Network optimization methods, apparatus, equipment and media for wireless networks

Abstract

This application relates to the field of network optimization, and provides a network optimization method, apparatus, device, and medium for wireless networks. The method includes: acquiring several historical network optimization process data within a first statistical period; updating a first acquisition method for historical root cause analysis data, a second acquisition method for historical network optimization schemes, and / or a third acquisition method for historical optimization evaluation data using the historical network optimization process data; and optimizing the network to be optimized within a second statistical period based on the latest methods, wherein the network optimization process data generated during the second statistical period is used for the next update of the latest first acquisition method, second acquisition method, and / or third acquisition method. The technical solution provided by this application updates the first acquisition method, second acquisition method, and / or third acquisition method based on historical network optimization process data, enabling long-term and accurate output of network optimization schemes in complex network environments.

Description

Technical Field

[0001] This application relates to the field of network optimization technology, specifically to a network optimization method, apparatus, device, and medium for wireless networks. Background Technology

[0002] As the coverage of mobile communication networks continues to expand, the number of user terminals continues to grow, and the types of services become increasingly diverse, the operating environment of wireless networks is becoming more and more complex. Various network problems (such as weak coverage, signal interference, and insufficient resource capacity) occur frequently, posing extremely high technical requirements for the timeliness, accuracy, and efficiency of network optimization.

[0003] Network optimization is a key step in ensuring communication quality and improving user experience. Solutions for network problems include: collecting data to be processed and drive test data, performing network diagnosis based on this data to obtain diagnostic results and generate optimization solutions, or performing performance tests on the optimization solutions and then fine-tuning the optimization solutions based on the test results.

[0004] This approach provides inaccurate network optimization solutions when wireless networks are constantly being updated and iterated, or when facing new network scenarios or novel network problems. Summary of the Invention

[0005] This application provides a network optimization method, apparatus, device, and medium for wireless networks, which solves the technical problem that the network optimization schemes given in the prior art are inaccurate when facing wireless network updates or new network scenarios and new network problems. It can output network optimization schemes accurately and continuously in complex network environments.

[0006] In a first aspect, embodiments of this application provide a network optimization method for wireless networks, comprising: acquiring several historical network optimization process data within a first statistical period, wherein each historical network optimization process data includes historical root cause analysis data of network problems, historical network optimization schemes, and historical optimization evaluation data; The first method of obtaining historical root cause analysis data, the second method of obtaining historical network optimization schemes, and / or the third method of obtaining historical optimization evaluation data are updated using data from each historical network optimization process. Based on the latest first acquisition method, second acquisition method, and third acquisition method, network optimization is performed on the network to be optimized within the second statistical period. The network optimization process data generated during the network optimization within the second statistical period is used to update the latest first acquisition method, second acquisition method, and / or third acquisition method in the next update.

[0007] In one embodiment, network optimization is performed on the network to be optimized within the second statistical period based on the latest first acquisition method, second acquisition method, and third acquisition method, including: Obtain network metrics data of the network to be optimized before optimization; Determine network indicator characteristics based on the network indicator data before optimization; Problem types are identified by analyzing network indicator characteristics. Based on the first acquisition method, root cause analysis is performed on the problem type and network indicator characteristics to obtain the target root cause analysis data; The target root cause analysis data is processed based on the second acquisition method to obtain the target network optimization scheme and the target optimization evaluation data corresponding to the target network optimization scheme determined based on the third acquisition method.

[0008] In one embodiment, the method further includes: If the target optimization evaluation data indicates that the network optimization result is successful, the current network optimization of the network to be optimized is considered to be complete. Alternatively, if the target optimization evaluation data indicates that the network optimization result is a failure, the network optimization iteration steps are executed until the iteration stop condition is met. The network optimization iteration steps include: processing the target optimization evaluation data and the target root cause analysis data based on the second acquisition method to obtain the updated target network optimization scheme, and determining the target optimization evaluation data corresponding to the updated target network optimization scheme based on the third acquisition method.

[0009] In one embodiment, the target root cause analysis data is processed based on the second acquisition method to obtain a target network optimization scheme, including: Obtain target rules that match the problem type and / or target root cause analysis data from the latest target knowledge base, and use the latest solution generation model to process the target root cause analysis data and target rules to obtain the target network optimization solution; The methods for updating the second acquisition method of historical network optimization schemes using historical network optimization process data include: Using historical network optimization process data, incremental learning is performed on the scheme generation model used in the first statistical period to obtain an updated scheme generation model; and / or, the target knowledge base used in the first statistical period is updated using historical network optimization process data to obtain an updated target knowledge base.

[0010] In one embodiment, historical optimization evaluation data can indicate whether the network optimization result is successful or unsuccessful. Using historical network optimization process data, incremental learning is performed on the scheme generation model used in the first statistical period to obtain an updated scheme generation model, including: Positive samples are obtained based on historical network optimization process data corresponding to successful network optimization results; and negative samples are obtained based on historical network optimization process data corresponding to unsuccessful network optimization results. Using positive and negative samples, incremental learning is performed on the scheme generation model used in the first statistical period to obtain an updated scheme generation model.

[0011] In one embodiment, historical optimization evaluation data can indicate whether the network optimization result is successful or unsuccessful. The target knowledge base includes a policy rule base and a taboo rule base. The target knowledge base used in the first statistical period is updated using historical network optimization process data to obtain an updated target knowledge base, including: Based on the historical network optimization process data where the network optimization result is successful, association rules are determined, and the strategy rule base used in the first statistical period is updated using the association rules. And / or, based on the historical network optimization process data where the network optimization result is a failure, determine taboo rules, and use the taboo rules to update the taboo rule library used in the first statistical period.

[0012] In one embodiment, determining the target optimization evaluation data corresponding to the target network optimization scheme based on a third acquisition method includes: Obtain network indicator data within a preset time period after the target network optimization scheme is implemented on the network to be optimized, and evaluate the network indicator data within the preset time period based on the latest reward function to obtain target optimization evaluation data; The methods for updating historical optimization evaluation data using data from various historical network optimization processes include: Obtain the reward function used in the first statistical period; The reward function used in the first statistical period is updated using historical network optimization process data to obtain the updated reward function.

[0013] In one embodiment, root cause analysis is performed on the problem type and network indicator characteristics based on the first acquisition method to obtain target root cause analysis data, including: The root cause analysis model is used to process the problem type and network indicator characteristics to obtain the target root cause analysis data; The methods for updating the primary acquisition method of historical root cause analysis data using historical network optimization process data include: By using historical network optimization process data, incremental learning is performed on the root cause analysis model used in the first statistical period to obtain an updated root cause analysis model.

[0014] In one embodiment, acquiring several historical network optimization process data within a first statistical period includes: Extract several historical network optimization process data from the data lake during the first statistical period; The method also includes: The network optimization process data generated during the second statistical period will be stored in the data lake.

[0015] In one embodiment, the network optimization process data generated during each network optimization in the second statistical period also includes at least one of the following: network indicator data of the network to be optimized before optimization, network indicator characteristics, problem type, reasoning basis used for obtaining the root cause analysis data of the target based on the first acquisition method, and thought chain used for obtaining the target network optimization scheme based on the second acquisition method.

[0016] Secondly, embodiments of this application provide a network optimization device for wireless networks, comprising: The data acquisition module is used to acquire several historical network optimization process data within the first statistical period. Each historical network optimization process data includes historical root cause analysis data of network problems, historical network optimization schemes, and historical optimization evaluation data. The update and iteration module is used to update the first method of obtaining historical root cause analysis data, the second method of obtaining historical network optimization schemes, and / or the third method of obtaining historical optimization evaluation data using data from each historical network optimization process. The network optimization module is used to optimize the network to be optimized within a second statistical period based on the latest first acquisition method, second acquisition method, and third acquisition method. The network optimization process data generated during the network optimization within the second statistical period is used to update the latest first acquisition method, second acquisition method, and / or third acquisition method in the next iteration.

[0017] Thirdly, embodiments of this application provide an electronic device, including a processor and a memory storing a computer program, wherein the processor executes the program to implement the steps of the network optimization method for wireless networks of the first aspect.

[0018] Fourthly, embodiments of this application provide a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the network optimization method for wireless networks of the first aspect.

[0019] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the steps of the network optimization method for wireless networks of the first aspect.

[0020] The network optimization method, apparatus, device, and medium for wireless networks provided in this application's embodiments acquire historical network optimization process data, including historical root cause analysis data, historical network optimization schemes, and historical optimization evaluation data, within a first statistical period. This data is then used to update the first acquisition method for root cause analysis, the second acquisition method for optimization schemes, and the third acquisition method for optimization evaluation. Based on these updated acquisition methods, network optimization is carried out within a second statistical period. Simultaneously, the network optimization process data generated in the second statistical period is used for the next round of updates. This effectively constructs a closed loop of data accumulation, iterative updates to network optimization methods, execution of network optimization, and data feedback. This allows network optimization capabilities to evolve autonomously with practical experience, enabling the development of accurate network optimization schemes as wireless networks evolve, new scenarios emerge, or new network problems arise. Ultimately, this achieves the effect of consistently, stably, and accurately outputting network optimization schemes in complex network environments. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is one of the flowcharts illustrating a network optimization method for wireless networks provided in an embodiment of this application; Figure 2 This is a second schematic flowchart of a network optimization method for wireless networks provided in an embodiment of this application; Figure 3 This is a schematic diagram of the architecture of the network optimization system provided in the embodiments of this application; Figure 4 This is a schematic diagram of the structure of a network optimization device for wireless networks provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] Figure 1This is one of the flowcharts illustrating a network optimization method for wireless networks provided in this application embodiment. The network optimization method for wireless networks in this embodiment can be applied to a network optimization device for wireless networks within a network optimization system built on the digital intelligent evolution factory model. The network optimization system may include the network optimization device and a data lake.

[0025] The device can be implemented by software and / or hardware. It can be configured in a computer device or a cloud server. The computer device can include, but is not limited to, terminals and servers. For example, the terminal can be a mobile phone, a PDA, etc.

[0026] The data lake stores network optimization process data generated by the device during each statistical period. Optionally, the data lake can be stored using a distributed columnar database to handle efficient writing and querying of massive amounts of process data. For example, several historical network optimization process data sets are configured with scenario identifiers as row keys, and different data categories within the historical network optimization process data are configured as column families.

[0027] Reference Figure 1 This application provides a network optimization method for wireless networks, which may include the following steps: Step 101: Obtain several historical network optimization process data within the first statistical period.

[0028] The first statistical period can be the period used to accumulate historical network optimization process data. Its setting method can be flexibly selected, including but not limited to monthly, weekly, cumulative completion of a preset number of network optimization tasks, or the system's continuous running time reaching a preset time threshold. In some application scenarios, steps 101 and 102 are executed periodically, such as at midnight every day or triggered based on data thresholds.

[0029] Each historical network optimization process includes historical root cause analysis data of network problems, historical network optimization schemes, and historical optimization evaluation data. Each network optimization may be an optimization performed for a specific network problem. The historical root cause analysis data of the network problem may include the causes of the network problem. The historical network optimization schemes include the network optimization methods formulated based on the causes of the network problem. The historical optimization evaluation results are the judgment results of the implementation of the historical network optimization scheme.

[0030] In some application scenarios, the data for each historical network optimization process can specifically include the complete data set generated from the start to the end of each network optimization task within the first statistical period. Specifically, several historical network optimization process data points within the first statistical period can be obtained from a data lake.

[0031] Step 102: Update the first method of obtaining historical root cause analysis data, the second method of obtaining historical network optimization schemes, and / or the third method of obtaining historical optimization evaluation data using the historical network optimization process data.

[0032] The first acquisition method refers to the method of generating the historical root cause analysis data during the execution of the network optimization task (i.e., the process of optimizing the network). The second acquisition method refers to the method of generating the historical network optimization scheme during the execution of the network optimization task. The third acquisition method refers to the method of generating the historical optimization evaluation data during the execution of the network optimization task.

[0033] Optionally, the first acquisition method may include, but is not limited to, methods based on first association rule matching or methods based on root cause analysis models. The second acquisition method may include, but is not limited to, methods based on second association rule matching, taboo rule avoidance, and / or methods based on solution generation models. The third acquisition method may include, but is not limited to, indicator comparison, threshold determination, or reward function-based benefit evaluation.

[0034] In step 102, at least one of the first, second, and third acquisition methods can be updated using historical network optimization process data. Specifically, updating the first acquisition method using historical network optimization process data can be achieved by adjusting the first association rule or updating the parameters of the root cause analysis model. Updating the second acquisition method using historical network optimization process data can be achieved by updating the second association rule, updating the tabu rules, and / or updating the parameters of the scheme generation model. Updating the third acquisition method using historical network optimization process data can be achieved by updating the indicator comparison method, updating the judgment threshold, or updating the reward function, etc.

[0035] Step 103: Based on the latest first acquisition method, second acquisition method and third acquisition method, perform network optimization on the network to be optimized within the second statistical period.

[0036] The root cause analysis data of the network problem to be optimized is determined based on the latest first acquisition method, the network optimization scheme of the network to be optimized is determined based on the second acquisition method, and the optimization evaluation data of the network optimization scheme can also be determined using the third acquisition method.

[0037] The network optimization process data generated during the second statistical period is used to update the latest first acquisition method, second acquisition method and / or third acquisition method in the next update.

[0038] In this embodiment, by acquiring historical network optimization process data containing historical root cause analysis data, historical network optimization schemes, and historical optimization evaluation data within the first statistical period, the first acquisition method for root cause analysis, the second acquisition method for optimization schemes, and the third acquisition method for optimization evaluation are updated. Based on the updated acquisition methods, network optimization is carried out within the second statistical period. At the same time, the network optimization process data generated in the second statistical period is used for the next round of updates. This is equivalent to constructing a closed loop of data accumulation, network optimization method updates and iterations, network optimization execution, and data feedback. This allows network optimization capabilities to evolve autonomously with practical accumulation, enabling accurate network optimization schemes to be proposed as wireless networks are updated and iterated, and new scenarios or new network problems emerge. This achieves the effect of long-term, stable, and accurate output of network optimization schemes in complex network environments.

[0039] In one embodiment, step 103 described above may include, for example: Figure 2 The following steps are shown: Step 1031: Obtain network metric data of the network to be optimized before optimization.

[0040] Network metric data refers to the operational status data of the network to be optimized before optimization, which may include, but is not limited to, signal-related metrics, service-related metrics, and resource-related metrics. Signal-related metrics include Reference Signal Received Power (RSRP) and Signal-to-Interference Plus Noise Ratio (SINR). Service-related metrics include throughput, call drop rate, and handover success rate. Resource-related metrics include Physical Resource Block (PRB) utilization and number of users. The sources of network metric data before optimization may include, but are not limited to, network management systems, drive test equipment, and / or user terminal feedback. Optionally, event trigger logs (such as alarm logs) may also be obtained. The network metric data can be in JSON or CSV stream format.

[0041] Step 1032: Determine network indicator characteristics based on the network indicator data before optimization.

[0042] The processing logic in step 1032 includes data preprocessing and scene identifier generation to obtain network indicator features. Data preprocessing includes noise filtering and normalization. For example, the obtained network indicator features could be {"RSRP":-110, "SINR": 5}.

[0043] Among them, network indicator features, after being associated with scene identifiers, can be stored in the data lake.

[0044] Step 1033: Identify the problem type by analyzing the network indicator features.

[0045] Problem type refers to the classification of network problems, such as network coverage problems, network interference problems, capacity problems, handover problems, dropped call problems, etc.

[0046] Step 1033 can be implemented as follows: The network indicator features are used as input to a machine learning classifier. The machine learning classifier processes the network indicator features to obtain the classification probability of each question type. Then, the question type with the highest classification probability can be selected as the question type corresponding to that network indicator feature. The machine learning classifier can be an extreme gradient boosting model (XGBoost). This extreme gradient boosting model is trained using several network indicator features as training samples and their corresponding question types as labels.

[0047] Alternatively, in some application scenarios, step 1033 can also be implemented as follows: historical network indicator features and their problem types related to the network indicator features can be obtained from the data lake as prior knowledge, and together with the network indicator features of the network to be optimized, they can be used as input to the machine learning classifier to obtain the final problem type.

[0048] Among these features, problem types, network indicator characteristics, and scenario identifiers can be stored in a data lake for associated storage.

[0049] Step 1034: Based on the first acquisition method, perform root cause analysis on the problem type and network indicator characteristics to obtain target root cause analysis data.

[0050] Target root cause analysis data refers to the specific root cause judgment results for the current network to be optimized, which may include root cause type, confidence level, etc. For example, target root cause analysis data could be {"Missing neighbor cell match": 0.7}. Missing neighbor cell match is the root cause type, and 0.7 is the confidence level.

[0051] Step 1034 can be implemented by processing the problem type and network indicator characteristics using a root cause analysis model to obtain the target root cause analysis data. For example, the problem type and network indicator characteristics can be used as input to the root cause analysis model, and the model can be used to process these characteristics to obtain the target root cause analysis data. The root cause analysis model can employ a hierarchical multi-classification structure, outputting the root cause type, probability distribution, and reasoning basis through Bayesian probabilistic inference to form the target root cause analysis data. The reasoning basis refers to the basis upon which the root cause analysis model obtains the root cause analysis data based on the problem type and network indicator characteristics.

[0052] Alternatively, step 1034 can be implemented by using historical problem types, historical network indicator characteristics, and their corresponding historical root cause analysis data from the data lake that match the problem type and network indicator characteristics as prior knowledge, along with the network indicator characteristics and the problem type, as input to the root cause analysis model. The latest root cause analysis model is then used to process the data to obtain the target root cause analysis data. This method of determining the root cause analysis data for the current network optimization task by referencing relevant historical network optimization process data ensures that the determined root cause analysis data is more accurate.

[0053] Furthermore, it can also determine the reasoning basis used to obtain the target root cause analysis data based on the first acquisition method. The reasoning basis, problem type, network indicator characteristics, and scenario identifier are then associated and stored in the data lake.

[0054] Step 1035: Process the target root cause analysis data based on the second acquisition method to obtain the target network optimization scheme and the target optimization evaluation data corresponding to the target network optimization scheme based on the third acquisition method.

[0055] The target network optimization scheme can be a set of executable operations formulated based on the target root cause analysis data. For example, it can include specific operation instructions, execution steps, parameter configurations, etc. For instance, the target network optimization scheme could be: adjusting the transmit power of the base stations in the network to be optimized from a first transmit power to a second transmit power.

[0056] The above-mentioned method of processing the target root cause analysis data based on the second acquisition method to obtain the target network optimization scheme can be as follows: the target root cause analysis data is used as the input of the scheme generation model, and the target root cause analysis data is processed by the scheme generation model to obtain the target network optimization scheme.

[0057] Alternatively, the method described above for processing the target root cause analysis data based on the second acquisition method to obtain the target network optimization scheme can also be as follows: Obtain target rules matching the problem type and / or the target root cause analysis data from the latest target knowledge base, and use the latest scheme generation model to process the target root cause analysis data and target rules to obtain the target network optimization scheme. The target knowledge base can be used to store network optimization rules, which may include association rules between root cause analysis data and successfully optimized network optimization schemes, or may also include taboo rules between root cause analysis data and unsuccessfully optimized network optimization schemes. The scheme generation model can be a trained large language model.

[0058] Optionally, the method for obtaining target rules from the target knowledge base that match the problem type and / or target root cause analysis data can be as follows: match the problem type and / or target root cause analysis data of the network to be optimized with each network optimization rule in the target knowledge base, and determine the target rules based on indicators such as matching similarity. Target rules can be taboo rules or association rules.

[0059] The above-mentioned method of using the latest scheme generation model to process the target root cause analysis data and target rules to obtain the target network optimization scheme can be as follows: the target rules are used as prior data and the target root cause analysis data are used together as input to the scheme generation model to obtain the target network optimization scheme.

[0060] The target optimization evaluation data may include the judgment results of the execution effect of the target network optimization scheme. For example, it may include optimization benefit scores, changes in indicators before and after optimization, and / or network optimization results (network optimization successful / failed / partially successful). The above-mentioned method of determining the target optimization evaluation data corresponding to the target network optimization scheme based on the third acquisition method may be: using a preset evaluation method to predict the changes in network indicator data after the network to be optimized implements the target network optimization scheme, and determining the target optimization evaluation data based on the changes in network indicator data.

[0061] Alternatively, the method described above for determining the target optimization evaluation data corresponding to the target network optimization scheme based on the third acquisition method can also be as follows: First, send the command corresponding to the target network optimization scheme to the network to be optimized. After receiving the command, the network to be optimized can perform operations such as adjusting network parameter configuration according to the target network optimization scheme. Obtain network indicator data within a preset time period after the network to be optimized executes the target network optimization scheme, and evaluate the network indicator data within the preset time period based on the latest reward function to obtain the target optimization evaluation data. Here, the preset time period can be a short period, such as 15 minutes. The reward function can be linear, piecewise, or polynomial in form. Use the reward function to determine the optimization benefit score for each network indicator data to obtain the target optimization evaluation data.

[0062] This embodiment continuously updates at least one of the first acquisition method, the second acquisition method, and the third acquisition method during the network optimization process, resulting in a more accurate target network optimization scheme determined using the latest first acquisition method, the second acquisition method, and the third acquisition method.

[0063] In one embodiment, the method further includes: if the target optimization evaluation data indicates that the network optimization result is successful, determining that the current network optimization of the network to be optimized has ended; or, if the target optimization evaluation data indicates that the network optimization result is unsuccessful, executing a network optimization iteration step until the iteration stop condition is met, wherein the network optimization iteration step includes: processing the target optimization evaluation data and the target root cause analysis data based on the second acquisition method to obtain an updated target network optimization scheme and determining the target optimization evaluation data corresponding to the updated target network optimization scheme based on the third acquisition method.

[0064] One way to indicate successful network optimization in the target optimization evaluation data is that the network optimization result is "successful". The evaluation criteria for success or failure can be determined based on the optimization benefit score. If the optimization benefit score is greater than or equal to a preset threshold, the network optimization is considered successful; otherwise, it is considered a failure. In some application scenarios, the evaluation criteria for successful network optimization can also be that the optimized network metrics meet preset improvement standards, such as the improvement in core metrics, the core metrics reaching a threshold, or the network problem being resolved; otherwise, the network optimization is considered a failure.

[0065] The iteration stopping conditions may include, but are not limited to: the objective optimization evaluation data indicating that the network optimization result is successful, the maximum number of iterations has been reached, or the network indicator data shows no trend of improvement.

[0066] The data used to obtain the target network optimization scheme in each network optimization iteration step includes target root cause analysis data and target optimization evaluation data determined in the previous network optimization iteration, or all target optimization evaluation data from all previous network optimization iterations of the network to be optimized. In other words, in each network optimization iteration step, all failed target network optimization schemes and their corresponding target optimization evaluation data can be combined, reducing the generation of duplicate target network optimization schemes in each iteration and promoting a faster generation of the final accurate target network optimization scheme.

[0067] Optionally, if it is determined that the current network optimization of the network to be optimized is complete, the network optimization process data during the optimization process can be stored in the data lake. If network optimization iteration steps were executed during the optimization process, the network optimization process data during the optimization process includes the network optimization process data from each iteration step. That is, the network optimization process data of failed optimizations can also be stored.

[0068] This embodiment solves the problem of ineffective remediation after the first optimization failure by executing network optimization iteration steps when network optimization is determined to have failed. This enables the network optimization scheme to have dynamic adjustment capabilities, which not only improves the overall optimization success rate, but also accumulates more diverse historical data (including failed schemes, iteration adjustment strategies, etc.) through the iteration process, providing richer samples for updating various acquisition methods, and further enhancing the depth and effectiveness of closed-loop optimization.

[0069] In one embodiment, processing the target root cause analysis data based on the second acquisition method to obtain a target network optimization scheme, as described above, includes: acquiring target rules matching the problem type and / or target root cause analysis data from the latest target knowledge base, and processing the target root cause analysis data and target rules using the latest scheme generation model to obtain the target network optimization scheme. Because the scheme generation model and / or target knowledge base can be continuously updated and iterated, the latest scheme generation model and target knowledge base can be selected to determine the target network optimization scheme.

[0070] Based on this, the second method of updating the historical network optimization scheme by using the historical network optimization process data includes: using the historical network optimization process data to incrementally learn the scheme generation model used in the first statistical period to obtain the updated scheme generation model; and / or, using the historical network optimization process data to update the target knowledge base used in the first statistical period to obtain the updated target knowledge base.

[0071] For example, the scheme generation model used in the first statistical period can be incrementally learned using historical network optimization process data to obtain an updated scheme generation model. Alternatively, the target knowledge base used in the first statistical period can be updated using historical network optimization process data to obtain an updated target knowledge base. Or, the scheme generation model used in the first statistical period can be incrementally learned using historical network optimization process data to obtain an updated scheme generation model, and the target knowledge base used in the first statistical period can be updated accordingly to obtain an updated target knowledge base.

[0072] Incremental learning refers to a learning method that uses newly added historical network optimization process data to optimize the parameters and correct the logic of the original scheme-generated model without retraining the entire model, so that the model's capabilities can be continuously improved as data accumulates.

[0073] The target knowledge base can be updated by using historical network optimization process data to add network optimization rules, delete invalid rules, and / or merge similar rules.

[0074] This embodiment generates network optimization schemes by combining a scheme generation model with a target knowledge base, and uses an incremental update mechanism based on historical network optimization process data to enable the second acquisition method to continuously absorb experience and generate more accurate network optimization schemes.

[0075] In one embodiment, each historical optimization evaluation data can indicate whether the network optimization result is successful or unsuccessful. The above-mentioned method of incrementally learning the scheme generation model used in the first statistical period using each historical network optimization process data to obtain an updated scheme generation model may include: obtaining positive samples based on each historical network optimization process data corresponding to a successful network optimization result; and obtaining negative samples based on each historical network optimization process data corresponding to a failed network optimization result; and using the positive and negative samples to incrementally learn the scheme generation model used in the first statistical period to obtain an updated scheme generation model.

[0076] Before incrementally learning the solution production model, training parameters can be configured, such as the learning rate, batch size, and the ratio of positive to negative samples. For example, the proportion of negative samples can be 30% to 50%.

[0077] The historical network optimization process data may include thought chains. These thought chains are the thought processes used to obtain the target network optimization solution based on the second acquisition method. The thought chains can indicate the thought process by which the solution generation model obtains historical network optimization solutions based on historical root cause analysis data. For example, positive samples can be determined by using the thought chains in the historical network optimization process data corresponding to successful network optimization as positive samples. Similarly, negative samples can be determined by using the thought chains in the historical network optimization process data corresponding to failed network optimization as negative samples.

[0078] In the incremental learning process of the scheme generation model used in the first statistical period, using positive and negative samples, online gradient descent combined with low-rank adaptation (LoRA) or efficient parameter tuning (PTuning) is employed. Only local weights of the model are updated to obtain the updated scheme generation model. Low-rank adaptation achieves model tuning by injecting a low-rank matrix, while efficient parameter tuning adds soft cues to optimize the thought process.

[0079] This embodiment uses positive and negative samples to perform balanced learning on the scheme generation model, making the scheme generation model more robust, effectively reducing repeated errors, and improving the success rate of wireless network optimization.

[0080] In one embodiment, historical optimization evaluation data can indicate whether the network optimization result is successful or unsuccessful. The target knowledge base includes a policy rule base and a taboo rule base. The association rules in the policy rules can be determined using the network optimization process data of successful optimizations; that is, policy rules are used to store successful experiences. The taboo rules in the taboo rule base can be determined using the network optimization process data of unsuccessful optimizations; they can store failure taboos, achieving structured accumulation of experience.

[0081] The aforementioned method of updating the target knowledge base used in the first statistical period using historical network optimization process data to obtain the updated target knowledge base may include: determining association rules based on historical network optimization process data where the network optimization result was successful, and updating the policy rule base used in the first statistical period using the association rules; and / or determining tabu rules based on historical network optimization process data where the network optimization result was unsuccessful, and updating the tabu rule base used in the first statistical period using the tabu rules.

[0082] This involves constructing association rules using knowledge graphs combined with rule mining algorithms. For example, association rules can be determined through association rule mining, template matching, and natural language processing. In some application scenarios, association rule mining algorithms are used to mine successful solutions (e.g., mining solutions with support greater than 0.5 and confidence greater than 0.8) to obtain association rules. For instance, a successful solution can be abstracted as "IF {coverage difference & SINR < 10} THEN {prioritize tilt adjustment, then PCI; expected return: 0.85}". The thought chain is integrated; if the thought chain displays "assess neighbor cell impact", a conditional clause is added. Each association rule is stored in a key-value pair strategy rule base, or only if the success rate of the association rule is greater than or equal to a preset success rate threshold (e.g., 0.85). The strategy rule base used in the first statistical period is updated using association rules, including adding rules, updating rule confidence, and adjusting rule priorities. In some application scenarios, after storing each association rule in a key-value pair structured policy rule base, the similarity between each association rule is determined based on the Levenstein distance between them, and association rules with similarity greater than or equal to a threshold are merged.

[0083] For negative samples, failure boundaries are extracted (e.g., "tilt angle > 10° leads to a 5% increase in neighboring cell interference"). A decision tree algorithm is then used to generate taboo rules (e.g., "IF {cell X is over-covered} THEN {prohibit increasing tilt angle; priority: high}"). That is, taboo rules can include triggering conditions, prohibited actions, and explanations.

[0084] In addition, it can periodically clean up low-frequency rules in the target knowledge base, such as those with a usage rate less than the preset usage rate (e.g., less than 10%).

[0085] In this embodiment, successful experiences are solidified into reusable strategies, and lessons learned from failures are solidified into safety taboos. This not only accelerates the generation of solutions but also reduces dangerous operations from the source, thereby improving system stability and security.

[0086] In one embodiment, the method for determining the target optimization evaluation data corresponding to the target network optimization scheme based on the third acquisition method described above may include: acquiring network indicator data within a preset time period after the network to be optimized executes the target network optimization scheme, and evaluating the network indicator data within the preset time period based on the latest reward function to obtain the target optimization evaluation data. Based on this, the method for updating the historical optimization evaluation data using the third acquisition method with historical network optimization process data includes: acquiring the reward function used in the first statistical period; updating the reward function used in the first statistical period using historical network optimization process data to obtain the updated reward function.

[0087] Obtain the original reward function for the first statistical period. Taking a linear reward function as an example, the reward function... . and Indicates the weight.

[0088] The reward function used in the first statistical period is updated using historical network optimization process data. Specifically, this can be achieved by updating the reward function used in the first statistical period using historical optimization evaluation data from each historical network optimization process. For example, historical optimization evaluation data includes changes in indicator data before and after optimization, optimization benefit scores, and network optimization results.

[0089] This can be achieved by using a combination of Genetic Algorithm (GA) / Particle Swarm Optimization (PSO) and nonlinear least squares to automatically optimize the weights and form of the reward function. This can automatically transform from a linear function to a piecewise / saturated function to adapt to the nonlinear balance of multi-indicator data. For example, the GA / PSO algorithm is used to automatically search for the optimal reward function form, gradually searching and switching from a simple linear form to more complex forms such as piecewise functions, saturated functions, and polynomial functions that better fit the nonlinear relationships of the current network, thus solving the problem of balancing multiple indicator data using linear relationships. After determining the function form, nonlinear least squares is used to accurately fit the weight coefficients of each indicator data based on historical data, minimizing the error between the score calculated by the reward function and the historical actual optimization effect, and maximizing the fit to the current network's patterns.

[0090] In this embodiment, the reward function is changed from a fixed manual setting to a data-driven, adaptive evolution, which can accurately adapt to the nonlinear balance relationship of multiple indicators in wireless networks and significantly improve the evaluation accuracy.

[0091] In one embodiment, the method of obtaining target root cause analysis data by performing root cause analysis on problem types and network indicator characteristics based on the first acquisition method as described above may include: processing problem types and network indicator characteristics using a root cause analysis model to obtain target root cause analysis data; and updating the first acquisition method of historical root cause analysis data using historical network optimization process data may include: incrementally learning the root cause analysis model used in the first statistical period using historical network optimization process data to obtain an updated root cause analysis model.

[0092] For example, positive samples of the root cause analysis model are obtained based on the historical network optimization process data corresponding to the successful network optimization results.

[0093] Furthermore, based on the network optimization results, negative samples of the root cause analysis model are obtained from the historical network optimization process data corresponding to network optimization failures.

[0094] Then, using the positive and negative samples of the root cause analysis model, incremental learning is performed on the root cause analysis model to obtain the updated root cause analysis model.

[0095] In this embodiment, the root cause analysis model is incrementally learned by using historical optimization process data, which enables the root cause analysis capability to continuously evolve with the accumulation of data, providing a high-quality basis for the generation of subsequent solutions and improving the overall optimization accuracy.

[0096] In one embodiment, obtaining several historical network optimization process data within a first statistical period includes: extracting several historical network optimization process data within the first statistical period from a data lake; the method further includes: storing each network optimization process data generated during network optimization in a second statistical period into a data lake.

[0097] A data lake can be a distributed columnar database that uses scenario identifiers as indexes to store all optimization process data and supports high-speed read / write and batch queries.

[0098] After each network optimization is completed within the second statistical period, all network optimization process data generated in this optimization is asynchronously stored in the data lake.

[0099] In this implementation, a data lake is used to achieve unified management, accumulation, and reuse of historical optimization process data, providing a data foundation for the network optimization system to achieve self-evolution and self-acceleration.

[0100] In one embodiment, the network optimization process data generated during each network optimization in the second statistical period also includes at least one of the following: network indicator data of the network to be optimized before optimization, network indicator characteristics, problem type, reasoning basis used to obtain the target root cause analysis data based on the first acquisition method, and thought chain used to obtain the target network optimization scheme based on the second acquisition method.

[0101] All network optimization process data is stored in a unified index based on scenario identifiers, forming a complete, traceable, explainable, auditable, and reusable data chain for diagnosis, decision-making, execution, and evaluation results, providing comprehensive support for incremental learning, rule mining, and retrospective auditing.

[0102] This embodiment significantly improves system interpretability and data reuse value by storing complete data across the entire process, thereby enhancing the data flywheel evolution effect.

[0103] In one embodiment, the network optimization system consists of software modules that can be deployed on a cloud platform or edge server, supporting distributed computing. The architecture of the network optimization system can be as follows: Figure 3 As shown, the network optimization system includes a network optimization device, a data lake, and a target knowledge base. Figure 3 (Not shown in the image). The network optimization device includes a data acquisition unit, a problem identification unit, a root cause analysis unit, a solution generation unit, a solution execution unit, a short-cycle evaluation unit, an incremental learning engine, a target knowledge base update engine, and a reward function optimizer. Figure 3 The dashed line represents offline feedback, the solid line can represent real-time data flow, and the arrow can indicate the direction of data interaction.

[0104] The data acquisition unit is responsible for collecting raw network indicator data, including network indicator data of the network to be optimized before optimization. It serves as the entry point for the network optimization system, simulating a "factory raw material warehouse." The input data for the data acquisition unit includes real-time network indicators (such as RSRP and SINR values) and event triggers (such as alarm logs), in JSON or CSV stream format. It performs data preprocessing such as noise filtering and normalization on the network indicator data, and obtains a scene identifier based on cell identifiers (e.g., ID) and timestamp hashes. Its output data includes structured network indicator features, such as {"RSRP":-110, "SINR": 5}.

[0105] The data acquisition unit outputs data unidirectionally to the problem identification unit in the form of a real-time data stream, and can also connect bidirectionally to the data lake. That is, the data acquisition unit can also read data from the data lake. The data acquisition unit can integrate two complete wireless communication protocol stacks, 4G and 5G, and is compatible with the signaling and data formats of both generations of networks. For example, it supports uplink and downlink time-division multiplexing frame structures, supports uplink and downlink frequency-division multiplexing continuous frame structures, adapts to dual-band simultaneous transmission and reception modes, and is compatible with different protocol stacks of higher-layer and lower-layer protocols.

[0106] The problem identification unit can identify problem types by analyzing network indicator features, simulating "raw material screening." Specifically, the input to the problem identification unit includes the network indicator features output by the data acquisition unit, and may also include several historical network indicator features extracted from the data lake that match these network indicator features, along with their corresponding problem types. These historical network indicator features and their corresponding problem types serve as prior information and can be used together with the network indicator features output by the data acquisition unit as input to the problem identification unit. The problem identification unit can use a machine learning classifier to calculate the classification probability of the problem, thus obtaining the problem type. The problem identification unit can output the network indicator features and their corresponding problem types to the root cause analysis unit and the data lake. The problem identification unit and the data lake can be bidirectionally connected.

[0107] The root cause analysis unit analyzes the root causes of problems, simulating "quality inspection finding the root cause." Its input includes network indicator features and their corresponding problem types output by the problem identification unit, or it may include historical problem types, historical network indicator features, and their corresponding historical root cause analysis data identified from the data lake that match the problem type and network indicator features. Historical problem types, historical network indicator features, and their corresponding historical root cause analysis data can be used as prior knowledge, along with the network indicator features and the problem type, as input to the root cause analysis model. The latest root cause analysis model is then used to process the data to obtain target root cause analysis data. This target root cause analysis data includes a probability distribution (root cause type and its confidence level) and can also output the reasoning basis. The target root cause analysis data can be transmitted to the solution generation unit. The reasoning basis and the target root cause analysis data can be stored in the data lake. That is, the root cause analysis unit and the data lake can be bidirectionally connected.

[0108] The solution generation unit can generate network optimization solutions, simulating "recipe development." For example, if target optimization evaluation data from the previous round of network optimization exists, the input to the solution generation unit includes the target root cause analysis data output by the root cause analysis unit and the target optimization evaluation data from the previous round of network optimization output by the short-cycle evaluation unit, or it may also include network optimization rules (such as association rules and / or taboo rules) from the target knowledge base. It generates prompt words and obtains the thought chain and target network optimization solution through a solution generation model (e.g., a large language model). The target network optimization solution can be output to the solution execution unit, while the thought chain and target network optimization solution can be saved to the data lake. That is, the solution generation unit and the data lake can be bidirectionally connected, and the solution generation unit and the target knowledge base can also be bidirectionally connected, meaning the target network optimization solution output by the solution generation unit can be used to update the target knowledge base.

[0109] The scheme execution unit executes the target network optimization scheme, simulating a "production line operation." Its inputs include the target network optimization scheme output by the scheme generation unit, and parameters adjusted in the network to be optimized (live or simulated environment) via an API interface, such as sending commands corresponding to the target network optimization scheme to the network to be optimized. No complex calculations are required; it only monitors the execution status of the network to be optimized, and the output is an execution log.

[0110] The short-cycle evaluation unit is used to assess the execution effect, simulating "product quality inspection." Its inputs include the execution logs output by the solution execution unit and network indicator data within a preset time period after the network to be optimized executes the target network optimization solution. It uses a reward function to determine the optimization benefit score, obtaining target optimization evaluation data. This target optimization evaluation data includes the optimization evaluation result, changes in network indicator data before and after optimization, and the optimization benefit score. The target optimization evaluation data can be stored in a data lake. Furthermore, if the target optimization evaluation data indicates that the network optimization result is a failure, the network optimization iteration steps are executed until the iteration termination condition is met.

[0111] The data lake stores high-value data, simulating an "expansion of raw material warehouses." It doesn't simply store raw network data, but centrally stores structured or semi-structured process data generated during network optimization, providing "raw materials" for system evolution. The system generates a globally unique scenario identifier for each network optimization scenario, linking data from problem identification units, root cause analysis units, solution generation units, solution execution units, and short-cycle evaluation units through this identifier, forming a complete "problem diagnosis-decision-result" data chain. The data lake uses a distributed columnar database for storage to handle efficient writing and querying of massive amounts of process data. Data uses the scenario identifier as the row key and different data categories as column families.

[0112] The incremental learning engine, target knowledge base update engine, and reward function optimizer simulate a "factory's R&D and training center." Their main function is to drive continuous iteration and adaptive improvement of the system by consuming network optimization process data accumulated in the data lake (including feature vectors, root cause analysis data, thought chains, and failed solutions and their network states). The incremental learning engine, target knowledge base update engine, and reward function optimizer run periodically (e.g., daily at midnight or triggered based on data thresholds) to handle offline tasks, ensuring the realization of the data flywheel effect: initially relying on external data for startup, but as data accumulates, the system's optimization efficiency accelerates automatically, improving wireless network performance (such as throughput, latency, packet loss rate, and handover success rate) and reducing manual intervention costs.

[0113] The incremental learning engine is responsible for the incremental learning of the root cause analysis model and the solution generation model. The target knowledge base update engine is responsible for updating the target knowledge base, and the reward function optimizer is used to update the reward function. If the system is deployed on a cloud platform, the incremental learning engine, the target knowledge base update engine, and the reward function optimizer can use a distributed computing framework to handle large amounts of data, ensuring efficient operation. These components support deployment on the live network.

[0114] The incremental learning engine utilizes new data from the data lake (especially negative samples and process data) to fine-tune model parameters, enhancing the agent's diagnostic and decision-making capabilities and enabling the model's gradual evolution. For example, it can correct probability biases in root cause analysis models or optimize the inference path of solution generation models, making the system more accurate and efficient when facing similar network problems.

[0115] For example, the input to the incremental learning engine includes datasets obtained from a data lake. Each dataset includes a set of historical network optimization process data, such as network metric features output by the problem identification unit, historical root cause analysis data and reasoning basis output by the root cause analysis unit, thought chains output by the solution generation unit, and historical network optimization solutions (or possibly prompt words), as well as historical optimization evaluation data output by the short-cycle evaluation unit. Datasets are associated by scenario ID, forming a complete "diagnosis-decision-result" chain. Optionally, negative samples can be prioritized, such as a 30%-50% failure rate, to reinforce boundary learning. Training parameters, such as learning rate, batch size, and positive / negative sample ratio, are configured. Positive samples are obtained using thought chains from successful optimizations, while negative samples are obtained using thought chains from failed optimizations. An incremental learning framework, such as online gradient descent combined with low-rank adaptation or efficient parameter fine-tuning, is employed to minimize resource consumption. Reinforcement learning is used to fine-tune from human feedback variants, focusing on reusing the "thinking process" (e.g., prompts such as "Based on historical failures, tilt angle adjustment > 10° is prohibited"). Its output includes updated model parameters or weight files: for example, the weights of a fine-tuned root cause analysis model or scheme generation model, and may also include training logs: including loss function values, accuracy improvements, and validation metrics.

[0116] The target knowledge base update engine is responsible for abstracting and accumulating knowledge from the success / failure data of the data lake, forming reusable optimization strategies and taboo rule bases, and solidifying experience to accelerate decision-making. This reduces the computational overhead of the solution generation module (e.g., eliminating the need for reasoning from scratch) and enhances security (reducing the reproduction of known errors), supporting rapid matching and experience sharing within the system. By placing historical experience into the target knowledge base, runtime matching can provide historically similar and reasonable solutions. Specifically, the input to the target knowledge base update engine includes filtered datasets (successful and failed solutions) from the data lake. Successful solutions include thought chains, historical network optimization solutions, and labels indicating successful network optimization. Failed solutions include historical network optimization solutions, network status, and reasons for failure. Each filtered dataset can be sorted according to scenario similarity. An abstraction threshold (e.g., only those with a success rate greater than a preset threshold are stored in the target knowledge base) and rule priority (e.g., taboo rules are prioritized over association rules) can be configured. For example, knowledge graphs combined with rule mining algorithms can be used to construct association rules. For example, association rules can be determined through association rule mining, template matching, and natural language processing. In some application scenarios, association rule mining algorithms are used to mine successful solutions (support greater than 0.5, confidence greater than 0.8) to obtain association rules. For example, a successful solution can be abstracted as "IF {coverage difference & SINR<10} THEN {prioritize tilt adjustment, then PCI; expected return: 0.85}". The thought chain is integrated; if the thought chain displays "evaluate neighbor cell impact", a conditional clause is added. Each association rule is stored in a key-value pair-based policy rule base, or only if the success rate of the association rule is greater than or equal to a preset success rate threshold (e.g., 0.85). The association rules are used to update the policy rule base used in the first statistical period, including adding rules, updating rule confidence, and adjusting rule priorities. In some application scenarios, after storing each association rule in a key-value pair-based policy rule base, the similarity between the association rules is determined based on the Levenstein distance, and association rules with similarity greater than or equal to the threshold are merged. For negative samples, failure boundaries are extracted (e.g., "tilt angle > 10° leads to a 5% increase in neighboring cell interference"). A decision tree algorithm is then used to generate taboo rules (e.g., "IF {cell X is over-covered} THEN {prohibit increasing tilt angle; priority: high}"). That is, taboo rules can include triggering conditions, prohibited actions, and explanations. Additionally, low-frequency rules in the target knowledge base can be periodically cleaned up, such as those with usage rates lower than a preset rate (e.g., less than 10%). The target knowledge base update engine outputs the updated target knowledge base and a query interface. The query interface includes an API endpoint for real-time invocation by the solution generation unit.

[0117] The reward function optimizer dynamically adjusts the reward function in reinforcement learning, making the evaluation criteria more aligned with the actual network needs and guiding the solution generation unit to generate more efficient solutions. This addresses the limitations of traditional fixed reward functions (such as ignoring cost-benefit balance) and improves the relevance of the evolution. The input to the reward function optimizer includes a set of historical optimization evaluation data from the data lake and the reward function to be optimized. Specifically, the reward function can be optimized using meta-learning optimization (such as genetic algorithms or particle swarm optimization) combined with regression analysis (such as nonlinear least squares). For example, the form of the reward function can be optimized; for instance, the form can be extended from linear to piecewise or polynomial. Its output includes the optimized reward function and an optimization report, which includes the fitted curve and performance improvements.

[0118] This embodiment realizes a data flywheel effect, from data acquisition to model optimization, then to performance improvement and the generation of better data, ultimately forming a self-evolving "digital intelligent factory." Specifically, the system's reinforcement learning closed-loop framework (problem identification, root cause analysis, solution generation, solution execution, and short-cycle evaluation) is deeply integrated with the data flywheel mechanism, enabling adaptive and sustainable evolution of wireless network optimization. Specifically, the system utilizes a data lake to store structured process data (such as feature vectors, root cause probability distributions, and solution thought chains) and negative samples during the optimization process. Through an evolution engine, it performs incremental learning, knowledge base construction, and reward function optimization, forming a flywheel effect that accelerates optimization efficiency, improves network performance (such as throughput, latency, packet loss rate, and handover success rate), and reduces manual intervention costs. By endowing the system with "factory-like" production evolution capabilities, it is suitable for complex and dynamic 4G / 5G environments.

[0119] The following describes a network optimization apparatus for wireless networks provided in the embodiments of this application. The network optimization apparatus for wireless networks described below can be referred to in correspondence with the network optimization method for wireless networks described above. For example... Figure 4 As shown, the network optimization device 400 for wireless networks may include the following modules: The data acquisition module 401 is used to acquire several historical network optimization process data within the first statistical period. Each historical network optimization process data includes historical root cause analysis data of network problems, historical network optimization schemes, and historical optimization evaluation data. The update iteration module 402 is used to update the first acquisition method of historical root cause analysis data, the second acquisition method of historical network optimization scheme and / or the third acquisition method of historical optimization evaluation data using data from each historical network optimization process. The network optimization module 403 is used to optimize the network to be optimized within a second statistical period based on the latest first acquisition method, second acquisition method and third acquisition method. The network optimization process data generated during the network optimization within the second statistical period is used to update the latest first acquisition method, second acquisition method and / or third acquisition method in the next cycle.

[0120] In one embodiment, the network optimization module 403 performs network optimization on the network to be optimized within a second statistical period based on the latest first acquisition method, second acquisition method, and third acquisition method, including: Obtain network metrics data of the network to be optimized before optimization; Determine network indicator characteristics based on the network indicator data before optimization; Problem types are identified by analyzing network indicator characteristics. Based on the first acquisition method, root cause analysis is performed on the problem type and network indicator characteristics to obtain the target root cause analysis data; The target root cause analysis data is processed based on the second acquisition method to obtain the target network optimization scheme and the target optimization evaluation data corresponding to the target network optimization scheme determined based on the third acquisition method.

[0121] In one embodiment, the network optimization module 403 is further configured to: If the target optimization evaluation data indicates that the network optimization result is successful, the current network optimization of the network to be optimized is considered to be complete. Alternatively, if the target optimization evaluation data indicates that the network optimization result is a failure, the network optimization iteration steps are executed until the iteration stop condition is met. The network optimization iteration steps include: processing the target optimization evaluation data and the target root cause analysis data based on the second acquisition method to obtain the updated target network optimization scheme, and determining the target optimization evaluation data corresponding to the updated target network optimization scheme based on the third acquisition method.

[0122] In one embodiment, the network optimization module 403 processes the target root cause analysis data based on the second acquisition method to obtain a target network optimization scheme, including: Obtain target rules that match the problem type and / or target root cause analysis data from the latest target knowledge base, and use the latest solution generation model to process the target root cause analysis data and target rules to obtain the target network optimization solution; The update iteration module 402 updates the second acquisition method of historical network optimization schemes using historical network optimization process data in the following ways: Using historical network optimization process data, incremental learning is performed on the scheme generation model used in the first statistical period to obtain an updated scheme generation model; and / or, the target knowledge base used in the first statistical period is updated using historical network optimization process data to obtain an updated target knowledge base.

[0123] In one embodiment, historical optimization evaluation data can indicate whether the network optimization result is successful or unsuccessful. The update iteration module 402 uses the historical network optimization process data to incrementally learn the scheme generation model used in the first statistical period to obtain an updated scheme generation model, including: Positive samples are obtained based on historical network optimization process data corresponding to successful network optimization results; and negative samples are obtained based on historical network optimization process data corresponding to unsuccessful network optimization results. Using positive and negative samples, incremental learning is performed on the scheme generation model used in the first statistical period to obtain an updated scheme generation model.

[0124] In one embodiment, historical optimization evaluation data can indicate whether the network optimization result is successful or unsuccessful. The target knowledge base includes a policy rule base and a taboo rule base. The update iteration module 402 updates the target knowledge base used in the first statistical period using historical network optimization process data to obtain an updated target knowledge base, including: Based on the historical network optimization process data where the network optimization result is successful, association rules are determined, and the strategy rule base used in the first statistical period is updated using the association rules. And / or, based on the historical network optimization process data where the network optimization result is a failure, determine taboo rules, and use the taboo rules to update the taboo rule library used in the first statistical period.

[0125] In one embodiment, the network optimization module 403 determines the target optimization evaluation data corresponding to the target network optimization scheme based on a third acquisition method, including: Obtain network indicator data within a preset time period after the target network optimization scheme is implemented on the network to be optimized, and evaluate the network indicator data within the preset time period based on the latest reward function to obtain target optimization evaluation data; The update iteration module 402 updates the historical optimization evaluation data using the third acquisition method based on the historical network optimization process data, including: Obtain the reward function used in the first statistical period; The reward function used in the first statistical period is updated using historical network optimization process data to obtain the updated reward function.

[0126] In one embodiment, the network optimization module 403 performs root cause analysis on the problem type and network indicator characteristics based on a first acquisition method to obtain target root cause analysis data, including: The root cause analysis model is used to process the problem type and network indicator characteristics to obtain the target root cause analysis data; The update iteration module 402 updates the first acquisition method of historical root cause analysis data using historical network optimization process data, including: By using historical network optimization process data, incremental learning is performed on the root cause analysis model used in the first statistical period to obtain an updated root cause analysis model.

[0127] In one embodiment, the data acquisition module 401 acquires several historical network optimization process data within a first statistical period, including: Extract several historical network optimization process data from the data lake during the first statistical period; Network optimization module 403 is also used for: The network optimization process data generated during the second statistical period will be stored in the data lake.

[0128] In one embodiment, the network optimization process data generated during each network optimization in the second statistical period also includes at least one of the following: network indicator data of the network to be optimized before optimization, network indicator characteristics, problem type, reasoning basis used for obtaining the root cause analysis data of the target based on the first acquisition method, and thought chain used for obtaining the target network optimization scheme based on the second acquisition method.

[0129] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5As shown, the electronic device may include a processor 510, a communication interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other via the communication bus 540. The processor 510 can call a computer program in the memory 530 to execute steps of a network optimization method for wireless networks, such as: acquiring several historical network optimization process data within a first statistical period, each of the historical network optimization process data including historical root cause analysis data of network problems, historical network optimization schemes, and historical optimization evaluation data; updating the first acquisition method of the historical root cause analysis data, the second acquisition method of the historical network optimization schemes, and / or the third acquisition method of the historical optimization evaluation data using each of the historical network optimization process data; and performing network optimization on the network to be optimized within a second statistical period based on the latest first acquisition method, second acquisition method, and third acquisition method, wherein the network optimization process data generated during the network optimization within the second statistical period is used for the next update of the latest first acquisition method, second acquisition method, and / or third acquisition method.

[0130] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0131] On the other hand, this application also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the steps of the network optimization method for wireless networks provided in the above embodiments, such as: acquiring a plurality of historical network optimization process data within a first statistical period, each of the historical network optimization process data including historical root cause analysis data of network problems, historical network optimization schemes, and historical optimization evaluation data; updating the first acquisition method of the historical root cause analysis data, the second acquisition method of the historical network optimization schemes, and / or the third acquisition method of the historical optimization evaluation data using each of the historical network optimization process data; and performing network optimization on the network to be optimized within a second statistical period based on the latest first acquisition method, second acquisition method, and third acquisition method, wherein the network optimization process data generated during the network optimization within the second statistical period is used for the next update of the latest first acquisition method, second acquisition method, and / or third acquisition method.

[0132] On the other hand, embodiments of this application also provide a processor-readable storage medium storing a computer program for causing a processor to execute the steps of the methods provided in the above embodiments, such as: acquiring a plurality of historical network optimization process data within a first statistical period, each of the historical network optimization process data including historical root cause analysis data of network problems, historical network optimization schemes, and historical optimization evaluation data; updating the first acquisition method of the historical root cause analysis data, the second acquisition method of the historical network optimization schemes, and / or the third acquisition method of the historical optimization evaluation data using each of the historical network optimization process data; performing network optimization on the network to be optimized within a second statistical period based on the latest first acquisition method, second acquisition method, and third acquisition method, wherein the network optimization process data generated during the network optimization within the second statistical period is used for the next update of the latest first acquisition method, second acquisition method, and / or third acquisition method.

[0133] The processor-readable storage medium can be any available medium or data storage device that the processor can access, including but not limited to magnetic memory (e.g., floppy disk, hard disk, magnetic tape, magneto-optical disk (MO)), optical memory (e.g., CD, DVD, BD, HVD), and semiconductor memory (e.g., ROM, EPROM, EEPROM, non-volatile memory (NAND FLASH), solid-state drive (SSD)).

[0134] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0135] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0136] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A network optimization method for wireless networks, characterized in that, include: Acquire several historical network optimization process data within the first statistical period. Each historical network optimization process data includes historical root cause analysis data of network problems, historical network optimization schemes, and historical optimization evaluation data. The first method of obtaining the historical root cause analysis data, the second method of obtaining the historical network optimization scheme, and / or the third method of obtaining the historical optimization evaluation data are updated using the historical network optimization process data. Based on the latest first acquisition method, second acquisition method and third acquisition method, network optimization is performed on the network to be optimized in the second statistical period. The network optimization process data generated during the network optimization in the second statistical period is used to update the latest first acquisition method, second acquisition method and / or third acquisition method in the next update.

2. The network optimization method for wireless networks according to claim 1, characterized in that, The network optimization based on the latest first, second, and third acquisition methods within the second statistical period includes: Obtain the network metric data of the network to be optimized before optimization; Determine network indicator characteristics based on the network indicator data before optimization; Problem types are obtained by identifying problems based on the network indicator features; Based on the first acquisition method, root cause analysis is performed on the problem type and the network indicator characteristics to obtain target root cause analysis data; The target root cause analysis data is processed based on the second acquisition method to obtain the target network optimization scheme and the target optimization evaluation data corresponding to the target network optimization scheme based on the third acquisition method.

3. The network optimization method for wireless networks according to claim 2, characterized in that, The method further includes: If the target optimization evaluation data indicates that the network optimization result is successful, then the current network optimization of the network to be optimized is determined to be complete. Alternatively, if the target optimization evaluation data indicates that the network optimization result is a network optimization failure, the network optimization iteration steps are executed until the iteration stop condition is met. The network optimization iteration steps include: processing the target optimization evaluation data and the target root cause analysis data based on the second acquisition method to obtain an updated target network optimization scheme and determining the target optimization evaluation data corresponding to the updated target network optimization scheme based on the third acquisition method.

4. The network optimization method for wireless networks according to claim 2, characterized in that, The step of processing the target root cause analysis data based on the second acquisition method to obtain the target network optimization scheme includes: Obtain target rules that match the problem type and / or the target root cause analysis data from the latest target knowledge base, and process the target root cause analysis data and the target rules using the latest scheme generation model to obtain the target network optimization scheme; The methods for updating the second acquisition method of the historical network optimization scheme using the historical network optimization process data include: Using the historical network optimization process data, incremental learning is performed on the scheme generation model used in the first statistical period to obtain an updated scheme generation model; and / or, the target knowledge base used in the first statistical period is updated using the historical network optimization process data to obtain an updated target knowledge base.

5. The network optimization method for wireless networks according to claim 4, characterized in that, The historical optimization evaluation data can indicate whether the network optimization result is successful or unsuccessful. The step of incrementally learning the scheme generation model used within the first statistical period using the historical network optimization process data to obtain an updated scheme generation model includes: Positive samples are obtained based on the historical network optimization process data corresponding to successful network optimization results; and negative samples are obtained based on the historical network optimization process data corresponding to failed network optimization results. Using the positive and negative samples, incremental learning is performed on the scheme generation model used in the first statistical period to obtain an updated scheme generation model.

6. The network optimization method for wireless networks according to claim 4, characterized in that, The historical optimization evaluation data can indicate whether the network optimization result is successful or unsuccessful. The target knowledge base includes a policy rule base and a taboo rule base. The step of updating the target knowledge base used in the first statistical period using the historical network optimization process data to obtain the updated target knowledge base includes: Based on the historical network optimization process data for each network optimization success, association rules are determined, and the strategy rule base used in the first statistical period is updated using the association rules. And / or, based on the historical network optimization process data where the network optimization result is a failure, determine taboo rules, and use the taboo rules to update the taboo rule library used in the first statistical period.

7. The network optimization method for wireless networks according to any one of claims 2 to 6, characterized in that, The step of determining the target optimization evaluation data corresponding to the target network optimization scheme based on the third acquisition method includes: Obtain network indicator data within a preset time period after the network to be optimized executes the target network optimization scheme, and evaluate the network indicator data within the preset time period based on the latest reward function to obtain target optimization evaluation data; The third method of updating the historical optimization evaluation data using the historical network optimization process data includes: Obtain the reward function used during the first statistical period; The reward function used in the first statistical period is updated using the historical network optimization process data to obtain the updated reward function.

8. The network optimization method for wireless networks according to any one of claims 2 to 6, characterized in that, The step of performing root cause analysis on the problem type and network indicator characteristics based on the first acquisition method to obtain target root cause analysis data includes: The root cause analysis model is used to process the problem type and the network indicator characteristics to obtain the target root cause analysis data; The methods for updating the first acquisition method of the historical root cause analysis data using the historical network optimization process data include: Using the historical network optimization process data, the root cause analysis model used in the first statistical period is incrementally learned to obtain the updated root cause analysis model.

9. The network optimization method for wireless networks according to any one of claims 2 to 6, characterized in that, The acquisition of several historical network optimization process data within the first statistical period includes: Extract several historical network optimization process data from the data lake during the first statistical period; The method further includes: The network optimization process data generated during the second statistical period are stored in the data lake.

10. The network optimization method for wireless networks according to claim 9, characterized in that, The network optimization process data generated during each network optimization in the second statistical period also includes at least one of the following: network indicator data of the network to be optimized before optimization, network indicator characteristics, problem type, reasoning basis used to obtain the target root cause analysis data based on the first acquisition method, and thought chain used to obtain the target network optimization scheme based on the second acquisition method.

11. A network optimization device for wireless networks, characterized in that, include: The data acquisition module is used to acquire several historical network optimization process data within the first statistical period. Each of the historical network optimization process data includes historical root cause analysis data of network problems, historical network optimization schemes, and historical optimization evaluation data. The update and iteration module is used to update the first acquisition method of the historical root cause analysis data, the second acquisition method of the historical network optimization scheme, and / or the third acquisition method of the historical optimization evaluation data using the historical network optimization process data. The network optimization module is used to perform network optimization on the network to be optimized within a second statistical period based on the latest first acquisition method, second acquisition method, and third acquisition method. The network optimization process data generated during the network optimization within the second statistical period is used to update the latest first acquisition method, second acquisition method, and / or third acquisition method in the next iteration.

12. An electronic device comprising a processor and a memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the network optimization method for wireless networks as described in any one of claims 1 to 10.

13. A non-transitory 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 network optimization method for wireless networks as described in any one of claims 1 to 10.

14. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the network optimization method for wireless networks as described in any one of claims 1 to 10.

Images