network planning method described in any of the foregoing example embodiments.

[0019] According to one aspect of this disclosure, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to perform the semantic understanding-based network planning method described in any of the foregoing example embodiments by executing the executable instructions.

[0020] This disclosure provides a semantic understanding-based network planning method. On one hand, it receives a query question sentence corresponding to the network to be planned, along with network context information, and generates a text semantic representation based on the query question sentence. Then, it generates a context feature representation based on the network context information and a domain knowledge representation based on the network planning domain knowledge corresponding to the network to be planned. Next, it fuses the text semantic representation, context feature representation, and domain knowledge representation to obtain an overall semantic feature vector, and performs logical-dimensional reasoning on the overall semantic feature vector to obtain a first reasoning result. Finally, it performs semantic-dimensional reasoning on the overall semantic feature vector to obtain a second reasoning result, and determines the target network planning result corresponding to the network to be planned based on the first and second reasoning results. This solves the problem in related technologies where the accuracy of network planning results is low due to a lack of deep fusion capabilities for multimodal domain knowledge, thus improving the accuracy of the target network planning result. On the other hand, it also solves the problem in related technologies where the accuracy of reasoning results is low because general language models often cannot accurately understand the deep meaning of network professional concepts and their interrelationships in specific planning scenarios.

[0021] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0022] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure. It is obvious that the drawings described below are merely some embodiments of this disclosure, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0023] Figure 1 The diagram illustrates a flowchart of a semantic understanding-based network planning method according to an exemplary embodiment of the present disclosure.

[0024] Figure 2The diagram illustrates a scenario example of a semantic understanding-based network planning system architecture according to an exemplary embodiment of the present disclosure.

[0025] Figure 3 The illustration shows a scenario example of a query question sentence according to an exemplary embodiment of the present disclosure.

[0026] Figure 4 The diagram illustrates a scenario example of network context information according to an exemplary embodiment of this disclosure.

[0027] Figure 5 The illustration shows a scenario example of a process for generating an overall semantic feature vector according to an exemplary embodiment of the present disclosure.

[0028] Figure 6 The diagram illustrates a scenario example of a target network planning result obtained according to an exemplary embodiment of the present disclosure.

[0029] Figure 7 The diagram schematically illustrates a structural example of a semantic understanding-based network planning apparatus according to an exemplary embodiment of the present disclosure.

[0030] Figure 8 An electronic device for implementing a semantic understanding-based network planning method is illustrated according to an example embodiment of the present disclosure. Detailed Implementation

[0031] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this disclosure more comprehensive and complete, and to fully convey the concept of the example embodiments to those skilled in the art. The described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a full understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced with one or more of the specific details omitted, or other methods, components, apparatus, steps, etc., can be employed. In other instances, well-known technical solutions are not shown or described in detail to avoid obscuring various aspects of this disclosure.

[0032] Furthermore, the accompanying drawings are merely illustrative of this disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0033] Currently, with the rapid development of next-generation communication technologies such as 5G and 6G, the field of network planning is facing unprecedented complexity and dynamic challenges. Against this backdrop, semantic understanding technology based on natural language processing has become a key means to improve the intelligence level of network planning. However, existing semantic understanding technologies for network planning have significant limitations, severely restricting their effectiveness in practical applications. These limitations manifest in several ways: First, traditional semantic understanding methods rely heavily on static knowledge base systems, which require manual maintenance and updates, making them ill-suited to the rapid evolution of network technologies. Consequently, when new technical standards, network protocols, or device types emerge, the static knowledge base often fails to keep pace, leading to a lag in the system's understanding capabilities. Furthermore, existing technologies lack the ability to deeply integrate multimodal domain knowledge, making it difficult to effectively handle the complex relationships between natural language descriptions, network topology, and technical parameters. Second, traditional natural language processing models have insufficient understanding of technical terms and contextual relationships. In other words, the network planning domain contains numerous technical terms and complex contextual dependencies, making it difficult for general language models to accurately understand the deeper meanings of these concepts and their interrelationships in specific planning scenarios. This often results in misunderstandings or incomplete understandings when processing technical queries, severely impacting the practicality and reliability of general language models in the network planning process.

[0034] This exemplary embodiment first provides a network planning method based on semantic understanding, which can run on servers, server clusters, or cloud servers, etc. Of course, those skilled in the art can also run the method disclosed herein on other platforms as needed, and this exemplary embodiment does not impose any special limitations on this. Specifically, refer to... Figure 1 As shown, the semantic understanding-based network planning method may include the following steps: Step S110. Receive the query question sentence corresponding to the network to be planned and the network context information, and generate a text semantic representation based on the query question sentence; Step S120. Generate a context feature representation based on the network context information, and generate a domain knowledge representation based on the network planning domain knowledge corresponding to the network to be planned; Step S130. The text semantic representation, context feature representation and domain knowledge representation are fused to obtain the overall semantic feature vector, and the overall semantic feature vector is used for logical dimension reasoning to obtain the first reasoning result; Step S140. Perform semantic dimension reasoning on the overall semantic feature vector to obtain a second reasoning result, and determine the target network planning result corresponding to the network to be planned based on the first reasoning result and the second reasoning result.

[0035] In the semantic understanding-based network planning method described above, on the one hand, it receives the query question sentence corresponding to the network to be planned and network context information, and generates a text semantic representation based on the query question sentence; then, it generates a context feature representation based on the network context information, and generates a domain knowledge representation based on the network planning domain knowledge corresponding to the network to be planned; then, it fuses the text semantic representation, context feature representation, and domain knowledge representation to obtain an overall semantic feature vector, and performs logical dimension reasoning on the overall semantic feature vector to obtain a first reasoning result; finally, it performs semantic dimension reasoning on the overall semantic feature vector to obtain a second reasoning result, and determines the target network planning result corresponding to the network to be planned based on the first reasoning result and the second reasoning result. This solves the problem in related technologies where the accuracy of the obtained network planning result is low due to the lack of deep fusion capability of multimodal domain knowledge, and improves the accuracy of the target network planning result. On the other hand, it also solves the problem in related technologies where the accuracy of the obtained reasoning result is low because general language models often cannot accurately understand the deep meaning of network professional concepts and their interrelationships in specific planning scenarios.

[0036] The semantic understanding-based network planning method described in the exemplary embodiments of this disclosure will be explained and illustrated in detail below with reference to the accompanying drawings.

[0037] First, the terms used in the exemplary embodiments of this disclosure will be explained.

[0038] Domain knowledge enhancement refers to using expertise from a specific domain (here, network planning) to enhance and improve the performance of a semantic understanding system. In practical applications, the knowledge enhancement process in the network planning domain does not use a general language model. Instead, it deeply integrates information such as professional terminology, technical parameters, topological relationships, and planning principles of communication networks into the system's understanding process. This enables the system to more accurately understand professional concepts such as "rain attenuation," "millimeter wave," and "capacity planning" in network planning scenarios, avoiding ambiguity.

[0039] Semantic understanding: In the example embodiments of this disclosure, semantic understanding specifically refers to enabling a computer to understand the true intent and contextual relationship behind a user's query input through natural language (such as "What rain attenuation factors need to be considered when deploying millimeter-wave base stations in residential areas?"). In this process, the ultimate goal of semantic understanding is not simply to perform lexical analysis, but to output a structured and actionable planning intent, which may include an intent type and a specific set of parameters corresponding to that intent type.

[0040] Network planning refers to a series of engineering design activities such as the deployment, configuration, and optimization of base stations in a communication network, including but not limited to network coverage planning, network capacity planning, and network frequency band planning. In the actual process of network planning, a large amount of professional knowledge and complex decisions are involved. Therefore, in order to simplify the network planning process, the example embodiments of this disclosure use natural language interaction to replace the traditional static knowledge base planning process.

[0041] Dynamic knowledge graphs: Unlike traditional static knowledge bases, dynamic knowledge graphs are a form of knowledge representation that can update and evolve itself. In practical applications, dynamic knowledge graphs can use concept drift detection algorithms to perceive new developments in network technology in real time (such as from 5G to 6G), and use incremental adversarial training mechanisms to automatically integrate new concepts and relationships into the graph. This ensures that its knowledge system is always in sync with cutting-edge technologies without relying entirely on manual reconstruction.

[0042] Neuro-symbolic reasoning encompasses two artificial intelligence paradigms: symbolic reasoning and neural reasoning. Symbolic reasoning, based on predefined domain rules and logic, yields results that are accurate and highly interpretable. Neural reasoning, on the other hand, leverages deep learning models (such as the Transformer) and excels at handling fuzzy and uncertain semantic information. In practical applications, neuro-symbolic reasoning technology integrates symbolic and neural reasoning through a dual-channel processing flow, employing a credibility-weighted fusion strategy to combine the strengths of both approaches. This ensures both professional accuracy and the flexibility to understand natural language.

[0043] Cross-modal semantic alignment refers to the unified understanding and mapping of information in different forms (modalities) at the semantic level. In the example embodiments of this disclosure, cross-modal semantic alignment refers to aligning and fusing natural language text (such as user-inputted query questions), contextual information (such as network type and current planning stage), and structured knowledge (such as network topology and technical parameters in dynamic knowledge graphs) to form a unified, semantically rich representation, laying the foundation for subsequent deep reasoning.

[0044] Concept drift detection: an algorithm used to identify changes in the definition or importance of concepts in a knowledge base over time. In practical applications, the system analyzes the degree of difference between newly emerging concepts and the existing knowledge base (calculating drift scores) to determine whether it is necessary to initiate a knowledge update process. It should be noted that concept drift detection is a key technology for realizing the self-evolution of dynamic knowledge graphs, which ensures that the system can automatically adapt to the dynamic development of technology.

[0045] Multimodal fusion: the process of integrating data information from different sources and forms; in the example embodiments of this disclosure, the system needs to integrate text embedding (e.g., the semantics of the query question), context embedding (e.g., network planning scenario information) and knowledge embedding (e.g., network planning domain knowledge), and on this basis, extract complementary information through multi-granularity attention mechanisms and other algorithms to generate a comprehensive representation of network planning that is more comprehensive and accurate than any single modal information.

[0046] Self-evolving knowledge base: the ultimate goal of the entire knowledge management system; the so-called self-evolving knowledge base refers to a knowledge base that is no longer a static database that requires manual maintenance, but a living system that can perceive multi-source knowledge flow in real time and automatically perform incremental updates; the self-evolving knowledge base involved in the example embodiments of this disclosure can learn from each query and reasoning result during the operation of the system, thereby continuously expanding and correcting itself to achieve the purpose of autonomous growth and evolution.

[0047] Context awareness: This refers to the system's ability to understand user queries by not only analyzing the literal text but also fully considering and utilizing the current context. The context involved in the example embodiments of this disclosure may include, but is not limited to, network type (such as 5G standalone networking), planning stage (such as network capacity planning), and technical parameters (such as network frequency band and network coverage radius). This awareness capability enables the system to more accurately grasp the user's true intention in the current specific scenario.

[0048] Secondly, the technical implementation principles of the exemplary embodiments of this disclosure will be explained and described. Specifically, the semantic understanding-based network planning method described in the exemplary embodiments of this disclosure relates to the interdisciplinary field of Natural Language Processing (NLP) and communication network planning, particularly a semantic understanding method that integrates dynamic domain knowledge graphs and context-aware neural symbolic reasoning. Specifically, the exemplary embodiments of this disclosure combine knowledge graph construction and evolution technology, neural symbolic reasoning, and cross-modal semantic alignment methods to improve the semantic understanding capability of natural language queries in network planning scenarios. Furthermore, the semantic understanding-based network planning method described in the exemplary embodiments of this disclosure aims to overcome the aforementioned deficiencies of the prior art and provide a domain knowledge-enhanced semantic understanding technology for network planning scenarios, comprehensively improving the accuracy, adaptability, and practicality of semantic understanding in network planning. Specifically, the exemplary embodiments disclosed herein aim to establish a knowledge management system capable of autonomous evolution, which achieves continuous updating and optimization of knowledge through dynamic knowledge graph technology, solving the problem that static knowledge bases cannot adapt to technological development; on this basis, by introducing a multi-source knowledge flow real-time perception architecture and a concept drift detection mechanism, the system can automatically identify and absorb emerging technology concepts, maintaining the cutting-edge nature and integrity of the knowledge system.

[0049] In one exemplary embodiment, another important objective of this disclosure is to construct a robust semantic understanding framework that integrates the accuracy of rule-based reasoning with the flexibility of neural network processing through neural symbolic joint reasoning technology. In practical applications, this framework can deeply understand the semantic connotations of technical terms and accurately grasp complex contextual relationships, thereby significantly improving the accuracy of understanding professional queries in the network planning field. Furthermore, this disclosure aims to achieve deep fusion of multimodal information by establishing a precise mapping relationship between natural language and network planning elements through cross-modal semantic alignment technology. This innovation enables the system to simultaneously process multi-source information such as text, technical parameters, and network topology, achieving true comprehensive semantic understanding. Ultimately, this disclosure aims to create an intelligent, adaptive, and highly efficient network planning semantic understanding system that not only significantly improves the accuracy of query understanding but also greatly reduces the design time for planning schemes, providing strong technical support for the intelligent planning and operation and maintenance of communication networks, and promoting the entire industry towards a higher level of intelligence.

[0050] The following will explain and describe the semantic understanding-based network planning system architecture involved in the exemplary embodiments of this disclosure. Specifically, the exemplary embodiments of this disclosure provide a complete solution for domain knowledge-enhanced semantic understanding technology for network planning scenarios; in practical applications, this technical solution adopts a three-layer architecture design, achieving deep understanding and intelligent processing of natural language queries in the network planning domain through the collaborative work of multiple innovative modules. Further, refer to... Figure 2 As shown, the semantic understanding-based network planning system architecture can include a dynamic knowledge management system 210, an intelligent reasoning and understanding system 220, and a multimodal fusion processing system 230. The dynamic knowledge management system employs a self-evolving domain knowledge graph construction technology, building a self-evolving knowledge base system through a real-time perception architecture of multi-source knowledge flows. In practical applications, the system achieves dynamic knowledge management through the following mechanisms: First, it establishes a knowledge evolution mechanism based on concept drift detection, automatically identifying emerging technology concepts and development trends by monitoring technical literature, standard documents, and practical cases in the field of network planning; second, it uses an incremental adversarial training method for graph updates, ensuring the consistency and accuracy of new knowledge additions through the adversarial training process of the generator and discriminator; finally, it implements transaction-based incremental updates of the knowledge base, ensuring the stability and reliability of the system during knowledge evolution. This knowledge management system can continuously adapt to the rapid development of new technologies such as 5G / 6G, solving the key problem of traditional static knowledge bases lagging behind technological evolution.

[0051] Furthermore, the intelligent reasoning and understanding system constructs a neural-symbolic joint reasoning framework, employing a dual-channel processing architecture to achieve accurate semantic understanding. In practical applications, the symbolic reasoning channel, based on descriptive logic and a rule engine, processes professional rules and constraints in the field of network planning, ensuring accurate parsing of professional terminology and standardized expressions. The neural network channel utilizes deep learning models such as Transformer to process fuzzy expressions and implicit semantics in natural language. On this basis, the two channels collaborate through an innovative fusion mechanism. For example, the contribution of each channel can be dynamically adjusted based on a weight calculation model of rule coverage and logical consistency. The context-aware relational path reasoning algorithm optimizes the reasoning strategy according to the specific planning scenario. The credibility-weighted result fusion strategy ensures the accuracy and reliability of the final output. This hybrid reasoning architecture significantly improves the ability to understand complex semantic relationships while maintaining interpretability.

[0052] Furthermore, the multimodal fusion processing system has developed a cross-modal semantic alignment model, achieving deep semantic fusion through a multi-level attention mechanism. For example, the system employs multi-head cross-modal attention technology to capture the complex relationships between different modal information from multiple representation subspaces; it constructs a learnable projection matrix to map heterogeneous features such as text, images, and structured data to a unified semantic space to achieve a multi-granularity attention mechanism, simultaneously processing semantic information at the lexical, phrase, and sentence levels; and the system places particular emphasis on the accurate mapping between natural language descriptions and professional information such as network topology and technical parameters, adaptively learning the optimal fusion strategy through an end-to-end training process to ensure a comprehensive understanding and effective utilization of multimodal information for network planning.

[0053] It should also be noted that the entire technical solution described above adopts a closed-loop learning architecture that organically integrates the three core layers. In practical applications, after processing each query, the system initiates a knowledge self-evolution process based on the reasoning results. This is achieved as follows: First, valuable new knowledge is identified through a concept importance assessment algorithm; second, the reliability of the new knowledge is verified using an adversarial training mechanism; and finally, the confirmed knowledge is integrated into the system using a transactional update method. This design forms a complete closed loop from knowledge acquisition, understanding and reasoning to evolutionary learning, enabling the system to continuously optimize. Simultaneously, the system also establishes a performance monitoring and tuning mechanism. By monitoring the operating status and processing effects of each module in real time, system parameters and strategies are dynamically adjusted to ensure optimal performance over the long term.

[0054] The following will be about Figure 1 The semantic understanding-based network planning method shown will be further explained and illustrated. Specifically: In step S110, the query question sentence corresponding to the network to be planned and the network context information are received, and a text semantic representation is generated based on the query question sentence.

[0055] In this example embodiment, firstly, the system receives the query question sentence and network context information corresponding to the network to be planned. Specifically, the query question sentence and network context information described here are input by the user (i.e., the network planning engineer) through the network planning interface of the terminal device. In practical applications, the user can input the query question sentence through the network planning interface via voice input or manual input, and input the network context information manually. After inputting, clicking the "OK" control generates the corresponding network planning generation request and sends it to the server or the backend of the terminal device. Based on this, parsing the network planning generation request yields the query question sentence and network context information. For example, taking the deployment of a 5G millimeter-wave base station as an example, a specific scenario example diagram of the query question sentence can be found in the image. Figure 3 As shown in the diagram, you can refer to the specific scenario examples of network context information. Figure 4 As shown.

[0056] Secondly, generating a text semantic representation based on the query question sentence includes: performing deep semantic parsing on the query question sentence based on a preset element extraction model to obtain the key semantic elements included in the query question; analyzing the semantic relationships between the key semantic elements based on a preset semantic analysis model to obtain the association relationships between the key semantic elements; and generating the text semantic representation based on the key semantic elements and the association relationships between the key semantic elements. Specifically, the preset element extraction model and the preset semantic analysis model described here can be the BERT model or other deep neural network models, and this example does not impose any special restrictions on them. At the same time, taking the query question sentence "What rain attenuation factors need to be considered when deploying millimeter-wave base stations in residential areas? What protective measures need to be taken?" as an example, the parsed key semantic elements may include, but are not limited to, "residential area, millimeter-wave base station, rain attenuation factors, protective measures," etc., and the association relationships between the key semantic elements may include, but are not limited to, limiting relationships, parallel relationships, judgment relationships, and logical relationships, etc.

[0057] In step S120, a context feature representation is generated based on the network context information, and a domain knowledge representation is generated based on the network planning domain knowledge corresponding to the network to be planned.

[0058] In this example embodiment, firstly, a contextual feature representation is generated based on network context information. Specifically, this can be achieved as follows: extracting network type information, network planning stage information, and network planning technical parameter information corresponding to the network to be planned from the network context information; wherein, the network planning technical parameter information includes at least one of the operating frequency band, network coverage area, and expected user density of the network to be planned; constructing a network planning background model corresponding to the network to be planned based on the network type information, network planning stage information, and network planning technical parameter information, and generating the contextual feature representation based on the network planning background model; specifically, the network type mentioned here can be, for example, 5G NSA (Non-Stand Alone) networking mode, the network planning stage mentioned here can be detailed planning and design, and the network planning technical parameter information mentioned here can be, for example, an operating frequency band of 3.5GHz, a base station height of 30 meters, a coverage radius of 200 meters, and a high expected user density, etc.; based on this, the obtained network planning background model can include the network type, network planning stage, operating frequency band, base station height, coverage radius, and expected user density of the network to be planned.

[0059] Secondly, a domain knowledge representation is generated based on the network planning domain knowledge corresponding to the network to be planned. Specifically, this can be achieved as follows: based on the key semantic elements in the query question sentence and the network context information, the network planning domain knowledge corresponding to the network to be planned is queried from a preset dynamic knowledge graph; a domain knowledge representation is generated based on the network planning domain knowledge; wherein, the network planning domain knowledge includes propagation characteristics, influencing factor calculation models, network deployment specifications, and climate influencing factors, etc. That is, in practical applications, the knowledge retrieval module can obtain the latest relevant domain knowledge from the dynamic knowledge graph based on key semantic elements and network context information; wherein, the domain knowledge recorded here may include, but is not limited to, professional knowledge such as millimeter wave propagation characteristics, rain attenuation calculation models, residential area deployment specifications, and climate influencing factors.

[0060] In step S130, the text semantic representation, context feature representation and domain knowledge representation are fused to obtain an overall semantic feature vector, and logical dimension reasoning is performed on the overall semantic feature vector to obtain a first reasoning result.

[0061] In this example embodiment, firstly, the text semantic representation, context feature representation, and domain knowledge representation are fused to obtain an overall semantic feature vector. Specifically, this can be achieved as follows: the text semantic representation is embedded and mapped based on a preset first embedding mapping model to obtain a first representation vector, and the context feature representation is embedded and mapped based on a preset second embedding mapping model to obtain a second representation vector; the domain knowledge representation is embedded and mapped based on a preset third embedding mapping model to obtain a third representation vector, and the first, second, and third representation vectors are projected onto a trained projection matrix for semantic alignment to obtain the overall semantic feature vector. In other words, in practical applications, the cross-modal alignment module can use a multi-head attention mechanism to deeply fuse text semantic representation, contextual feature representation, and domain knowledge embedding representation to obtain an overall semantic feature vector. Furthermore, in the specific fusion process, a learnable projection matrix can be used to map features from different modalities to a unified semantic space to generate an enhanced semantic representation (i.e., an overall semantic feature vector) rich in domain knowledge. It should be noted here that in the specific projection process, special attention should be paid to establishing a precise mapping relationship between natural language concepts and professional technical parameters to ensure the accuracy of subsequent inference.

[0062] In one possible example embodiment, reference is made to... Figure 5 As shown, the cross-modal alignment module described here may include a first embedding mapping model, a second embedding mapping model, a third embedding mapping model, and a trained projection matrix; wherein, the first embedding model and the second embedding model may be multi-layer bidirectional Transformers, and the third embedding mapping model may be a combination of multi-layer bidirectional Transformers and Vision Transformers; wherein, the multi-layer bidirectional Transformer may serve as a text encoder, and the Vision Transformer may serve as an image encoder; the text encoder encodes the text portion of the text semantic representation, context feature representation, and domain knowledge embedding representation to obtain the corresponding text representation vector; the image encoder encodes the image portion of the domain knowledge embedding representation to obtain the corresponding image representation vector.

[0063] Secondly, logical reasoning is performed on the overall semantic feature vector to obtain the first reasoning result; specifically, this can be achieved as follows: matching the standard network planning rule corresponding to the network to be planned from the preset network planning rule base according to the overall semantic feature vector; and performing logical reasoning on the network planning details information corresponding to the network to be planned in the standard network planning rule based on the preset symbol model to obtain the first reasoning result. Specifically, the logical reasoning described here can be implemented based on a symbolic reasoning channel. In practical applications, the symbolic reasoning channel uses a predefined network planning rule base to perform precise logical reasoning to obtain the first reasoning result. Simultaneously, when performing logical reasoning based on the symbolic reasoning channel, the focus is on handling tasks requiring precise calculation, such as technical parameter calculation, specification compliance checks, and constraint verification. Furthermore, the pre-defined network planning rule base described here can include, but is not limited to, authoritative rules such as ITU recommendations, 3GPP (3rd Generation Partnership Project) technical specifications, and industry deployment guidelines. Under this premise, the obtained overall semantic feature vector can be matched with corresponding network planning recommendations, network planning technical specifications, and base station deployment guidelines to obtain the first reasoning result. It should also be noted that the specific generation process of the corresponding feature vector in the pre-defined network planning rule base is similar to the specific generation process of the overall semantic feature vector, and will not be elaborated further here.

[0064] In step S140, semantic dimension reasoning is performed on the overall semantic feature vector to obtain a second reasoning result, and the target network planning result corresponding to the network to be planned is determined based on the first reasoning result and the second reasoning result.

[0065] In this example embodiment, firstly, semantic dimension reasoning is performed on the overall semantic feature vector to obtain a second reasoning result. Specifically, this can be achieved as follows: based on the overall semantic feature vector, matched with planned network cases associated with the network to be planned from a preset network planning database, and extracting historical network planning details from the planned network cases; the overall semantic feature vector, historical network planning details, and preset first parameter prompts are input into a preset content generation model to obtain the second reasoning result. Specifically, semantic dimension reasoning can be implemented based on a neural network channel; the neural network channel described here can be implemented using a pre-trained deep learning model; that is, based on the pre-trained deep learning model (i.e., the preset content generation model) in the neural network channel, implicit needs and fuzzy semantics in the query are processed, and combined with complex factors such as historical deployment cases, climate influencing factors, and user density correlations, potentially relevant considerations and optimization suggestions are identified to obtain the second reasoning result.

[0066] In one possible example embodiment, the content generation big model described herein may include an embedding mapping layer, an encoding layer, and a hybrid expert model. Based on this, the second inference result is obtained by inputting the overall semantic feature vector, historical network planning details, and preset first parameter prompts into the preset content generation big model. This can be achieved as follows: First, generate the basic information to be predicted based on the overall semantic feature vector and historical network planning details, and generate the context information for the big prediction based on the preset first parameter prompts. Second, perform embedding mapping processing on the basic information to be predicted using the embedding mapping layer (e.g., the Embedding embedding mapping layer) to obtain the network features to be planned, and perform embedding mapping processing on the context information to be predicted using the embedding mapping layer (e.g., the Bert embedding mapping layer) to obtain the context sequence. Then, encode the network features to be planned and the context sequence using the encoding layer (e.g., a multi-layer bidirectional Transformer) to obtain the overall context representation, and generate the second inference result by performing content generation on the context sequence and the overall context representation using the hybrid expert model.

[0067] Secondly, the target network planning result corresponding to the network to be planned is determined based on the first inference result and the second inference result, including: inputting the first inference result, the second inference result, and preset second parameter prompt information into a preset weight value prediction model to obtain the first weight value and the second weight value of the first inference result and the second inference result in the rule coverage dimension, the third weight value and the fourth weight value in the logical consistency dimension, and the fifth weight value and the sixth weight value in the confidence dimension; determining the first target weight value corresponding to the first inference result based on the first weight value, the third weight value, and the fifth weight value, and determining the second target weight value corresponding to the second inference result based on the second weight value, the fourth weight value, and the sixth weight value; and determining the target network planning result corresponding to the network to be planned based on the first inference result, the first target weight value, the second inference result, and the second target weight value. In other words, in practical applications, the processing results of the two channels (i.e., the first inference result and the second inference result) can be input into the fusion module. Then, the weights of the outputs of each channel are calculated based on factors such as rule coverage, logical consistency, and confidence. The weighted fusion strategy is used to generate the final structured inference result (i.e., the target network planning result). Meanwhile, the model structure of the preset weight value prediction model described here is largely similar to the model structure of the large model generated by the content described above, and the specific prediction process is also largely similar. No further details will be provided here.

[0068] Furthermore, the system identifies the query intent as "millimeter-wave rain attenuation analysis and protection planning," and outputs detailed parameter analysis results (i.e., target network planning results). These include a list of specific rain attenuation influencing factors (e.g., rainfall intensity, polarization, transmission distance, climate region, etc.), quantitative analysis data to provide detailed implementation parameters and configuration suggestions, protection measure suggestions (e.g., increasing rain attenuation margin by 6dB, adopting adaptive modulation and coding technology, deploying smart beamforming, etc.), and implementation priority assessment. The obtained target network planning results can be used as a reference. Figure 6 As shown in the figure. Practical application verification shows that the system achieves an accuracy rate of 97.3% in processing this type of query, a 47.3% improvement compared to traditional methods, and reduces solution design time by 62%, significantly improving the efficiency and quality of network planning. This specific embodiment fully demonstrates the powerful functions and processing capabilities of this patented technology in practical applications, reflecting its significant value and broad prospects in the field of network planning.

[0069] In one example embodiment, the semantic understanding-based network planning method further includes: dynamically evolving a preset dynamic knowledge graph based on the target network planning result; wherein the dynamic evolution is achieved by: determining the current network planning technology concept included in the target network planning result, and matching historical technology concepts corresponding to the current network planning technology concept in the preset dynamic knowledge graph; determining new technology concepts from the current network planning technology concepts based on the matched historical technology concepts, and extracting technology concepts to be added from the new technology concepts; evaluating the reliability and consistency of the technology concepts to be added to obtain a concept evaluation result, and when it is determined that the concept evaluation result indicates that the technology concepts to be added have reliability and consistency, integrating the technology concepts to be added into the preset dynamic knowledge graph to achieve the dynamic evolution.

[0070] In one example embodiment, the extraction of the technical concept to be added from the newly added technical concept can be achieved as follows: The importance of the newly added technical concept in the network planning scenario is evaluated based on a preset importance evaluation model to obtain an importance evaluation result; based on the importance evaluation result, the technical concept to be added is extracted from the newly added technical concept; wherein, if the importance evaluation result is greater than or equal to a preset importance threshold, the newly added technical concept corresponding to the importance evaluation result is determined as the technical concept to be added. The importance evaluation model described herein may include, but is not limited to, decision tree models, convolutional neural network models, and recurrent neural network models, etc., and this example does not impose any special limitations on this.

[0071] In one example embodiment, the reliability and consistency of the technical concept to be added are evaluated to obtain a concept evaluation result, which can be achieved as follows: the technical concept to be added is input into the generator in a preset adversarial generative network to obtain a concept generation result corresponding to the technical concept to be added; based on the discriminator in the preset adversarial generative network, the reliability and consistency of the technical concept to be added and the concept generation result corresponding to the technical concept to be added are judged to obtain the concept evaluation result.

[0072] In one example embodiment, the technical concept to be added is integrated into the preset dynamic knowledge graph to achieve dynamic evolution. This can be achieved as follows: determine the network planning entity included in the technical concept to be added and the entity attribute information of the network planning entity, and determine the mounting position of the technical concept to be added in the preset dynamic knowledge graph based on the network planning entity and the entity attribute information; based on the mounting position of the technical concept to be added in the preset dynamic knowledge graph, integrate the technical concept to be added into the preset dynamic knowledge graph to achieve dynamic evolution.

[0073] The following section will further explain and illustrate the dynamic evolution process of the pre-defined dynamic knowledge graph. Specifically, in practical applications, the system initiates a knowledge self-evolution process based on the reasoning results of this processing. The concept drift detection module analyzes the technical concepts involved in the results, compares them with the existing knowledge base, and identifies potential new knowledge or content that needs updating. The knowledge importance assessment algorithm evaluates the value of the identified concepts and filters out new knowledge with significant importance. For new knowledge that passes the evaluation, the system initiates an adversarial training verification process to ensure the reliability and consistency of the new knowledge. After successful verification, the new knowledge is integrated into the knowledge graph using a transactional update approach, including operations such as adding entities, updating attributes, and establishing relationships. Simultaneously, the system records detailed logs of this processing for subsequent performance analysis and optimization.

[0074] Thus, the semantic understanding-based network planning method described in the exemplary embodiments of this disclosure has been fully implemented. Based on the foregoing description, the semantic understanding-based network planning method described in the exemplary embodiments of this disclosure has the following key points: Firstly, a self-evolving domain knowledge graph construction method; specifically, this technical solution requires protection of a dynamic knowledge management mechanism based on a multi-source knowledge flow real-time perception architecture. This mechanism can continuously and automatically acquire and update professional knowledge from technical documents, industry standards, technical white papers, practical cases, and real-time network data in the field of network planning through web crawlers, application programming interfaces, and data flow processing technologies. Furthermore, it specifically requires protection of a graph update method based on incremental adversarial training. This method adopts the idea of ​​generative adversarial networks, ensuring the accuracy and consistency of knowledge updates through the adversarial training process of the generator and discriminator. Further, in the specific self-evolution process, a concept drift detection algorithm is also involved. This algorithm monitors the degree of change in the semantic representation of concepts and uses drift scores to quantify the degree of evolution of concept definitions. Based on this, the knowledge update process is automatically triggered when an important concept change is detected. The incremental update operation of the knowledge graph adopts a transactional processing mechanism to ensure that the data remains consistent and complete during the update process. These technological innovations enable the knowledge base to autonomously identify new concepts and trends in technological development and automatically complete the evolution and update of the knowledge system, ensuring that the system always has the latest and most comprehensive domain knowledge.

[0075] On the other hand, a joint reasoning framework encompassing neural symbols is required. In practical applications, this technical solution requires the protection of a dual-channel processor that integrates a Transformer deep learning model and a rule-based symbolic reasoning system. The symbolic reasoning channel employs a reasoning method based on descriptive logic and a rule engine, capable of handling specialized rules and constraints in network planning to ensure the accuracy and interpretability of the reasoning results. Simultaneously, the neural network channel utilizes deep learning and attention mechanisms to handle fuzzy semantics and complex contextual relationships in natural language. Specifically, a context-aware relational path reasoning algorithm is required, capable of dynamically adjusting the reasoning path and strategy based on the current network planning scenario and contextual information (e.g., matching the target neural network model from multiple expert neural network models in a hybrid expert model to perform corresponding content generation). Furthermore, an innovative result fusion strategy is included, comprising a credibility-weighted fusion method based on rule coverage and logical consistency. This method intelligently balances the results of symbolic reasoning and neural reasoning, fully leveraging the advantages of both reasoning methods. These technologies achieve an effective combination of the accuracy of symbolic reasoning and the flexibility of neural networks, significantly improving the ability to deeply understand specialized terminology and complex semantic relationships.

[0076] On the other hand, there is the cross-modal semantic alignment model. In practical applications, this technical solution requires the protection of a method for semantic mapping between natural language descriptions and multimodal information such as network topology and technical parameters. Specifically, the cross-modal semantic alignment model can simultaneously process multiple forms of information, such as text, images, and structured data, and identify semantic relationships between different modalities through an attention mechanism. Furthermore, the cross-modal fusion mechanism employs multi-head attention technology and a learnable projection matrix to map the feature representations of different modalities into a unified semantic space. This part specifically requires the protection of the multi-head cross-modal attention technology, which can capture complex relationships between modalities from multiple different representation subspaces, while the learnable cross-modal attention mechanism adaptively learns the optimal modal fusion strategy through an end-to-end training process. These technologies can effectively process multi-source heterogeneous information such as text descriptions, technical parameters, network topology diagrams, and performance indicators, achieving true multimodal semantic understanding and alignment, and providing comprehensive and accurate information support for network planning.

[0077] Based on the foregoing description, the exemplary embodiments of this disclosure have the following beneficial effects: Firstly, in terms of knowledge management and updating: The system achieves dynamic evolution and autonomous updating capabilities of the knowledge base, completely solving the industry problem that traditional static knowledge bases cannot adapt to the rapid evolution of network technology; simultaneously, through multi-source knowledge flow perception and concept drift detection, the system can automatically capture and absorb emerging technology concepts such as 5G and 6G, maintaining the cutting-edge nature and integrity of the knowledge system; furthermore, the update mechanism based on incremental adversarial training ensures the consistency and accuracy of knowledge addition, avoiding errors and omissions that may result from manual updates. This innovation significantly reduces the manual and time costs of knowledge maintenance, improves the efficiency and reliability of knowledge management, and provides a solid and accurate knowledge foundation for network planning.

[0078] Secondly, in terms of semantic understanding and reasoning: a qualitative leap has been achieved through the neural-symbolic joint reasoning framework, significantly improving the accuracy, depth, and reliability of query understanding. Simultaneously, the dual-channel processing mechanism fully leverages the respective advantages of symbolic reasoning and neural reasoning. Symbolic reasoning ensures the precision and interpretability of technical terms and rule processing, while neural reasoning provides the ability to handle the complexity and ambiguity of natural language. Context-aware reasoning algorithms enable the system to provide more accurate and relevant reasoning results based on specific planning scenarios and needs. Practical application data shows that this technology improves the accuracy of network planning query understanding by 47.3%, greatly enhancing the practicality and reliability of semantic understanding and providing strong technical support for network planning decisions.

[0079] Thirdly, in terms of multimodal information processing: cross-modal semantic alignment technology has achieved breakthrough progress, solving the problem of deep fusion and effective utilization of multi-source heterogeneous information. The system can simultaneously process and understand multiple forms of information such as text descriptions, technical parameters, network topology diagrams, and performance data, and establish accurate semantic mapping relationships through an attention mechanism. This deep fusion capability enables the system to understand planning requirements from multiple angles and levels, providing more comprehensive and accurate analysis results. Multimodal processing not only enhances the system's understanding capabilities but also expands its application scope, enabling it to handle more complex and diverse network planning scenarios.

[0080] Fourthly, regarding overall system performance and adaptability: the innovative closed-loop learning architecture endows the system with powerful capabilities for continuous optimization and self-evolution. The system not only efficiently processes current natural language queries but also learns from each interaction through a dynamic knowledge evolution mechanism, continuously enriching and improving its knowledge system and reasoning abilities. This self-evolutionary capability ensures the system maintains high performance over the long term, adapting to the rapidly changing technological environment of network planning and avoiding the problem of traditional systems gradually falling behind due to technological advancements. System performance continuously improves with usage time, forming a virtuous cycle of learning and evolution.

[0081] Fifthly, regarding practical application value and economic benefits: This technology brings significant practical value and economic benefits. By providing accurate and efficient natural language understanding capabilities and intelligent planning support, the system can greatly improve the efficiency and quality of network planning. Practical application data shows that the system can reduce network planning design time by up to 62%, while improving the accuracy and reliability of planning schemes and reducing subsequent problems caused by planning errors and insufficient optimization. These improvements bring direct economic benefits to the network construction and operation and maintenance work of telecommunications companies, improve network resource utilization efficiency, reduce operating costs, and enhance market competitiveness. Furthermore, this technology also helps promote the accumulation and inheritance of network planning knowledge, improving the overall technical level and development speed of the industry.

[0082] The following are embodiments of the apparatus disclosed herein, which can be used to execute embodiments of the method disclosed herein. For details not disclosed in the apparatus embodiments of this disclosure, please refer to the embodiments of the method disclosed herein.

[0083] This disclosure also provides an example embodiment of a network planning apparatus based on semantic understanding. Specifically, refer to... Figure 7 As shown, the semantic understanding-based network planning device may include a text semantic representation generation module 710, a domain knowledge representation generation module 720, an overall semantic feature vector generation module 730, and a target network planning result determination module 740. Wherein: The text semantic representation generation module 710 can be used to receive a query question sentence corresponding to the network to be planned and network context information, and generate a text semantic representation based on the query question sentence; the domain knowledge representation generation module 720 can be used to generate a context feature representation based on the network context information, and generate a domain knowledge representation based on the network planning domain knowledge corresponding to the network to be planned; the overall semantic feature vector generation module 730 can be used to fuse the text semantic representation, context feature representation and domain knowledge representation to obtain an overall semantic feature vector, and perform logical dimension reasoning on the overall semantic feature vector to obtain a first reasoning result; the target network planning result determination module 740 can be used to perform semantic dimension reasoning on the overall semantic feature vector to obtain a second reasoning result, and determine the target network planning result corresponding to the network to be planned based on the first reasoning result and the second reasoning result.

[0084] In one example embodiment of this disclosure, generating a text semantic representation based on the query question sentence includes: performing deep semantic parsing on the query question sentence based on a preset element extraction model to obtain key semantic elements included in the query question; analyzing the semantic relationships between the key semantic elements based on a preset semantic analysis model to obtain the association relationships between the key semantic elements; and generating the text semantic representation based on the key semantic elements and the association relationships between the key semantic elements.

[0085] In one example embodiment of this disclosure, generating a context feature representation based on network context information includes: extracting network type information, network planning stage information, and network planning technical parameter information corresponding to the network to be planned from the network context information; wherein, the network planning technical parameter information includes at least one of the operating frequency band, network coverage area, and expected user density of the network to be planned; constructing a network planning background model corresponding to the network to be planned based on the network type information, network planning stage information, and network planning technical parameter information, and generating the context feature representation based on the network planning background model.

[0086] In one example embodiment of this disclosure, generating a domain knowledge representation based on network planning domain knowledge corresponding to the network to be planned includes: querying network planning domain knowledge corresponding to the network to be planned from a preset dynamic knowledge graph based on key semantic elements in the query question sentence and network context information; generating a domain knowledge representation based on the network planning domain knowledge; wherein the network planning domain knowledge includes at least one of propagation characteristics, influence factor calculation models, network deployment specifications, and climate influence factors.

[0087] In one example embodiment of this disclosure, the fusion of text semantic representation, context feature representation, and domain knowledge representation to obtain an overall semantic feature vector includes: performing embedding mapping processing on the text semantic representation based on a preset first embedding mapping model to obtain a first representation vector, and performing embedding mapping processing on the context feature representation based on a preset second embedding mapping model to obtain a second representation vector; performing embedding mapping processing on the domain knowledge representation based on a preset third embedding mapping model to obtain a third representation vector, and projecting the first representation vector, the second representation vector, and the third representation vector onto a trained projection matrix for semantic alignment to obtain the overall semantic feature vector.

[0088] In one example embodiment of this disclosure, the first reasoning result is obtained by logically reasoning the overall semantic feature vector, including: matching the standard network planning rule corresponding to the network to be planned from a preset network planning rule base according to the overall semantic feature vector; and performing logical reasoning on the network planning details information corresponding to the network to be planned in the standard network planning rule based on a preset symbol model to obtain the first reasoning result.

[0089] In one example embodiment of this disclosure, the second reasoning result is obtained by performing semantic dimension reasoning on the overall semantic feature vector, including: matching planned network cases associated with the network to be planned from a preset network planning database based on the overall semantic feature vector, and extracting historical network planning details from the planned network cases; inputting the overall semantic feature vector, historical network planning details, and preset first parameter prompt information into a preset content generation big model to obtain the second reasoning result.

[0090] In one example embodiment of this disclosure, determining the target network planning result corresponding to the network to be planned based on the first inference result and the second inference result includes: inputting the first inference result, the second inference result, and preset second parameter prompt information into a preset weight value prediction model to obtain the first weight value and the second weight value of the first inference result and the second inference result in the rule coverage dimension, the third weight value and the fourth weight value in the logical consistency dimension, and the fifth weight value and the sixth weight value in the confidence dimension; determining the first target weight value corresponding to the first inference result based on the first weight value, the third weight value, and the fifth weight value, and determining the second target weight value corresponding to the second inference result based on the second weight value, the fourth weight value, and the sixth weight value; and determining the target network planning result corresponding to the network to be planned based on the first inference result, the first target weight value, the second inference result, and the second target weight value.

[0091] In one example embodiment of this disclosure, the semantic understanding-based network planning device further includes a dynamic evolution module, which can be used to dynamically evolve a preset dynamic knowledge graph based on the target network planning result; wherein the dynamic evolution is achieved by: determining the current network planning technology concept included in the target network planning result, and matching the historical technology concept corresponding to the current network planning technology concept in the preset dynamic knowledge graph; determining a new technology concept from the current network planning technology concept based on the matched historical technology concept, and extracting the technology concept to be added from the new technology concept; evaluating the reliability and consistency of the technology concept to be added to obtain a concept evaluation result, and when it is determined that the concept evaluation result indicates that the technology concept to be added has reliability and consistency, integrating the technology concept to be added into the preset dynamic knowledge graph to achieve the dynamic evolution.

[0092] In one example embodiment of this disclosure, extracting a technical concept to be added from the newly added technical concept includes: evaluating the importance of the newly added technical concept in a network planning scenario based on a preset importance assessment model to obtain an importance assessment result; extracting a technical concept to be added from the newly added technical concept according to the importance assessment result; wherein, if the importance assessment result is greater than or equal to a preset importance threshold, then the newly added technical concept corresponding to the importance assessment result is determined as a technical concept to be added.

[0093] In one example embodiment of this disclosure, the reliability and consistency of the technical concept to be added are evaluated to obtain a concept evaluation result, including: inputting the technical concept to be added into a generator in a preset adversarial generative network to obtain a concept generation result corresponding to the technical concept to be added; judging the reliability and consistency of the technical concept to be added and the concept generation result corresponding to the technical concept to be added based on the discriminator in the preset adversarial generative network to obtain the concept evaluation result.

[0094] In one example embodiment of this disclosure, the technical concept to be added is integrated into the preset dynamic knowledge graph to achieve dynamic evolution, including: determining the network planning entity included in the technical concept to be added and the entity attribute information of the network planning entity, and determining the mounting position of the technical concept to be added in the preset dynamic knowledge graph based on the network planning entity and the entity attribute information; and integrating the technical concept to be added into the preset dynamic knowledge graph based on the mounting position of the technical concept to be added in the preset dynamic knowledge graph to achieve dynamic evolution.

[0095] The specific details of each module in the above semantic understanding-based network planning device have been described in detail in the corresponding semantic understanding-based network planning method, so they will not be repeated here.

[0096] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0097] Furthermore, although the steps of the method in this disclosure are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps may be omitted, multiple steps may be combined into one step, and / or a step may be broken down into multiple steps.

[0098] In an exemplary embodiment of this disclosure, an electronic device capable of implementing the above-described method is also provided.

[0099] Those skilled in the art will understand that various aspects of this disclosure can be implemented as a system, method, or program product. Therefore, various aspects of this disclosure can be specifically implemented in the following forms: a completely hardware implementation, a completely software implementation (including firmware, microcode, etc.), or a combination of hardware and software aspects, collectively referred to herein as a "circuit," "module," or "system."

[0100] The following reference Figure 8 To describe an electronic device 800 according to such an embodiment of the present disclosure. Figure 8 The electronic device 800 shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein. Figure 8 As shown, the electronic device 800 is manifested in the form of a general-purpose computing device. The components of the electronic device 800 may include, but are not limited to: at least one processing unit 810, at least one storage unit 820, a bus 830 connecting different system components (including storage unit 820 and processing unit 810), and a display unit 840.

[0101] The storage unit stores program code that can be executed by the processing unit 810, causing the processing unit 810 to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure. For example, the processing unit 810 can perform actions such as... Figure 1 The steps shown are as follows: Step S110: Receive the query question sentence corresponding to the network to be planned and network context information, and generate a text semantic representation based on the query question sentence; Step S120: Generate a context feature representation based on the network context information, and generate a domain knowledge representation based on the network planning domain knowledge corresponding to the network to be planned; Step S130: Fuse the text semantic representation, context feature representation and domain knowledge representation to obtain an overall semantic feature vector, and perform logical dimension reasoning on the overall semantic feature vector to obtain a first reasoning result; Step S140: Perform semantic dimension reasoning on the overall semantic feature vector to obtain a second reasoning result, and determine the target network planning result corresponding to the network to be planned based on the first reasoning result and the second reasoning result.

[0102] Storage unit 820 may include a readable medium in the form of a volatile storage unit, such as random access memory (RAM) 8201 and / or cache memory 8202, and may further include a read-only memory (ROM) 8203.

[0103] The storage unit 820 may also include a program / utility 8204 having a set (at least one) of program modules 8205, such program modules 8205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0104] Bus 830 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the various bus structures.

[0105] Electronic device 800 can also communicate with one or more external devices 900 (e.g., keyboard, pointing device, Bluetooth device, etc.), and with one or more devices that enable a user to interact with electronic device 800, and / or with any device that enables electronic device 800 to communicate with one or more other computing devices (e.g., router, modem, etc.). This communication can be performed via input / output (I / O) interface 850. Furthermore, electronic device 800 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 860. As shown, network adapter 860 communicates with other modules of electronic device 800 via bus 830. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 800, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0106] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, terminal device, or network device, etc.) to execute the methods according to the embodiments of this disclosure.

[0107] In exemplary embodiments of this disclosure, a computer-readable storage medium is also provided, on which a program product capable of implementing the methods described above is stored. In some possible implementations, various aspects of this disclosure may also be implemented as a program product including program code that, when the program product is run on a terminal device, causes the terminal device to perform the steps of the various exemplary embodiments of this disclosure described in the "Exemplary Methods" section above.

[0108] The program product for implementing the above-described method according to embodiments of the present disclosure may employ a portable compact disc read-only memory (CD-ROM) and include program code, and may run on a terminal device, such as a personal computer. However, the program product of the present disclosure is not limited thereto. In this document, the readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.

[0109] The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0110] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of sending, propagating, or transmitting programs for use by or in conjunction with an instruction execution system, apparatus, or device.

[0111] The program code contained on the readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0112] Program code for performing the operations of this disclosure can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0113] Furthermore, the foregoing figures are merely illustrative of the processes included in the methods according to exemplary embodiments of this disclosure and are not intended to be limiting. It is readily understood that the processes shown in the foregoing figures do not indicate or limit the temporal order of these processes. Additionally, it is readily understood that these processes may be performed synchronously or asynchronously, for example, in multiple modules. Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not invented by this disclosure. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the claims.

Claims

1. A network planning method based on semantic understanding, characterized in that, include: Receive the query question sentence corresponding to the network to be planned and the network context information, and generate a text semantic representation based on the query question sentence; Context feature representations are generated based on network context information, and domain knowledge representations are generated based on network planning domain knowledge corresponding to the network to be planned. The text semantic representation, context feature representation, and domain knowledge representation are fused to obtain an overall semantic feature vector, and logical reasoning is performed on the overall semantic feature vector to obtain a first reasoning result; The semantic dimension reasoning of the overall semantic feature vector is used to obtain a second reasoning result, and the target network planning result corresponding to the network to be planned is determined based on the first reasoning result and the second reasoning result.

2. The network planning method based on semantic understanding according to claim 1, characterized in that, Generate a text semantic representation based on the query sentence, including: Based on a preset element extraction model, deep semantic parsing is performed on the query question sentence to obtain the key semantic elements included in the query question; Based on a preset semantic analysis model, the semantic relationships between the key semantic elements are analyzed to obtain the association relationships between the key semantic elements, and the text semantic representation is generated according to the key semantic elements and the association relationships between the key semantic elements.

3. The network planning method based on semantic understanding according to claim 1, characterized in that, The system generates contextual feature representations based on network context information, including: Extract the network type information, network planning stage information, and network planning technical parameter information corresponding to the network to be planned from the network context information; wherein, the network planning technical parameter information includes at least one of the operating frequency band, network coverage area, and expected user density of the network to be planned; Based on the network type information, network planning stage information, and network planning technical parameter information, a network planning background model corresponding to the network to be planned is constructed, and the context feature representation is generated based on the network planning background model.

4. The network planning method based on semantic understanding according to claim 1, characterized in that, Generate a domain knowledge representation based on the network planning domain knowledge corresponding to the network to be planned, including: Based on the key semantic elements in the query question sentence and the network context information, query the network planning domain knowledge corresponding to the network to be planned from the preset dynamic knowledge graph; Domain knowledge representation is generated based on knowledge of network planning; wherein, the knowledge of network planning includes at least one of propagation characteristics, influence factor calculation models, network deployment specifications, and climate influence factors.

5. The network planning method based on semantic understanding according to claim 1, characterized in that, The overall semantic feature vector is obtained by fusing text semantic representation, context feature representation, and domain knowledge representation, including: The text semantic representation is embedded and mapped based on a preset first embedding mapping model to obtain a first representation vector, and the context feature representation is embedded and mapped based on a preset second embedding mapping model to obtain a second representation vector; The domain knowledge representation is embedded and mapped based on a preset third embedding mapping model to obtain a third representation vector. The first, second, and third representation vectors are then projected onto a trained projection matrix for semantic alignment to obtain the overall semantic feature vector.

6. The network planning method based on semantic understanding according to claim 1, characterized in that, The first reasoning result is obtained by performing logical-dimensional reasoning on the overall semantic feature vector, including: Based on the overall semantic feature vector, standard network planning rules corresponding to the network to be planned are matched from a preset network planning rule base. Based on a preset symbolic model, logical reasoning is performed on the network planning details information corresponding to the network to be planned in the standard network planning rules to obtain the first reasoning result.

7. The network planning method based on semantic understanding according to claim 1, characterized in that, The second reasoning result is obtained by performing semantic dimension reasoning on the overall semantic feature vector, including: Based on the overall semantic feature vector, the network planning database is used to match the planned network cases associated with the network to be planned, and the historical network planning details in the planned network cases are extracted. The overall semantic feature vector, historical network planning details, and preset first parameter prompts are input into a preset content generation model to obtain the second inference result.

8. The network planning method based on semantic understanding according to claim 1, characterized in that, Based on the first and second inference results, the target network planning result corresponding to the network to be planned is determined, including: The first inference result, the second inference result, and the preset second parameter prompt information are input into the preset weight value prediction model to obtain the first weight value and the second weight value of the first inference result and the second inference result in the rule coverage dimension, the third weight value and the fourth weight value in the logical consistency dimension, and the fifth weight value and the sixth weight value in the confidence dimension. A first target weight value corresponding to the first inference result is determined based on the first weight value, the third weight value, and the fifth weight value; and a second target weight value corresponding to the second inference result is determined based on the second weight value, the fourth weight value, and the sixth weight value. Based on the first inference result and the first target weight value, the second inference result and the second target weight value, the target network planning result corresponding to the network to be planned is determined.

9. The network planning method based on semantic understanding according to claim 1, characterized in that, The semantic understanding-based network planning method also includes: The preset dynamic knowledge graph is dynamically evolved based on the target network planning results; wherein the dynamic evolution is achieved in the following manner: Identify the current network planning technology concepts included in the target network planning results, and match the historical technology concepts corresponding to the current network planning technology concepts in the preset dynamic knowledge graph; Based on the matched historical technical concepts, new technical concepts are determined from the current network planning technical concepts, and new technical concepts to be added are extracted from the new technical concepts. The reliability and consistency of the proposed technical concept are evaluated to obtain a concept evaluation result. When the concept evaluation result determines that the proposed technical concept has reliability and consistency, the proposed technical concept is integrated into the preset dynamic knowledge graph to achieve the dynamic evolution.

10. The network planning method based on semantic understanding according to claim 9, characterized in that, Extracting the technical concepts to be added from the newly added technical concepts, including: The importance of the newly added technical concept in the network planning scenario is evaluated based on a preset importance evaluation model to obtain the importance evaluation result; Based on the importance assessment results, technical concepts to be added are extracted from the newly added technical concepts; wherein, if the importance assessment results are greater than or equal to a preset importance threshold, the newly added technical concept corresponding to the importance assessment results is determined as the technical concept to be added.

11. The network planning method based on semantic understanding according to claim 9, characterized in that, The reliability and consistency of the proposed technical concepts are evaluated to obtain concept evaluation results, including: The technical concept to be added is input into the generator in the preset adversarial generative network to obtain the concept generation result corresponding to the technical concept to be added; The discriminator in the preset adversarial generative network judges the reliability and consistency of the technical concept to be added and the concept generation results corresponding to the technical concept to be added, and obtains the concept evaluation result.

12. The network planning method based on semantic understanding according to claim 9, characterized in that, Integrating the proposed technological concepts into the preset dynamic knowledge graph to achieve dynamic evolution includes: The network planning entity and its entity attribute information included in the technical concept to be added are determined, and the mounting position of the technical concept to be added in the preset dynamic knowledge graph is determined based on the network planning entity and its entity attribute information. Based on the mounting position of the technical concept to be added in the preset dynamic knowledge graph, the technical concept to be added is integrated into the preset dynamic knowledge graph to achieve dynamic evolution.

13. A network planning device based on semantic understanding, characterized in that, include: The text semantic representation generation module is used to receive the query question sentence corresponding to the network to be planned and the network context information, and generate a text semantic representation based on the query question sentence; The domain knowledge representation generation module is used to generate a context feature representation based on network context information, and to generate a domain knowledge representation based on the network planning domain knowledge corresponding to the network to be planned. The overall semantic feature vector generation module is used to fuse the text semantic representation, context feature representation and domain knowledge representation to obtain an overall semantic feature vector, and to perform logical dimension reasoning on the overall semantic feature vector to obtain a first reasoning result; The target network planning result determination module is used to perform semantic dimension reasoning on the overall semantic feature vector to obtain a second reasoning result, and determine the target network planning result corresponding to the network to be planned based on the first reasoning result and the second reasoning result.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the semantic understanding-based network planning method according to any one of claims 1-12.

15. An electronic device, characterized in that, include: processor; as well as Memory for storing the executable instructions of the processor; The processor is configured to execute the semantic understanding-based network planning method according to any one of claims 1-12 by executing the executable instructions.

Images