Network optimization method and device and electronic equipment

Abstract

The invention provides a network optimization method and device and electronic equipment, and relates to the technical field of communication network optimization and artificial intelligence, and the method comprises the steps: receiving first information inputted by a user, the first information being used for representing a network optimization demand of the user; performing intention analysis on the first information to generate target performance information of network optimization; cutting a pre-acquired action space to obtain at least one application of which the adaptation degree with the target performance information is greater than a preset threshold value, the pre-acquired action space comprising a plurality of applications; and performing network optimization by using the at least one application. In the embodiment of the invention, the user can express the network optimization requirement through the natural language, the natural language requirement of the user can be automatically understood, the executable network strategy can be quickly generated, and the generation efficiency of the network strategy can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of communication network optimization and artificial intelligence, and particularly relates to a network optimization method and device and electronic equipment. BACKGROUND

[0002] In a traditional wireless network management system, network optimization relies on pre-defined rule configuration or manual strategy arrangement. However, in actual application, the network optimization requirements of different users often change, and the fixedness of pre-defined rules and the hysteresis of manual arrangement make it difficult for network optimization strategies to quickly respond to the requirements of users, thereby resulting in low efficiency of network optimization strategy generation. SUMMARY

[0003] Embodiments of the present application provide a network optimization method, device and electronic equipment to solve the problem of low efficiency of network optimization strategy generation.

[0004] To solve the above technical problems, the present application is implemented as follows:

[0005] In a first aspect, the embodiments of the present application provide a network optimization method, and the method comprises the following steps:

[0006] receiving first information input by a user, the first information being used to represent network optimization requirements of the user;

[0007] performing intent analysis on the first information to generate target performance information of network optimization;

[0008] performing pruning on a pre-acquired action space to obtain at least one application with an adaptation degree greater than a preset threshold to the target performance information, the pre-acquired action space comprising a plurality of applications;

[0009] performing network optimization by using the at least one application.

[0010] In a second aspect, the embodiments of the present application provide a network optimization device, and the device comprises:

[0011] a receiving module configured to receive first information input by a user, the first information being used to represent network optimization requirements of the user;

[0012] an analysis module configured to perform intent analysis on the first information to generate target performance information of network optimization;

[0013] a pruning module configured to perform pruning on a pre-acquired action space to obtain at least one application with an adaptation degree greater than a preset threshold to the target performance information, the pre-acquired action space comprising a plurality of applications;

[0014] an optimization module configured to perform network optimization using the at least one application.

[0015] In a third aspect, an embodiment of the present application provides an electronic device, comprising a processor, a memory, and a program stored in the memory and capable of running on the processor, and when the program is executed by the processor, the steps of the network optimization method in the first aspect are implemented.

[0016] In a fourth aspect, an embodiment of the present application provides a computer readable storage medium, and the computer readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the network optimization method in the first aspect are implemented.

[0017] In a fifth aspect, a computer program product is provided, comprising computer instructions, and when the computer instructions are executed by a processor, the steps of the network optimization method in the first aspect are implemented.

[0018] In the embodiments of the present application, the user can express the network optimization requirement through natural language, and the natural language requirement of the user can be automatically understood to quickly generate an executable network policy, thereby reducing the manual configuration workload, improving the network policy generation efficiency, and improving the network optimization efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the description of the embodiments of the present application will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0020] Figure 1 is a flowchart of a network optimization method provided by an embodiment of the present application;

[0021] Figure 2 is a network optimization flowchart provided by an embodiment of the present application;

[0022] Figure 3 is a structural schematic diagram of a network optimization device provided by an embodiment of the present application;

[0023] Figure 4 is a structural schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0024] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described, obviously, the described embodiments are some of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the protection scope of the present application.

[0025] The embodiments of the present application provide a network optimization method, device and electronic equipment to solve the problem of low efficiency of network optimization strategy generation.

[0026] Referring to Figure 1 , Figure 1 is a flowchart of a network optimization method provided by the embodiments of the present application, as shown in Figure 1 , the method comprises the following steps:

[0027] Step 101, receiving first information input by a user, the first information being used to represent the network optimization demand of the user;

[0028] Step 102, performing intent analysis on the first information to generate target performance information of network optimization;

[0029] Step 103, performing pruning on a pre-acquired action space to obtain at least one application with an adaptation degree greater than a preset threshold to the target performance information, the pre-acquired action space comprising a plurality of applications;

[0030] Step 104, performing network optimization by using the at least one application.

[0031] The first information can be information input by the user in a natural language manner through a terminal, for example, the user inputs “improve high-definition video experience”, “reduce remote meeting lag rate” and the like. The user can express the network optimization demand in a colloquial manner through a terminal device, the system receives the user text input by using a standardized input interface, and performs preliminary preprocessing on the input content, including removing special characters, converting to lowercase and the like, so as to form structured first information input.

[0032] The intent analysis on the first information can obtain the intent of the user, and generate target performance information of network optimization based on the intent of the user.

[0033] In order to convert the semantic input corresponding to the first information into an executable strategy target, an input sequence constructed based on few-shot prompting is adopted, and the input content comprises the following three items:

[0034] Intent (i): network target stated by the user (natural language description);

[0035] Type (Type) (λ): Intended type (such as throughput, energy consumption, delay);

[0036] Keywords (Keywords) (λ): Technical keywords extracted in the intention;

[0037] The prompt is sent to a pre-trained language model (such as Mini-BERT or distilled BERT), and the output is a structured tuple: ;

[0038] The output will be used as key control information for the downstream network optimization phase.

[0039] For example, the output includes:

[0040] Intent: Reduce network energy consumption by 10%;

[0041] Type: Energy consumption;

[0042] Keywords: Power consumption, energy saving, green.

[0043] The specific algorithm includes the following process:

[0044] Algorithm 1 Intent resolution based on Large Language Model (LLM) Require: E = {( ), ( ), …} / / Few-shot prompt example set I = {i} / / set of intents to be resolved LLM / / light-weight language model Ensure: / / Structured intent output set Procedure CLASSIFY_AND_EXTRACT(I, E, LLM)  1:  ← ∅ 2: for each do 3: p ← CREATE_PROMPT(i, E) 4: response ← QUERY_LLM(p, LLM) 5:  ( ) ← PARSE_RESPONSE(response) 6:   ← Ξ ∪ {( )} 7: end for 8: return Ξ end procedure Procedure CREATE_PROMPT(i, E)  1: p ← "" 2: for each do 3: p ← p + "Example:" 4:  p ← p + "Intent: " + 5:  p ← p + "Type: " + 6:  p ← p + "Keywords: " + 7: end for 8: p ← p + "New Intent: " + 9: p ← p + "Type, Keywords" 10: return p end procedure Procedure QUERY_LLM(p, LLM) 1: return LLM.query(p) end procedure Procedure PARSE_RESPONSE(response) 1: ty ← EXTRACT_TYPE(response) 2: λ ← EXTRACT_KEYWORDS(response) 3: return (ty, λ) end procedure

[0045] The above process can be implemented by a hierarchical deep Q network (h-DQN), which contains two control levels: a policy guidance layer and a policy execution layer.

[0046] Among them, the policy guidance layer (Meta-Controller) located in the network strategy controller is used to receive structured intention input and output target performance vector , and initialize the action space filtering threshold . Its tasks include:

[0047] 1. Set target KPI according to intention type, such as (up, down, up) (throughput, latency, power consumption);

[0048] 2. Manage the calling order and target decomposition structure of multiple low-level policy modules.

[0049] The policy execution layer (Controller) deployed in the tactical controller is used to receive target performance information from the Meta layer, combine the current network state , and select the optimal optimization application combination . The action space of this layer is pruned and optimized through attention mechanism.

[0050] ​In some embodiments, the network optimization application is applied as an initial constructed action space, an attention mechanism is used to calculate the adaptation degree score of each application relative to the user intent, and by setting a filtering threshold (i.e., the above-mentioned preset threshold), applications with low adaptation degree scores are cut off, and at least one application with an adaptation degree score greater than the preset threshold (i.e., the adaptation degree with the target performance information is greater than the preset threshold) is obtained. For example, for the "high-definition video experience" intent, the system calculates App1 (traffic steering) score 0.87, App3 (beamforming) score 0.82, and App4 (power allocation) score 0.79, and when ε = 0.7, the candidate set (i.e., the application subset) A att = {App1, App3, App4}. That is, the Controller first performs attention scoring α(s t , i, a) on the action set A s ᵃᵗᵗ 。

[0051] In some embodiments, an application preference matrix is constructed based on historical optimization results. The system analyzes historical optimization data to count the success rate of each type of application under different intent types, and establishes an "intent-application" preference matrix. The calculation method is: preference score = (number of times of success of the application under the intent) / (total number of applications of the application under all intents). When the preference score is greater than a preset threshold (such as 0.6), the application is included in the candidate set. For example, if the "throughput" intent "traffic steering" success rate is 80%, it is included in the candidate set.

[0052] In some embodiments, the above-mentioned candidate set can be used as at least one application, and in some embodiments, one step of screening can be performed in the candidate set to obtain at least one application.

[0053] Since each application in the above-mentioned at least one application has a high adaptation degree with the target performance information, the network optimization is executed by using the combination of the above-mentioned at least one application, or is executed in turn according to a preset priority order.

[0054] In the embodiments of the present application, the user can express the network optimization demand through natural language, without professional knowledge, and the network configuration threshold is greatly reduced. By cutting the action space to obtain the optimal action for combination execution, compared with single application optimization, the network throughput can be improved.

[0055] Optionally, the cutting of the pre-acquired action space to obtain at least one application with an adaptation degree greater than a preset threshold with the target performance information comprises:

[0056] constructing the action space comprising the plurality of applications;

[0057] determine a fitness score of each application in the plurality of applications to the target performance information, the fitness score indicating a degree of fitness of the each application to the target performance information;

[0058] prune the action space to obtain an application subset, each application in the application subset having a fitness score to the target performance information greater than a preset threshold;

[0059] determine an application combination including the at least one application in the application subset by a reward mechanism.

[0060] construct an action space including a plurality of applications The plurality of applications can include different types of optimization applications. Action space pruning optimization is performed through an attention mechanism.

[0061] First, define an attention score function

[0062]

[0063] wherein, is a score function obtained through supervised training, and the output is a fitness score of a current action a to a user intent i (i.e., a fitness score to the target performance information).

[0064] Prune the action space by the following formula to obtain an application subset:

[0065]

[0066] When pruning the action space, set a filtering threshold Filter the applications less than or equal to the filtering threshold to obtain an application subset, that is, the score of each application in the application subset is greater than the filtering threshold .

[0067] In some embodiments, the input includes the current network state and the target performance information obtained through intent analysis. For each application a∈{App1, App2, App3, App4, App5}, the score is calculated through the attention mechanism, for example: for the "improve high-definition video experience" intent, App1 (traffic steering) scores 0.87, App3 (beamforming) scores 0.82, and App4 (power allocation) scores 0.79. The score reflects the matching degree of the application to the target performance information, and a high score indicates that the application is more in line with the user's demand. Obtain the above-mentioned applications with a score greater than a preset threshold, and use these applications for network optimization. Through action space pruning, the candidate application set is reduced, and the efficiency of application combination determination is improved.

[0068] In the application subset obtained through action space pruning In the middle, the corresponding reward mechanism selects the action to be executed by the following strategy function:

[0069]

[0070] Wherein, Q( , a) represents the state-action value.

[0071] For example, for the application combination {App1, App3} corresponding to at least one application, if the throughput is improved by 12% and the power consumption is increased by 2%, the comprehensive reward value of the strategy execution is The score is higher than that of triggering App1 alone. Therefore, the combined strategy is preferentially executed.

[0072] In the above manner, the application combination with a higher comprehensive reward is further selected in the application subset obtained after spatial pruning, which can improve the effect of network optimization.

[0073] Optionally, the plurality of applications includes at least two of traffic steering, base station dormancy, beamforming, power allocation, and energy-saving handover management.

[0074] The plurality of applications can include two or more of the above-mentioned applications, and in some embodiments, can include the above-mentioned five kinds.

[0075] Among them, traffic steering can be used to redirect data packets to a more optimal transmission path;

[0076] Base station dormancy can be used to turn off part of the base station to save energy under low load;

[0077] Beamforming can be used to optimize the signal transmission direction to improve channel quality;

[0078] Power allocation can be used to adjust the transmission power under the premise of ensuring coverage;

[0079] Energy-saving handover management can be used to optimize user equipment switching strategy to reduce energy consumption.

[0080] The specific information of the above-mentioned five applications is shown in the following table:

[0081] Application Typical triggering scenarios Input features Key performance improvement App1 traffic steering Macro cell congestion, micro cell idle 3.5 GHz load 95% ↑ throughput 15% App2 base station dormancy Nighttime flow < 15% of peak Queue = 0 ↓ power consumption 20% App3 beamforming Augmented Reality (AR) head-mounted display movement UE (x, y, z) ↑ Signal to Interference plus Noise Ratio (SINR) 8 dB App4 power allocation Edge user weak coverage RSRP -110 dBm ↑ Cell-Edge Throughput (Cell-Edge TP) 18% App5 energy saving handover Internet of Things (IoT) low-speed roaming Speed 0.9 m / s ↓ handover failure 30%

[0082] The above-mentioned multiple applications can perform network optimization from different dimensions, which can improve the optimization effect.

[0083] Optionally, the network optimization using the at least one application comprises:

[0084] Constructing a triple of network optimization based on the at least one application, the triple comprising a set of initial states, a termination condition and an optimization strategy;

[0085] perform network optimization based on the initial state set according to the optimization strategy, and terminate the network optimization when the termination condition is met.

[0086] For each action in at least one application , it can be regarded as a "Markov option", which is defined by a triple, including an initial state set , a termination condition , and an optimization strategy function .

[0087] The option controller updates the strategy by accumulating the combination of intrinsic rewards and extrinsic target rewards. The intrinsic reward can be calculated by the following formula:

[0088]

[0089] wherein, is the intrinsic reward, is the positive reward, is the penalty term.

[0090] Based on the initial state set, the network optimization is performed using the optimization strategy function, and the network optimization is terminated when the termination condition is met, which can be used to monitor the execution of network optimization.

[0091] The present embodiment is applicable to application combination scheduling under multi-objective optimization tasks. For example, the termination condition β of App2 (base station sleep) is TrafficLoad > 40 % or the end of the period, and when the above termination condition is met, the base station sleep is terminated.

[0092] In the above manner, network optimization can be automatically performed based on the strategy function, and terminated when the termination condition is met, which can improve the efficiency of network optimization.

[0093] Optionally, the first information is analyzed for intent, and target performance information for network optimization is generated, including:

[0094] The first information is analyzed to obtain intent information and technical keywords for network optimization;

[0095] The intent information and the technical keywords are used to generate the target performance information for network optimization.

[0096] Using a lightweight large language model based on few-shot prompting for intent analysis, the intent information for network optimization, such as throughput, energy consumption, etc., and the technical keywords, such as power consumption, energy saving, etc., can be obtained.

[0097] With the above intention information, a target Key Performance Indicator (KPI) such as throughput can be inferred; and a key parameter can be determined through a technical keyword, so as to generate target performance information. For example, a target performance vector G = {throughput↑, latency↓, energy consumption→}.

[0098] In this way, intention analysis is performed based on information for input, so as to generate target performance information for network optimization, which can reduce the threshold of professional knowledge of a user and improve the effect of network performance optimization.

[0099] Optionally, the method further comprises:

[0100] predicting future traffic based on the intention information and the technical keyword to obtain predicted traffic;

[0101] comparing the predicted traffic with a preset load threshold range to obtain a comparison result;

[0102] the network optimization by using the at least one application comprises:

[0103] in a case where the comparison result indicates that the network optimization is executable, performing network optimization by using the at least one application.

[0104] Since the user intention may deviate from the network state at the time of input, to avoid invalid or unstable strategy execution, a “prediction verification submodule” is added to verify the feasibility of each intention target in a future time slot.

[0105] This module uses a traffic prediction network based on an Autoformer architecture to predict future traffic in a t+1 time slot, and sets an upper threshold and a lower threshold corresponding to the load threshold range:

[0106] : high load threshold (trigger performance enhancement optimization);

[0107] : low load threshold (trigger energy consumption saving optimization).

[0108] The predicted traffic is compared with the two thresholds of the above load threshold range, and if the comparison result indicates that the network optimization is executable, the above network optimization is performed; if it indicates that the network optimization is not executable, a prompt information is output.

[0109] For example, if it is located in the above threshold range, it indicates that it has feasibility. If it exceeds the above threshold range and the target performance and the current performance differ greatly, it indicates that the feasibility is low or not executable.

[0110] The effect of network optimization execution is improved through feasibility verification.

[0111] Optionally, in the case that the comparison result indicates that the network optimization is executable, the network optimization is performed by using the at least one application, and the method further includes:

[0112] In the case that the predicted traffic flow exceeds the load threshold range, a deviation between the target performance information and the current performance is obtained.

[0113] In the case that the deviation is less than or equal to a preset value, the network optimization is performed by using the at least one application.

[0114] The method further includes:

[0115] In the case that the deviation is greater than the preset value, prompt information indicating that the network optimization is not executable is output.

[0116] In the case that the predicted traffic flow exceeds the load threshold range, a KPI index set corresponding to a QoS class is called to perform deviation calculation on the intent.

[0117]

[0118] In the case that the predicted traffic flow exceeds the load threshold range, a KPI index set corresponding to a QoS class is called to perform deviation calculation on the intent. In the case that the predicted traffic flow exceeds the load threshold range, a KPI index set corresponding to a QoS class is called to perform deviation calculation on the intent.

[0119] In some embodiments, when the deviation is small, i.e., less than or equal to a preset value, it is indicated that the network optimization is executable.

[0120] When the deviation is large, i.e., greater than the preset value, it is indicated that the network optimization is less feasible or not executable, and prompt information indicating that the network optimization is not executable is output. Or prompt information indicating whether to reduce the target performance is output, so that a user reduces the network optimization requirement based on the prompt information.

[0121] The above process is shown as follows:

[0122] Intent verification 1. Predict network traffic at time t+1 , call algorithm 1 to extract structured intent . 2. If > or < then the following steps are performed: 3. Find intent corresponding service category QoS parameter set . 4. If the intent i affects the QoS parameter set : 5. For each performance parameter in the set of performance parameters :​ 6. Calculate the QoS drift of this parameter . 7. If : 8. Mark the intent as invalid and exit the loop. 9. If all parameters = 0, or > : 10. Mark the intent as "verified". 11. Otherwise: 12. Recompute threshold With . 13. Return the intent verification result.

[0123] The execution flow of the above embodiments of the application can be referred to Figure 2 .

[0124] In order to facilitate understanding of the embodiments, the following is described by a specific example.

[0125] If the user intent is "reduce the high-definition video packet loss rate to below 1%", the LLM parses the Type = throughput, and the target KPI is "Packet Loss ≤ 1%". The attention block will assign higher weights to App1 (traffic steering), App3 (beamforming), and App4 (power allocation), while assigning lower weights to App2 / 5 and eliminating them. The Controller then selects a single or combined strategy (such as {App1+App3}) from {App1, App3, App4} based on the Q-value function, continuously executes, and updates the network with real-time KPI feedback.

[0126] Its operation flow is as follows:

[0127] 1. Meta-Controller receives intent i, parses its optimization target type (such as throughput↑, latency↓, energy consumption↓), and generates target vector G.

[0128] 2. Construct the initial action space A = {App1, App2,..., AppN}.

[0129] 3. For each action a∈A, use the attention mechanism to calculate its score:

[0130]

[0131] 4. Construct the action subset (i.e. application subset) :

[0132] = {a∈A |α( , i, a) >ε}

[0133] 5. For each action in , calculate its state-action value Q( , a).

[0134] 6. Select the optimal action from :

[0135] π( ) = argmax Q( , a)

[0136] 7. Send the selected action to the tactical controller, call the corresponding optimization application module to execute:

[0137] - App1: traffic steering

[0138] - App2: base station sleep

[0139] - App3: beamforming

[0140] - App4: Power allocation

[0141] - App5: Handover management

[0142] 8. Collect KPIs in real time = {Throughput, Delay, Energy}.

[0143] 9. If the target is not achieved (e.g., throughput improvement is less than 10%), go back to step 1 to adjust the target vector G or expand .

[0144] 10. Return the current optimal strategy execution result.

[0145] In the embodiments of the application, an intent analysis mechanism combining a few-shot prompt and a large language model is used: a structured few-shot prompt is constructed to guide a lightweight large language model to complete type classification and keyword extraction of natural language intent. The above method has high universality and expansibility and can be widely adapted to diversified intent input in operator scenarios. The natural language driven semantic analysis capability can be supported, and the method is different from traditional methods that rely on rule parameter input. The method can analyze strategy intent expressed by a user in a colloquial or task target manner, significantly improves human-computer interaction efficiency, and simplifies a network configuration process.

[0146] An intent feasibility verification mechanism based on traffic prediction is used: an Autoformer or other time series model is introduced to predict future traffic trends, QoS drift calculation is performed in combination with an intent target, and it is determined whether the target can be executed under the condition that resources are reachable, thereby effectively avoiding uncontrollable strategy issuing. By introducing the intent verification process based on time series prediction, the problem that a strategy target does not match reality and causes execution failure is solved, and the stability and controllability of the system are enhanced.

[0147] An attention mechanism based hierarchical reinforcement learning control framework (Attention-hDQN) is used: a Meta-Controller is used to decompose a target and initialize an action space, an Attention Block is used to perform action pruning, and a Controller is used to perform high-value strategy selection. The structure significantly improves action space exploration efficiency, reduces training complexity, and improves the pertinence and generalization ability of strategy decision-making. State information and intent features are combined to dynamically score candidate actions, high-relevance actions are preferentially explored, the number of redundant strategy attempts is effectively reduced, and the convergence speed and performance upper limit of reinforcement learning are improved.

[0148] Optimization application combination execution mechanism under Markov option definition: five types of network optimization applications (App1-App5) are modeled as Markov options, combined with a goal-oriented strategy function π(s t ) to achieve combined scheduling and adaptive control of optimization tasks, and to improve the performance of network performance collaboration.

[0149] Performance feedback driven closed loop self-tuning mechanism: real-time monitoring of KPI and feedback of performance deviation to the strategy generation layer to realize online fine-tuning and strategy compensation of the control target, and to guarantee the consistency and robustness of target execution.

[0150] Traditional strategy distribution and resource management methods cannot meet the dynamic, multi-objective, and multi-role scheduling requirements.

[0151] For example, in some embodiments, the cell parameter configuration is automatically optimized and the wireless network performance is improved by generating experience tuples and updating the strategy based on global / local rewards. This method adaptively selects the optimal action through the reinforcement learning mechanism, but lacks the ability to process user semantic intent and still relies on explicitly modeled strategy goals. In other embodiments, device layer integration and performance improvement are used, but the strategy generation and semantic interaction are still insufficient, especially lacking support for user natural language intent.

[0152] The embodiments of the present application combine the network strategy optimization method of large language model (LLM) and attention mechanism, express the user's network target (intent) in a form, and hand it over to the network system for automatic analysis, verification and execution, which simplifies the management process and improves the adaptive ability of network service. It is suitable for intent-driven wireless network automation management, especially for strategy control and resource scheduling optimization under 5G and future evolution network architecture. It solves the problems of how to construct an executable strategy path through semantic analysis and strategy mapping mechanism under the fuzzy expression of user natural language intent, how to real-time check the feasibility of strategy intent under the condition of dynamic change of network resources, and how to reasonably allocate resources among various optimization applications, and realize the optimal strategy combination and scheduling combined with the attention mechanism.

[0153] Specifically, by intent analysis, policy verification, and optimized policy scheduling and execution, a lightweight large language model (LLM) is used for semantic abstraction processing, and an attention mechanism is embedded into a hierarchical deep Q-network (h-DQN) architecture to achieve efficient policy execution. It can automatically understand the natural language requirements of operation and maintenance personnel or users, quickly generate executable network policies, greatly reduce manual configuration workload, and improve service delivery efficiency. It is a key supporting module in the intent-driven network system. From semantic analysis, prediction verification to policy execution, each step has explicit intermediate data records (such as G, , KPI trajectory), which facilitates post-analysis, optimization and auditing, and meets the management needs of operators for transparent AI.

[0154] Referring to Figure 3 , Figure 3 is a structural schematic diagram of a network optimization device provided by an embodiment of the present application, as shown in Figure 3 , the network optimization device 300 comprises:

[0155] A receiving module 301 is configured to receive first information input by a user, wherein the first information is used to represent the user's network optimization requirements.

[0156] An analysis module 302 is configured to perform intent analysis on the first information to generate target performance information for network optimization.

[0157] A pruning module 303 is configured to prune a pre-acquired action space to obtain at least one application with an adaptation degree greater than a preset threshold to the target performance information, wherein the pre-acquired action space comprises a plurality of applications.

[0158] An optimization module 304 is configured to perform network optimization using the at least one application.

[0159] Optionally, the pruning module comprises:

[0160] A first construction sub-module is configured to construct the action space comprising the plurality of applications.

[0161] A first determination sub-module is configured to determine an adaptation degree score of each application in the plurality of applications to the target performance information, wherein the adaptation degree score is used to represent the degree of adaptation of the each application to the target performance information.

[0162] A pruning sub-module is configured to prune the action space to obtain an application sub-set, wherein the adaptation degree score of each application in the application sub-set to the target performance information is greater than a preset threshold.

[0163] A second determining sub-module is configured to determine, through a reward mechanism, an application combination including the at least one application from the application subset.

[0164] Optionally, the plurality of applications include at least two of traffic steering, base station dormancy, beamforming, power allocation, and energy-saving handover management.

[0165] Optionally, the optimization module includes:

[0166] A second constructing sub-module is configured to construct a triple of network optimization based on the at least one application, the triple including a set of initial states, a termination condition, and an optimization strategy.

[0167] An optimization sub-module is configured to perform network optimization based on the set of initial states according to the optimization strategy, and terminate the network optimization when the termination condition is met.

[0168] Optionally, the parsing module is specifically configured to:

[0169] Parse the first information to obtain intention information and technical keywords of network optimization.

[0170] Generate the target performance information of network optimization by using the intention information and the technical keywords.

[0171] Optionally, the apparatus further includes:

[0172] A prediction module is configured to predict future traffic based on the intention information and the technical keywords to obtain predicted traffic.

[0173] A comparison module is configured to compare the predicted traffic with a preset load threshold range to obtain a comparison result.

[0174] The optimization module is specifically configured to:

[0175] In a case where the comparison result indicates that the network optimization is executable, perform network optimization by using the at least one application.

[0176] Optionally, the optimization module is specifically configured to:

[0177] In a case where the predicted traffic exceeds the load threshold range, obtain a deviation between the target performance information and a current performance.

[0178] In a case where the deviation is less than or equal to a preset value, perform network optimization by using the at least one application.

[0179] The method further includes:

[0180] In a case where the deviation is greater than the preset value, outputting prompt information that the network optimization is not executable.

[0181] The network optimization apparatus can implement Figure 1 the various processes implemented in the method embodiments and achieve the same technical effects. To avoid repetition, details are not described herein.

[0182] As shown in Figure 4 , the embodiments of the present application also provide an electronic device 400, which comprises a processor 401, a memory 402, and a program stored in the memory 402 and capable of running on the processor 401. The program, when executed by the processor 401, implements the various processes of the above network optimization method embodiments and achieves the same technical effects. To avoid repetition, details are not described herein.

[0183] The embodiments of the present application also provide a computer readable storage medium, which stores a computer program. The computer program, when executed by a processor, implements the various processes of the above network optimization method embodiments and achieves the same technical effects. To avoid repetition, details are not described herein. The computer readable storage medium is, for example, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk.

[0184] The embodiments of the present application also provide a computer program product, which comprises computer instructions. The computer instructions, when executed by a processor, implement the various processes of the above network optimization method embodiments and achieve the same technical effects. To avoid repetition, details are not described herein. Figure 1

[0185] It should be noted that, in this document, the terms "comprising", "including", or any other variant thereof are intended to cover non-exclusive inclusion, so that processes, methods, articles, or apparatuses that include a series of elements not only include those elements, but also include other elements not explicitly listed, or inherent to such processes, methods, articles, or apparatuses. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0186] ​Those skilled in the art can clearly understand the above-mentioned embodiment method can be realized by means of software and the necessary general hardware platform, of course, also can be through hardware, but many cases the former is the better embodiment. Based on such understanding, the technical solutions of the present application essentially or say the part of the contribution to the prior art can be embodied in the form of software products, the computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disc), including a number of instructions to make a terminal (may be a mobile phone, computer, server, air conditioner, or network equipment, etc.) executes the method described in various embodiments of the present application.

[0187] The embodiments of the present application are described above in conjunction with the drawings, but the present application is not limited to the above-mentioned specific embodiments, the above-mentioned specific embodiments are only illustrative, but not limited, those skilled in the art can make many forms without departing from the purpose of the present application and the scope of the claims under the inspiration of the present application, all belong to the protection of the present application.

Claims

1. A network optimization method, characterized by, The method comprises: receiving first information input by a user, the first information being used to represent network optimization requirements of the user; performing intention analysis on the first information to generate target performance information of network optimization; trimming a pre-acquired action space to obtain at least one application with an adaptation degree to the target performance information greater than a preset threshold, the pre-acquired action space comprising a plurality of applications; performing network optimization using the at least one application.

2. The method of claim 1, wherein, The trimming of the pre-acquired action space to obtain at least one application with an adaptation degree to the target performance information greater than a preset threshold comprises: constructing the action space comprising the plurality of applications; determining an adaptation degree score of each application in the plurality of applications to the target performance information, the adaptation degree score being used to represent a degree of adaptation of the each application to the target performance information; trimming the action space to obtain an application subset, each application in the application subset having an adaptation degree score to the target performance information greater than a preset threshold; determining an application combination comprising the at least one application in the application subset through a reward mechanism.

3. The method according to claim 1 or 2, characterized in that, The plurality of applications comprise at least two of traffic steering, base station dormancy, beamforming, power allocation, and energy-saving handover management.

4. The method of claim 1, wherein, The network optimization using the at least one application comprises: constructing a triple of network optimization based on the at least one application, the triple comprising a set of initial states, a termination condition, and an optimization strategy; performing network optimization based on the set of initial states according to the optimization strategy, and terminating the network optimization when the termination condition is met.

5. The method of claim 1, wherein, The intention analysis on the first information to generate target performance information of network optimization comprises: analyzing the first information to obtain intention information and technical keywords of network optimization; generating the target performance information of network optimization using the intention information and the technical keywords.

6. The method of claim 5, wherein, The method further comprises: predicting future traffic based on the intention information and the technical keywords to obtain predicted traffic; comparing the predicted traffic with a preset load threshold range to obtain a comparison result; The network optimization using the at least one application comprises: performing network optimization using the at least one application when the comparison result indicates that the network optimization is executable.

7. The method of claim 6, wherein, The network optimization using the at least one application when the comparison result indicates that the network optimization is executable comprises: obtaining a deviation between the target performance information and a current performance when the predicted traffic exceeds the load threshold range; performing network optimization using the at least one application when the deviation is less than or equal to a preset value; The method further comprises: outputting prompt information indicating that the network optimization is not executable when the deviation is greater than the preset value.

8. A network optimization apparatus, characterized by, comprise: a receiving module configured to receive first information input by a user, the first information being used to represent network optimization requirements of the user; an analysis module configured to perform intention analysis on the first information to generate target performance information of network optimization; a clipping module configured to clip a pre-acquired action space to obtain at least one application whose fitness to the target performance information is greater than a preset threshold, the pre-acquired action space comprising a plurality of applications; an optimization module configured to perform network optimization using the at least one application.

9. An electronic device, comprising: comprising: a processor, a memory, and a program stored on the memory and executable on the processor, the program, when executed by the processor, implements the steps of the network optimization method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, a computer readable storage medium having stored thereon a computer program, the computer program, when executed by a processor, implements the steps of the network optimization method according to any one of claims 1 to 7.

11. A computer program product, characterised in that, computer instructions, which, when executed by a processor, implement the steps of the network optimization method according to any one of claims 1 to 7.

Images