Adaptive traffic distribution method and system based on clustering and quantum computing

Abstract

The present invention provides an adaptive traffic distribution method and system based on clustering and quantum computing, which relates to the field of network technology, including collecting network behavior data, constructing sample clusters, taking values ​​of the total number of clusters corresponding to the sample clusters for multiple times, and obtaining the optimal total number of clusters and initial clustering results based on a similarity profile method; determining the average position of all sample points for each cluster, updating the cluster center, reallocating all sample points, obtaining new clustering results, and repeating iterations to obtain network behavior clusters; determining network behavior characteristics, defining a comprehensive objective function, mapping decision variables into quantum states through binary coding, constructing an initial quantum population, performing quantum crossover and quantum mutation on the quantum states, iteratively updating the quantum population, and realizing adaptive traffic distribution in combination with the calculation results of the comprehensive objective function.

Description

Technical Field

[0001] The present invention relates to the field of network technology, and in particular to an adaptive traffic distribution method and system based on clustering and quantum computing. Background Art

[0002] With the rapid development of the Internet, Internet applications are everywhere in our lives. With the increasing number of users and the continuous growth of network data demand, the number of application visits has increased sharply. A large amount of request traffic has caused serious application load. In severe cases, the application server has crashed and cannot provide services to the outside world, which has a significant impact on user experience.

[0003] CN202110439567.9 discloses a traffic distribution method based on HTTP request content, including: obtaining HTTP requests; analyzing and aggregating the content of the HTTP requests based on a preset content analysis engine; forwarding the HTTP requests to the corresponding backend processing program according to the preset forwarding logic based on the preset content analysis engine. The present invention adopts the above technical solution, and by introducing a content analysis engine implemented by advanced conditional expressions, it can analyze the HTTP request content, and then realize traffic distribution according to business scenarios and risk control strategies;

[0004] To sum up, the existing technologies are all targeted at their own business scenarios, and different traffic distribution strategies are planned in advance, and fixed distribution strategies are used to adapt to actual distribution scenarios. They are not adaptable enough to changes and cannot expand the scope of application. Therefore, an adaptive traffic distribution method is needed to dynamically adjust the distribution strategy. The application of the present invention can at least solve some of the problems of the existing technology. Summary of the invention

[0005] The embodiments of the present invention provide an adaptive traffic distribution method and system based on clustering and quantum computing, which can at least solve some problems in the prior art.

[0006] According to a first aspect of the embodiments of the present invention,

[0007] An adaptive traffic distribution method based on clustering and quantum computing is provided, comprising:

[0008] Collect network behavior data, use the network behavior data as sample points, construct sample clusters, take values ​​of the total number of clusters corresponding to the sample clusters multiple times according to a preset number of iterations, determine the maximum value of the average silhouette coefficient of the sample points based on the similarity silhouette method, and obtain the optimal total number of clusters and the initial clustering result corresponding to the optimal total number of clusters;

[0009] Based on the initial clustering result, for each cluster, determine the average position of all sample points in the cluster, update the cluster center according to the average position, redistribute all sample points according to the principle of proximity, obtain a new clustering result, repeat the iteration until the cluster center no longer changes, and obtain the network behavior clustering;

[0010] Based on the network behavior clustering, the network behavior characteristics are determined, and a comprehensive objective function is defined according to the network behavior characteristics. The decision variables corresponding to the comprehensive objective function are mapped into corresponding quantum states through binary coding, and an initial quantum population is constructed. The quantum states in the initial quantum population are subjected to quantum crossover and quantum mutation, and the quantum population is iteratively updated. Combined with the calculation result of the comprehensive objective function, the optimal traffic distribution strategy is obtained to achieve adaptive traffic distribution.

[0011] In an optional embodiment,

[0012] The total number of clusters corresponding to the sample clustering is taken multiple times according to a preset number of iterations, and the maximum value of the average silhouette coefficient of the sample points is determined based on the similarity silhouette method, and the optimal total number of clusters is obtained, including:

[0013] A value is randomly selected as the total number of clusters. Based on the total number of clusters, a cluster center selection algorithm is used to iteratively calculate the probability of each sample point being selected as a cluster center, initialize the cluster center, and assign all sample points to the corresponding cluster center according to the principle of proximity to obtain an initial clustering result.

[0014] Based on the intra-class difference and inter-class similarity of each sample point, a silhouette coefficient of each sample point is calculated, and the average silhouette coefficient of all the sample points in the total number of clusters is counted;

[0015] The similarity profile method is repeated until a preset number of iterations is met, and the maximum value of the average profile coefficient is selected. The total number of clusters corresponding to the maximum value is taken as the optimal total number of clusters.

[0016] In an optional embodiment,

[0017] Based on the intra-class difference and inter-class similarity of each sample point, the silhouette coefficient of each sample point is calculated, and the average silhouette coefficient of all the sample points in the total number of clusters is counted, including:

[0018] The intra-class difference of each sample point is calculated, and the formula is as follows:

[0019]

[0020] Where x represents a sample point, y represents another sample point, a(x) represents the difference between sample point x and other sample points in the same cluster, and Lx represents all sample points that are in the same cluster as the sample point x, |L x | represents the total number of all sample points in the same cluster as sample point x, and d(x,y) represents the distance between sample point x and sample point y;

[0021] The inter-class similarity of each sample point is calculated, and the formula is as follows:

[0022]

[0023] Among them, b(x) represents the similarity between the sample point x and the sample point of the nearest cluster, z represents a sample point, and Lz represents the number of samples in addition to L x For clusters other than , d(x,z) represents the distance between sample point x and sample point z;

[0024] The silhouette coefficient of each sample point is calculated using the following formula:

[0025]

[0026] Among them, s(x) represents the silhouette coefficient of sample point x.

[0027] In an optional embodiment,

[0028] Based on the total number of clusters, a cluster center selection algorithm is used to iteratively calculate the probability of each sample point being selected as a cluster center. Initializing the cluster center includes:

[0029] Set a cluster center set, randomly select a sample point, add it to the cluster center set, and the remaining sample points constitute the candidate set, where each sample point in the candidate set is used as a candidate point, and each sample point in the cluster center set is used as a center point, and the cluster center selection algorithm iteration begins:

[0030] The probability of selection is calculated for the selected point:

[0031] Select a candidate point as the current candidate point, calculate the distance between the current candidate point and each center point in the cluster center set, determine the minimum distance value corresponding to the current candidate point, and calculate the probability of the current candidate point being selected as the center point based on the minimum distance value;

[0032] Until the probability of selection of each candidate point is obtained, the selection probability calculation is stopped, the candidate point corresponding to the maximum selection probability is selected as the optimal candidate point, the optimal candidate point is removed from the candidate set, and the optimal candidate point is added to the cluster center set;

[0033] Until the total number of elements in the cluster center set is equal to the total number of clusters, the iteration of the cluster center selection algorithm is stopped to complete the initialization of the cluster centers.

[0034] In an optional embodiment,

[0035] Performing a selection probability calculation on the candidate point also includes:

[0036] Calculate the distance between each of the candidate points and each of the center points, where the candidate points and the center point are both sample points, containing n features, and the formula is as follows:

[0037]

[0038] Among them, dist 2 (p,c) represents the square of the Euclidean distance between the candidate point p and the center point c, i represents the feature index, n represents the total number of features of the sample point, p i represents the i-th feature of the point to be selected, c i Represents the i-th feature of the center point;

[0039] The probability of the selected point being selected is calculated, and the formula is as follows:

[0040]

[0041] Among them, prob(p) represents the probability of the selected point p, c n Represents the center point closest to the selected point p.

[0042] In an optional embodiment,

[0043] Based on the network behavior clustering, the network behavior characteristics are determined, and a comprehensive objective function is defined according to the network behavior characteristics. The decision variables corresponding to the comprehensive objective function are mapped into corresponding quantum states through binary coding, and an initial quantum population is constructed. The quantum states in the initial quantum population are subjected to quantum crossover and quantum mutation, and the quantum population is iteratively updated. In combination with the calculation result of the comprehensive objective function, an optimal traffic distribution strategy is obtained to realize adaptive traffic distribution, which includes:

[0044] According to the network behavior characteristics, a comprehensive objective function is constructed by taking minimizing response time, maximizing content popularity, and adapting to the constraints of server load balancing as goals;

[0045] Determine the decision variables corresponding to the result update of the comprehensive objective function, binary encode the decision variables, map them into quantum bits, combine the quantum bits into quantum states, randomly select a group of the quantum states, initialize the quantum population, and start quantum iteration:

[0046] Performing quantum state measurement on the quantum state to obtain a quantum state probability, selecting a shortlisted quantum state whose quantum state probability is greater than a preset screening probability, performing quantum crossover on the shortlisted quantum state through a quantum gate to generate a daughter quantum state, performing quantum mutation on the shortlisted quantum state through a bit quantum gate to generate a mutant quantum state, and using the shortlisted quantum state, the daughter quantum state and the mutant quantum state to construct an updated population;

[0047] Until the number of iterations meets the preset maximum number of iterations, the quantum iteration is stopped to obtain a final quantum population, and all quantum states in the final quantum population are decomposed and binary decoded to obtain a corresponding final decision variable set;

[0048] The final decision variable set is applied to the comprehensive objective function to calculate the fitness result set, and the optimal decision variable corresponding to the maximum fitness value is obtained by selecting the maximum fitness value from the fitness result set to generate the optimal traffic distribution strategy.

[0049] In an optional embodiment,

[0050] Binary encoding the decision variables, mapping them into quantum bits, and combining the quantum bits into quantum states comprises:

[0051] The formula is as follows:

[0052]

[0053] Among them, |ψ〉 represents the quantum state, k represents the total number of decision variables, j represents a state of the decision variable, and q j represents the probability amplitude corresponding to the decision variable state j, |j〉 represents the basis quantum state corresponding to the decision variable state j;

[0054] The quantum state is measured to obtain the quantum state probability, which is expressed as follows:

[0055] PB j =|<j|ψ> | 2 ;

[0056] Among them, PB j represents the probability of the quantum state corresponding to the decision variable state j, 〈j|ψ〉 represents the inner product of the quantum state |ψ〉 on the basis quantum state corresponding to the decision variable state j, |〈j|ψ〉| 2 represents the square of the modulus of the inner product.

[0057] According to a second aspect of the embodiments of the present invention,

[0058] Provided is an adaptive traffic distribution system based on clustering and quantum computing, comprising:

[0059] The first unit is used to collect network behavior data, use the network behavior data as sample points, construct sample clusters, take values ​​of the total number of clusters corresponding to the sample clusters multiple times according to a preset number of iterations, determine the maximum value of the average silhouette coefficient of the sample points based on the similarity silhouette method, and obtain the optimal total number of clusters and the initial clustering result corresponding to the optimal total number of clusters;

[0060] The second unit is used to determine the average position of all sample points in each cluster based on the initial clustering result, update the cluster center according to the average position, redistribute all sample points according to the principle of proximity, obtain a new clustering result, repeat the iteration until the cluster center no longer changes, and obtain the network behavior clustering;

[0061] The third unit is used to determine the network behavior characteristics based on the network behavior clustering, define a comprehensive objective function according to the network behavior characteristics, map the decision variables corresponding to the comprehensive objective function into corresponding quantum states through binary coding, construct an initial quantum population, perform quantum crossover and quantum mutation on the quantum states in the initial quantum population, iteratively update the quantum population, and combine the calculation results of the comprehensive objective function to obtain the optimal traffic distribution strategy and realize adaptive traffic distribution.

[0062] According to a third aspect of the embodiments of the present invention,

[0063] An electronic device is provided, comprising:

[0064] processor;

[0065] a memory for storing processor-executable instructions;

[0066] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0067] A fourth aspect of the embodiments of the present invention is:

[0068] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0069] In the embodiment of the present invention, dynamic selection of the total number of clusters is achieved through multiple iterations and silhouette coefficient calculations, which helps to avoid the limitation of the preset total number of clusters on the clustering effect, making the algorithm more flexible and adaptable; by calculating the silhouette coefficients of sample points, the compactness and separation of clusters are evaluated, and maximizing the average silhouette coefficient helps to obtain better clustering results; by calculating the probability of each candidate point becoming a center point, a probabilistic component is introduced, which helps to more flexibly select cluster centers, making the algorithm more adaptable to different data sets; by iteratively updating cluster centers and redistributing sample points, the accuracy of clustering can be gradually improved; the nature of quantum computing allows multiple possible solutions to be processed at the same time, and through the parallelism of quantum states, the fitness of multiple solutions can be evaluated simultaneously, accelerating the search process; through binary encoding of decision variables and dynamic updating of quantum states, the algorithm is adaptive and can adapt to the complexity and changes of problems. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 A schematic diagram of a flow chart of an adaptive traffic distribution method based on clustering and quantum computing according to an embodiment of the present invention;

[0071] Figure 2 It is a structural diagram of an adaptive traffic distribution system based on clustering and quantum computing according to an embodiment of the present invention. DETAILED DESCRIPTION

[0072] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0073] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0074] Figure 1 FIG. 1 is a flow chart of an adaptive traffic distribution method based on clustering and quantum computing according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0075] S101. Collect network behavior data, use the network behavior data as sample points, construct sample clusters, take values ​​of the total number of clusters corresponding to the sample clusters multiple times according to a preset number of iterations, determine the maximum value of the average silhouette coefficient of the sample points based on the similarity silhouette method, and obtain the optimal total number of clusters and the initial clustering result corresponding to the optimal total number of clusters;

[0076] The network behavior data specifically refers to user information and environment information collected from network activities, including user behavior related information such as user clicks, searches, and browsing, as well as network environment related information such as user location, network speed, and device type. User behavior related information and network environment related information together describe the interactive behavior of the network, which is used to analyze user trends, regional characteristics, and adjust service forms, and is the data basis for realizing adaptive traffic distribution;

[0077] Obtain user behavior and device environment data from the network to form network behavior data. Take each set of network behavior data as a sample point. Use a clustering algorithm to divide the sample points into different clusters to form sample clusters. Preset the number of iterations. Take multiple values ​​for the total number of clusters according to the set number of times. Each value is used as the target of sample point division. Calculate the silhouette coefficient of each sample point based on the similarity silhouette method. Then calculate the average silhouette coefficient of the overall sample to form clustering results under different total number of clusters. After multiple iterations, find the total number of clusters corresponding to the maximum value of the average silhouette coefficient. Obtain the sample clustering result corresponding to the total number of clusters as the initial clustering result for subsequent operations.

[0078] In this embodiment, through multiple iterations and similarity profile method, the total number of sample clusters is dynamically selected to ensure that the total number of clusters finally selected is closer to the internal structure of the data, thereby improving the accuracy of clustering; by calculating the maximum value of the average profile coefficient, the clustering result corresponding to the optimal total number of clusters is found, which helps to ensure that the clustering result has high similarity within the sample and high isolation between samples, thereby improving the quality of clustering;

[0079] In an optional embodiment, the total number of clusters corresponding to the sample cluster is multiplexed according to a preset number of iterations, and based on the similarity profile method, the maximum value of the average profile coefficient of the sample point is determined, and the optimal total number of clusters is obtained, which includes:

[0080] A value is randomly selected as the total number of clusters. Based on the total number of clusters, a cluster center selection algorithm is used to iteratively calculate the probability of each sample point being selected as a cluster center, initialize the cluster center, and assign all sample points to the corresponding cluster center according to the principle of proximity to obtain an initial clustering result.

[0081] Based on the intra-class difference and inter-class similarity of each sample point, a silhouette coefficient of each sample point is calculated, and the average silhouette coefficient of all the sample points in the total number of clusters is counted;

[0082] The similarity profile method is repeated until a preset number of iterations is met, and the maximum value of the average profile coefficient is selected. The total number of clusters corresponding to the maximum value is taken as the optimal total number of clusters.

[0083] The proximity principle specifically means that in clustering, sample points are assigned to the cluster center that is closest to them, ensuring that each sample point belongs to the closest cluster;

[0084] The intra-class difference specifically refers to measuring the similarity between sample points in the same cluster, usually measured by Euclidean distance. The smaller the intra-class difference, the more similar the sample points in the same cluster are.

[0085] The inter-class similarity specifically refers to measuring the similarity between different clusters, and is usually measured using the Euclidean distance. The greater the inter-class similarity, the less similar the sample points between different clusters are, and the greater the difference.

[0086] Randomly select a value as the total number of initial clusters. Based on the selected total number of initial clusters, iteratively calculate the probability of each sample point being selected as the cluster center, and assign all sample points to the corresponding cluster centers according to the principle of proximity to form the initial clustering result.

[0087] Calculate the intra-class difference of each sample point, and calculate the inter-class similarity of each sample point at the same time. According to the results, calculate the silhouette coefficient of each sample point, and use the average silhouette coefficient to evaluate the overall compactness and separation of the cluster.

[0088] We keep trying different total number of clusters, and get the average silhouette coefficients corresponding to the corresponding total number of clusters. Among all the average silhouette coefficients, we select the maximum value. The total number of clusters corresponding to the maximum value of the average silhouette coefficient is the total number of clusters that can clearly divide the clusters. The closer the connection between the sample points in each cluster is, the greater the difference between the sample points between clusters, and the optimal total number of clusters is obtained.

[0089] In this embodiment, through multiple iterations and silhouette coefficient calculations, dynamic selection of the total number of clusters is achieved, which helps to avoid the limitation of the pre-set total number of clusters on the clustering effect, making the algorithm more flexible and adaptable; according to the principle of proximity, the sample points are assigned to the nearest cluster center to obtain more accurate clustering results, which helps to ensure that samples within the same cluster are similar and samples between different clusters are significantly different; by calculating the silhouette coefficient of the sample points, the compactness and separation of the clusters are evaluated, and maximizing the average silhouette coefficient helps to obtain better clustering results.

[0090] In an optional embodiment, based on the intra-class difference and inter-class similarity of each sample point, the silhouette coefficient of each sample point is calculated, and the average silhouette coefficient of all the sample points in the total number of clusters is calculated, including:

[0091] The intra-class difference of each sample point is calculated, and the formula is as follows:

[0092]

[0093] Where x represents a sample point, y represents another sample point, a(x) represents the difference between sample point x and other sample points in the same cluster, and L x represents all sample points that are clustered with sample point x, |L x | represents the total number of all sample points in the same cluster as sample point x, and d(x,y) represents the distance between sample point x and sample point y;

[0094] The inter-class similarity of each sample point is calculated, and the formula is as follows:

[0095]

[0096] Among them, b(x) represents the similarity between the sample point x and the sample point of the nearest cluster, z represents a sample point, and Lz represents the number of samples in addition to L x For clusters other than , d(x,z) represents the distance between sample point x and sample point z;

[0097] The silhouette coefficient of each sample point is calculated using the following formula:

[0098]

[0099] Among them, s(x) represents the silhouette coefficient of sample point x.

[0100] As described in the intra-class difference calculation formula, the average difference between the sample point x and other sample points in the same cluster is calculated. By calculating the distance between all sample points in the same cluster as x, the distribution uniformity of the sample point x in its cluster can be measured;

[0101] As described in the inter-class similarity calculation formula, the average similarity between the sample point x and the nearest other clusters is calculated. By finding other clusters that are different from the cluster where x is located and calculating the average distance between x and all sample points in these clusters, the similarity between the sample point x and other clusters can be measured;

[0102] As described in the calculation formula of the silhouette coefficient, the intra-class difference and inter-class similarity of the sample point x are comprehensively considered, and the silhouette coefficient of the sample point x is obtained by standardizing their differences. The silhouette coefficient reflects the compactness and separation of the cluster where the sample point is located. The value range is between [-1, 1]. The closer to 1, the better the clustering effect.

[0103] According to the formula, the intra-class difference a(x) and inter-class similarity b(x) in the formula provide quantitative measurements of the sample distribution inside and outside the cluster where the sample point is located, which helps to more accurately understand the structure of the cluster; the silhouette coefficient s(x) provides a comprehensive evaluation index for each sample point by comprehensively considering the intra-class difference and inter-class similarity, helping to determine the quality of the clustering; the calculation of the silhouette coefficient takes into account the local structure of each sample point, and the evaluation of the clustering effect is more accurate and comprehensive.

[0104] In an optional embodiment, based on the total number of clusters, a cluster center selection algorithm is used to iteratively calculate the probability of each sample point being selected as a cluster center, and initializing the cluster center includes:

[0105] Set a cluster center set, randomly select a sample point, add it to the cluster center set, and the remaining sample points constitute the candidate set, where each sample point in the candidate set is used as a candidate point, and each sample point in the cluster center set is used as a center point, and the cluster center selection algorithm iteration begins:

[0106] The probability of selection of the selected point is calculated as follows:

[0107] Select a candidate point as the current candidate point, calculate the distance between the current candidate point and each center point in the cluster center set, determine the minimum distance value corresponding to the current candidate point, and calculate the probability of the current candidate point being selected as the center point based on the minimum distance value;

[0108] Until the probability of selection of each candidate point is obtained, the selection probability calculation is stopped, the candidate point corresponding to the maximum selection probability is selected as the optimal candidate point, the optimal candidate point is removed from the candidate set, and the optimal candidate point is added to the cluster center set;

[0109] Until the total number of elements in the cluster center set is equal to the total number of clusters, the iteration of the cluster center selection algorithm is stopped to complete the initialization of the cluster centers.

[0110] The cluster center set specifically means that in the cluster, each cluster has a center point, the cluster center set stores all cluster center points, and the cluster center set includes sample points that have been selected as cluster centers;

[0111] The candidate set specifically refers to selecting candidate points from the candidate set as new cluster centers during the process of initializing cluster centers, and the candidate set includes sample points that have not yet been selected as cluster centers;

[0112] Create an empty cluster center set and an empty candidate set, randomly select a sample point from all sample points as the first center point, put it into the cluster center set, and construct the candidate set together with the remaining sample points except the sample point selected as the center point, where the sample points in the cluster center set are called center points, and the sample points in the candidate set are called candidate points;

[0113] Select a candidate point from the candidate set, calculate the distance between the candidate point and each center point in the cluster center set to obtain a set of distance values, select the minimum distance value from the set of distance values, calculate the probability of the candidate point being selected as the center point, wherein the probability of being selected as the center point is called the selected probability, when the selected probabilities corresponding to each candidate point in the candidate set are calculated, select the maximum probability from all the selected probabilities, and the candidate point corresponding to the maximum probability is selected as the second center point and added to the cluster center set, and is removed from the candidate set at the same time, and the calculation of the candidate points in the candidate set is restarted;

[0114] And so on, until the number of elements in the cluster center set meets the total number of clusters determined in the previous step, the cluster center is initialized, and each center point in the cluster center set corresponds to a cluster center;

[0115] In this embodiment, by calculating the probability of each candidate point becoming the center point, a probabilistic component is introduced, which helps to select the cluster center more flexibly and makes the algorithm better adaptable to different data sets; the probabilistic center selection helps to achieve a uniform distribution of cluster centers and ensure that the center point of each cluster has better representativeness and coverage in the data set; the probabilistic selection mechanism selects the best candidate point in each iteration instead of simply relying on the global optimum, which helps to avoid falling into the local optimal solution.

[0116] In an optional embodiment, performing a selection probability calculation on the candidate point further includes:

[0117] Calculate the distance between each of the candidate points and each of the center points, where the candidate points and the center point are both sample points, containing n features, and the formula is as follows:

[0118]

[0119] Among them, dist 2(p,c) represents the square of the Euclidean distance between the candidate point p and the center point c, i represents the feature index, n represents the total number of features of the sample point, p i represents the i-th feature of the point to be selected, c i Represents the i-th feature of the center point;

[0120] The probability of the selected point being selected is calculated, and the formula is as follows:

[0121]

[0122] Among them, prob(p) represents the probability of the selected point p, c n Represents the center point closest to the selected point p.

[0123] As described in the distance calculation formula, it is used to calculate the square of the Euclidean distance between the candidate point p and the center point c;

[0124] As described in the probability calculation formula of the selected point, the probability of the selected point p being selected as the cluster center is calculated, and the probability is related to the distance from the selected point to the nearest center point c. n The square of the distance is inversely proportional to the probability that the smaller the distance is, the greater the probability that the point to be selected is selected as the center.

[0125] According to the formula, the Euclidean distance measures the similarity or difference between the selected point and the center point, and provides a comprehensive evaluation of the spatial relationship between sample points by considering the sum of the squared differences of all features; the probabilistic selection mechanism makes it more likely that points with closer distances will be selected as centers, increasing the flexibility and adaptability of the algorithm; the square of the distance is used as the denominator of the probability, so that a large distance has a greater impact on the probability, which helps to reduce the impact of outliers on the selection of the center point and improve the robustness of the algorithm; the square of the distance in the probability calculation is inversely proportional, so that points with smaller distances have a higher probability of being selected, which better reflects the concentration of sample points in the data space and is conducive to the selection of representative centers.

[0126] S102. Based on the initial clustering result, for each cluster, determine the average position of all sample points in the cluster, update the cluster center according to the average position, redistribute all sample points according to the principle of proximity, obtain a new clustering result, repeat the iteration until the cluster center no longer changes, and obtain the network behavior clustering;

[0127] Based on the initial clustering results obtained in the previous step, the average position of all sample points in each cluster is calculated, and the average position is used as the current cluster center;

[0128] Check whether the current cluster center and the cluster center obtained in the previous step have changed. If so, redistribute the sample points according to the current cluster center and update the clustering result. The redistribution is based on the principle of proximity, so that similar sample points are allocated in one cluster.

[0129] If there is no change, the iteration stops and the final result is obtained, which is the network behavior clustering;

[0130] A maximum number of updates can be set for cluster updates to prevent endless iterative updates;

[0131] In this embodiment, by iteratively updating the cluster centers and reallocating sample points, the accuracy of clustering can be gradually improved, and the clustering results can be continuously optimized to ensure better division of network behaviors; by recalculating the cluster centers and reallocating sample points, the algorithm is adaptive and can flexibly respond to changes in different network behavior patterns, thereby improving the versatility of the method; stopping the iteration when the cluster centers no longer change significantly helps improve the stability of the algorithm, avoid excessive iterations, and reduce unnecessary computational overhead.

[0132] S103. Determine the network behavior characteristics based on the network behavior clustering, define a comprehensive objective function according to the network behavior characteristics, map the decision variables corresponding to the comprehensive objective function into corresponding quantum states through binary coding, construct an initial quantum population, perform quantum crossover and quantum mutation on the quantum states in the initial quantum population, iteratively update the quantum population, and combine the calculation results of the comprehensive objective function to obtain the optimal traffic distribution strategy and realize adaptive traffic distribution;

[0133] The comprehensive objective function specifically refers to defining a comprehensive objective function that comprehensively considers different network behavior characteristics, such as response time, content popularity, server load balancing, etc., and is a quantitative representation of network behavior clustering results;

[0134] The decision variables specifically refer to adjustable parameters that affect the traffic distribution strategy, including weights, thresholds, various service parameters, and optimization strategies, etc., which are used to adjust the influence of different network behavior characteristics on optimal traffic distribution;

[0135] The quantum state specifically refers to mapping the decision variables into quantum states through binary coding. In quantum computing, the quantum state is a mathematical representation of the state of the system. Here, the quantum state is used to represent the combination of different decision variables. The characteristics of quantum computing can be used to process multiple possibilities in parallel.

[0136] Based on the network behavior clustering results, comprehensively analyze various factors affecting traffic distribution and define a comprehensive objective function;

[0137] The decision variables contained in the comprehensive objective function are binary-encoded, and each decision variable is represented by a set of binary bits, forming a set of binary-encoded decision variables; each binary bit is mapped to a quantum bit, so that the combination of decision variables is mapped to a quantum state;

[0138] Randomly generate or use some pre-sets to construct the initial quantum population, perform quantum crossover and quantum mutation operations on the quantum states in the initial quantum population, and use quantum computing to form quantum operations similar to classical crossover and mutation operations; by iteratively updating the quantum population, gradually optimize the quantum state and find a better traffic distribution strategy;

[0139] According to each quantum state in the quantum population, the corresponding comprehensive objective function is calculated to evaluate the performance of each possible strategy. Combined with the calculation results, the quantum state with the optimal comprehensive objective function value is selected to obtain the optimal traffic distribution strategy.

[0140] In this embodiment, the comprehensive objective function takes into account multiple network behavior characteristics, making the traffic distribution strategy more comprehensive and flexible to meet different needs; the parallel nature of quantum computing allows multiple possible decision variable combinations to be calculated simultaneously, thereby accelerating the process of searching for the optimal solution. The characteristics of quantum computing help reduce the computational complexity of certain problems and improve computational efficiency; through quantum crossover and quantum mutation, the algorithm can more efficiently search the decision variable space and find a better traffic distribution strategy; comprehensive consideration of multiple objectives and utilization of the characteristics of quantum computing make the algorithm more robust to different network behavior scenarios.

[0141] In an optional embodiment, the network behavior characteristics are determined based on the network behavior clustering, a comprehensive objective function is defined according to the network behavior characteristics, the decision variables corresponding to the comprehensive objective function are mapped into corresponding quantum states through binary coding, an initial quantum population is constructed, and the quantum states in the initial quantum population are quantum crossed and quantum mutated, and the quantum population is iteratively updated. The optimal traffic distribution strategy is obtained in combination with the calculation result of the comprehensive objective function, and the adaptive traffic distribution is realized, which includes:

[0142] According to the network behavior characteristics, a comprehensive objective function is constructed by taking minimizing response time, maximizing content popularity, and adapting to the constraints of server load balancing as goals;

[0143] Determine the decision variables corresponding to the result update of the comprehensive objective function, binary encode the decision variables, map them into quantum bits, combine the quantum bits into quantum states, randomly select a group of the quantum states, initialize the quantum population, and start quantum iteration:

[0144] Performing quantum state measurement on the quantum state to obtain a quantum state probability, selecting a shortlisted quantum state whose quantum state probability is greater than a preset screening probability, performing quantum crossover on the shortlisted quantum state through a quantum gate to generate a daughter quantum state, performing quantum mutation on the shortlisted quantum state through a bit quantum gate to generate a mutant quantum state, and using the shortlisted quantum state, the daughter quantum state and the mutant quantum state to construct an updated population;

[0145] Until the number of iterations meets the preset maximum number of iterations, the quantum iteration is stopped to obtain a final quantum population, and all quantum states in the final quantum population are decomposed and binary decoded to obtain a corresponding final decision variable set;

[0146] The final decision variable set is applied to the comprehensive objective function to calculate the fitness result set, and the optimal decision variable corresponding to the maximum fitness value is obtained by selecting the maximum fitness value from the fitness result set to generate the optimal traffic distribution strategy.

[0147] The qubit specifically refers to the basic unit of quantum computing, which is similar to the bit in classical computing. In particular, a qubit can be in a superposition state of 0 and 1 at the same time, and parallel computing is performed using the quantum superposition property;

[0148] The quantum state measurement specifically refers to the state of a quantum system that can be obtained through measurement in quantum computing, and the measured system will collapse to the state corresponding to the measurement result;

[0149] The quantum state probability specifically refers to the probability of obtaining a result by measuring a quantum state in quantum computing, and the probability is proportional to the square of the modulus of the amplitude of the quantum state;

[0150] The quantum gate specifically refers to an operation unit in quantum computing, which is used to interact between quantum bits. Common quantum gates include Hadamard gates, CNOT gates, etc.

[0151] The quantum crossover specifically refers to a crossover operation similar to that in classical computing, which is used to generate new solutions. In quantum computing, quantum gates are used to achieve mutual influence between different quantum bits, thereby generating new quantum states.

[0152] The bit quantum gate specifically refers to a special quantum computing operation unit used to perform operations on a single quantum bit, including a phase gate, a bit flip gate, etc.;

[0153] The quantum mutation specifically refers to a mutation operation similar to that in classical computing, which generates new solutions by introducing randomness. In quantum computing, this can be achieved by bit flipping, etc.

[0154] Taking minimizing response time as the first goal and maximizing content popularity as the second goal, the goals and constraints are integrated to construct a comprehensive objective function based on the constraints of server load balancing.

[0155] The decision variables corresponding to the comprehensive objective function are binary-encoded and mapped into quantum bits. Multiple decision variables are combined to form a strategy for dealing with the goal, so the corresponding quantum bits are combined to form a quantum state. A set of quantum states is randomly selected to initialize the quantum population.

[0156] Start quantum population iteration:

[0157] The quantum states in the current quantum population are measured to obtain the probability distribution of each quantum state. According to the measured probability distribution, the shortlisted quantum states are selected, wherein the quantum states whose shortlisted probability is greater than the preset screening probability are selected as the shortlisted quantum states; quantum gate operations are applied to the shortlisted quantum states to perform quantum crossover and generate daughter quantum states for integrating the characteristics of the shortlisted quantum states; bit quantum gate operations are applied to the shortlisted quantum states to perform quantum mutation and generate mutant quantum states for introducing new individual differences; the shortlisted quantum states, daughter quantum states and mutant quantum states are used to construct an updated quantum population so that the new quantum population has rich characteristics;

[0158] Until the number of iterations meets the preset maximum number of iterations, the quantum population iteration is stopped;

[0159] Decomposing and binary decoding all quantum states in the final quantum population, that is, using the inverse operation from decision variables to quantum states, to obtain the final decision variable set, where the elements in the final decision variable set make groups of decision variables, that is, each group represents a specific strategy, that is, the final decision variable set is a set of solutions to the comprehensive objective function;

[0160] The final decision variable set is applied to the comprehensive objective function, the fitness result set is calculated, the maximum fitness value is selected from the fitness result set, and the optimal decision variable corresponding to the maximum fitness is obtained. The optimal decision variable is used to generate the optimal traffic distribution strategy to meet the needs of network behavior clustering.

[0161] In this embodiment, the quantum iteration process utilizes the global search characteristics of quantum computing. By searching for possible solutions in the quantum space, it is expected to find the global optimal solution rather than just the local optimal solution. The nature of quantum computing allows multiple possible solutions to be processed at the same time. Through the parallelism of quantum states, the fitness of multiple solutions can be evaluated simultaneously, accelerating the search process. The application of quantum gates enables quantum states to interact in a more complex way, which helps to explore the solution space more efficiently. Through the binary encoding of decision variables and the dynamic updating of quantum states, the algorithm is adaptive and can adapt to the complexity and changes of the problem. Finally, the optimal traffic distribution strategy is generated by selecting the optimal decision variable corresponding to the maximum fitness value, ensuring that the best traffic distribution solution is obtained in the context of network behavior clustering.

[0162] In an optional embodiment, binary encoding the decision variables, mapping them into quantum bits, and combining the quantum bits into quantum states comprises:

[0163] The formula is as follows:

[0164]

[0165] Among them, |ψ〉 represents the quantum state, k represents the total number of decision variables, j represents a state of the decision variable, and q j represents the probability amplitude corresponding to the decision variable state j, |j〉 represents the basis quantum state corresponding to the decision variable state j;

[0166] The quantum state is measured to obtain the quantum state probability, which is expressed as follows:

[0167] PB j =|<j|ψ> | 2 ;

[0168] Among them, PB j represents the probability of the quantum state corresponding to the decision variable state j, 〈j|ψ〉 represents the inner product of the quantum state |ψ〉 on the basis quantum state corresponding to the decision variable state j, |〈j|ψ〉| 2 represents the square of the modulus of the inner product.

[0169] As described in the formula for combining the quantum bits into a quantum state, the formula represents the way in which the quantum state |ψ〉 is constructed, which is formed by weighted superposition of the basis quantum states |j〉 of different decision variable states according to their probability amplitudes;

[0170] As described in the formula for obtaining the probability of the quantum state by performing quantum state measurement on the quantum state, it is shown that the probability of the measurement result is related to the projection of the quantum state on the corresponding ground state, so that the system will collapse to a certain state during measurement, reflecting the uncertainty of the quantum system and the intervention caused by the measurement;

[0171] According to the formula, quantum states are used for representation, and each decision variable state corresponds to a basic quantum state. The superposition method allows multiple decision variable states to be represented simultaneously in one quantum state, reflecting the super-position characteristics of quantum computing. By adjusting the amplitude, the distribution of the quantum state can be affected, thereby affecting the direction of problem solving. The process of quantum state measurement produces a specific decision variable state, which corresponds to the collapse of the quantum state. The probability of collapse is given by the square of the modulus of the probability amplitude of the quantum state on a specific basic state, representing the possibility of obtaining each decision variable state during measurement. The probability obtained from quantum state measurement can be used to guide the evolutionary direction of the search or optimization algorithm, increasing the flexibility of the algorithm in navigating the search space.

[0172] Figure 2 FIG. 1 is a schematic diagram of the structure of an adaptive traffic distribution system based on clustering and quantum computing according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0173] The first unit is used to collect network behavior data, use the network behavior data as sample points, construct sample clusters, take values ​​of the total number of clusters corresponding to the sample clusters multiple times according to a preset number of iterations, determine the maximum value of the average silhouette coefficient of the sample points based on the similarity silhouette method, and obtain the optimal total number of clusters and the initial clustering result corresponding to the optimal total number of clusters;

[0174] The second unit is used to determine the average position of all sample points in each cluster based on the initial clustering result, update the cluster center according to the average position, redistribute all sample points according to the principle of proximity, obtain a new clustering result, repeat the iteration until the cluster center no longer changes, and obtain the network behavior clustering;

[0175] The third unit is used to determine the network behavior characteristics based on the network behavior clustering, define a comprehensive objective function according to the network behavior characteristics, map the decision variables corresponding to the comprehensive objective function into corresponding quantum states through binary coding, construct an initial quantum population, perform quantum crossover and quantum mutation on the quantum states in the initial quantum population, iteratively update the quantum population, and combine the calculation results of the comprehensive objective function to obtain the optimal traffic distribution strategy and realize adaptive traffic distribution.

[0176] According to a third aspect of the embodiments of the present invention,

[0177] An electronic device is provided, comprising:

[0178] processor;

[0179] a memory for storing processor-executable instructions;

[0180] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0181] A fourth aspect of the embodiments of the present invention is:

[0182] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0183] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0184] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An adaptive traffic distribution method based on clustering and quantum computing, characterized in that: include: Collect network behavior data, use the network behavior data as sample points, construct sample clusters, take values ​​of the total number of clusters corresponding to the sample clusters multiple times according to a preset number of iterations, determine the maximum value of the average silhouette coefficient of the sample points based on the similarity silhouette method, and obtain the optimal total number of clusters and the initial clustering result corresponding to the optimal total number of clusters; Based on the initial clustering result, for each cluster, determine the average position of all sample points in the cluster, update the cluster center according to the average position, redistribute all sample points according to the principle of proximity, obtain a new clustering result, repeat the iteration until the cluster center no longer changes, and obtain the network behavior clustering; Determine network behavior characteristics based on the network behavior clustering, define a comprehensive objective function according to the network behavior characteristics, map the decision variables corresponding to the comprehensive objective function into corresponding quantum states through binary coding, construct an initial quantum population, perform quantum crossover and quantum mutation on the quantum states in the initial quantum population, iteratively update the quantum population, and combine the calculation results of the comprehensive objective function to obtain the optimal traffic distribution strategy and realize adaptive traffic distribution; Based on the network behavior clustering, the network behavior characteristics are determined, and a comprehensive objective function is defined according to the network behavior characteristics. The decision variables corresponding to the comprehensive objective function are mapped into corresponding quantum states through binary coding, and an initial quantum population is constructed. The quantum states in the initial quantum population are subjected to quantum crossover and quantum mutation, and the quantum population is iteratively updated. In combination with the calculation result of the comprehensive objective function, an optimal traffic distribution strategy is obtained to realize adaptive traffic distribution, which includes: According to the network behavior characteristics, a comprehensive objective function is constructed by taking minimizing response time, maximizing content popularity, and adapting to the constraints of server load balancing as goals; Determine the decision variables corresponding to the result update of the comprehensive objective function, binary encode the decision variables, map them into quantum bits, combine the quantum bits into quantum states, randomly select a group of the quantum states, initialize the quantum population, and start quantum iteration: Performing quantum state measurement on the quantum state to obtain a quantum state probability, selecting a shortlisted quantum state whose quantum state probability is greater than a preset screening probability, performing quantum crossover on the shortlisted quantum state through a quantum gate to generate a daughter quantum state, performing quantum mutation on the shortlisted quantum state through a bit quantum gate to generate a mutant quantum state, and using the shortlisted quantum state, the daughter quantum state and the mutant quantum state to construct an updated population; Until the number of iterations meets the preset maximum number of iterations, the quantum iteration is stopped to obtain a final quantum population, and all quantum states in the final quantum population are decomposed and binary decoded to obtain a corresponding final decision variable set; The final decision variable set is applied to the comprehensive objective function to calculate the fitness result set, and the optimal decision variable corresponding to the maximum fitness value is obtained by selecting the maximum fitness value from the fitness result set to generate the optimal traffic distribution strategy.

2. The method according to claim 1, characterized in that The total number of clusters corresponding to the sample clustering is taken multiple times according to a preset number of iterations, and the maximum value of the average silhouette coefficient of the sample points is determined based on the similarity silhouette method, and the optimal total number of clusters is obtained, including: A value is randomly selected as the total number of clusters. Based on the total number of clusters, a cluster center selection algorithm is used to iteratively calculate the probability of each sample point being selected as a cluster center, initialize the cluster center, and assign all sample points to the corresponding cluster center according to the principle of proximity to obtain an initial clustering result. Based on the intra-class difference and inter-class similarity of each sample point, a silhouette coefficient of each sample point is calculated, and the average silhouette coefficient of all the sample points in the total number of clusters is counted; The similarity profile method is repeated until a preset number of iterations is met, and the maximum value of the average profile coefficient is selected. The total number of clusters corresponding to the maximum value is taken as the optimal total number of clusters.

3. The method according to claim 2, characterized in that Based on the intra-class difference and inter-class similarity of each sample point, the silhouette coefficient of each sample point is calculated, and the average silhouette coefficient of all the sample points in the total number of clusters is counted, including: The intra-class difference of each sample point is calculated, and the formula is as follows: Where x represents a sample point, y represents another sample point, a(x) represents the difference between sample point x and other sample points in the same cluster, and L x represents all sample points that are in the same cluster as the sample point x, |L x | represents the total number of all sample points in the same cluster as sample point x, and d(x,y) represents the distance between sample point x and sample point y; The inter-class similarity of each sample point is calculated, and the formula is as follows: Among them, b(x) represents the similarity between the sample point x and the sample point of the nearest cluster, z represents a sample point, and Lz represents the number of samples in addition to L x For clusters other than , d(x,z) represents the distance between sample point x and sample point z; The silhouette coefficient of each sample point is calculated using the following formula: Among them, s(x) represents the silhouette coefficient of sample point x.

4. The method according to claim 2, characterized in that: Based on the total number of clusters, a cluster center selection algorithm is used to iteratively calculate the probability of each sample point being selected as a cluster center. Initializing the cluster center includes: Set a cluster center set, randomly select a sample point, add it to the cluster center set, and the remaining sample points constitute the candidate set, where each sample point in the candidate set is used as a candidate point, and each sample point in the cluster center set is used as a center point, and the cluster center selection algorithm iteration begins: The probability of selection is calculated for the selected point: Select a candidate point as the current candidate point, calculate the distance between the current candidate point and each center point in the cluster center set, determine the minimum distance value corresponding to the current candidate point, and calculate the probability of the current candidate point being selected as the center point based on the minimum distance value; Until the probability of selection of each candidate point is obtained, the selection probability calculation is stopped, the candidate point corresponding to the maximum selection probability is selected as the optimal candidate point, the optimal candidate point is removed from the candidate set, and the optimal candidate point is added to the cluster center set; Until the total number of elements in the cluster center set is equal to the total number of clusters, the iteration of the cluster center selection algorithm is stopped to complete the initialization of the cluster centers.

5. The method according to claim 4, characterized in that Performing a selection probability calculation on the candidate point also includes: Calculate the distance between each of the candidate points and each of the center points, where the candidate points and the center point are both sample points, containing n features, and the formula is as follows: Among them, dist 2 (p,c) represents the square of the Euclidean distance between the candidate point p and the center point c, i represents the feature index, n represents the total number of features of the sample point, p i represents the i-th feature of the point to be selected, c i Represents the i-th feature of the center point; The probability of the selected point being selected is calculated, and the formula is as follows: Among them, prob(p) represents the probability of the selected point p, c n Represents the center point closest to the selected point p.

6. The method according to claim 1, characterized in that Binary encoding the decision variables, mapping them into quantum bits, and combining the quantum bits into quantum states comprises: The formula is as follows: Among them, |ψ〉 represents the quantum state, k represents the total number of decision variables, j represents a state of the decision variable, and q j represents the probability amplitude corresponding to the decision variable state j, |j〉 represents the basis quantum state corresponding to the decision variable state j; The quantum state is measured to obtain the quantum state probability, which is expressed as follows: PB j =|<j|ψ>| 2 ; Among them, PB j represents the probability of the quantum state corresponding to the decision variable state j, 〈j|ψ〉 represents the inner product of the quantum state |ψ〉 on the basis quantum state corresponding to the decision variable state j, |〈j|ψ〉| 2 represents the square of the modulus of the inner product.

7. An adaptive traffic distribution system based on clustering and quantum computing, used to implement the adaptive traffic distribution method based on clustering and quantum computing described in any one of claims 1 to 6, characterized in that: include: The first unit is used to collect network behavior data, use the network behavior data as sample points, construct sample clusters, take values ​​of the total number of clusters corresponding to the sample clusters multiple times according to a preset number of iterations, determine the maximum value of the average silhouette coefficient of the sample points based on the similarity silhouette method, and obtain the optimal total number of clusters and the initial clustering result corresponding to the optimal total number of clusters; The second unit is used to determine the average position of all sample points in each cluster based on the initial clustering result, update the cluster center according to the average position, redistribute all sample points according to the principle of proximity, obtain a new clustering result, repeat the iteration until the cluster center no longer changes, and obtain the network behavior clustering; The third unit is used to determine the network behavior characteristics based on the network behavior clustering, define a comprehensive objective function according to the network behavior characteristics, map the decision variables corresponding to the comprehensive objective function into corresponding quantum states through binary coding, construct an initial quantum population, perform quantum crossover and quantum mutation on the quantum states in the initial quantum population, iteratively update the quantum population, and combine the calculation results of the comprehensive objective function to obtain the optimal traffic distribution strategy and realize adaptive traffic distribution.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images