QoS optimization method and device based on long and short term prediction
By combining time series analysis of high-frequency fluctuations and low-frequency trends, more comprehensive QoS prediction results are generated and QoS configuration strategies are dynamically optimized, which solves the problem that existing technology is difficult to respond to changes in network load and diversified user needs in a highly dynamic environment, and achieves efficient resource utilization and user experience improvement.
Patent Information
- Application Number
- CN202411881631.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-19
- Publication Date
- 2025-03-25
AI Technical Summary
The prior art is difficult to respond dynamically to changes in network load and diversified user needs in a highly dynamic environment, resulting in low resource utilization and a decline in user experience.
The QoS optimization method based on long-term and short-term prediction is adopted, and by combining time series analysis of high-frequency fluctuations and low-frequency trends, more comprehensive QoS prediction results are generated, and QoS configuration strategies are dynamically optimized to achieve forward-looking scheduling of resources.
It improves the effect of QoS optimization, ensures efficient utilization of network resources and improves user experience, and can provide high-quality service support in different scenarios.
Smart Images
Figure CN119697668A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of communications, and in particular to the optimization of network parameter QoS. Background Art
[0002] In modern communication networks, with the diversification of user needs and the complexity of network environments, traditional statically configured QoS mechanisms can no longer meet the growing high-dynamic needs. Although the introduction of 5G networks has greatly improved the support capabilities of bandwidth, latency, and connection density, complex multi-dimensional application scenarios, such as real-time interactive applications (video conferencing, telemedicine), streaming services (video on demand, live broadcast), big data transmission, and IoT device monitoring, still put forward higher requirements for network resource management.
[0003] Existing technologies mainly rely on static QoS parameter configuration and fixed resource allocation strategies, which cannot respond to the dynamic changes of network load and the diversified characteristics of user needs in real time. This method is prone to resource waste or insufficient resources in a highly dynamic environment, resulting in a decline in service quality. In addition, existing prediction models are usually simplistic and cannot capture short-term fluctuations and long-term trends at the same time, making it difficult to accurately respond to holiday peak traffic or sudden demand scenarios. At the same time, the processing method of user feedback data is too simple and has not been effectively integrated into the closed-loop optimization system, resulting in insufficient response of resource adjustment to user experience and difficulty in meeting the strict service quality requirements of highly sensitive businesses. Therefore, because existing technologies usually adopt fixed resource allocation strategies, it is difficult to dynamically adapt to short-term fluctuations and long-term trend changes in network load, resulting in problems such as low resource utilization and reduced user experience.
[0004] Therefore, there is an urgent need for a QoS optimization method and device that can solve the above problems. Summary of the invention
[0005] The purpose of the present invention is to provide a QoS optimization method and device based on long-term and short-term predictions, which combines long-term and short-term predictions to capture short-term dynamic changes and identify periodic trends, thereby generating more comprehensive QoS prediction results and achieving better QoS optimization effects.
[0006] In order to achieve the above-mentioned purpose, the present invention provides a QoS optimization method based on long-term and short-term predictions, comprising: obtaining current network status information, wherein the network status information includes link load, signal strength and delay information; collecting QoS parameters and recording user equipment behavior data, wherein the QoS parameters include one or more combinations of delay, bandwidth utilization and packet loss rate, and the user equipment behavior data includes one or more combinations of access frequency, connection duration and data usage; storing the network status information, QoS parameters and user equipment behavior data, and stratifying the current and historical network status information, QoS parameters and user equipment behavior data according to time series to separate high-frequency fluctuation parts and low-frequency trend parts, transmitting the high-frequency fluctuation parts to a preset first model to obtain high-frequency prediction results, transmitting the low-frequency trend parts to a preset second model to obtain long-term periodic trends, and generating QoS prediction parameters in a future preset time period according to the high-frequency prediction results and the long-term periodic trends; dynamically optimizing the QoS configuration strategy according to the QoS prediction parameters, and executing the optimized QoS configuration strategy.
[0007] Preferably, QoS parameters are collected in real time, the QoS parameters are compared with QoS prediction parameters, and the QoS prediction parameters are adjusted in real time and / or the parameters of the first model and the second model are optimized according to the comparison results. The present invention also adjusts the QoS parameters according to the current QoS parameter comparison prediction results to achieve closed-loop control by real-time feedback.
[0008] Preferably, user experience data is also collected in real time, and network performance data is correlated with the user experience data for analysis to generate core optimization suggestions, and the resource allocation strategy in the QoS configuration strategy is adjusted according to the core optimization suggestions, and the network performance data includes delay fluctuation information and link load. This solution enables the present invention to drive dynamic resource allocation through user experience, ensure that network configuration is highly matched with user needs, thereby providing high-quality service support for different scenarios, and continuously improving service quality and user experience through closed-loop optimization.
[0009] Preferably, the network fluctuation characteristics are also analyzed based on the network status information, and the frequency of collecting the network status information, QoS parameters and user device behavior data is adjusted based on the network fluctuation characteristics, so that the collection frequency adapts to the current network situation and is adjusted in time without wasting network resources.
[0010] Preferably, the network demand characteristics at peak and valley periods are also identified, and the QoS prediction parameters at peak and valley periods are determined and adjusted according to the network demand characteristics. This solution makes the prediction values of QoS parameters more accurate.
[0011] Preferably, the ARIMA model is used as a time series analysis tool to stratify the current and historical network status information, QoS parameters and user equipment behavior data according to the time series, the first model is an LSMT model, and the second model is a Prophet model. Of course, the time series analysis tool, the first model and the second model are not limited to the above models.
[0012] Preferably, resource allocation reminders are also issued to operators based on the long-term periodic trend, so that the operators can provide sufficient network demand during holidays and the like.
[0013] Preferably, before the current and historical network status information, QoS parameters and user equipment behavior data are stratified according to the time series, the network status information, QoS parameters and user equipment behavior data are preprocessed to obtain sampling data with consistent time series, and redundant features of the sampling data are eliminated according to the relationship analysis, and the key features of the sampling data are screened by using information gain, and the highly correlated features in the key features are aggregated using PCA technology to generate comprehensive features, thereby generating multi-dimensional structured feature data; the NWDAF network element stratifies the multi-dimensional structured feature data according to the time series to separate the high-frequency fluctuation part and the low-frequency trend part, and then transmits the multi-dimensional structured feature data to the corresponding first model and second model for prediction according to its stratification results. This makes the prediction results of the first model and the second model more accurate and prevents prediction interference.
[0014] Specifically, the time sequence synchronization of the sampled data is achieved by using time stamp standardization and interpolation technology to obtain sampled data with consistent time sequence.
[0015] Preferably, the QoS configuration strategy is dynamically optimized according to the QoS prediction parameters, and the execution of the optimized QoS configuration strategy specifically includes: according to the QoS prediction parameters and the current network status information, the priority queue weights of different applications are quickly adjusted, low-latency resources are allocated to high-priority applications, and the resource occupancy of low-priority data flows is reduced, and bandwidth resources are allocated in multiple orbital satellite data transmission services using real-time dynamic broadband aggregation and allocation optimization methods; and the bandwidth occupancy of low-priority traffic is controlled by traffic shaping technology to release resources for preset key services. Priorities are quickly adjusted according to the QoS prediction parameters and the current network status information to quickly achieve QoS optimization.
[0016] The present invention also provides a QoS optimization device based on long-term and short-term prediction, including a base station, a UPF network element, a NWDAF network element and a PCF network element, wherein the base station, the UPF network element, the NWDAF network element and the PCF network element are used to execute the instructions of the QoS optimization method based on long-term and short-term prediction as described above: The base station collects the current network status information and uploads it to the NWDAF network element. The network status information includes link load, signal strength and delay information. The UPF network element collects QoS parameters and records user device behavior data, and transmits it to the NWDAF network element. The QoS parameters include delay, bandwidth utilization and packet loss rate. The user device behavior data includes access frequency, connection duration and data usage. The NWDAF network element stores the network status information, QoS parameters and user device behavior data, and stratifies the current and historical network status information, QoS parameters and user device behavior data according to time series to separate high-frequency fluctuations. part and a low-frequency trend part, the high-frequency fluctuation part is transmitted to a preset first model to obtain a high-frequency prediction result, the low-frequency trend part is transmitted to a preset second model to obtain a long-term periodic trend, and QoS prediction parameters in a future preset time period are generated according to the high-frequency prediction result and the long-term periodic trend, and the QoS prediction parameters are transmitted to the PCF network element; the PCF network element dynamically optimizes the QoS configuration strategy according to the QoS prediction parameters, and generates corresponding adjustment instructions according to the optimized QoS configuration strategy to send to the UPF network element; the UPF network element allocates bandwidth resources according to the adjustment instructions to execute the optimized QoS configuration strategy.
[0017] Compared with the existing technology, the present invention separates and processes the high-frequency dynamics and low-frequency trends in the time series separately by combining the dynamic multi-dimensional QoS demand prediction framework of short-term fluctuation capture and long-term trend prediction, ensuring the accuracy of short-term response and long-term planning, supporting the forward-looking scheduling of network resources, and generating more comprehensive QoS prediction results, so as to achieve better QoS optimization effect. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] Figure 1 The present invention is a flow chart of a QoS optimization method based on long-term and short-term predictions executed by a QoS optimization device based on long-term and short-term predictions.
[0019] Figure 2 This is a partial flow chart of a preferred embodiment of a QoS optimization method based on long-term and short-term predictions executed by a QoS optimization device based on long-term and short-term predictions of the present invention. DETAILED DESCRIPTION
[0020] In order to explain the technical content, structural features, achieved objectives and effects of the present invention in detail, the following is a detailed description in conjunction with the implementation methods and the accompanying drawings.
[0021] refer to Figure 1The present invention discloses a QoS optimization device based on long-term and short-term prediction, including a base station 10, a UPF network element 20, a NWDAF network element 30 and a PCF network element 40. The base station 10, the UPF network element 20, the NWDAF network element 30 and the PCF network element 40 are used to execute instructions of a QoS optimization method based on long-term and short-term prediction.
[0022] refer to Figure 1 The QoS optimization method based on long-term and short-term prediction includes steps S1 to S6.
[0023] S1. The base station 10 collects the current network status information and uploads it to the NWDAF network element 30. The network status information includes link load, signal strength and delay information. The UPF network element 20 collects QoS parameters and records user equipment behavior data, and transmits the QoS parameters and user equipment behavior data to the NWDAF network element 30. The QoS parameters include one or more combinations of delay, bandwidth utilization and packet loss rate, and the user equipment behavior data includes one or more combinations of access frequency, connection duration and data usage.
[0024] The base station 10 collects network status information, including link load, signal strength and delay, through wireless measurement reports, and uploads it to the NWDAF network element 30 for performance evaluation and dynamic optimization analysis of the wireless access layer.
[0025] The UPF network element 20 collects basic QoS parameters (including latency, bandwidth utilization, and packet loss rate) from the network layer, and records user device behavior data (including access frequency, connection duration, and data usage), and transmits the integrated data to the NWDAF network element 30.
[0026] S2. The NWDAF network element 30 stores the network status information, QoS parameters and user equipment behavior data, and stratifies the current and historical network status information, QoS parameters and user equipment behavior data according to the time series to separate the high-frequency fluctuation part and the low-frequency trend part.
[0027] Among them, after receiving the network status information, QoS parameters and user equipment behavior data, the NWDAF network element 30 uses the triple standard deviation method and Isolation Forest algorithm in statistics to remove outliers and noise data, and uses the difference technology to repair missing data to ensure data integrity and accuracy, and unify the data scale through normalization. Specifically, the NWDAF network element 30 performs data alignment and feature processing, realizes the time synchronization of multi-source data through timestamp standardization and interpolation technology, uses downsampling technology for high-frequency data, and interpolates and completes low-frequency data; uses sliding window matching to ensure time alignment. In feature processing, the NWDAF network element 30 also removes redundant features through correlation analysis, uses the information gain method to select the features most relevant to QoS optimization, and aggregates highly correlated features through PCA technology to generate comprehensive features. Finally, the NWDAF network element 30 selects the most critical comprehensive features for analysis and normalizes them, thereby generating multi-dimensional structured feature data. The above can ensure the structural consistency and processing efficiency of multi-dimensional data, and provide high-quality data support for subsequent QoS optimization and prediction.
[0028] The NWDAF network element 30 utilizes the ARIMA model as a time series analysis tool to stratify the current and historical network status information, QoS parameters and user equipment behavior data according to the time series.
[0029] The NWDAF network element 30 also analyzes network fluctuation characteristics based on the network status information, and adjusts the frequency of collecting the network status information, QoS parameters and user equipment behavior data based on the network fluctuation characteristics, so that the collection frequency adapts to the current network situation.
[0030] Specifically, the NWDAF network element 30 dynamically adjusts the sampling frequency according to the network fluctuation characteristics through a multi-source data fusion and cleaning optimization mechanism, increases the sampling frequency in high-volatility scenarios (peaks of bandwidth utilization) to ensure that key changes are captured; and reduces the frequency in stable scenarios (troughs of bandwidth utilization). Reducing the sampling frequency reduces the pressure of collecting redundant data.
[0031] S3. The NWDAF network element 30 transmits the high-frequency fluctuation part to a preset first model to obtain a high-frequency prediction result, and transmits the low-frequency trend part to a preset second model to obtain a long-term periodic trend.
[0032] Specifically, the NWDAF network element 30 removes redundant features of the sampled data based on phase relationship analysis. Among them, after the NWDAF network element 30 uses information gain to screen the key features of the sampled data, and uses PCA technology to aggregate highly correlated features in the key features to generate comprehensive features, thereby generating multidimensional structured feature data, the NWDAF network element 30 stratifies the multidimensional structured feature data according to the time series to separate the high-frequency fluctuation part and the low-frequency trend part, and then transmits the multidimensional structured feature data to the corresponding first model and second model for prediction according to its stratification results, so that the prediction results of the first model and the second model are more accurate and prediction interference is prevented.
[0033] In this embodiment, the first model is an LSMT model, and the second model is a Prophet model. The time series analysis tool, the first model, and the second model are not limited to the above models.
[0034] The NWDAF network element 30 also sends resource allocation reminders to operators based on the long-term periodic trend, so that operators can provide sufficient network demand during holidays and other times. For example, before holidays, link load distribution is actively optimized, key link bandwidth resources are reserved, or redundant transmission capabilities are deployed to ensure the stability and continuity of service quality during peak periods. This mechanism significantly improves the foresight and accuracy of network resource scheduling and enhances the overall efficiency of QoS management.
[0035] Among them, the high-frequency fluctuation part may include burst traffic, delay spikes and other parts with peak bandwidth utilization. The first model (LSMT model) is used to capture the dynamic changes of short-term QoS requirements, especially in high-load scenarios (such as traffic surges). It can quickly respond to changes and generate high-frequency prediction results, providing real-time support for responding to sudden demands.
[0036] The low-frequency trend part is the part where the bandwidth utilization has a periodic and long-term change trend, including the traffic difference between holidays, weekdays and weekends, or the impact of seasonal changes on network load. The second model (Prophet model) focuses on identifying long-term and periodic trends.
[0037] S4. The NWDAF network element 30 generates QoS prediction parameters in a future preset time period according to the high-frequency prediction results and the long-term periodic trend.
[0038] Among them, the prediction result of QoS prediction parameters in the future preset time period is output through a dynamic multi-dimensional QoS demand prediction framework to generate multiple QoS dimension data covering bandwidth requirements, delay fluctuation range and packet loss rate trends in different time periods in the future, and present them in a structured form, which is convenient for core network functional modules (such as PCF network element 40, SMF network element) to call and quickly call and adjust network policies.
[0039] Preferably, the NWDAF network element 30 also identifies the network demand characteristics during peak and trough periods, determines and adjusts the QoS prediction parameters during peak and trough periods based on the network demand characteristics, and generates QoS prediction parameters in a future preset time period based on the high-frequency prediction results, long-term periodic trends, and the QoS prediction parameters during peak and trough periods.
[0040] S5. The PCF network element 40 dynamically optimizes the QoS configuration strategy according to the QoS prediction parameters.
[0041] The PCF network element 40 dynamically optimizes the QoS configuration strategy based on the QoS prediction parameters, specifically including: quickly adjusting the priority queue weights of different applications based on the QoS prediction parameters and the current network status information, allocating low-latency resources to high-priority applications, reducing the resource occupancy of low-priority data flows, and utilizing real-time dynamic broadband aggregation and allocation optimization methods. PCF network element 40 uses a QoS dynamic adjustment mechanism driven by scenario-based requirements. According to the requirements of different business scenarios such as real-time interaction and streaming media, PCF network element quickly adjusts the priority queue weights, allocates low-latency resources to high-priority applications (such as real-time interaction), and reduces the resource usage of low-priority data streams.
[0042] S6. The UPF network element 20 allocates bandwidth resources according to the adjustment instruction to execute the optimized QoS configuration strategy. The UPF network element 20 allocates bandwidth resources in multiple orbital satellite data transmission services; it controls the bandwidth occupancy of low-priority traffic through traffic shaping technology to release resources for preset key services.
[0043] Specifically, PCF network element 40 transmits the specific adjustment instructions corresponding to the QoS configuration strategy to UPF network element 20. UPF network element 20 uses real-time dynamic bandwidth aggregation and allocation optimization methods to flexibly allocate bandwidth resources between LEO, MEO and GEO (multi-orbit satellite data transmission services) links, and controls the bandwidth occupancy of low-priority traffic through traffic shaping technology to release resources for key services. This collaborative mechanism ensures efficient resource allocation and stable network performance.
[0044] The better, reference Figure 2 The UPF network element 20 also collects QoS parameters in real time, and the NWDAF network element 30 compares the QoS parameters with QoS prediction parameters, and adjusts the QoS prediction parameters and / or optimizes the parameters of the first model and the second model in real time according to the comparison results.
[0045] The better, reference Figure 2The UPF network element 20 also collects user experience data in real time and transmits it to the NWDAF network element 30. The NWDAF network element 30 analyzes the user experience data in association with the network performance data, generates a core optimization suggestion and sends it to the PCF network element 40. The PCF network element 40 adjusts the resource allocation strategy in the QoS configuration strategy according to the core optimization suggestion. The network performance data includes delay fluctuation information and link load.
[0046] Specifically, after the NWDAF network element 30 obtains the user experience data, it first performs word segmentation, sentiment analysis and keyword extraction on the subjective evaluation provided by the user in the user experience data (such as "video playback freezes" or "slow connection speed"), and performs correlation analysis with implicit feedback (such as user ratings and complaint frequency). Subsequently, the root cause of the problem is verified by matching network performance data (such as delay fluctuations and link loads). For example, when a user reports video freezes, the NWDAF network element 30 confirms whether it is caused by insufficient resource allocation by correlating bandwidth and link delay data, and generates accurate core optimization suggestions. The PCF network element 40 obtains the above core optimization suggestions and dynamically adjusts the resource allocation strategy based on these suggestions, such as increasing the priority of real-time interactive services or allocating more bandwidth to streaming media services, while reducing the resource occupancy of low-priority data streams. After the adjustment, the NWDAF network element 30 continuously monitors changes in user experience data and verifies the adjustment effect, thereby forming a dynamic closed loop between user experience and QoS optimization, and achieving the accuracy of resource allocation and continuous improvement of service quality. After the adjustment, the NWDAF network element continuously monitors changes in user experience and network performance, and verifies the optimization effect by comparing the service quality differences before and after optimization.
[0047] Among them, NWDAF network element 30 obtains users' subjective feedback data on service quality (including scores and service evaluations) through the application layer interface, and extracts implicit feedback (including abnormal event records and resource usage patterns) in combination with log analysis. UPF (User Plane Function): User plane function in 5G core network, responsible for data forwarding, traffic management and QoS policy execution. NWDAF (Network Data Analytics Function): Network data analysis function in 5G core network, used to collect and analyze network performance and user behavior data, generate predictions and optimization suggestions. QoS (Quality of Service): Quality of service is an indicator to measure network transmission performance, including bandwidth, latency, packet loss rate, etc., used to ensure user experience. PCF (Policy Control Function): Policy control function in 5G core network, responsible for dynamic management of QoS policy and network resources. LEO / MEO / GEO (Low / Middle / Geostationary Earth Orbit): Different orbital satellite types, used to provide coverage and data transmission services. LEO is suitable for low latency scenarios, and GEO is suitable for high bandwidth requirements. LSMT (Long Short-Term Memory): A deep learning algorithm suitable for processing time series data and can capture short-term dynamic changes. Prophet model: A time series analysis tool. PCA (Principal Component Analysis): Principal component analysis, a data dimensionality reduction technique used to extract key features from high-dimensional data. IsolationForest: A machine learning algorithm for outlier detection that identifies noisy data by separating samples. ARIMA (AutoRegressive Integrated Moving Average): A time series prediction model suitable for capturing long-term trends and seasonal characteristics in historical data. Sliding Window Matching: A data alignment method that ensures synchronization of multi-source data in the time dimension by sliding time windows. Dynamic Bandwidth Aggregation: A network optimization technology that improves transmission efficiency by integrating bandwidth resources across multiple track links.
[0048] The above disclosure is only the preferred embodiment of the present invention, which certainly cannot be used to limit the scope of rights of the present invention. Therefore, equivalent changes made according to the scope of the patent application of the present invention are still within the scope covered by the present invention.
Claims
1. A QoS optimization method based on long-term and short-term prediction, Other features include: Obtaining current network status information, wherein the network status information includes link load, signal strength, and delay information; Collecting QoS parameters and recording user device behavior data, wherein the QoS parameters include one or more combinations of latency, bandwidth utilization, and packet loss rate, and the user device behavior data includes one or more combinations of access frequency, connection duration, and data usage; The network status information, QoS parameters and user equipment behavior data are stored, and the current and historical network status information, QoS parameters and user equipment behavior data are layered according to the time series to separate the high-frequency fluctuation part and the low-frequency trend part, and the high-frequency fluctuation part is transmitted to the preset first model to obtain a high-frequency prediction result, and the low-frequency trend part is transmitted to the preset second model to obtain a long-term periodic trend, and the QoS prediction parameters in the future preset time period are generated according to the high-frequency prediction result and the long-term periodic trend; The QoS configuration strategy is dynamically optimized according to the QoS prediction parameters, and the optimized QoS configuration strategy is executed.
2. The QoS optimization method based on long-term and short-term predictions as described in claim 1, further characterized by: collecting QoS parameters in real time, comparing the QoS parameters with QoS prediction parameters, and adjusting the QoS prediction parameters and / or optimizing the parameters of the first model and the second model in real time according to the comparison results.
3. The QoS optimization method based on long-term and short-term prediction as described in claim 1 has other characteristics: user experience data is also collected in real time, network performance data is associated with the user experience data for analysis, core optimization suggestions are generated, and the resource allocation strategy in the QoS configuration strategy is adjusted according to the core optimization suggestions, and the network performance data includes delay fluctuation information and link load.
4. The QoS optimization method based on long-term and short-term prediction as described in claim 1, further characterized in that: the network fluctuation characteristics are also analyzed based on the network status information, and the frequency of collecting the network status information, QoS parameters and user device behavior data is adjusted based on the network fluctuation characteristics.
5. The QoS optimization method based on long-term and short-term prediction as claimed in claim 1, further characterized by: The network demand characteristics during peak and valley periods are also identified, and the QoS prediction parameters during peak and valley periods are determined and adjusted according to the network demand characteristics.
6. The QoS optimization method based on long-term and short-term prediction as described in claim 1 is further characterized in that: the ARIMA model is used as a time series analysis tool to stratify the current and historical network status information, QoS parameters and user equipment behavior data according to the time series, the first model is the LSMT model, and the second model is the Prophet model.
7. The QoS optimization method based on long-term and short-term prediction as described in claim 1, further characterized in that: resource allocation reminders are also issued to operators based on the long-term periodic trend.
8. The QoS optimization method based on long-term and short-term prediction as claimed in claim 1, further characterized by: Before stratifying the current and historical network status information, QoS parameters and user device behavior data according to the time series, the network status information, QoS parameters and user device behavior data are also preprocessed to obtain sampling data with consistent time series, and redundant features of the sampling data are eliminated based on correlation analysis, and key features of the sampling data are screened using information gain, and highly correlated features in the key features are aggregated using PCA technology to generate comprehensive features, thereby generating multi-dimensional structured feature data; stratifying the current and historical network status information, QoS parameters and user device behavior data according to the time series is specifically as follows: stratifying the multi-dimensional structured feature data according to the time series to separate the high-frequency fluctuation part and the low-frequency trend part.
9. The QoS optimization method based on long-term and short-term prediction as claimed in claim 1, further characterized by: Dynamically optimize the QoS configuration strategy based on the QoS prediction parameters, and execute the optimized QoS configuration strategy specifically including: based on the QoS prediction parameters and the current network status information, quickly adjust the priority queue weights of different applications, allocate low-latency resources for high-priority applications, reduce the resource occupancy of low-priority data flows, and use real-time dynamic broadband aggregation and allocation optimization methods to allocate bandwidth resources in multiple orbital satellite data transmission services; control the bandwidth occupancy of low-priority traffic through traffic shaping technology to release resources for preset key services.
10. A QoS optimization device based on long-term and short-term prediction, characterized in that: It includes a base station, a UPF network element, a NWDAF network element and a PCF network element, and the base station, the UPF network element, the NWDAF network element and the PCF network element are used to execute the instructions of the QoS optimization method based on long-term and short-term prediction according to any one of claims 1-10: The base station collects current network status information and uploads it to the NWDAF network element, the network status information including link load, signal strength and delay information; The UPF network element collects QoS parameters and records user device behavior data, and transmits it to the NWDAF network element. The QoS parameters include delay, bandwidth utilization and packet loss rate. The user device behavior data includes access frequency, connection duration and data usage. The NWDAF network element stores the network status information, QoS parameters and user equipment behavior data, and stratifies the current and historical network status information, QoS parameters and user equipment behavior data according to the time series to separate the high-frequency fluctuation part and the low-frequency trend part, transmits the high-frequency fluctuation part to a preset first model to obtain a high-frequency prediction result, transmits the low-frequency trend part to a preset second model to obtain a long-term periodic trend, generates QoS prediction parameters in a future preset time period according to the high-frequency prediction result and the long-term periodic trend, and transmits the QoS prediction parameters to the PCF network element; The PCF network element dynamically optimizes the QoS configuration strategy according to the QoS prediction parameters, and generates corresponding adjustment instructions according to the optimized QoS configuration strategy to send to the UPF network element; The UPF network element allocates bandwidth resources according to the adjustment instruction to execute the optimized QoS configuration strategy.