Wireless network operation optimization method and system based on machine learning

Abstract

The invention provides a wireless network operation optimization method and system based on machine learning. The method comprises the following steps: collecting multi-mode operation data of a wireless network in real time; based on the operation optimization decision model, an operation optimization strategy is decided according to the multi-modal operation data; wherein the operation optimization decision model is obtained based on machine learning pre-training; and performing corresponding operation optimization on the wireless network based on the operation optimization strategy. According to the method, the operation optimization decision model is pre-trained based on machine learning, the operation optimization strategy is decided based on the operation optimization decision model according to the multi-mode operation data of the wireless network, and corresponding operation optimization is performed on the wireless network based on the operation optimization strategy. Therefore, the system can adapt to rapid changes of network requirements and respond to fluctuation of a network environment in real time, the processing capacity of the system for complex network topology and mass data traffic is improved, and network resource waste and service quality reduction are avoided.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and particularly relates to a method and system for optimizing the operation of a wireless network based on machine learning. Background Art

[0002] At present, with the development of technologies such as 5G, the application scenarios of the network have gradually become diversified. The popularization of Internet of Things (IoT) devices has led to an exponential growth in the number of devices, resulting in drastic fluctuations in network traffic and requirements for low latency and high real-time performance. In addition to traditional mobile communication requirements, emerging application scenarios such as industrial Internet, intelligent transportation, autonomous driving, and smart home have put forward more stringent performance requirements for wireless networks. In particular, the performance in terms of bandwidth, latency, reliability, etc. is crucial for these key applications. At the same time, the personalization and diversification of user needs have further exacerbated the complexity of the network. Especially in the same network, different quality of service (QoS) guarantees need to be provided for different application scenarios and services.

[0003] Under this background, traditional wireless network optimization methods are facing unprecedented challenges. Static spectrum allocation strategies often fail to adapt to the rapid changes in network demands. The method of manually configuring base station parameters cannot respond in real time to the fluctuations in the network environment. In addition, traditional methods have limited processing capabilities for complex network topologies and massive data traffic, and often perform poorly in high-density user scenarios, resulting in waste of network resources and degradation of service quality.

[0004] Therefore, a solution is urgently needed. Summary of the Invention

[0005] One of the purposes of the present invention is to provide a method for optimizing the operation of a wireless network based on machine learning. Based on a pre-trained operation optimization decision model of machine learning, according to the multi-modal operation data of the wireless network, an operation optimization strategy is determined. Based on the operation optimization strategy, the corresponding operation optimization of the wireless network is carried out, enabling the system to adapt to the rapid changes in network demands and respond in real time to the fluctuations in the network environment, improving the system's processing capabilities for complex network topologies and massive data traffic, and avoiding waste of network resources and degradation of service quality.

[0006] A method for optimizing the operation of a wireless network based on machine learning provided by an embodiment of the present invention includes:

[0007] Collecting multi-modal operation data of the wireless network in real time;

[0008] Based on the operation optimization decision model, determining an operation optimization strategy according to the multi-modal operation data; wherein, the operation optimization decision model is pre-trained based on machine learning.

[0009] Based on the operation optimization strategy, the wireless network is correspondingly optimized in operation.

[0010] Preferably, the method for optimizing the operation of a wireless network based on machine learning further includes:

[0011] Obtain the operation optimization history of the wireless network;

[0012] Analyze the operation optimization history to determine the triggering degree;

[0013] When the triggering degree exceeds the triggering degree threshold, trigger the activation of the application circle of the wireless network;

[0014] Obtain the first transfer information generated by the application circle within the first time period after triggering activation;

[0015] Based on the first transfer information, decide the feature map representation template;

[0016] Based on the feature map representation template, perform feature map representation processing on the operation optimization history to obtain a historical feature map;

[0017] Guide the interaction between the application circle and the historical feature map;

[0018] Obtain the second transfer information generated within the second time period after the interaction between the application circle and the historical feature map;

[0019] Add the second transfer information to the newly collected real-time multi-modal operation data.

[0020] Preferably, the analyzing the operation optimization history to determine the triggering degree includes:

[0021] Perform a situation analysis on the operation optimization history to obtain a set of situations;

[0022] Match the first situation in the situation set with the second situation in the triggering situation library;

[0023] When the match is in line, the triggering degree is counted as the first threshold; wherein, the first threshold exceeds the triggering degree threshold;

[0024] When none of them match, the triggering degree is counted as the sum of the first quantization values corresponding to the situation types of each first situation in the situation set in the first quantization value library.

[0025] Preferably, the deciding the feature map representation template based on the first transfer information includes:

[0026] Perform a time series representation on the first transfer information to obtain an information time series;

[0027] Based on the sequence item cluster division constraint, divide multiple sequence item clusters in the information time series;

[0028] Parse each sequence item cluster, and determine the template decision basis and transfer weight of each sequence item cluster;

[0029] Determine the template decision rule corresponding to the template decision basis of each sequence item cluster in the first J sequence item clusters in the information time series from the template decision rule library; wherein, the difference between the sum of the transfer weights of the first J + 1 sequence item clusters in the information time series and the sum of the transfer weights of the first J sequence item clusters is the largest;

[0030] Integrate each template decision rule to determine the feature map representation template;

[0031] Wherein, the sequence item cluster division constraint includes:

[0032] The sequence items in the sequence item cluster are arranged continuously in sequence in the information time series;

[0033] And, the data type sets of the sequence items in the sequence item cluster match the standard data type set.

[0034] Preferably, the parsing each sequence item cluster, determining the template decision basis and transfer weight of each sequence item cluster includes:

[0035] Based on the feature extraction template, perform feature extraction on the sequence item cluster to obtain a feature set;

[0036] Use the feature set as the template decision basis corresponding to the sequence item cluster;

[0037] When 1 ≤ i ≤ K, the transfer weight of the i-th sequence item cluster in the information time series is counted as the second quantization value corresponding to the basis type of the template decision basis of the i-th sequence item cluster in the second quantization value library; where K is the downward integer value of the product of the total number of sequence item clusters in the information time series and the proportionality coefficient;

[0038] When K < i ≤ N, the transfer weight of the i-th sequence item cluster in the information time series is counted as the third quantization value jointly corresponding to the basis type of the template decision basis of the i-th sequence item cluster and the number of template decision bases in the first i - 1 sequence item clusters in the information time series that have a standard basis association relationship with the template decision basis of the i-th sequence item cluster in the third quantization value library; where N is the total number of sequence item clusters in the information time series.

[0039] Preferably, the interaction between the guided application circle and the historical feature map includes:

[0040] Based on the connection guidance template, perform connection guidance on the application circle;

[0041] Obtain multiple connection relationships established after the application circle accepts the connection guidance;

[0042] Slice the application circle into multiple circle slices based on each connection relationship; among them, multiple application targets in the same circle slice jointly belong to at least one connection relationship;

[0043] Push the historical feature map to each circle slice;

[0044] Obtain the internal behavior set generated after each circle slice receives the historical feature map;

[0045] Based on the internal behavior set, match each pair of circle slices; among them, the similarity between the internal behavior sets generated by the paired circle slices exceeds the similarity threshold.

[0046] A wireless network operation optimization system based on machine learning provided by an embodiment of the present invention includes:

[0047] A collection module for collecting multi-modal operation data of the wireless network in real time;

[0048] A decision module for making an operation optimization strategy decision based on the operation optimization decision model and according to the multi-modal operation data; among them, the operation optimization decision model is pre-trained based on machine learning;

[0049] An optimization module for performing corresponding operation optimization on the wireless network based on the operation optimization strategy.

[0050] Optionally, the wireless network operation optimization system based on machine learning further includes:

[0051] An auxiliary module for:

[0052] Obtain the operation optimization history of the wireless network;

[0053] Analyze the operation optimization history and determine the triggering degree;

[0054] When the triggering degree exceeds the triggering degree threshold, trigger the activation of the application circle of the wireless network;

[0055] Obtain the first transfer information generated within the first time period after the application circle is triggered and activated;

[0056] Based on the first transfer information, make a decision on the feature map representation template;

[0057] Based on the feature map representation template, perform feature map representation processing on the operation optimization history to obtain the historical feature map;

[0058] Guide the interaction between the application circle and the historical feature map;

[0059] Obtain the second transfer information generated within the second time period after the interaction between the application circle and the historical feature map;

[0060] Add the second transfer information to the newly and real-time collected multimodal operation data.

[0061] Optionally, the auxiliary module analyzes the operation optimization history to determine the triggering degree, including:

[0062] Perform scenario analysis on the operation optimization history to obtain a set of scenarios;

[0063] Match the first scenario in the set of scenarios with the second scenario in the triggering scenario library;

[0064] When the match is met, the triggering degree is counted as the first threshold; wherein, the first threshold exceeds the triggering degree threshold;

[0065] When none of the matches are met, the triggering degree is counted as the sum of the first quantization values corresponding to the scenario types of each first scenario in the set of scenarios in the first quantization value library.

[0066] Optionally, the auxiliary module determines the template for representing the decision feature map based on the first transfer information, including:

[0067] Perform a time series representation on the first transfer information to obtain an information time series;

[0068] Based on the sequence item cluster division constraint, divide multiple sequence item clusters in the information time series;

[0069] Analyze each sequence item cluster to determine the template decision basis and transfer weight of each sequence item cluster;

[0070] Determine the template decision rule corresponding to the template decision basis of each of the first J sequence item clusters in the information time series from the template decision rule library; wherein, the difference between the sum of the transfer weights of the first J + 1 sequence item clusters and the sum of the transfer weights of the first J sequence item clusters in the information time series is the largest;

[0071] Integrate each template decision rule to determine the template for representing the feature map;

[0072] Wherein, the sequence item cluster division constraint includes:

[0073] Each sequence item in the sequence item cluster is arranged continuously in sequence in the information time series;

[0074] And, the data type set of each sequence item in the sequence item cluster matches the standard data type set.

[0075] Other features and advantages of the present invention will be described in the subsequent description, and some will be obvious from the description, or understood by implementing the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the structures specifically pointed out in the written description and the drawings.

[0076] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Description of the Drawings

[0077] The accompanying drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention, and do not constitute a limitation to the present invention. In the accompanying drawings:

[0078] Figure 1 It is a schematic diagram of a method for optimizing the operation of a wireless network based on machine learning in an embodiment of the present invention;

[0079] Figure 2 It is a schematic diagram of a system for optimizing the operation of a wireless network based on machine learning in an embodiment of the present invention. Detailed Embodiments

[0080] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0081] An embodiment of the present invention provides a method for optimizing the operation of a wireless network based on machine learning, as Figure 1 shown, including:

[0082] S1. Collect multi-modal operation data of the wireless network in real time;

[0083] S2. Based on the operation optimization decision model, make a decision on the operation optimization strategy according to the multi-modal operation data; wherein, the operation optimization decision model is obtained based on machine learning;

[0084] S3. Based on the operation optimization strategy, perform corresponding operation optimization on the wireless network.

[0085] The working principle and beneficial effects of the above technical solution are:

[0086] The multi-modal operation data at least includes various network state parameters such as network signal strength, traffic data, latency, network load, spectrum usage, etc.; the multi-modal operation data reflects the network state of the wireless network in real time, so operation optimization decisions can be made based on this. Based on machine learning, model training is carried out according to a large amount of empirical information on optimizing the operation of the wireless network (such as: records of optimizing the operation of the wireless network in history and producing positive optimization effects, etc.) to obtain an operation optimization decision model; the operation optimization decision model can automatically decide how to optimize the operation of the wireless network according to the multi-modal operation data, that is, decide the operation optimization strategy; through learning and adaptive adjustment, the operation optimization decision model can dynamically adapt to different network environments and load changes, and then generate a personalized optimization plan for the current wireless network. Finally, based on the operation optimization strategy, the corresponding operation optimization is carried out on the wireless network, and the optimization measures at least include dynamic allocation of spectrum resources, wireless power adjustment, load balancing, interference management, etc.

[0087] In an embodiment of the present invention, based on machine learning, an operation optimization decision model is pre-trained, and based on the operation optimization decision model, according to the multi-modal operation data of the wireless network, the operation optimization strategy is decided, and based on the operation optimization strategy, the corresponding operation optimization is carried out on the wireless network, so that the system can adapt to the rapid changes in network requirements and respond to the fluctuations of the network environment in real time, improving the system's processing ability for complex network topologies and massive data traffic, and avoiding waste of network resources and degradation of service quality.

[0088] In one embodiment, the method for optimizing the operation of a wireless network based on machine learning further includes:

[0089] Obtain the operation optimization history of the wireless network;

[0090] Analyze the operation optimization history to determine the triggering degree;

[0091] When the triggering degree exceeds the triggering degree threshold, trigger the activation of the application circle of the wireless network;

[0092] Obtain the first transfer information generated by the application circle within the first time period after triggering activation;

[0093] Based on the first transfer information, decide the feature map representation template;

[0094] Based on the feature map representation template, perform feature map representation processing on the operation optimization history to obtain a historical feature map;

[0095] Guide the interaction between the application circle and the historical feature map;

[0096] Obtain the second transfer information generated within the second time period after the interaction between the application circle and the historical feature map;

[0097] Add the second transfer information to the newly and real-time collected multi-modal operation data.

[0098] The working principle and beneficial effects of the above technical solution are as follows:

[0099] The operation optimization history at least includes the historical operation optimization strategies used for the operation optimization of the wireless network in history and the information on the effects of the wireless network after the use of the historical operation optimization strategies, etc. The application circle includes multiple application targets of the wireless network, such as users and devices provided with communication services by the wireless network, operation and maintenance experts of the wireless network, etc.; the triggering degree represents the degree to which the operation optimization history needs to be concerned by the application circle, and the triggering degree threshold is a threshold representing a relatively large triggering degree, such as 8. When the triggering degree exceeds the triggering degree threshold, it means that the operation optimization history must be concerned by the application circle, triggering the activation of the application circle of the wireless network. When triggered and activated, it is prompted that each application target in the application circle is ready to interact. The first time period can be a time period with a duration of 10 minutes. After the application circle is triggered and activated, each application target among them can actively exchange information, and the information exchanged by them will circulate in the application circle to form the first transfer information, that is, the first transfer information at least includes the feedback information on the historical use of the wireless network by the application target, the demand information on the future use of the wireless network by the application target, and the demand information on the future collaborative use of the wireless network by multiple application targets, etc. The decision feature map representation template is a template for the system to perform feature map representation processing on the operation optimization history. The first transfer information reflects the comprehensive application service situation of the application circle. Therefore, based on its decision feature map representation template. After obtaining the historical feature map, guide the interaction between the application circle and the historical feature map. At this time, each application entity in the application circle will understand the historical operation optimization situation of the wireless network through the historical feature map and interact with the historical feature map, such as operating to view the historical feature map, etc. Similarly, after the interaction between the application circle and the historical feature map, each application target among them can actively exchange information, then the second transfer information is generated. The second transfer information at least includes the feedback information on the historical use of the wireless network further proposed by each application target after understanding the historical operation optimization situation of the wireless network through the historical feature map, the demand information on the future use of the wireless network, and the demand information on the future collaborative use of the wireless network by multiple application targets, etc. At this time, based on the second transfer information, the operation of the wireless network can be further optimized, that is, add the second transfer information to the newly and real-time collected multi-modal operation data. The operation optimization decision model will make a decision on a new operation optimization strategy based on the multi-modal operation data after the addition, and perform corresponding operation optimization on the wireless network based on the new operation optimization strategy.

[0100] In an embodiment of the present invention, by introducing an interaction mechanism between the application circle and the historical feature map, the dynamic adjustment and precise decision-making of the wireless network optimization strategy are realized; by combining the operation optimization history and real-time transfer information of the wireless network, the optimization requirements can be captured more accurately, ensuring that the timing of triggering and activating the application circle is appropriate, thereby improving the system working efficiency and reducing resource consumption; within the application circle, through interactive communication among all parties, feedback information and requirements can be shared in a timely manner, so as to comprehensively understand the operation status of the wireless network, and based on the historical feature map, deeply analyze the past optimization effects, providing strong support for future optimization decisions, not only enhancing the system adaptability, but also improving the accuracy and real-time performance of the optimization decisions; through continuous interaction and feedback, the operation optimization decision-making model can formulate more forward-looking optimization strategies, not only improving the network operation efficiency, but also realizing more efficient resource scheduling and service guarantee in complex environments, significantly enhancing the overall performance and user experience of the wireless network.

[0101] In one embodiment, parsing the operation optimization history and determining the triggering degree includes:

[0102] Performing scenario analysis on the operation optimization history to obtain a scenario set;

[0103] Matching the first scenario in the scenario set with the second scenario in the trigger scenario library;

[0104] When the matching is in line, the triggering degree is counted as the first threshold; wherein, the first threshold exceeds the triggering degree threshold;

[0105] When none of them match, the triggering degree is counted as the sum of the first quantization values corresponding to the scenario types of the first scenarios in the scenario set in the first quantization value library.

[0106] The working principle and beneficial effects of the above technical solution are as follows:

[0107] The scenario set contains multiple operation optimization scenarios reflected by the operation optimization history, that is, the first scenario; the second scenario is the operation optimization scenario that represents that the operation optimization history must require the attention of the application circle, such as: the scenario of performing targeted operation optimization on the wireless network in response to the needs of multiple application entities to jointly use the wireless network; therefore, if the first scenario matches the second scenario, the triggering degree is counted as the first threshold that exceeds the triggering degree threshold, so that the application circle of the wireless network can be triggered and activated. There are first quantization values corresponding to different scenario types in the first quantization value library, which can be set by technicians according to the degree of attention required by the operation optimization scenarios of different scenario types to represent the operation optimization history. Therefore, when none of them match, the triggering degree is counted as the sum of the first quantization values corresponding to the scenario types of the first scenarios in the scenario set in the first quantization value library.

[0108] In an embodiment of the present invention, if the first situation meets a specific trigger condition, the trigger level is automatically set to the first threshold to quickly activate the wireless network application circle and ensure the efficient use of network resources; if not matched, it is optimized through the sum of the quantization values of the situation types, so that the network can respond to diverse requirements more comprehensively, improving the flexibility of the wireless network operation optimization and further enhancing the applicability of the system.

[0109] In one embodiment, the decision feature map representation template based on the first transfer information includes:

[0110] Perform a time series representation on the first transfer information to obtain an information time series;

[0111] Based on the sequence item cluster division constraint, divide multiple sequence item clusters in the information time series;

[0112] Parse each sequence item cluster to determine the template decision basis and transfer weight of each sequence item cluster;

[0113] Determine the template decision rule corresponding to the template decision basis of each of the first J sequence item clusters in the information time series from the template decision rule library; wherein, the difference between the sum of the transfer weights of the first J + 1 sequence item clusters in the information time series and the sum of the transfer weights of the first J sequence item clusters is the largest; J is a positive integer;

[0114] Integrate each template decision rule to determine the feature map representation template;

[0115] Wherein, the sequence item cluster division constraint includes:

[0116] The sequence items in the sequence item cluster are arranged continuously in sequence in the information time series;

[0117] And, the data type set of the sequence items in the sequence item cluster matches the standard data type set.

[0118] The working principle and beneficial effects of the above technical solution are:

[0119] The information time series includes multiple information items in the first circulation information arranged in sequence according to the time sequence of information generation. Based on the sequence item cluster division constraint, multiple sequence item clusters are divided in the information time series. The template decision basis of the sequence item cluster is the basis for the decision feature graph representation template. The circulation weight of the sequence item cluster represents the degree to which the sequence item cluster cannot jointly decide on the feature graph representation template with its previous sequence item cluster. Therefore, the difference between the sum of the circulation weights of the first J+1 sequence item clusters in the information time series and the sum of the circulation weights of the first J sequence item clusters is set to be the largest, so that the J value is unique, and each of the first J sequence item clusters of the template decision rule corresponding to the template decision basis is determined from the template decision rule library to be most suitable for jointly deciding on the feature graph representation template with its previous sequence item cluster. Since the first time period cannot be accurately set to obtain the first circulation information of the most suitable decision feature graph representation template, each application subject may have different stage information exchange ideas in the first time period after the application circle is triggered and activated, which affects the decision of the feature graph representation template. Therefore, the embodiment of the present invention accurately obtains the template decision basis of each sequence item cluster in the first J sequence item clusters of the most suitable decision feature graph representation template through the division of sequence item clusters and the determination of circulation weights, integrates the corresponding template decision rules, and determines the feature graph representation template. The template decision rule is the rule that the template decision basis reflects the need to represent the operation optimization history with a feature graph. For example, if the template decision basis is the need to perform Internet of Vehicles network services, the template decision rule is to filter out the operation optimization history related to the Internet of Vehicles network services and highlight them. After integrating the template decision rules to determine the feature graph representation template, when using the feature graph representation template, the template decision rules on it are executed sequentially on the operation optimization history. In the sequence item cluster partitioning constraints, each sequence item in the sequence item cluster is arranged consecutively in sequence in the information time sequence, and the data type set of each sequence item in the sequence item cluster matches the standard data type set. The standard data type set represents a data type set that can be used by each sequence item to simultaneously determine the template decision basis to determine the corresponding template decision rule, and can be preset in advance by the technical staff.

[0120] The embodiments of the present invention effectively improve the intelligence and accuracy of the feature graph representation template decision process by dividing the information time series, determining the flow weights, and analyzing the template decision basis; in the application process, the feature graph template can be automatically optimized, and the template decision rule library can be introduced to ensure that the decision basis is highly consistent with the template rules. When dealing with the impact of information exchanges at different stages, the flow weights are dynamically determined so that the selection of the first J sequence item clusters best meets actual needs, thereby achieving the best decision effect; in addition, the sequence item cluster division constraints based on the standard data type set improve the work efficiency of the system.

[0121] In one embodiment, parsing each sequence item cluster and determining the template decision basis and transfer weight of each sequence item cluster includes:

[0122] Based on the feature extraction template, perform feature extraction on the sequence item cluster to obtain a feature set;

[0123] Use the feature set as the template decision basis for the corresponding sequence item cluster;

[0124] When 1 ≤ i ≤ K, the transfer weight of the i-th sequence item cluster in the information time series is counted as the second quantization value corresponding to the basis type of the template decision basis of the i-th sequence item cluster in the second quantization value library; where K is the floor value of the product of the total number of sequence item clusters in the information time series and the proportionality coefficient;

[0125] When K < i ≤ N, the transfer weight of the i-th sequence item cluster in the information time series is counted as the third quantization value jointly corresponding to the basis type of the template decision basis of the i-th sequence item cluster and the number of template decision bases having a standard basis association relationship between the template decision basis of the i-th sequence item cluster and the template decision bases of the previous i - 1 sequence item clusters in the information time series in the third quantization value library; where N is the total number of sequence item clusters in the information time series.

[0126] The working principle and beneficial effects of the above technical solution are:

[0127] Each sequence item in the sequence item cluster can be used simultaneously to determine the template decision basis to determine the corresponding template decision rule. Therefore, by extracting features from it, the obtained feature set can be used as the template decision basis, and the features in the feature set at least include data type, data association relationship, etc. For the i-th sequence item cluster in the time series (where i ranges from 1 to K), that is, the sequence item is generated earlier in time and is less affected by the possible different phased information exchange ideas of each application entity. Its transfer weight is related to the basis type of the template decision basis of the sequence item cluster. The transfer weight is equal to the second quantization value corresponding to the basis type of the template decision basis of the sequence item cluster in the second quantization value library. There are second quantization values corresponding to different basis types in the second quantization value library. The degree to which the basis type represents the inability to jointly decide the feature map representing the template with the sequence item clusters before the sequence item cluster can be preset by the technical personnel. When i = 1, the second quantization value corresponding to the basis type of the template decision basis of the i-th sequence item cluster in the second quantization value library can be 0. The proportionality coefficient can be, for example, 0.3, or can be preset by the technical personnel in advance. When K < i ≤ N, its transfer weight is related to the basis type of its template decision basis and the number of template decision bases among the template decision bases of the previous i - 1 sequence item clusters that have a standard basis association relationship with the template decision basis of the i-th sequence item cluster. The standard basis association relationship is a basis association relationship that represents the inability to jointly decide the feature map representing the template with the sequence item clusters before the sequence item cluster. For example, there is conflicting content between the two template decision bases. Therefore, the more the number of template decision bases among the template decision bases of the previous i - 1 sequence item clusters that have a standard basis association relationship with the template decision basis of the i-th sequence item cluster, the larger the corresponding third quantization value; there are third quantization values corresponding to different basis types and the number of template decision bases that have a standard basis association relationship in the third quantization value library, which can be preset by the technical personnel in advance.

[0128] In the embodiment of the present invention, each sequence item cluster uses the feature set as the template decision basis, which can comprehensively consider multi-dimensional features such as data type and association relationship, thereby enhancing the scientificity and flexibility of the decision-making process; at the same time, combining the transfer weight with the basis type of the template decision basis and the association relationship of the previous sequence item clusters optimizes the dynamic adjustment ability of the decision rule, significantly improving the accuracy and efficiency of determining the transfer weight; in addition, through the quantization value library, the refined management of different basis types and the number of template decision bases with a standard basis association relationship enables the system to automatically adjust and optimize the decision-making process, further improving the applicability and intelligent level of the system.

[0129] In one embodiment, the interaction between the guided application circle and the historical feature map includes:

[0130] Based on the connection guidance template, conduct connection guidance on the application circle;

[0131] Obtain multiple connection relationships established after the application circle accepts connection guidance;

[0132] Based on each connection relationship, slice the application circle into multiple circle slices; among them, multiple application targets in the same circle slice jointly belong to at least one connection relationship;

[0133] Push the historical feature map to each circle slice;

[0134] Obtain the internal behavior set generated after each circle slice receives the historical feature map;

[0135] Based on the internal behavior set, pairwise match each circle slice; among them, the similarity between the internal behavior sets generated by the pairwise-matched circle slices exceeds the similarity threshold.

[0136] The working principle and beneficial effects of the above technical solution are as follows:

[0137] The connection guidance template is a template for the system to conduct connection guidance for the application circle. After the application circle accepts the connection guidance, connection relationships will be established. The connection relationship is a relationship representing the potential dependence or interaction between different application targets. Based on this, the application circle is sliced into multiple circle slices, and each circle slice consists of multiple application targets. Multiple application targets in the same circle slice jointly belong to at least one connection relationship. Push the historical feature map to each circle slice. After each circle slice receives the historical feature map, a content behavior set will be generated, and this set contains multiple content behaviors, such as further information exchange behaviors generated based on the historical feature map, etc. Based on the internal behavior set, pairwise match each circle slice. The similarity between the internal behavior sets generated by the pairwise-matched circle slices exceeds the similarity threshold. The similarity threshold can be, for example, 80%. The application targets in the matched circle slices can then conduct new information exchanges, etc., to achieve further interaction with the historical feature map.

[0138] In the embodiment of the present invention, through the guidance of the application circle based on the connection guidance template, the system can automatically identify and establish the potential dependence and interaction relationships between multiple application targets, and hierarchically group the application targets according to their relationships through the circle slicing technology to ensure that related targets are clustered together; in addition, by pushing the historical feature map to each circle slice and generating the internal behavior set according to the historical behavior data, the system can deeply analyze the behavior patterns and interaction trends of the targets, thereby providing strong support for subsequent behavior prediction and decision-making; the pairwise matching mechanism between the circle slices, based on the similarity of the behavior sets, conducts intelligent matching to achieve dynamic collaboration and information flow between different targets, effectively promoting the system to make precise adjustments and optimizations in a changing environment.

[0139] An embodiment of the present invention provides a wireless network operation optimization system based on machine learning, as Figure 2 shown, including:

[0140] A collection module 1 for collecting multi-modal operation data of a wireless network in real time;

[0141] A decision module 2 for making a decision on an operation optimization strategy based on an operation optimization decision model according to the multi-modal operation data; wherein, the operation optimization decision model is pre-trained based on machine learning;

[0142] An optimization module 3 for performing corresponding operation optimization on the wireless network based on the operation optimization strategy.

[0143] The wireless network operation optimization system based on machine learning further includes:

[0144] An auxiliary module for:

[0145] Obtaining the operation optimization history of the wireless network;

[0146] Analyzing the operation optimization history to determine the triggering degree;

[0147] When the triggering degree exceeds the triggering degree threshold, triggering the activation of the application circle of the wireless network;

[0148] Obtaining the first transfer information generated within the first time period after the application circle is triggered and activated;

[0149] Making a decision on the feature map representation template based on the first transfer information;

[0150] Performing feature map representation processing on the operation optimization history based on the feature map representation template to obtain a historical feature map;

[0151] Guiding the interaction between the application circle and the historical feature map;

[0152] Obtaining the second transfer information generated within the second time period after the interaction between the application circle and the historical feature map;

[0153] Adding the second transfer information to the newly collected multi-modal operation data in real time.

[0154] The auxiliary module analyzes the operation optimization history to determine the triggering degree, including:

[0155] Performing situation analysis on the operation optimization history to obtain a situation set;

[0156] Matching the first situation in the situation set with the second situation in the triggering situation library;

[0157] When the matching is in line, the triggering degree is counted as the first threshold; wherein, the first threshold exceeds the triggering degree threshold;

[0158] When none of them match, the triggering degree is the sum of the first quantization values corresponding to the case types of each first case in the case set in the first quantization value library.

[0159] The auxiliary module, based on the first transfer information, the decision feature map representation template includes:

[0160] Perform a time series representation on the first transfer information to obtain an information time series;

[0161] Based on the sequence item cluster division constraint, divide the information time series into multiple sequence item clusters;

[0162] Parse each sequence item cluster to determine the template decision basis and transfer weight of each sequence item cluster;

[0163] Determine the template decision rule corresponding to the template decision basis of each of the first J sequence item clusters in the information time series from the template decision rule library; wherein, the difference between the sum of the transfer weights of the first J + 1 sequence item clusters in the information time series and the sum of the transfer weights of the first J sequence item clusters is the largest;

[0164] Integrate each template decision rule to determine the feature map representation template;

[0165] Wherein, the sequence item cluster division constraint includes:

[0166] Each sequence item in the sequence item cluster is arranged continuously in sequence in the information time series;

[0167] And, the data type set of each sequence item in the sequence item cluster matches the standard data type set.

[0168] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and its equivalent technologies, the present invention is also intended to include these changes and modifications.

Claims

1. A wireless network operation optimization method based on machine learning, characterized in that: include: Collect multi-modal operation data of wireless networks in real time; Based on the operation optimization decision model, the operation optimization strategy is decided according to the multimodal operation data; wherein the operation optimization decision model is obtained based on machine learning pre-training; Based on the operation optimization strategy, the wireless network is optimized accordingly.

2. The wireless network operation optimization method based on machine learning according to claim 1, characterized in that: Also includes: Obtain the operation optimization history of the wireless network; Analyze the optimization operation history and determine the trigger degree; When the trigger degree exceeds the trigger degree threshold, the application circle of the wireless network is triggered to activate; Acquire first flow information generated by the application circle within a first time period after the application circle is triggered and activated; Based on the first flow information, a feature graph representation template is determined; Based on the feature graph representation template, feature graph representation processing is performed on the operation optimization history to obtain a historical feature graph; Guide the interaction between the application circle and the historical feature graph; Acquire second flow information generated in a second time period after the application circle interacts with the historical feature graph; The second flow information is added to the newly real-time collected multimodal operation data.

3. The wireless network operation optimization method based on machine learning according to claim 2, characterized in that: The analysis and operation optimization history is used to determine the triggering degree, including: Perform scenario analysis on the operation optimization history to obtain a scenario set; Matching a first situation in the situation set with a second situation in the trigger situation library; When the match is met, the trigger degree is counted as a first threshold; wherein the first threshold exceeds the trigger degree threshold; When no match is found, the trigger degree is calculated as the sum of the first quantization values ​​corresponding to the situation types of the first situations in the situation set in the first quantization value library.

4. The wireless network operation optimization method based on machine learning according to claim 2, characterized in that: The decision feature graph representation template based on the first flow information includes: Performing time series representation on the first flow information to obtain an information time series; Based on the sequence item cluster partitioning constraints, multiple sequence item clusters are divided in the information time series; Analyze each sequence item cluster and determine the template decision basis and flow weight of each sequence item cluster; Determine the template decision rule corresponding to the template decision basis of each sequence item cluster in the first J sequence item clusters in the information time series from the template decision rule library; wherein the difference between the sum of the circulation weights of the first J+1 sequence item clusters in the information time series and the sum of the circulation weights of the first J sequence item clusters is the largest; Integrate the decision rules of each template and determine the feature map representation template; The sequence item cluster partitioning constraints include: The sequence items in the sequence item cluster are arranged consecutively in the information time series; Also, the data type set of each sequence item in the sequence item cluster matches the standard data type set.

5. The wireless network operation optimization method based on machine learning according to claim 4, characterized in that: The step of parsing each sequence item cluster and determining the template decision basis and flow weight of each sequence item cluster includes: Based on the feature extraction template, feature extraction is performed on the sequence item cluster to obtain a feature set; The feature set is used as the template decision basis for the corresponding sequence item cluster; When 1≤i≤K, the circulation weight of the i-th sequence item cluster in the information time series is calculated as the second quantized value corresponding to the basis type of the template decision basis of the i-th sequence item cluster in the second quantized value library; wherein K is the rounded-down value of the product of the total number of sequence item clusters in the information time series multiplied by the proportional coefficient; When K<i≤N, the circulation weight of the i-th sequence item cluster in the information time series is calculated as the third quantitative value corresponding to the basis type of the template decision basis of the i-th sequence item cluster and the number of template decision basis of the first i-1 sequence item clusters in the information time series that have a standard basis association relationship with the template decision basis of the i-th sequence item cluster in the third quantitative value library; wherein N is the total number of sequence item clusters in the information time series.

6. The wireless network operation optimization method based on machine learning according to claim 2, characterized in that: The guiding application circle to interact with the historical feature graph includes: Conduct contact guidance for application circles based on the contact guidance template; Acquire multiple contact relationships established after the application circle accepts contact guidance; Based on each connection relationship, the application circle is sliced ​​into multiple circle layer slices; wherein multiple application targets in the same circle layer slice jointly belong to at least one connection relationship; Push the historical feature map to each circle slice; Obtain the internal behavior set generated after each circle slice receives the historical feature graph; Based on the internal behavior set, each circle slice is matched pairwise; wherein the similarity between the internal behavior sets generated by the pairwise matched circle slices exceeds the similarity threshold.

7. A wireless network operation optimization system based on machine learning, characterized in that: include: A collection module, used for collecting multi-modal operation data of the wireless network in real time; A decision module is used to decide on an operation optimization strategy based on an operation optimization decision model and multimodal operation data; wherein the operation optimization decision model is obtained based on machine learning pre-training; The optimization module is used to perform corresponding operation optimization on the wireless network based on the operation optimization strategy.

8. The wireless network operation optimization system based on machine learning as claimed in claim 7, characterized in that: Also includes: Auxiliary modules for: Obtain the operation optimization history of the wireless network; Analyze the optimization operation history and determine the trigger degree; When the trigger degree exceeds the trigger degree threshold, the application circle of the wireless network is triggered to activate; Acquire first flow information generated by the application circle within a first time period after the application circle is triggered and activated; Based on the first flow information, a feature graph representation template is determined; Based on the feature graph representation template, feature graph representation processing is performed on the operation optimization history to obtain a historical feature graph; Guide the interaction between the application circle and the historical feature graph; Acquire second flow information generated in a second time period after the application circle interacts with the historical feature graph; The second flow information is added to the newly real-time collected multimodal operation data.

9. The wireless network operation optimization system based on machine learning as claimed in claim 8, characterized in that: The auxiliary module analyzes the optimization operation history and determines the triggering degree, including: Perform scenario analysis on the operation optimization history to obtain a scenario set; Matching a first situation in the situation set with a second situation in the trigger situation library; When the match is met, the trigger degree is counted as a first threshold; wherein the first threshold exceeds the trigger degree threshold; When no match is found, the trigger degree is calculated as the sum of the first quantization values ​​corresponding to the situation types of the first situations in the situation set in the first quantization value library.

10. The wireless network operation optimization system based on machine learning according to claim 8, characterized in that: The auxiliary module determines a feature graph representation template based on the first flow information, including: Performing time series representation on the first flow information to obtain an information time series; Based on the sequence item cluster partitioning constraints, multiple sequence item clusters are divided in the information time series; Analyze each sequence item cluster and determine the template decision basis and flow weight of each sequence item cluster; Determine the template decision rule corresponding to the template decision basis of each sequence item cluster in the first J sequence item clusters in the information time series from the template decision rule library; wherein the difference between the sum of the circulation weights of the first J+1 sequence item clusters in the information time series and the sum of the circulation weights of the first J sequence item clusters is the largest; Integrate the decision rules of each template and determine the feature map representation template; The sequence item cluster partitioning constraints include: The sequence items in the sequence item cluster are arranged consecutively in the information time series; Also, the data type set of each sequence item in the sequence item cluster matches the standard data type set.

Images