Network performance index prediction method and device, electronic equipment and storage medium

Abstract

The invention provides a network performance index prediction method and apparatus, an electronic device and a storage medium. The method comprises the steps of obtaining current to-be-predicted data including a current network state; inputting the current network state and the current to-be-predicted data into a feature selection model to obtain current key features output by the feature selection model; the feature selection model is constructed based on a reinforcement learning algorithm; and inputting the current key features into a performance prediction model to obtain a performance prediction result output by the performance prediction model. According to the method and the device provided by the invention, the current to-be-predicted data including the current network state is acquired; inputting the current network state and the current to-be-predicted data into a feature selection model constructed based on a reinforcement learning algorithm to obtain a current key feature; and inputting the current key features into the performance prediction model to obtain a performance prediction result, thereby realizing network performance index prediction with high accuracy, high timeliness and high prediction speed, and further ensuring stable and efficient operation of the network.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a network performance indicator prediction method, device, electronic device and storage medium. Background Art

[0002] The rapid development and widespread adoption of information technology has placed increasing demands on networks, making network performance prediction and optimization a crucial research area. Network performance typically encompasses metrics such as reliability, availability, response time, and throughput. These metrics directly impact user experience and the quality of network services. Accurately predicting network performance can optimize network configuration and resource allocation. Traditional network performance indicator prediction methods primarily rely on analyzing collected data against fixed thresholds to achieve predictions.

[0003] However, the method of making predictions directly based on the original collected data makes the amount of calculation extremely large, which not only greatly reduces the timeliness of the prediction results, but may also be affected by non-critical data, reducing the accuracy of the prediction results. Summary of the Invention

[0004] The present invention provides a network performance indicator prediction method, device, electronic device and storage medium, which are used to solve the defects of low accuracy and timeliness of network performance indicator prediction in the prior art.

[0005] The present invention provides a network performance indicator prediction method, comprising: Acquire current data to be predicted, where the current data to be predicted includes a current network state; Inputting the current network state and the current data to be predicted into a feature selection model to obtain current key features output by the feature selection model; the feature selection model is constructed based on a reinforcement learning algorithm; The current key features are input into a performance prediction model to obtain a performance prediction result output by the performance prediction model.

[0006] According to a network performance indicator prediction method provided by the present invention, the current network state and the current data to be predicted are input into a feature selection model to obtain the current key features output by the feature selection model, including: Based on the previous network state and the previous key feature, calculate the current feature reward; Based on the feature selection model, the feature selection reward, the current network state, and the current data to be predicted are applied to output the key features.

[0007] According to a network performance indicator prediction method provided by the present invention, the feature selection model includes multiple feature selection sub-models; The step of applying the feature selection reward, the current network state, and the current data to be predicted based on the feature selection model to output the key features includes: Based on the scheduler, the multiple feature selection sub-models are scheduled to respectively perform vector encoding on the current network state to obtain a current network state vector, and the feature selection reward, the current network state vector and the current data to be predicted are applied to output the key features.

[0008] According to a network performance indicator prediction method provided by the present invention, the scheduler is a Dify scheduler.

[0009] According to a network performance indicator prediction method provided by the present invention, the performance prediction model includes a prediction knowledge base; the prediction knowledge base includes multiple document types of prediction knowledge.

[0010] According to a network performance indicator prediction method provided by the present invention, inputting the current key features into a performance prediction model to obtain a performance prediction result output by the performance prediction model includes: Based on the search text in the key feature, a key document segment matching the search text is retrieved from a prediction knowledge base; Based on the performance prediction model, feature encoding is performed on the key features to obtain a key feature vector; Matching similar knowledge texts of the key feature vector from the prediction knowledge base; Obtaining enhanced similar text based on the key document segment and / or the similar knowledge text; Based on the performance prediction model, the key feature vector and the enhanced similar text are applied to output the performance prediction result.

[0011] The present invention also provides a network performance indicator prediction device, comprising: an acquiring unit, configured to acquire current data to be predicted, wherein the current data to be predicted includes a current network state; a feature selection unit, inputting the current network state and the current data to be predicted into a feature selection model to obtain current key features output by the feature selection model; the feature selection model is constructed based on a reinforcement learning algorithm; The prediction unit inputs the current key feature into the performance prediction model to obtain a performance prediction result output by the performance prediction model.

[0012] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any of the above-described network performance indicator prediction methods is implemented.

[0013] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described network performance indicator prediction methods.

[0014] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned network performance indicator prediction methods.

[0015] The network performance indicator prediction method, device, electronic device and storage medium provided by the present invention obtain the current data to be predicted including the current network status; input the current network status and the current data to be predicted into a feature selection model constructed based on a reinforcement learning algorithm to obtain the current key features output by the feature selection model; input the current key features into the performance prediction model to obtain the performance prediction results output by the performance prediction model, thereby achieving network performance indicator prediction with high accuracy, high timeliness and fast prediction speed, thereby ensuring stable and efficient operation of the network. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 It is a flow chart of the network performance indicator prediction method provided by the present invention; Figure 2 It is a flow chart of the communication network performance indicator prediction method based on dify arrangement provided by the present invention; Figure 3 It is a structural diagram of the network performance indicator prediction device provided by the present invention; Figure 4 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0018] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0019] In response to the above problems, the present invention provides a network performance indicator prediction method to achieve accurate and low-latency network performance indicator prediction. Figure 1 FIG. 1 is a flow chart of the network performance indicator prediction method provided by the present invention, such as Figure 1 As shown, the method includes: Step 110: obtaining current data to be predicted, wherein the current data to be predicted includes a current network state; Here, the current data to be predicted refers to a collection of various raw data related to the network's operating status, acquired through network monitoring and collection methods at the current moment. This includes, for example, the current network status. The current network status reflects the overall operating status and characteristics of the network at the current moment.

[0020] Specifically, various data during network operation can be collected in real time through network monitoring tools, sensors, log systems, etc., including but not limited to network traffic data, link quality data, and network topology information. Then, the acquired data can be used as the current data to be predicted.

[0021] For example, traffic logs can be collected through sFlow / NetFlow, device configurations such as MTU values ​​can be obtained through the SNMP protocol, and business layer performance indicators can be synchronized through the HTTP API. In addition, obtaining the current data to be predicted here supports multi-mode data source expansion. File upload formats include 15+ formats such as PDF, PPT, TXT, etc., and single files are ≤15MB. They can be automatically segmented and vector indexed. In addition, real-time document synchronization can be achieved by binding a Notion account. After updating, clicking the synchronization button triggers the incremental indexing of the current data to be predicted. For sFlow / NetFlow to collect network device traffic logs, the current data to be predicted can be accessed and forwarded through the UDP port, and indicator visualization can be achieved in combination with Prometheus.

[0022] Among them, network traffic data includes, for example, the number of data packets and traffic rate; device status data includes, for example, CPU usage, memory occupancy, and device temperature; link quality data includes, for example, packet loss rate, delay, and jitter.

[0023] It's clear that real-time data collection and preprocessing ensures data integrity and accuracy, providing a reliable foundation for subsequent feature selection and performance prediction. Furthermore, by collecting data from multiple sources and implementing a unified orchestration mechanism, the efficiency of feature fusion is improved, thereby enhancing the accuracy of network performance indicator predictions.

[0024] Step 120: Input the current network state and the current data to be predicted into a feature selection model to obtain the current key features output by the feature selection model; the feature selection model is constructed based on a reinforcement learning algorithm; Here, current key features refer to the subset of features selected from the current data to be predicted that have a significant impact on the predicted network performance metrics. As can be understood, using current key features as input to the performance prediction model reduces the interference of redundant features on the prediction process, improving the efficiency and accuracy of the performance prediction model. Furthermore, the dynamic selection of key features enables the model to adapt to changes in network status and maintain the timeliness of predictions.

[0025] Specifically, a feature selection model based on a reinforcement learning algorithm can be constructed. This feature selection model can be based on the Q-Learning algorithm, also known as the reinforcement learning algorithm. This model interacts with the network state and learns how to dynamically select the most relevant features based on the current network state. During training, the reinforcement learning agent gradually optimizes its feature selection strategy by trying different feature combinations and receiving rewards or penalties based on the accuracy of the prediction results.

[0026] In practical applications, the current network state and the current data to be predicted can be input into a trained feature selection model. The feature selection model can then select the most critical feature set for the prediction performance indicator from the current data to be predicted, and output the current key features.

[0027] It should be noted that the reinforcement learning-based feature selection model can adaptively select key features based on the dynamic changes in network status, avoiding the computational overhead of using redundant or irrelevant features. This improves the accuracy and efficiency of feature selection, helps reduce model complexity, and enhances the prediction accuracy of subsequent performance prediction models. Furthermore, it addresses the issue of low prediction accuracy for network performance indicators due to insufficient single-dimensional modeling and neglect of the interactive effects of multiple indicators. By interacting with multiple sources of data on the current network status, dynamic key feature selection is achieved, improving prediction accuracy.

[0028] It should also be noted that compared to the prior art which usually predicts network performance indicators through fixed thresholds, the method provided in the embodiment of the present invention dynamically selects key features through the current network status, and then predicts network performance indicators based on the dynamic key features to cope with fluctuations in dynamic traffic, thereby ensuring the accuracy of network performance indicator prediction.

[0029] Step 130: Input the current key features into a performance prediction model to obtain a performance prediction result output by the performance prediction model.

[0030] Here, performance prediction results refer to the estimated network performance metrics, such as throughput and latency, output by the performance prediction model based on current key features over a period of time. As you can see, performance prediction results can provide decision support for network operations and maintenance personnel, helping them identify potential performance bottlenecks or potential faults in advance, allowing them to implement appropriate optimization measures to ensure stable and efficient network operation.

[0031] Specifically, a performance prediction model can be constructed by using machine learning models, deep learning models, etc. The performance prediction model can be trained based on historical data and selected key features to learn the mapping relationship between features and network performance indicators.

[0032] In practical applications, the current key features output by the feature selection model can be input into the trained performance prediction model. The performance prediction model outputs the prediction results of network performance indicators such as throughput, delay, packet loss rate, etc., and the performance prediction results are obtained.

[0033] It should be noted that by using key features for prediction, the performance prediction model can more efficiently capture changing trends in network performance, improving the accuracy and timeliness of predictions. Furthermore, by eliminating unnecessary feature inputs, the model's computational complexity is reduced, inference speed is accelerated, and it can better meet the needs of real-time predictions.

[0034] The method provided by the embodiment of the present invention obtains the current data to be predicted including the current network status; inputs the current network status and the current data to be predicted into a feature selection model constructed based on a reinforcement learning algorithm to obtain the current key features output by the feature selection model; inputs the current key features into a performance prediction model to obtain the performance prediction results output by the performance prediction model, thereby achieving network performance indicator prediction with high accuracy, high timeliness and fast prediction speed, thereby ensuring stable and efficient operation of the network.

[0035] Based on any of the above embodiments, step 120 includes: Based on the previous network state and the previous key feature, calculate the current feature reward; Based on the feature selection model, the feature selection reward, the current network state, and the current data to be predicted are applied to output the key features.

[0036] Specifically, first, the feature selection model can be used to calculate the current feature reward at the current moment based on the previous network state and the action taken at the previous moment, that is, the previous key feature. Here, the current feature reward can be calculated using the following formula, as shown below: Where, represents the current feature reward, Indicates the previous network status; Indicates the previous key feature; represents the learning rate, (0< ≤ 1), controlling the update amplitude; Indicates immediate reward; represents the discount factor, (0 ≤ γ<1), which measures the importance of future rewards; Indicates the current network status Next, all possible actions The maximum reward Q value can be used to reflect the current network status Next, according to the currently known information, select the Q value corresponding to the action that can obtain the maximum expected long-term reward.

[0037] It's understandable that the goal of a feature selection model built on a reinforcement learning algorithm is to maximize cumulative rewards—that is, to maximize the current feature reward. Consequently, the feature selection model adjusts its action selection strategy based on the reward. For example, if the feature selection model finds that taking a certain action consistently yields a higher reward, it will tend to choose that action again in similar situations, thereby selecting the optimal key features for the current data to be predicted, thereby improving the accuracy of subsequent network performance metric predictions.

[0038] Furthermore, the feature selection model can be guided by feature selection rewards to select key features from the current data to be predicted based on the current network state, thereby achieving accurate and key feature selection.

[0039] Based on any of the above embodiments, the feature selection model includes multiple feature selection sub-models; The step of applying the feature selection reward, the current network state, and the current data to be predicted based on the feature selection model to output the key features includes: Based on the scheduler, the multiple feature selection sub-models are scheduled to respectively perform vector encoding on the current network state to obtain a current network state vector, and the feature selection reward, the current network state vector and the current data to be predicted are applied to output the key features.

[0040] Specifically, the feature selection model includes multiple feature selection sub-models, each of which can be an AI model built based on a reinforcement learning algorithm. A scheduler can schedule multiple feature selection sub-models to vector-encode the current network state to obtain the current network state vector. Furthermore, the feature selection sub-models used for feature selection are scheduled to apply the feature selection reward, the current network state vector, and the current data to be predicted to output key features.

[0041] It is understandable that, considering the complexity of the current data to be predicted and the frequency of feature selection tasks, different data and prediction algorithms are collected according to the prediction scenario. Through the scheduler, multiple feature selection sub-models can be combined into a complete workflow, and different tasks can be automatically completed by configuring different feature selection sub-models to finally output key features, so that complex feature selection task flows can be processed efficiently and accurately, thereby improving the accuracy and timeliness of subsequent network performance indicator predictions.

[0042] Based on any of the above embodiments, the scheduler is a Dify scheduler It's important to note that the Dify scheduler enables modular assembly of components such as LLMs, RAGs, and Agents based on visual workflows. This allows developers to combine different functional modules like building blocks to build AI applications that meet specific needs. Furthermore, it supports topological sorting for executing complex task flows. Topological sorting is an algorithm that sorts directed acyclic graphs, ensuring that tasks are executed in the correct order and avoiding dependency errors. In Dify, this mechanism enables complex task flows to be processed efficiently and accurately.

[0043] In one embodiment, the present invention also provides an intelligent process orchestration method. This method includes: initially, data access is performed, that is, the current data to be predicted is obtained. Then, dynamic feature extraction is performed on the current data to be predicted. After extracting the key features, a performance prediction model or a fault prediction analysis model is scheduled through conditional branch judgment to perform corresponding task prediction. It should be noted that the Dify scheduler here can also schedule performance prediction models and fault prediction analysis models. In other words, the performance prediction model and the fault prediction analysis model can also be constructed based on the AI ​​model.

[0044] Based on any of the above embodiments, the performance prediction model includes a prediction knowledge base; the prediction knowledge base includes multiple document types of prediction knowledge.

[0045] Specifically, the prediction knowledge base supports uploading files in formats such as TXT, PDF, Word, Excel, and HTML, with a maximum file size of 15MB. It also supports syncing content from Notion and websites for dynamic updates. Furthermore, uploaded initial data can be automatically segmented into text using common patterns and parent-child segmentation, processing redundant symbols to obtain more accurate prediction background knowledge. This addresses the data silos inherent in traditional prediction methods and enables highly accurate network performance prediction.

[0046] Based on any of the above embodiments, step 130 includes: Based on the search text in the key feature, a key document segment matching the search text is retrieved from a prediction knowledge base; Based on the performance prediction model, feature encoding is performed on the key features to obtain a key feature vector; Matching similar knowledge texts of the key feature vector from the prediction knowledge base; Obtaining enhanced similar text based on the key document segment and / or the similar knowledge text; Based on the performance prediction model, the key feature vector and the enhanced similar text are applied to output the performance prediction result.

[0047] Here, key document segments refer to text snippets retrieved from the prediction knowledge base that are semantically similar or related to the search text in the current key feature. Key document segments provide direct historical data or solution references for performance prediction, helping the performance prediction model understand the background and context of the current key feature, thereby improving the accuracy and relevance of predictions.

[0048] Furthermore, similar knowledge text refers to knowledge text that has semantically similar key features, selected by calculating the similarity between key feature vectors and text vectors in the knowledge base. This enriches the contextual information for performance prediction, providing more knowledge and experience relevant to the current network state, and helping the performance prediction model make more comprehensive and accurate predictions.

[0049] Therefore, enhanced similar text here refers to a more accurate and comprehensive text collection obtained by combining key document segments and / or similar knowledge text. It can be understood that enhanced similar text improves the utilization efficiency of the predictive knowledge base and the input quality of the performance prediction model by optimizing the display order of search results and integrating multi-source information, providing more reliable and powerful support for performance prediction.

[0050] Specifically, first, text information for retrieval can be extracted from key features. The text information may be keywords, phrases or descriptive sentences related to network performance, thereby realizing the extraction of retrieval text.

[0051] Then, using information retrieval techniques such as TF-IDF, BM25, or deep learning-based semantic retrieval models, the predicted knowledge base can be searched for document segments that are semantically similar or related to the search text. It should be noted that by searching the search text, key document segments matching the search text are retrieved from the predicted knowledge base, enabling fast and comprehensive search for relevant text and providing valuable contextual information for performance prediction. Furthermore, the retrieval process improves knowledge base utilization, avoids analyzing network status from scratch, and conserves computing resources.

[0052] Alternatively, key features can be encoded based on the text encoder in the performance prediction model to obtain key feature vectors. For example, the encoder portion of the performance prediction model, such as BERT or Transformer, can be used to encode key features and convert them into high-dimensional vector representations, thereby obtaining key feature vectors. It is understood that key feature vectors can be used to reflect the textual semantic information of key features, ensuring that the encoded vectors capture the semantic and relational information in the key features, facilitating subsequent similarity calculations and predictions.

[0053] It is understandable that feature encoding converts discrete key features into a continuous vector space, which facilitates mathematical operations and similarity comparisons in the model. In addition, high-quality feature vectors help improve the accuracy of subsequent similar knowledge text matching and performance prediction.

[0054] Furthermore, similar knowledge texts to the key feature vector can be matched from the predicted knowledge base. For example, the similarity between the key feature vector and all text vectors in the predicted knowledge base can be calculated using metrics such as cosine similarity and Euclidean distance. Based on a set similarity threshold, knowledge texts with high similarity to the key feature vector are screened. These screened knowledge texts are then reordered to obtain optimized similar knowledge texts.

[0055] It should be noted that matching similar knowledge texts can further enrich the contextual information of performance prediction, improving the comprehensiveness and accuracy of prediction. In addition, through similarity screening, irrelevant or low-quality knowledge texts can be excluded, reducing noise interference.

[0056] It should be noted that in the above process, the execution order of retrieving the key document segments matching the retrieval text from the prediction knowledge base based on the retrieval text in the key features, and the execution order of the similar knowledge text matching the key feature vector from the prediction knowledge base is not specifically limited in the embodiment of the present invention. They can be executed simultaneously or one of them can be executed first.

[0057] Next, the key document segments and similar knowledge texts are re-ranked to generate enhanced similar texts. For example, a re-ranking algorithm (such as Learning to Rank) can be designed by combining factors such as the search scores, timestamps, and authority of the key document segments and similar knowledge texts. Based on the re-ranking results, the key document segments and similar knowledge texts are then integrated to generate an enhanced similar text set.

[0058] It should be noted that the reranking process optimizes the display order of search results, placing more relevant and authoritative knowledge texts at the top, improving the efficiency of information retrieval. Enhancing similar texts provides more accurate and comprehensive contextual support for subsequent performance predictions.

[0059] Finally, based on the performance prediction model, the key feature vectors and the enhanced similar text are applied to output the performance prediction results. For example, the key feature vectors and the vector representations of the enhanced similar text can be fused, such as by concatenation or weighted summation. The fused features are then input into the prediction layer of the performance prediction model, which outputs the predicted network performance metrics.

[0060] It should be noted that combining key features with enhanced similar text for prediction fully leverages historical data and contextual information in the knowledge base, improving prediction accuracy and robustness. The prediction results can provide network operations and maintenance personnel with more precise decision support, helping them proactively address potential network performance issues.

[0061] Based on any of the above embodiments, Figure 2 This is a flow chart of the communication network performance indicator prediction method based on dify arrangement provided by the present invention, such as Figure 2 As shown, the method includes: data collection and preprocessing, that is, realizing unified access management of relational data, traffic logs, device status and other data. Then, dynamic feature selection is performed. In detail, key features are dynamically selected according to changes in network status to improve computing efficiency and prediction accuracy. Furthermore, hybrid model prediction is orchestrated through intelligent workflow. In detail, different data and prediction algorithms are collected according to the prediction scenario to improve resource utilization; in addition, through the support of RAG-enhanced prediction knowledge base, vector retrieval, full-text retrieval and hybrid mode are supported to quickly and easily obtain historical faults and operation and maintenance knowledge, thereby improving fault prediction efficiency. Then, feedback optimization is performed based on the prediction results to end the prediction process.

[0062] Based on any of the above embodiments, Figure 3 This is a schematic diagram of the structure of the network performance indicator prediction device provided by the present invention. Figure 3 As shown, the device includes: An acquiring unit 310 acquires current data to be predicted, where the current data to be predicted includes a current network state; The feature selection unit 320 inputs the current network state and the current data to be predicted into a feature selection model to obtain the current key features output by the feature selection model; the feature selection model is constructed based on a reinforcement learning algorithm; The prediction unit 330 inputs the current key feature into a performance prediction model to obtain a performance prediction result output by the performance prediction model.

[0063] The device provided by the embodiment of the present invention obtains the current data to be predicted including the current network status; inputs the current network status and the current data to be predicted into a feature selection model constructed based on a reinforcement learning algorithm to obtain the current key features output by the feature selection model; inputs the current key features into a performance prediction model to obtain the performance prediction results output by the performance prediction model, thereby achieving network performance indicator prediction with high accuracy, high timeliness and fast prediction speed, thereby ensuring stable and efficient operation of the network.

[0064] Based on any of the above embodiments, the feature selection unit is specifically used to: Based on the previous network state and the previous key feature, calculate the current feature reward; Based on the feature selection model, the feature selection reward, the current network state, and the current data to be predicted are applied to output the key features.

[0065] Based on any of the above embodiments, the feature selection model includes multiple feature selection sub-models; The feature selection unit is also specifically used for: Based on the scheduler, the multiple feature selection sub-models are scheduled to respectively perform vector encoding on the current network state to obtain a current network state vector, and the feature selection reward, the current network state vector and the current data to be predicted are applied to output the key features.

[0066] Based on any of the above embodiments, the scheduler is a Dify scheduler.

[0067] Based on any of the above embodiments, the performance prediction model includes a prediction knowledge base; the prediction knowledge base includes multiple document types of prediction knowledge.

[0068] Based on any of the above embodiments, the prediction unit is specifically configured to: Based on the search text in the key feature, a key document segment matching the search text is retrieved from a prediction knowledge base; Based on the performance prediction model, feature encoding is performed on the key features to obtain a key feature vector; Matching similar knowledge texts of the key feature vector from the prediction knowledge base; Obtaining enhanced similar text based on the key document segment and / or the similar knowledge text; Based on the performance prediction model, the key feature vector and the enhanced similar text are applied to output the performance prediction result.

[0069] Figure 4 An example of a physical structure diagram of an electronic device is shown below. Figure 4As shown, the electronic device may include: a processor 410, a communication interface 420, a memory 430, and a communication bus 440, wherein the processor 410, the communication interface 420, and the memory 430 communicate with each other via the communication bus 440. The processor 410 may call logic instructions in the memory 430 to execute a network performance indicator prediction method, which includes: obtaining current data to be predicted, the current data to be predicted including the current network state; inputting the current network state and the current data to be predicted into a feature selection model to obtain current key features output by the feature selection model; the feature selection model is constructed based on a reinforcement learning algorithm; and inputting the current key features into a performance prediction model to obtain a performance prediction result output by the performance prediction model.

[0070] Furthermore, the logic instructions in the aforementioned memory 430 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 the present invention, or the portion 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, stored in a storage medium, includes instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0071] On the other hand, the present invention also provides a computer program product, which includes a computer program, and the computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the network performance indicator prediction method provided by the above methods, and the method includes: obtaining current data to be predicted, and the current data to be predicted includes the current network status; inputting the current network status and the current data to be predicted into a feature selection model to obtain the current key features output by the feature selection model; the feature selection model is constructed based on a reinforcement learning algorithm; inputting the current key features into a performance prediction model to obtain the performance prediction results output by the performance prediction model.

[0072] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the network performance indicator prediction method provided by the above-mentioned methods, the method comprising: obtaining current data to be predicted, the current data to be predicted including the current network state; inputting the current network state and the current data to be predicted into a feature selection model to obtain the current key features output by the feature selection model; the feature selection model is constructed based on a reinforcement learning algorithm; inputting the current key features into a performance prediction model to obtain the performance prediction results output by the performance prediction model.

[0073] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0074] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion 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, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0075] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A network performance indicator prediction method, characterized in that: include: Acquire current data to be predicted, where the current data to be predicted includes a current network state; Inputting the current network state and the current data to be predicted into a feature selection model to obtain current key features output by the feature selection model; The feature selection model is constructed based on the reinforcement learning algorithm; The current key features are input into a performance prediction model to obtain a performance prediction result output by the performance prediction model.

2. The network performance indicator prediction method according to claim 1, characterized in that: Inputting the current network state and the current data to be predicted into a feature selection model to obtain the current key features output by the feature selection model includes: Based on the previous network state and the previous key feature, calculate the current feature reward; Based on the feature selection model, the feature selection reward, the current network state, and the current data to be predicted are applied to output the key features.

3. The network performance indicator prediction method according to claim 2, characterized in that: The feature selection model includes multiple feature selection sub-models; The step of applying the feature selection reward, the current network state, and the current data to be predicted based on the feature selection model to output the key features includes: Based on the scheduler, the multiple feature selection sub-models are scheduled to respectively perform vector encoding on the current network state to obtain a current network state vector, and the feature selection reward, the current network state vector and the current data to be predicted are applied to output the key features.

4. The network performance indicator prediction method according to claim 3, characterized in that: The scheduler is a Dify scheduler.

5. The network performance indicator prediction method according to any one of claims 1 to 4, characterized in that: The performance prediction model includes a prediction knowledge base; the prediction knowledge base includes multiple document types of prediction knowledge.

6. The network performance indicator prediction method according to claim 5, characterized in that: Inputting the current key features into the performance prediction model to obtain a performance prediction result output by the performance prediction model includes: Based on the search text in the key feature, a key document segment matching the search text is retrieved from a prediction knowledge base; Based on the performance prediction model, feature encoding is performed on the key features to obtain a key feature vector; Matching similar knowledge texts of the key feature vector from the prediction knowledge base; Obtaining enhanced similar text based on the key document segment and / or the similar knowledge text; Based on the performance prediction model, the key feature vector and the enhanced similar text are applied to output the performance prediction result.

7. A network performance indicator prediction device, characterized in that: include: an acquiring unit, configured to acquire current data to be predicted, wherein the current data to be predicted includes a current network state; A feature selection unit, inputting the current network state and the current data to be predicted into a feature selection model to obtain a current key feature output by the feature selection model; The feature selection model is constructed based on the reinforcement learning algorithm; The prediction unit inputs the current key feature into the performance prediction model to obtain a performance prediction result output by the performance prediction model.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the network performance indicator prediction method according to any one of claims 1 to 6 is implemented.

9. 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, the network performance indicator prediction method according to any one of claims 1 to 6 is implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the network performance indicator prediction method according to any one of claims 1 to 6 is implemented.

Images