CDN multi-source station

Abstract

The invention, which belongs to the CDN field, discloses a CDN multi-source station comprising a line detection module, an automatic switching module, a manual selection module, a line cache module and a priority management module. Through a time sequence algorithm, a clustering algorithm and a collaborative filtering algorithm, the line detection module monitors network performance in real time and predicts a fluctuation trend, and a switching decision is provided for the automatic switching module; and the automatic switching module is combined with an intelligent cache preloading mechanism to load the line content to be switched in advance so as to improve the switching efficiency. The manual selection module is combined with the priority management module, and the line priority is dynamically adjusted and the user experience is optimized by analyzing multi-dimensional data such as user behaviors, real-time network environments and equipment states. The line caching and synchronization module utilizes a blockchain-based content verification and synchronization mechanism to ensure that the content updating process of the CDN node is transparent and cannot be tampered, and the consistency is ensured. The method has the beneficial effects that the real-time response, personalized selection and content security of the system are improved through intelligent prediction, priority dynamics and the like.

Description

Technical Field

[0001] The present invention relates to the CDN field, and more specifically, to a CDN multi-source station. Background Art

[0002] CDN multi-source station technology improves the flexibility and stability of content distribution by supporting multiple source servers, which plays a key role in dealing with high concurrent traffic. Traditional CDN architecture usually relies on a single source station. If the source station fails or the traffic is too high, it may cause content distribution interruption or delay, affecting user experience. Multi-source station technology configures multiple source servers, allowing CDN to automatically switch to the best source station based on traffic demand, server status, etc., thereby ensuring continuous and efficient content distribution, greatly enhancing the reliability and fault tolerance of the system. The advantage of this technology is that it not only improves access speed and content availability, but also prevents single point failures.

[0003] However, its biggest drawback is the complexity of management and cost: configuring and maintaining multiple source stations requires coordinating different source station versions, content consistency, and cache management, which increases technical difficulty and operation and maintenance costs. In addition, content synchronization of different source stations may cause version lag, affecting the user's real-time experience. Summary of the invention

[0004] The purpose of the present invention is to provide a CDN multi-source station to solve the problems raised in the above background technology: configuring and maintaining multi-source stations requires coordinating different source station versions, content consistency and cache management, etc., which increases technical difficulty and operation and maintenance costs. In addition, content synchronization of different source stations may cause version lag, affecting the real-time experience of users.

[0005] Technical solution: A CDN multi-source station, the CDN multi-source station includes a line detection module, an automatic switching module, a manual selection module, a line cache module and a priority management module; The line detection module and the automatic switching module introduce time series algorithms, clustering algorithms and collaborative filtering algorithms; time series algorithms (such as ARIMA, LSTM, etc.) are mainly used to predict future network performance trends based on historical data. Specifically, the line detection module collects historical network data (such as delay, bandwidth, packet loss rate, etc.), converts these data into time series, and uses deep learning models such as **autoregressive model (ARIMA) or long short-term memory network (LSTM)** for training. These models can identify long-term trends and short-term fluctuations in network performance, and then predict the performance changes of each CDN line in the next few minutes or hours.

[0006] Clustering algorithms (such as K-means, DBSCAN, etc.) are mainly used to group different CDN lines to simplify the decision-making process. The line detection module can use clustering algorithms to cluster lines based on multi-dimensional performance indicators such as geographical location, latency, and bandwidth of each line. After clustering, lines with similar performance are divided into the same group, which can play an important role in subsequent decision-making.

[0007] Collaborative filtering algorithms are often used in personalized recommendation systems. In the route detection module, collaborative filtering algorithms can predict and recommend the most suitable CDN routes based on the user's historical behavior and the network experience of other similar users. The algorithm analyzes the user's behavior patterns at a specific time, location, and network environment, and combines other user data in big data to infer the most suitable route selection for the current user.

[0008] Based on machine learning and big data analysis, the line detection module monitors performance in real time and predicts network fluctuation trends, providing switching decisions for the automatic switching module; the automatic switching module loads the content of the line to be switched in advance by combining with the intelligent cache preloading mechanism; The manual selection module and the priority management module adjust the priority strategy in real time through user behavior analysis. The priority management module dynamically adjusts the line priority by considering historical selections and analyzing the user's real-time network environment and device status multi-dimensional data; The line cache and synchronization module is responsible for the consistency of content, introduces a blockchain-based content verification and synchronization mechanism, and ensures the transparency and non-tamperability of the content update and transmission process on each CDN node through blockchain technology; Preferably, the line detection module detects and evaluates the connection performance of each CDN source station line in real time, and provides the best line selection basis; By regularly performing network detection on the client, using Ping, TCP handshake, and HTTP request response time methods, the latency, packet loss rate, bandwidth, and response time performance indicators of different CDN lines are evaluated; based on the evaluated data, the comprehensive performance score of each line is calculated, and the line priority is dynamically adjusted; the Ping test is used to measure latency, and the HTTP request response time is used to detect the processing speed of the line; the measured performance data is fed back to the client.

[0009] Preferably, the line detection module uses the time series algorithm, and the time series prediction formula based on the ARIMA model is as follows: = + + in, is the network performance delay and bandwidth at the current moment; μ is the average value of the time series; is the parameter of the AR (autoregressive) model, indicating the dependence on the previous moments; is the parameter of the MA (moving average) model, which represents the dependence on the error terms of previous periods; is the error term, reflecting the residual error of the model.

[0010] Preferably, the long short-term memory network calculates the network performance through the following activation function: = . + . + ) = . + .tanh( . + ) in, is the prediction result of the hidden state network performance of the long short-term memory network unit; It is the memory state of the LSTM network, capturing long-term dependencies; , , is the weight matrix, is bias; , They are the input gate and forget gate, which are used to control the information flow.

[0011] Preferably, when the automatic switching module detects that the current line is unavailable and the quality is degraded, it automatically switches to the next CDN line with a higher priority; By continuously monitoring the playback status and network quality, if the current line is found to respond slowly or errors occur, including timeout and disconnection, it will automatically switch to the next line with better performance according to the line priority list; call the real-time performance data of the line detection module to determine whether it is necessary to switch the line; when the switch occurs, the client pauses the current video stream and re-requests a new line data stream to ensure uninterrupted playback; based on dynamic evaluation of network quality, through error detection and priority adjustment mechanisms, the client is always connected to the available and best performing CDN line.

[0012] Preferably, based on the predicted network performance and user needs, the optimal CDN line switching decision formula is as follows: Automatic switching decisions based on predicted line performance , switching is performed taking into account latency, bandwidth, and user needs: + ) in, , , are the predicted delay, bandwidth, and packet loss rate; and is a weight coefficient used to measure the impact of each performance indicator on the switching decision; by minimizing the weighted indicator value, the automatic switching module selects the optimal CDN line.

[0013] Preferably, the manual selection module provides a function for the user to manually select a CDN line, and when the automatic selection fails to meet the needs, the user switches the line by himself; The manual selection module displays a line selection menu in the player UI, listing all available CDN lines and real-time performance data including latency and bandwidth; users view the status of each line and manually select the best line for playback; if the user selects a new line, the player will re-request the content stream of the line and start playing; the line priority selected by the user will be temporarily saved, and if the user manually switches too much, the system will adjust the priority based on historical selections.

[0014] Preferably, the line cache and synchronization module keeps the content on all CDN nodes consistent and synchronizes new content to all line nodes in a timely manner; The line cache and synchronization module manages the content cache of each source station through a distributed cache system; using incremental update technology, when the content of the source station changes, only the changed part is updated to the cache node; each node is synchronized with the source station regularly; during synchronization, a hash comparison and verification mechanism is used; the monitoring system checks the synchronization progress so that the content is updated to all CDN nodes in a timely manner.

[0015] Preferably, the priority management module dynamically adjusts the priority of the CDN line according to the user network environment and line performance; Dynamic priority is assigned to each CDN line based on the user's IP geographic location, operator information, and historical connection data. The system selects the best CDN node based on the user's operator and network latency, and adjusts line priority based on the user's historical usage. If a line has performed well in history, the best-performing route will be assigned a higher priority. Priority adjustment is based on the client's real-time network status.

[0016] Preferably, the formula for dynamically adjusting the priority of CDN lines based on user demand and network environment is as follows: The priority management module can calculate the priority of each line based on the user's historical selection and network conditions : = + in, is the historical selection frequency and score of line k; is the predicted performance delay and bandwidth of line k; and is the weight coefficient, which is used to adjust the influence of historical selection frequency on prediction performance.

[0017] Compared with the prior art, the advantages of the present invention are: (1) The CDN multi-source station designed in this paper predicts the future performance of each line through time series prediction models (such as ARIMA or LSTM), thereby achieving more forward-looking line switching decisions. Compared with the traditional line selection method that relies on simple rules, this prediction mechanism can detect potential line degradation problems in advance, effectively reduce delays and fluctuations, and significantly improve the system's response speed and service stability.

[0018] (2) By using clustering algorithms (such as K-means), the CDN system can dynamically group lines with similar network performance, simplify switching decisions, and enable line selection to adapt to changes in user geographic location and network environment. Especially during peak hours, the optimal line group can be given priority to alleviate load pressure. This method is more flexible than the traditional method of equally selecting all CDN lines.

[0019] (3) Compared with the traditional method that relies on network performance indicators, this system analyzes user historical behavior data through a collaborative filtering algorithm and recommends the optimal route based on user preferences, device type, and network environment, making route switching more personalized, improving user satisfaction, and reducing the unstable experience caused by frequent switching.

[0020] (4) The CDN multi-source station system designed in this paper introduces machine learning and big data analysis, and continuously optimizes the line selection and switching decisions based on the latest network feedback data, enhancing the CDN system's adaptability and fault tolerance when the network environment fluctuates. It can adjust more dynamically than traditional fixed strategies, improving the efficiency of traffic load management and service quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a schematic diagram of an overall system of a CDN multi-source station of the present invention; DETAILED DESCRIPTION

[0022] For examples, see Figure 1 ,A CDN multi-source station,A CDN multi-source station includes a line detection module, an automatic switching module, a manual selection module, a line cache module and a priority management module; The line detection module and the automatic switching module introduce time series algorithms, clustering algorithms, and collaborative filtering algorithms. Based on machine learning and big data analysis, the line detection module monitors performance in real time and predicts network fluctuation trends, providing switching decisions for the automatic switching module. The automatic switching module combines with the intelligent cache preloading mechanism to load the content of the line to be switched in advance. Specifically, whenever the time series algorithm detects fluctuations in network performance, it analyzes these fluctuations and outputs a prediction value to help the automatic switching module determine in advance whether the performance of a certain line will decline in the short term. For example, if it is predicted that the delay of a certain line will increase significantly in the next few minutes, the automatic switching module can switch to a better performing line in advance to avoid affecting the user experience.

[0023] Specifically, when the system detects changes in the network environment, the clustering algorithm will classify lines with similar performance. For example, based on the user's location and network environment, the system will automatically select the line group that best matches the current user's network conditions. When there is high load or high traffic, the system can identify the line group that may have bottlenecks in advance and adjust traffic distribution or switch lines in advance.

[0024] Specifically, by analyzing the data and selection history of a large number of users, the collaborative filtering algorithm can identify which CDN routes perform better under similar network conditions. For example, when users in a certain area select a specific route many times, the system will learn that the route has better adaptability to the network environment in that area, and will automatically recommend the route to other users in similar environments in the future.

[0025] Specifically, the clustering algorithm calculation formula is as follows; The goal of the K-means algorithm is to cluster by minimizing the distance from the sample point to the cluster center and defining the cluster center for: = in, is the center of the kth cluster; is the set of sample points belonging to the kth cluster; is the i-th data point belonging to cluster k (performance indicator of a certain line) Each sample point Will be assigned to the closest cluster center ; arg ( ) in, It is the Euclidean distance from the sample point to the cluster center, indicating the similarity of network performance.

[0026] The manual selection module and the priority management module adjust the priority strategy in real time through user behavior analysis. The priority management module considers historical selections and dynamically adjusts line priorities by analyzing the user's real-time network environment and device status multi-dimensional data; The line cache and synchronization module is responsible for content consistency and introduces a blockchain-based content verification and synchronization mechanism. The blockchain technology is used to ensure the transparency and immutability of the content update and transmission process on each CDN node. Specifically, the collaborative filtering algorithm predicts the CDN route that users are most likely to choose based on their historical behaviors, thus improving the personalized recommendation effect; Based on the similarity calculation between users and lines, cosine similarity can be used: S(u,v)= Among them, S(u,v) is the similarity between user u and user v; and are the ratings or behavior data of user u and user v on route i, such as the selected route and delay experience; I is the set of routes rated by all users.

[0027] Specifically, the collaborative filtering algorithm is used to predict the preference of user u for the unselected route j, which can be expressed as a weighted average: + is the predicted rating of user u for route j; is a set of users similar to user u; Let user v year old Enelu J's rating.

[0028] The line detection module detects and evaluates the connection performance of each CDN source station line in real time, providing the best basis for line selection; By regularly performing network detection on the client side, using Ping, TCP handshake, and HTTP request response time methods, the latency, packet loss rate, bandwidth, and response time performance indicators of different CDN lines are evaluated. Based on the evaluation data, the comprehensive performance score of each line is calculated, and the line priority is dynamically adjusted. Ping tests are used to measure latency, and HTTP request response time is used to detect the processing speed of the line. The measured performance data is fed back to the client.

[0029] The line detection module uses a time series algorithm. The time series prediction formula based on the ARIMA model is as follows: = + + in, is the network performance delay and bandwidth at the current moment; μ is the average value of the time series; is the parameter of the AR (autoregressive) model, indicating the dependence on the previous moments; is the parameter of the MA (moving average) model, which represents the dependence on the error terms of previous periods; is the error term, reflecting the residual error of the model.

[0030] The LSTM network calculates network performance using the following activation function: = . + . + ) = . + .tanh( . + ) in, is the prediction result of the hidden state network performance of the long short-term memory network unit; It is the memory state of the LSTM network, capturing long-term dependencies; , , is the weight matrix, is bias; , They are the input gate and forget gate, which are used to control the information flow.

[0031] When the automatic switching module detects that the current line is unavailable and the quality is degraded, it automatically switches to the next CDN line with higher priority; By continuously monitoring the playback status and network quality, if the current line is found to be responding slowly or errors occur, including timeout and disconnection, it will automatically switch to the next line with better performance according to the line priority list; call the real-time performance data of the line detection module to determine whether the line needs to be switched; when the switch occurs, the client pauses the current video stream and re-requests a new line data stream to ensure uninterrupted playback; based on dynamic evaluation of network quality, through error detection and priority adjustment mechanisms, the client is always connected to the available and best-performing CDN line.

[0032] Based on the predicted network performance and user needs, the optimal CDN line switching decision formula is as follows: Automatic switching decisions based on predicted line performance , switching is performed taking into account latency, bandwidth, and user needs: + ) in, , , is the predicted delay, bandwidth, and packet loss rate; and is a weight coefficient used to measure the impact of each performance indicator on the switching decision; by minimizing the weighted indicator value, the automatic switching module selects the optimal CDN line.

[0033] The manual selection module allows users to manually select CDN lines. When the automatic selection fails to meet the needs, users can switch lines by themselves. The manual selection module displays a line selection menu in the player UI, listing all available CDN lines and real-time performance data including latency and bandwidth; users view the status of each line and manually select the best line for playback; if the user selects a new line, the player will re-request the content stream of the line and start playing; the line priority selected by the user will be temporarily saved, and if the user manually switches too much, the system will adjust the priority based on historical selections.

[0034] The line cache and synchronization module keeps the content on all CDN nodes consistent and synchronizes new content to all line nodes in a timely manner; The line cache and synchronization module manages the content cache of each source station through a distributed cache system; using incremental update technology, when the content of the source station changes, only the changed part is updated to the cache node; each node is synchronized with the source station regularly; during synchronization, a hash comparison and verification mechanism is used; the monitoring system checks the synchronization progress so that the content is updated to all CDN nodes in a timely manner.

[0035] Specifically, incremental update technology means that when the source site content changes, only the changed part is updated to the CDN cache node, rather than the entire content is repeatedly transmitted. The main steps of this technology include content change detection, differential data extraction and transmission optimization: Content change detection: When the source site content changes, the system quickly detects the changed parts of the content through hash value comparison of files or databases, timestamp checks, etc., to ensure accurate identification of the scope of the updated content. This step avoids the redundancy of repeated transmission and improves the efficiency of content updates.

[0036] Differential data extraction: After change detection, the incremental update technology extracts the changed data differences, and only transmits the different parts between the source station and the cache node through binary difference algorithms (such as rsync) or HTTP-based Range requests. This method reduces bandwidth usage and system overhead, and effectively improves content transmission speed.

[0037] Node cache update and synchronization: The cache node updates the existing cache content based on the received differential data to ensure data consistency. Each node will perform regular synchronization checks with the source station to ensure that all node content can be synchronized to the latest version within a certain period of time after the source station is updated. Incremental update technology not only reduces bandwidth usage, but also avoids the load pressure caused by full updates, allowing the CDN system to respond quickly and maintain a high quality of service when the content changes frequently.

[0038] Specifically, in the incremental update technology, the hash comparison and verification mechanism is used to efficiently detect content changes and ensure data integrity. The main process of this mechanism is as follows: Hash comparison: When the source site content is updated, the system will generate a new hash value (such as MD5 or SHA-256) for the content block or file and compare it with the hash value on the CDN node. By comparing the hash value of each content block, the system can quickly find the difference, avoiding byte-by-byte scanning of the entire file, greatly improving the efficiency of content change detection. Only content blocks with different hash values ​​will be marked as needing to be updated, thereby reducing the amount of data transmission.

[0039] Checksum mechanism: During data transmission and update, the checksum mechanism ensures content integrity. Each updated content block will generate a new checksum after transmission. The CDN node will compare the checksum with the source station to confirm whether the data has errors or damage during transmission, ensuring the accuracy of the cached content. This mechanism prevents incomplete or inconsistent data while efficiently detecting errors.

[0040] Synchronization progress monitoring system: In order to ensure that all CDN nodes obtain the latest content in a timely manner, the system is equipped with a synchronization progress monitoring module. This monitoring module continuously tracks the update status of each node and records whether the content is successfully synchronized to each node. Through the automatic alarm and status feedback mechanism, once a synchronization progress delay or data inconsistency is detected, the monitoring system will notify the maintenance personnel in real time, or automatically trigger retry and repair operations to ensure that the content update is not missed and is distributed to all CDN nodes in a timely manner.

[0041] The priority management module dynamically adjusts the priority of CDN lines according to the user's network environment and line performance; Dynamic priority is assigned to each CDN line based on the user's IP geographic location, operator information, and historical connection data. The system selects the best CDN node based on the user's operator and network latency, and adjusts line priority based on the user's historical usage. If a line has performed well in history, the best-performing line will be assigned a higher priority. Priority adjustment is based on the client's real-time network status.

[0042] The formula for dynamically adjusting the priority of CDN lines based on user needs and network environment is as follows: The priority management module can calculate the priority of each line based on user historical selection and network conditions. : = + in, is the historical selection frequency and score of line k; is the predicted performance delay and bandwidth of line k; and is the weight coefficient, which is used to adjust the influence of historical selection frequency on prediction performance.

[0043] The above shows and describes the basic principles, main features and advantages of the present invention; those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only preferred examples of the present invention and are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected; the scope of protection claimed by the present invention is defined by the attached claims and their equivalents.

Claims

1. A CDN multi-source station, characterized in that: The CDN multi-source station includes a line detection module, an automatic switching module, a manual selection module, a line cache module and a priority management module; The line detection module and the automatic switching module introduce time series algorithms, clustering algorithms and collaborative filtering algorithms; based on machine learning and big data analysis, the line detection module monitors performance in real time and predicts network fluctuation trends, providing switching decisions for the automatic switching module; the automatic switching module loads the content of the line to be switched in advance by combining with the intelligent cache preloading mechanism; The manual selection module and the priority management module adjust the priority strategy in real time through user behavior analysis. The priority management module dynamically adjusts the line priority by considering historical selections and analyzing the user's real-time network environment and device status multi-dimensional data; The line cache and synchronization module is responsible for the consistency of the content, introduces a blockchain-based content verification and synchronization mechanism, and ensures the transparency and non-tamperability of the content update and transmission process on each CDN node through blockchain technology.

2. According to claim 1, a CDN multi-source station is characterized in that: The line detection module detects and evaluates the connection performance of each CDN source station line in real time, and provides the best line selection basis; By regularly performing network detection on the client side, using Ping, TCP handshake, and HTTP request response time methods, the latency, packet loss rate, bandwidth, and response time performance indicators of different CDN lines are evaluated; based on the evaluation data, the comprehensive performance score of each line is calculated, and the line priority is dynamically adjusted; The Ping test is used to measure the delay, and the HTTP request response time is used to detect the processing speed of the line; The measured performance data is fed back to the client.

3. According to claim 1, a CDN multi-source station is characterized in that: The line detection module uses the time series algorithm, and the time series prediction formula based on the ARIMA model is as follows: = + + in, is the network performance delay and bandwidth at the current moment; μ is the average value of the time series; is the parameter of the AR (autoregressive) model, indicating the dependence on the previous moments; is the parameter of the MA (moving average) model, which represents the dependence on the error terms of previous periods; is the error term, reflecting the residual error of the model.

4. According to claim 3, a CDN multi-source station is characterized in that: The LSTM network calculates network performance using the following activation function: = . + . + ) = . + . fishy( . + ) in, is the prediction result of the hidden state network performance of the long short-term memory network unit; It is the memory state of the LSTM network, capturing long-term dependencies; , , is the weight matrix, is bias; , They are the input gate and forget gate, which are used to control the information flow.

5. The CDN multi-source station according to claim 1, characterized in that: When the automatic switching module detects that the current line is unavailable and the quality is degraded, it automatically switches to the next CDN line with higher priority; By continuously monitoring the playback status and network quality, if the current line is found to be responding slowly or having errors, including timeouts and disconnections, it will automatically switch to the next line with better performance according to the line priority list; The real-time performance data of the line detection module is called to determine whether the line needs to be switched; when the switch occurs, the client pauses the current video stream and re-requests a new line data stream so that the playback is not interrupted; Based on dynamic evaluation of network quality, through error detection and priority adjustment mechanisms, clients are always connected to the available and best performing CDN line.

6. A CDN multi-source station according to claim 5, characterized in that: Based on the predicted network performance and user needs, the optimal CDN line switching decision formula is as follows: Automatic switching decisions based on predicted line performance , switching is performed taking into account latency, bandwidth, and user needs: + ) in, , , is the predicted delay, bandwidth, and packet loss rate; and is a weight coefficient used to measure the impact of each performance indicator on the switching decision; by minimizing the weighted indicator value, the automatic switching module selects the optimal CDN line.

7. The CDN multi-source station according to claim 1, characterized in that: The manual selection module provides the user with the function of manually selecting a CDN line. When the automatic selection fails to meet the needs, the user switches the line by himself. The manual selection module displays a line selection menu in the player UI, listing all available CDN lines and real-time performance data including latency and bandwidth; The user checks the status of each line and manually selects the best line for playback; If the user selects a new line, the player will re-request the content stream of the line and start playing; the line priority selected by the user will be temporarily saved, and if the user manually switches too much, the system will adjust the priority based on historical selections.

8. The CDN multi-source station according to claim 1, characterized in that: The line cache and synchronization module keeps the content on all CDN nodes consistent and synchronizes new content to all line nodes in a timely manner; The line cache and synchronization module manages the content cache of each source station through a distributed cache system; Using incremental update technology, when the source site content changes, only the changed part is updated to the cache node; each node is synchronized with the source site regularly; during synchronization, a hash comparison and verification mechanism is used; The monitoring system checks the synchronization progress and ensures that the content is updated to all CDN nodes in a timely manner.

9. The CDN multi-source station according to claim 1, characterized in that: The priority management module dynamically adjusts the priority of the CDN line according to the user network environment and line performance; Combine the user's IP location, operator information, and historical connection data to assign dynamic priority to each CDN line; The system selects the best CDN node based on the user's carrier and network latency, and adjusts line priority based on the user's historical usage; If a line has performed well historically, the best performing line will be assigned a higher priority; the priority is adjusted based on the client's real-time network status.

10. The CDN multi-source station according to claim 1, characterized in that: The formula for dynamically adjusting the priority of CDN lines based on user needs and network environment is as follows: The priority management module can calculate the priority of each line based on the user's historical selection and network conditions : = + in, is the historical selection frequency and score of line k; is the predicted performance delay and bandwidth of line k; and is the weight coefficient, which is used to adjust the influence of historical selection frequency on prediction performance.

Images