An optical network wireless controller system based on TR069 and PON protocols

Abstract

The invention belongs to the technical field of wireless networks. The invention discloses an optical network wireless controller system based on TR069 and PON protocols. The system comprises the following steps: obtaining real-time relevant data of each wireless access point; calculating the performance score of each wireless access point through a preset performance evaluation model according to the real-time relevant data; determining the abnormal evaluation factor of each wireless access point based on the performance score of each wireless access point, and determining the suspected abnormal point and the monitoring neighborhood radius of the suspected abnormal point according to the abnormal evaluation factor; constructing spatiotemporal sample points in the monitoring neighborhood, performing outlier detection on the spatiotemporal sample points, and obtaining the local outlier factor corresponding to the suspected abnormal point; determining the abnormal wireless access point based on the local outlier factor corresponding to the suspected abnormal point, optimizing and adjusting the parameters of the abnormal wireless access point through the TR069 protocol and the PON protocol, and significantly improving the performance and management level of the wireless network.

Description

Technical Field

[0001] The present invention relates to the field of wireless network technology, and more specifically, to an optical network wireless controller system based on TR069 and PON protocols. Background Art

[0002] The patent application publication number CN108668303A discloses an incremental outlier detection method for a wireless sensor network data stream, including: using a sliding window to sample to obtain the latest data stream information; calculating the local distribution density of all data in the data stream based on the nearest neighbor idea; judging whether the data is an outlier based on the local distribution density of the data; the local outlier factor calculation method is: in the data set determined by the sliding window, the local outlier factor of each data is equal to the ratio of the mean of the local distribution density of all data in the K-order nearest neighbor, K-order reverse neighbor and K-order shared neighbor of the data to the local distribution density of the data itself; introducing new data or deleting data based on step three, and using an incremental local outlier factor update method to update the above data and calculate outliers, which can effectively reduce the computational complexity of online detection of outliers in wireless sensor network data streams and process faster.

[0003] However, traditional network management and optimization are still highly dependent on manual experience, lacking the ability to perceive and evaluate the overall performance status of the network in real time. Once performance anomalies occur, it is often difficult to detect and accurately locate the anomaly in a timely manner, causing the problem to worsen and spread, seriously affecting network quality. Secondly, the wireless network environment is complex and dynamic, and the optimization and adjustment of network parameters need to fully consider the combined impact of multiple factors. However, most traditional optimization methods are based on simple empirical rules, which are difficult to adapt to complex environments and have limited optimization effects. At the same time, manual optimization decision-making cycles are long and inefficient, and cannot meet the needs of real-time optimization. Furthermore, in the existing network management and optimization process, the degree of manual intervention is high, which not only increases the workload of management and operation, but also easily leads to human operational errors, affecting network stability. In addition, the decentralized management model also reduces the collaborative efficiency of each link.

[0004] In view of this, the present invention proposes an optical network wireless controller system based on TR069 and PON protocols to solve the above problems. Summary of the invention

[0005] In order to overcome the above defects of the prior art and to achieve the above objectives, the present invention provides the following technical solutions: an optical network wireless controller system based on TR069 and PON protocols, comprising: a data acquisition module for acquiring real-time relevant data of each wireless access point;

[0006] The performance scoring module is used to calculate the performance score of each wireless access point based on real-time relevant data through a preset performance evaluation model;

[0007] An anomaly assessment module determines an anomaly assessment factor of each wireless access point based on the performance score of each wireless access point, and determines a suspected anomaly point and a monitoring neighborhood radius of the suspected anomaly point according to the anomaly assessment factor;

[0008] The outlier detection module is used to construct the spatiotemporal sample points of the suspected outlier points within the monitoring neighborhood formed by the monitoring neighborhood radius of the suspected outlier points, and perform outlier value detection on the spatiotemporal sample points to obtain the local outlier factors corresponding to the suspected outlier points;

[0009] The optimization and adjustment module determines the abnormal wireless access point based on the local outlier factor corresponding to the suspicious abnormal point, and optimizes and adjusts the parameters of the abnormal wireless access point through the TR069 protocol and the PON protocol; each module is connected through wired and / or wireless means.

[0010] Furthermore, the real-time related data includes the load of the wireless access point, the number of users, the throughput, the delay and the bit error rate.

[0011] Furthermore, the preset method of the performance evaluation model includes:

[0012] Collecting historical operation data of N groups of wireless access points; representing each group of historical operation data of the wireless access points as a multi-dimensional state vector; and using the multi-dimensional state vector as a state corresponding to the historical operation data;

[0013] Define a series of parameter adjustment operations that can be performed by the wireless access point, discretize the parameter adjustment operations, and the obtained discrete actions constitute a finite discrete action set; regard the parameter adjustment operations as continuous values, that is, a continuous action set; the discrete action set, the continuous action set and the state constitute a training data set;

[0014] Define the reward and punishment function ;

[0015] ;in, , and is the adaptive weight; For status, For action, Status Next action The throughput after Status Next action After the delay, Status Next action The bit error rate after is the state value function;

[0016] State Value Function ;in, is the discount factor, For the status Next select action The probability of is the state-action-value function; is the entropy regularization coefficient; For strategy In Status The entropy value under

[0017] State-Action-Value Function ;in, is the strategy discount factor; For the status Next action Then transfer to the next state The transition probability of For the next state The state value function of For the status Next action The reward value immediately obtained after

[0018] ;in, For the status Next action The weight of

[0019] Build a policy network and define a performance evaluation function ; The policy network and performance evaluation function Integration, that is, obtaining a performance evaluation model;

[0020] The performance score is calculated by:

[0021] Input the current state of each wireless access point into the policy network, output an optimal action sequence, and substitute the current state of each wireless access point and the optimal action sequence output by the policy network into the performance evaluation function , calculate and obtain the performance score corresponding to each wireless access point.

[0022] Furthermore, the transition probability Ways to obtain include:

[0023] Build a differentiable environment model, use M sets of fixed strategies to interact with the environment in a real or simulated environment, and record N1 sets of state-action-next-state triplets As training data for a differentiable environment model;

[0024] The fully connected network is used as the backbone network of the differentiable environment model, and the state and actions As input, the next state The probability distribution of is used as output; train the differentiable environment model using training data; define the objective function of the differentiable environment model , use the back propagation algorithm to update the parameters of the backbone network, repeat the training until the output value of the objective function no longer changes, and then the differentiable environment model is built; use the constructed differentiable environment model to fit the transition probability in the real environment ;

[0025] Objective Function The expression is:

[0026] ;in, is the next state output by the differentiable environment model The probability distribution of The next state is true The probability distribution of is the divergence loss; is a differentiable environment model, outputting the next state The predicted value of for The entropy value of and is a hyperparameter.

[0027] Furthermore, the method of constructing the policy network includes:

[0028] Define the input of the policy network as the state, and the output as the probability distribution of each executable action and form the optimal action sequence;

[0029] The network structure of the policy network is defined as follows: an input layer, M1 fully connected hidden layers, and an output layer; the input layer is used to receive a multi-dimensional state vector; each of the M1 fully connected hidden layers is connected to BatchNorm and ReLU activation; for the space of discrete actions, the output layer uses SoftMax to output the probability of each action; for the space of continuous actions, the output layer uses the mean and variance of the Gaussian distribution as output;

[0030] Initialize the parameters of the policy network; divide the states and actions in the training data set into P training batches to train the policy network, and update the parameters of the policy network during the training process to minimize the policy objective function ;

[0031] LU=E[min(r( i ) ∙ R,cl(r( i ),1 - ϵ ,1+ ϵ ) ∙ R)] ;in, E [ ∙] For expectations; is the strategy ratio, are the parameters of the policy network,

[0032] Strategy Ratio ;in, The parameters of the policy network are new parameters When the policy network is in state Next select action The probability of The parameters of the policy network are the old parameters When the policy network is in state Next select action The probability of is the reward and punishment function, To update the hyperparameters, the policy network is repeatedly trained until the policy network converges and the policy network is constructed;

[0033] The performance evaluation function The expression is:

[0034] ;in, is the throughput of the wireless access point when executing the actions in the optimal action sequence in the current state, is the delay of the wireless access point when executing the actions in the optimal action sequence in the current state, is the bit error rate of the wireless access point when executing actions in the optimal action sequence in the current state; is the signal-to-noise ratio of the wireless access point when executing the actions in the optimal action sequence in the current state, It is the load level of the wireless access point when executing the actions in the optimal action sequence under the current state.

[0035] Furthermore, the abnormality assessment factor is determined in the following manner:

[0036] Calculate the distance between any two wireless access points and The performance score similarity between , construct a similarity matrix , similarity matrix Elements within ;

[0037] ;in, For wireless access points Performance rating, For wireless access points Performance rating, for The weighted norm of is a diagonal weight matrix; for The weighted norm of is the position weight coefficient, For wireless access points The physical location coordinates of For wireless access points The physical location coordinates of For the environment function ;in, To control the weight of the directional information, For wireless access points and The angle between the directions, To control the weight of environment similarity; For wireless access points The environmental feature vector, For wireless access points The environmental feature vector of is the bandwidth parameter;

[0038] For wireless access points , define its most similar partition value K; in the similarity matrix Find the top K most similar wireless access points;

[0039] calculate The average similarity of the performance scores of these K wireless access points ; Wireless Access Point The abnormal assessment factor of

[0040] The method of determining the suspicious outlier point and the monitoring neighborhood radius of the suspicious outlier point includes:

[0041] Sort the abnormality assessment factors in descending order, take the top N3% of wireless access points, and mark them as suspicious abnormal points. For each suspicious abnormal point , in the similarity matrix Find the The most similar K wireless access points; calculate The maximum distance to these K wireless access points , maximum distance That is The monitoring neighborhood radius.

[0042] Furthermore, in the similarity matrix Ways to find the top K most similar wireless access points include:

[0043] These K wireless access points are connected to Construct weighted undirected edges, the weight of which is K wireless access points and The performance score similarity between them; repeatedly construct the weighted undirected edges of all wireless access points, merge all the weighted undirected edges, and obtain the weighted K nearest neighbor graph G;

[0044] Obtaining Gaussian graph kernel function based on weighted K nearest neighbor graph G , use the Gaussian kernel function to the similarity matrix Smooth it and get the smoothed similarity matrix ;

[0045] Gaussian kernel function ;in, For wireless access points and The shortest path distance on the weighted K-nearest-neighbor graph G, For wireless access points and The temperature difference between the surroundings, For wireless access points and The difference in network traffic between is the smoothing parameter, is the temperature difference weight parameter, is the traffic difference weight parameter;

[0046] Based on the smoothed similarity matrix , each wireless access point As a single cluster, calculate the distance matrix DLP between all pairs of clusters;

[0047] Elements in the distance matrix DLP ; is the smoothed similarity matrix Elements of

[0048] Find the element with the minimum distance in the distance matrix DLP , at this time and Each corresponds to a separate cluster and , the cluster and Merge into a new cluster and remove the cluster involved from DLP and The rows and columns of

[0049] Compute new clusters With other clusters Distance ;in, For new cluster The number of elements of For other clusters The number of elements of ;

[0050] Distance Add to the distance matrix DLP and build a hierarchical tree according to the order of merging;

[0051] The arithmetic mean of the distance between elements in a cluster is used as the inter-cluster distance value, and the inter-cluster distance values ​​are arranged in the order of merging to construct a curve, where the horizontal axis of the curve is the number of clusters and the vertical axis of the curve is the corresponding merged inter-cluster distance value; two straight lines are fitted on the curve, and the intersection of the two straight lines is the inflection point; the merged inter-cluster distance value at the inflection point is used as the cutting threshold;

[0052] The hierarchical tree is cut at the cutting threshold to form the final clustering cluster. For each wireless access point , within its cluster; according to the smoothed similarity matrix , select the value of The top K most similar wireless access points.

[0053] Furthermore, the construction method of the spatiotemporal sample points includes:

[0054] For each suspected anomaly ,by Centered on As the radius, a monitoring neighborhood sphere area is constructed, namely the monitoring neighborhood; in the monitoring neighborhood, historical performance data within a fixed time is collected and the performance data at each time point is The multidimensional performance index data is used as a sample point ;

[0055] For each sample point Extract spatiotemporal feature vectors, which are obtained by concatenating time features and spatial features; time features include hourly, periodic, and holiday sample points The periodic characteristics of the spatial features include the sample points The environmental characteristics of the space-time feature vector As a space-time sample point.

[0056] Furthermore, the method for performing outlier detection includes:

[0057] All spatiotemporal sample points constitute a spatiotemporal sample point set; define the key parameters of the isolation forest, including the number of trees ntr and the subsampling size sam;

[0058] For each tree, sam spatiotemporal sample points are randomly selected from the spatiotemporal sample point set as the training set;

[0059] For each spatiotemporal sample point y in the training set, randomly select a feature v of one dimension from the spatiotemporal feature vector of y, and randomly generate a segmentation value u within its value range for the selected feature v;

[0060] According to the relationship between the value of y on feature v and the segmentation value u, y is assigned to the left child node or the right child node, and the left child node and the right child node are repeatedly assigned until all the spatiotemporal sample points in the training set are isolated as separate nodes; that is, an isolation path is constructed for each spatiotemporal sample point y;

[0061] Define the length of the isolation path of the spatiotemporal sample point y, pa(y), as the number of splits required from the root node to y; record the length of the isolation path of each spatiotemporal sample point, recorded as the path length; normalize the path length to the range [0, 1]; that is, construct an isolation tree; repeatedly construct ntr isolation trees;

[0062] For each spatiotemporal sample point y, calculate its average path length on all isolation trees ;

[0063] Define the outlier function, the expression of the outlier function is:

[0064] ;in, is the average path length of ntr isolation trees, is the outlier score of the spatiotemporal sample point y;

[0065] For suspicious anomalies , calculate the mean of the outlier scores of all spatiotemporal sample points in its monitoring neighborhood and standard deviation ;

[0066] Suspicious abnormal point The local outlier factor ;in, Suspected abnormal point The maximum outlier score of spatiotemporal sample points in the monitoring neighborhood;

[0067] The method of determining abnormal wireless access points is: presetting an outlier threshold, if a suspected abnormal point The local outlier factor If it is greater than the outlier threshold, a suspicious outlier point is determined It is an abnormal wireless access point.

[0068] Furthermore, the method of optimizing and adjusting the parameters of the abnormal wireless access point includes:

[0069] Use the TR069 protocol to establish a remote management session, connect to the abnormal wireless access point, obtain the current configuration parameters and performance data of the abnormal wireless access point, and analyze the root cause of the problem of the abnormal wireless access point;

[0070] The root causes of the problem include improper parameter configuration, changes in the RF environment, network congestion, and malicious attacks;

[0071] Determine the parameters that need to be optimized and adjusted based on the root cause of the problem; send new optimized parameter configurations to the abnormal wireless access point through the TR069 protocol;

[0072] The abnormal wireless access point accesses the core network through the PON network, and the PON protocol is used to manage and optimize the uplink bandwidth and latency of the abnormal wireless access point. According to the root cause of the problem, the bottleneck point of the PON network resources is identified, which is insufficient uplink bandwidth resources, too high PON network latency, or too high PON network packet loss rate.

[0073] If the identified bottleneck is insufficient uplink bandwidth resources, the uplink bandwidth share of the abnormal wireless access point is increased and the bandwidth of other wireless access points is reduced;

[0074] If the identified bottleneck is that the PON network latency is too high, the scheduling and caching strategies of the PON network are optimized for the uplink traffic of the abnormal wireless access point;

[0075] If the identified bottleneck is that the packet loss rate of the PON network is too high, the abnormal wireless access point will be migrated to other PON network resources.

[0076] The technical effects and advantages of the optical network wireless controller system based on TR069 and PON protocol of the present invention are as follows:

[0077] The present invention can evaluate and monitor the performance status of each access point in the wireless network in real time, discover abnormal points in time, avoid the deterioration and spread of performance problems, and accurately and efficiently locate the real abnormal access points through the intelligent abnormal detection and identification mechanism, reduce false alarms and missed alarms, and improve management efficiency; secondly, with the help of reinforcement learning methods, it can autonomously learn and perceive complex and changeable network environments, adaptively optimize the parameter configuration of wireless access points, and realize intelligent management and optimization of wireless network performance; this autonomous optimization capability can not only reduce the workload of network management personnel, but more importantly, it can quickly make accurate optimization decisions and effectively improve the overall performance of the wireless network; furthermore, by combining with the network management protocol, remote automatic adjustment of abnormal access point parameters is realized, which greatly simplifies the necessity of manual intervention and can significantly improve the performance and management level of the wireless network. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 A schematic diagram of an optical network wireless controller system based on TR069 and PON protocols of the present invention;

[0079] Figure 2 The present invention is a schematic diagram of an operation method of an optical network wireless controller system based on TR069 and PON protocols. DETAILED DESCRIPTION

[0080] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions 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.

[0081] Example 1

[0082] See also Figure 1 As shown, the optical network wireless controller system based on TR069 and PON protocol described in this embodiment includes:

[0083] Data collection module, used to obtain real-time relevant data of each wireless access point;

[0084] The performance scoring module is used to calculate the performance score of each wireless access point based on real-time relevant data through a preset performance evaluation model;

[0085] An anomaly assessment module determines an anomaly assessment factor of each wireless access point based on the performance score of each wireless access point, and determines a suspected anomaly point and a monitoring neighborhood radius of the suspected anomaly point according to the anomaly assessment factor;

[0086] The outlier detection module is used to construct the spatiotemporal sample points of the suspected outlier points within the monitoring neighborhood formed by the monitoring neighborhood radius of the suspected outlier points, and perform outlier value detection on the spatiotemporal sample points to obtain the local outlier factors corresponding to the suspected outlier points;

[0087] The optimization and adjustment module determines the abnormal wireless access points based on the local outlier factors corresponding to the suspicious abnormal points, and optimizes and adjusts the parameters of the abnormal wireless access points through the TR069 protocol and the PON protocol; each module is connected through wired and / or wireless means to realize data transmission between modules.

[0088] Furthermore, the real-time related data includes the load of the wireless access point, the number of users, the throughput, the delay and the bit error rate.

[0089] The load data of the wireless access point is obtained by regularly reporting the load status using the load monitoring function of the wireless access point itself or by querying and obtaining it through the management system of the wireless access point; the number of users of the wireless access point is directly obtained using the user equipment data associated with the wireless access point; the throughput data of the wireless access point is obtained by reporting and obtaining it using the throughput statistics function of the wireless access point itself; the delay data of the wireless access point is obtained by using the delay measurement mechanism, such as the Ping test; the bit error rate data of the wireless access point is obtained by reporting and obtaining it using the signal quality monitoring function of the wireless access point.

[0090] Furthermore, the preset method of the performance evaluation model includes:

[0091] Collect historical operation data of N groups of wireless access points. The data types of historical operation data and real-time related data are the same, including five indicators: load, number of users, throughput, delay and bit error rate.

[0092] Each set of historical operation data of the wireless access point is represented as a multidimensional state vector. The dimension of the multidimensional state vector is equal to the total number of indicators. Each dimension corresponds to an indicator, that is, the dimension of the multidimensional state vector is 5; the multidimensional state vector is used as the state corresponding to the historical operation data.

[0093] Define a series of parameter adjustment operations that can be performed by the wireless access point, discretize the parameter adjustment operations, such as increasing the transmit power, decreasing the transmit power, switching channels, etc.; the obtained discrete actions constitute a finite discrete action set; regard the parameter adjustment operation as a continuous value, such as adjusting the specific value of the transmit power within a certain range; that is, a continuous action set, and the spatial dimension of the continuous action is equal to the number of parameter adjustment operations; for example, if the two actions of adjusting the transmit power and switching channels are considered, the space of discrete actions is {increase power, decrease power, do not adjust power, switch channels}; the space of continuous actions is [transmit power adjustment value, channel index], and the two dimensions correspond to power and channel adjustments respectively; that is, the training data set is obtained.

[0094] Define the reward and punishment function ;

[0095] ;in, , and It is an adaptive weight, which is learned through the attention mechanism; For status, For action, Status Next action The throughput after Status Next action After the delay, Status Next action The bit error rate after is the state value function.

[0096] State Value Function ;in, is the discount factor, which controls the degree of attenuation of future returns. For the status Next select action The probability of is the state action value function, calculated in state Next action long-term returns after is the entropy regularization coefficient, which controls the exploration and exploitation trade-off of the strategy; is the policy (the strategy for choosing an action in each state) In Status The entropy value under reflects the randomness of the strategy.

[0097] State-Action-Value Function ;in, is the strategy discount factor, which is a value between 0 and 1 and is used to control the degree of attenuation; For the status Next action Then transfer to the next state The transition probability reflects the dynamic process of the environment and is a probability distribution; For the next state The state value function of For the status Next action The reward value obtained immediately after the action is taken (a scalar value, usually a positive or negative value used to evaluate how well the agent performs an action in the current state).

[0098] ;in, For the status Next action The weights are used to adjust the contribution of different states and actions to entropy.

[0099] Transition probability Ways to obtain include:

[0100] Construct a differentiable environment model, use M groups of fixed strategies (such as random strategies) to interact with the environment in a real environment or a simulated environment, and record N1 groups of state-action-next-state triplets As training data for a differentiable environment model.

[0101] The fully connected network is used as the backbone network of the differentiable environment model, and the state and actions As input, the next state The probability distribution of is used as output; train the differentiable environment model using training data; define the objective function of the differentiable environment model , use the back propagation algorithm to update the parameters of the backbone network, repeat the training until the output value of the objective function no longer changes, and then the differentiable environment model is built; use the constructed differentiable environment model to fit the transition probability in the real environment .

[0102] Objective Function The expression is:

[0103] ;in, is the next state output by the differentiable environment model The probability distribution of The next state is true The probability distribution of (manually set labels); is the divergence loss; is a differentiable environment model, outputting the next state The predicted value of for The entropy value (calculated in the same way as strategy In Status The entropy value is calculated in the same way as above); and is a hyperparameter used to weigh the importance of different loss terms.

[0104] Construct a policy network. The goal of the policy network is to learn a mapping function. The input of the policy network is defined as the state, and the output is the probability distribution of each executable action and constitutes the optimal action sequence.

[0105] The network structure of the policy network is defined as an input layer, M1 fully connected hidden layers and an output layer; the input layer is used to receive a multi-dimensional state vector; each of the M1 fully connected hidden layers is connected to BatchNorm and ReLU activation; for the space of discrete actions, the output layer uses SoftMax to output the probability of each action; for the space of continuous actions, the output layer uses the mean and variance of the Gaussian distribution as output.

[0106] Initialize the parameters of the policy network (initialization method such as Xavier or He initialization); divide the states and actions in the training data set into P training batches to train the policy network.

[0107] And update the parameters of the policy network during the training process to minimize the policy objective function ;

[0108] LU=E[min(r( i ) ∙ R,cl(r( i ),1 - ϵ ,1+ ϵ ) ∙ R)] ;in, E [ ∙] is the expectation, that is, averaged over all state-action pairs; is the strategy ratio, are the parameters of the policy network.

[0109] Strategy Ratio ;in, The parameters of the policy network are new parameters When the policy network is in state Next select action The probability of The parameters of the policy network are the old parameters When the policy network is in state Next select action The probability of is the reward and punishment function, In order to update the hyperparameters and control the amplitude of the policy update, the policy network is repeatedly trained until the policy network converges (the accumulated reward and punishment values ​​no longer increase significantly); the policy network outputs an optimal action sequence based on the current state.

[0110] Defining performance evaluation functions ; The policy network and performance evaluation function After integration, we get the performance evaluation model.

[0111] ;in, is the throughput of the wireless access point when executing the actions in the optimal action sequence in the current state, is the delay of the wireless access point when executing the actions in the optimal action sequence in the current state, is the bit error rate of the wireless access point when executing the actions in the optimal action sequence in the current state, and the value range is [0, 1]; is the signal-to-noise ratio of the wireless access point when executing the actions in the optimal action sequence in the current state, It is the load level of the wireless access point when executing the actions in the optimal action sequence under the current state, and its value range is [0, 1].

[0112] The current state of each wireless access point is input into the policy network, an optimal action sequence is output, and the current state of each wireless access point (a multidimensional state vector composed of real-time related data) and the optimal action sequence output by the policy network are substituted into the performance evaluation function , calculate and obtain the performance score corresponding to each wireless access point.

[0113] Furthermore, the abnormality assessment factor is determined in the following manner:

[0114] Calculate the distance between any two wireless access points and The performance score similarity between , construct a similarity matrix , similarity matrix Elements within .

[0115] ;in, For wireless access points Performance rating, For wireless access points Performance rating, for The weighted norm of is a diagonal weight matrix used to adjust the importance of different dimensions; for The weighted norm of is the position weight coefficient, For wireless access points The physical location coordinates of For wireless access points The physical location coordinates of For the environment function ; in, To control the weight of the directional information, For wireless access points and The angle between the directions, To control the weight of environment similarity; For wireless access points The environmental feature vector, For wireless access points The environmental feature vector of is the bandwidth parameter, which controls the smoothness of the function.

[0116] It should be noted that the existing measurement data of the wireless access point, such as signal strength, interference level, user distribution, etc., are used to extract features from these measurement data to construct an environmental feature vector.

[0117] For wireless access points , define its most similar partition value K; in the similarity matrix Find the top K most similar wireless access points.

[0118] Specifically, these K wireless access points are connected to Construct weighted undirected edges, the weight of which is K wireless access points and The performance score similarity between them is calculated; weighted undirected edges of all wireless access points are repeatedly constructed, and all weighted undirected edges are merged to obtain a weighted K nearest neighbor graph G.

[0119] Obtaining Gaussian graph kernel function based on weighted K nearest neighbor graph G , use the Gaussian kernel function to the similarity matrix Smooth it and get the smoothed similarity matrix .

[0120] Gaussian kernel function ;in, For wireless access points and The shortest path distance on the weighted K-nearest-neighbor graph G, For wireless access points and The temperature difference between the environments (reflecting the impact of ambient temperature on wireless transmission), For wireless access points and The difference in network traffic between (reflecting the impact of network load on performance); is the smoothing parameter, which controls the smoothness of the Gaussian kernel function. is the temperature difference weight parameter (controls the effect of temperature difference on the Gaussian kernel function), is the flow difference weight parameter (controls the impact of flow difference on the Gaussian kernel function).

[0121] Based on the smoothed similarity matrix , each wireless access point As a single cluster, calculate the distance matrix DLP between all pairs of clusters.

[0122] Elements in the distance matrix DLP ; is the smoothed similarity matrix elements.

[0123] Find the element with the minimum distance in the distance matrix DLP , at this time and Each corresponds to a separate cluster and , the cluster and Merge into a new cluster and remove the cluster involved from DLP and rows and columns.

[0124] Compute new clusters With other clusters Distance ;in, For new cluster The number of elements of For other clusters The number of elements.

[0125] Distance Add to the distance matrix DLP and build a hierarchical tree (dendrogram or hierarchical tree) according to the order of merging.

[0126] The arithmetic mean of the distances between elements within a cluster is used as the inter-cluster distance value, and the inter-cluster distance values ​​are arranged in the order of merging to construct a curve with the horizontal axis of the curve being the number of clusters and the vertical axis of the curve being the corresponding merged inter-cluster distance value; two straight lines are fitted on the curve, and the intersection of the two straight lines is the inflection point; the merged inter-cluster distance value at the inflection point is used as the cutting threshold.

[0127] The hierarchical tree is cut at the cutting threshold to form the final clustering cluster. For each wireless access point , within its cluster; according to the smoothed similarity matrix , select the value of The top K most similar wireless access points.

[0128] The definition of the most similar partition value K includes:

[0129] A multi-dimensional environmental feature vector of a wireless access point is defined. The multi-dimensional environmental feature vector includes a vector composed of geographical location, interference noise level, user distribution density, and weather conditions (artificially divided into condition level numbers).

[0130] Based on the multi-dimensional environmental feature vectors of the wireless access points, the environmental similarity between any two wireless access points (Euclidean distance between the multi-dimensional environmental feature vectors) is calculated.

[0131] For each wireless access point, find the first M4 wireless access points with the highest environmental similarity among the remaining wireless access points, where M4 is a preset maximum value; calculate the average of the performance score similarity between these M4 wireless access points and the current wireless access point, which is recorded as Sic.

[0132] Define an adaptive function FUL, whose input is Sic and output is K value; the adaptive function can be a model obtained by supervised learning or unsupervised learning, such as Gaussian process regression model.

[0133] For each wireless access point, its Sic is substituted into the adaptive function FUL to obtain the corresponding K value.

[0134] calculate The average similarity of the performance scores of these K wireless access points ; Wireless Access Point The abnormal assessment factor.

[0135] Sort the abnormality assessment factors in descending order, take the top N3% of wireless access points, and mark them as suspicious abnormal points. For each suspicious abnormal point , in the similarity matrix Find the The most similar K wireless access points; calculate The maximum distance to these K wireless access points , maximum distance That is The monitoring neighborhood radius.

[0136] Furthermore, the construction method of the spatiotemporal sample points includes:

[0137] For each suspected anomaly ,by Centered on As the radius, a monitoring neighborhood sphere area is constructed, namely the monitoring neighborhood; in the monitoring neighborhood, historical performance data within a fixed time is collected and the performance data at each time point is The multidimensional performance index data is used as a sample point ; Historical performance data includes overall throughput, air interface latency, uplink and downlink bit error rate / packet loss rate, reference signal received power and operating frequency band.

[0138] For each sample point Extract spatiotemporal feature vectors, which are obtained by concatenating time features and spatial features; time features include hourly, periodic, and holiday sample points The periodic characteristics of the spatial features include the sample points Environmental characteristics (using multi-source heterogeneous data such as geographic information system (GIS) data, meteorological data, demographic data, network planning data, etc., to extract the corresponding environmental characteristics and encode them into environmental feature vectors of sample points through feature engineering); the sample points endowed with spatiotemporal feature vectors As a space-time sample point.

[0139] Furthermore, the method for performing outlier detection includes:

[0140] All spatiotemporal sample points constitute a spatiotemporal sample point set; define the key parameters of the isolation forest, including the number of trees ntr and the subsampling size sam.

[0141] For each tree, sam spatiotemporal sample points are randomly selected from the spatiotemporal sample point set as the training set;

[0142] For each spatiotemporal sample point y in the training set, a feature v of one dimension is randomly selected from the spatiotemporal feature vector of y, and for the selected feature v, a segmentation value u is randomly generated within its value range.

[0143] According to the relationship between the value of y on feature v and the split value u, y is assigned to the left child node or the right child node (for example, if the value of y on feature v is greater than the split value u, y is assigned to the left child node). The left and right child nodes are assigned repeatedly until all spatiotemporal sample points in the training set are isolated as separate nodes or the preset path length limit is reached; that is, an isolation path is constructed for each spatiotemporal sample point y.

[0144] Define the length of the isolation path of the spatiotemporal sample point y, pa(y), as the number of splits required from the root node to y; record the length of the isolation path of each spatiotemporal sample point, recorded as the path length; normalize the path length to the range [0, 1]; that is, build an isolation tree; repeatedly build ntr isolation trees; each tree uses a different random number seed, thus generating a different feature / segmentation value selection sequence; for each spatiotemporal sample point y, calculate its average path length on all isolation trees .

[0145] Define the outlier function, the expression of the outlier function is:

[0146] ;in, is the average path length of ntr isolation trees, used for normalization, is the outlier fraction of the spatiotemporal sample point y.

[0147] For suspicious anomalies , calculate the mean of the outlier scores of all spatiotemporal sample points in its monitoring neighborhood and standard deviation ; then the suspicious abnormal point The local outlier factor ;in, Suspected abnormal point The maximum outlier score of spatiotemporal sample points in the monitoring neighborhood.

[0148] Preset outlier threshold, if suspicious outliers The local outlier factor If it is greater than the outlier threshold, a suspicious outlier point is determined It is an abnormal wireless access point.

[0149] Furthermore, the method of optimizing and adjusting the parameters of the abnormal wireless access point includes:

[0150] Use the TR069 protocol to establish a remote management session, connect to the abnormal wireless access point, obtain the current configuration parameters and performance data of the abnormal wireless access point, and analyze the root cause of the problem of the abnormal wireless access point.

[0151] Specifically, the current configuration parameters include wireless parameters, network parameters, security parameters and system parameters; wireless parameters include working frequency band, channel, bandwidth, transmission power, antenna parameters, etc.; network parameters include IP address, gateway, VLAN configuration, QoS policy, etc.; security parameters include encryption method, key update policy, etc.; system parameters include device model, software version, startup time, etc.

[0152] Performance data includes radio frequency data, throughput data, latency data, error data, and fault data; error data includes packet loss rate, bit error rate, etc.; fault data includes the number of abnormal restarts, memory usage, etc.

[0153] Compare and analyze the obtained configuration parameters and performance data to check whether the configuration parameters are reasonable, such as whether there are conflicts in power, channel, etc.; analyze the performance data to check whether there are abnormal fluctuations, such as abnormally increased latency or packet loss rate; compare the performance of adjacent wireless access points to check whether there are external factors such as environmental interference.

[0154] The root causes of the problem include improper parameter configuration, changes in the RF environment, network congestion and malicious attacks; improper parameter configuration, such as excessive power causing interference, channel overlap, etc., changes in the RF environment, such as new interference sources, obstructions affecting coverage, etc.; network congestion, such as insufficient uplink bandwidth resources, routing loops, etc.; malicious attacks, such as attacks on wireless air interfaces, etc.

[0155] According to the root cause of the problem, determine the parameters that need to be optimized, such as channel, power, bandwidth, etc.; use the TR069 protocol to send new optimization parameter configurations to abnormal wireless access points, and the wireless access points apply the new parameter configurations to achieve performance optimization.

[0156] Abnormal wireless access points access the core network through the PON network, and use the PON protocol to manage and optimize the uplink bandwidth and latency of abnormal wireless access points; dynamically adjust PON network resource allocation based on the root cause of the problem.

[0157] Specifically, according to the root cause of the problem, the bottleneck point of the PON network resources is identified, and the bottleneck point is insufficient upstream bandwidth resources, too high PON network delay or too high PON network packet loss rate.

[0158] If the identified bottleneck is insufficient uplink bandwidth resources, the uplink bandwidth share of the abnormal wireless access point is increased, and the bandwidth of other low-priority wireless access points is reduced;

[0159] If the identified bottleneck is that the PON network latency is too high, the scheduling and caching strategies of the PON network are optimized for the uplink traffic of the abnormal wireless access point;

[0160] If the identified bottleneck is that the packet loss rate of the PON network is too high, the abnormal wireless access point will be migrated to other PON network resources.

[0161] This embodiment can evaluate and monitor the performance status of each access point in the wireless network in real time, discover abnormal points in time, avoid the deterioration and spread of performance problems, and accurately and efficiently locate the real abnormal access points through the intelligent abnormal detection and identification mechanism, reduce false alarms and missed alarms, and improve management efficiency; secondly, with the help of reinforcement learning methods, it can autonomously learn and perceive complex and changeable network environments, adaptively optimize the parameter configuration of wireless access points, and realize intelligent management and optimization of wireless network performance; this autonomous optimization capability can not only reduce the workload of network management personnel, but more importantly, it can quickly make accurate optimization decisions and effectively improve the overall performance of the wireless network; furthermore, by combining with the network management protocol, remote automatic adjustment of abnormal access point parameters is realized, which greatly simplifies the necessity of manual intervention and can significantly improve the performance and management level of the wireless network.

[0162] Example 2

[0163] See also Figure 2 As shown, the part not described in detail in this embodiment is described in Example 1, which provides an operation method of an optical network wireless controller system based on TR069 and PON protocols, including:

[0164] S1. Obtain real-time relevant data of each wireless access point;

[0165] S2. Calculate the performance score of each wireless access point based on the real-time relevant data and through a preset performance evaluation model;

[0166] S3, based on the performance score of each wireless access point, determining the abnormality assessment factor of each wireless access point, and determining the suspected abnormal point and the monitoring neighborhood radius of the suspected abnormal point according to the abnormality assessment factor;

[0167] S4. Constructing spatiotemporal sample points of the suspected abnormal point within the monitoring neighborhood formed by the monitoring neighborhood radius of the suspected abnormal point, and performing outlier detection on the spatiotemporal sample points to obtain the local outlier factor corresponding to the suspected abnormal point;

[0168] S5. Based on the local outlier factors corresponding to the suspicious abnormal points, the abnormal wireless access points are determined, and the parameters of the abnormal wireless access points are optimized and adjusted through the TR069 protocol and the PON protocol.

[0169] Example 3

[0170] The present embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the operation mode of the optical network wireless controller system operation method based on TR069 and PON protocols provided above is implemented.

[0171] Since the electronic device introduced in this embodiment is an electronic device used to implement the optical network wireless controller system operation method based on TR069 and PON protocols in the embodiment of the present application, based on the optical network wireless controller system operation method based on TR069 and PON protocols introduced in the embodiment of the present application, the technical personnel of the field can understand the specific implementation of the electronic device of the present embodiment and its various variations, so how the electronic device implements the method in the embodiment of the present application is not described in detail here. As long as the technical personnel of the field implement the electronic device used in the optical network wireless controller system operation method based on TR069 and PON protocols in the embodiment of the present application, it belongs to the scope of protection of this application.

[0172] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0173] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. An optical network wireless controller system, characterized in that: include: Data collection module, used to obtain real-time relevant data of each wireless access point; The performance scoring module is used to calculate the performance score of each wireless access point based on real-time relevant data through a preset performance evaluation model; An anomaly assessment module determines an anomaly assessment factor of each wireless access point based on the performance score of each wireless access point, and determines a suspected anomaly point and a monitoring neighborhood radius of the suspected anomaly point according to the anomaly assessment factor; The outlier detection module is used to construct the spatiotemporal sample points of the suspected outlier points within the monitoring neighborhood formed by the monitoring neighborhood radius of the suspected outlier points, and perform outlier value detection on the spatiotemporal sample points to obtain the local outlier factors corresponding to the suspected outlier points; The optimization and adjustment module determines the abnormal wireless access points based on the local outlier factors corresponding to the suspicious abnormal points, and optimizes and adjusts the parameters of the abnormal wireless access points; Each module is connected via wired and / or wireless means; The preset method of the performance evaluation model includes: Collecting historical operation data of N groups of wireless access points; N is a positive integer greater than 1; representing each group of historical operation data of the wireless access point as a multi-dimensional state vector; and using the multi-dimensional state vector as a state corresponding to the historical operation data; Define a series of parameter adjustment operations that can be performed by the wireless access point, discretize the parameter adjustment operations, and the obtained discrete actions constitute a finite discrete action set; regard the parameter adjustment operations as continuous values, that is, a continuous action set; the discrete action set, the continuous action set and the state constitute a training data set; Define the reward and punishment function R; R = β1·log(T(s,a))+β2·exp(-D(s,a))+β3·(1-P(s,a))-V(s); where β1, β2 and β3 are adaptive weights; s is the state, a is the action, T(s,a) is the throughput after action a is performed in state s; D(s,a) is the delay after action a is performed in state s, P(s,a) is the bit error rate after action a is performed in state s; V(s) is the state value function; State Value Function ; Where γ is the discount factor, π(a|s) is the probability of selecting action a in state s, and Q(s, a) is the state-action value function; is the entropy regularization coefficient; H(π(·|s)) is the entropy value of strategy π in state s; State action value function Q(s,a)=f(s,a)+δ×∑(PL(s′|s,a)×V(s′)); where, is the strategy discount factor; PL(s′|s, a) is the transition probability of transferring to the next state s′ after executing action a in state s; V(s′) is the state value function of the next state s′; f(s, a) is the reward value immediately obtained after executing action a in state s; H(π(·|s))=-∑(π(a|s)×log(π(a|s)×ω(s,a)));where ω(s,a) is the weight of executing action a in state s; Construct a policy network and define a performance evaluation function FL; integrate the policy network and the performance evaluation function FL to obtain a performance evaluation model; The performance score is calculated by: The current state of each wireless access point is input into the policy network, an optimal action sequence is output, and the current state of each wireless access point and the optimal action sequence output by the policy network are substituted into the performance evaluation function FL to calculate the performance score corresponding to each wireless access point.

2. The optical network wireless controller system according to claim 1, characterized in that: The real-time related data includes the load of the wireless access point, the number of users, throughput, delay and bit error rate.

3. The optical network wireless controller system according to claim 2, characterized in that: The acquisition method of the transition probability PL(s′|s, a) includes: Construct a differentiable environment model. In a real environment or a simulated environment, use M groups of fixed strategies to interact with the environment, where M is a positive integer greater than 1. Record N1 groups of state-action-next-state triplets (s, a, s′) as training data for the differentiable environment model. N1 is a positive integer greater than 1. The fully connected network is used as the backbone network of the differentiable environment model, the state s and action a are used as input, and the probability distribution of the next state s′ is used as output; the differentiable environment model is trained using training data; the objective function LP of the differentiable environment model is defined, and the parameters of the backbone network are updated using the back propagation algorithm. The training is repeated until the output value of the objective function no longer changes, and the differentiable environment model is constructed; the constructed differentiable environment model is used to fit the transition probability PL(s′|s, a) in the real environment; The expression of the objective function LP is: LP = ∑(||s′-Di(s, a)|| 2 +α1·KL(P′(s′|s,a)||QL(s′|s,a))+α2·H(P′(s′|s,a))); where P′(s′|s,a) is the probability distribution of the next state s′ output by the differentiable environment model; QL(s′|s,a) is the probability distribution of the actual next state s′; KL(·) is the divergence loss; Di(s,a) is the differentiable environment model, outputting the predicted value of the next state s′; H(P′(s′|s,a)) is the entropy value of P′(s′|s,a); α1 and α2 are hyperparameters.

4. The optical network wireless controller system according to claim 3, characterized in that: The method of constructing the policy network includes: Define the input of the policy network as the state, and the output as the probability distribution of each executable action and form the optimal action sequence; The network structure of the policy network is defined as follows: an input layer, M1 fully connected hidden layers, and an output layer; the input layer is used to receive a multi-dimensional state vector; each of the M1 fully connected hidden layers is connected to BatchNorm and ReLU activation; for the discrete action space, the output layer uses SoftMax to output the probability of each action; for the continuous action space, the output layer uses the mean and variance of the Gaussian distribution as output; M1 is a positive integer greater than 1; Initialize the parameters of the policy network; divide the states and actions in the training data set into P training batches to train the policy network, and update the parameters of the policy network during the training process to minimize the policy objective function LU; P is a positive integer greater than 1; LU = E[min(r(θ)·R, cl(r(θ), 1-∈, 1+∈)·R)]; where E[·] is the expectation; r(θ) is the policy ratio, θ is the parameter of the policy network, and cl() is a truncation function used to limit the range of the policy ratio, limiting the value of the policy ratio to the interval (1-∈, 1+∈); Strategy Ratio Among them, π(a|s, θ_new) is the probability that the policy network selects action a in state s when the policy network parameter is the new parameter θ_new; π(a|s, θ_old) is the probability that the policy network selects action a in state s when the policy network parameter is the old parameter θ_old; R is the reward and punishment function, ∈ is the update hyperparameter, and the policy network is repeatedly trained until the policy network converges and the policy network is constructed; The expression of the performance evaluation function FL is: FL=w1×log(1+th)-w2×dy-w3×pr+w4×sr-w5×ld 2 ; Wherein, th is the throughput of the wireless access point when executing the actions in the optimal action sequence under the current state, dy is the delay of the wireless access point when executing the actions in the optimal action sequence under the current state, pr is the bit error rate of the wireless access point when executing the actions in the optimal action sequence under the current state; sr is the signal-to-noise ratio of the wireless access point when executing the actions in the optimal action sequence under the current state, ld is the load level of the wireless access point when executing the actions in the optimal action sequence under the current state; w1, w2, w3, w4 and w5 are respectively the weight parameters of the corresponding items of the performance evaluation function FL, which are used to regulate the influence of the corresponding items on the function.

5. The optical network wireless controller system according to claim 4, characterized in that: The abnormal assessment factor is determined in the following manner: Calculate the performance score similarity sim(i, j) between any two wireless access points i and j, and construct a similarity matrix S, where the element S(i, j) in the similarity matrix S=sim(i, j); Where sc(i) is the performance score of wireless access point i, sc(j) is the performance score of wireless access point j, ||sc(i)||_W is the weighted norm of sc(i), W is the diagonal weight matrix; ||sc(j)||_W is the weighted norm of sc(j), T is the location weight coefficient, loc(i) is the physical location coordinate of wireless access point i, loc(j) is the physical location coordinate of wireless access point j, and FD(i, j) is the environment function in, To control the weight of the directional information, PY(i, j) is the directional angle between wireless access points i and j. is the weight for controlling the environmental similarity; env(i) is the environmental feature vector of wireless access point i, env(j) is the environmental feature vector of wireless access point j; ρ1 is the bandwidth parameter; For wireless access point i, define its most similar partition value K; find the first K most similar wireless access points in the similarity matrix S; K is a positive integer greater than or equal to 1; Calculate the average avg(i) of the similarity between i and the performance scores of these K wireless access points; avg(i) is the abnormal evaluation factor of wireless access point i; The method of determining the suspicious outlier point and the monitoring neighborhood radius of the suspicious outlier point includes: Sort the anomaly assessment factors in descending order, and take the top N3% of wireless access points, where N3 is a positive integer greater than 0 and less than 100; mark these wireless access points as suspicious anomaly points, and for each suspicious anomaly point l, find the K wireless access points that are most similar to l in the similarity matrix S; calculate the maximum distance max(l) between l and these K wireless access points, and the maximum distance max(l) is the monitoring neighborhood radius of l.

6. The optical network wireless controller system according to claim 5, characterized in that: The method of finding the top K most similar wireless access points in the similarity matrix S includes: Construct weighted undirected edges between the K wireless access points and i, where the weight of the weighted undirected edge is the performance score similarity between the K wireless access points and i; K is a positive integer greater than or equal to 1; repeatedly construct weighted undirected edges for all wireless access points, merge all weighted undirected edges, and obtain a weighted K nearest neighbor graph G; Based on the weighted K-nearest neighbor graph G, the Gaussian graph kernel function ker is obtained, and the similarity matrix S is smoothed using the Gaussian graph kernel function to obtain a smoothed similarity matrix S′; Gaussian graph kernel function ker=exp(-μ1·(dist(i, j)+ε1·tem(i, j)+ε2·traf(i, j))); where dist(i, j) is the shortest path distance between wireless access points i and j on the weighted K nearest neighbor graph G, tem(i, j) is the temperature difference between the environments where wireless access points i and j are located, and traf(i, j) is the network traffic difference between wireless access points i and j; μ1 is a smoothing parameter, ε1 is a temperature difference weight parameter, and ε2 is a traffic difference weight parameter; Based on the smoothed similarity matrix S′, each wireless access point i is regarded as a separate cluster, and the distance matrix DLP between all pairs of clusters is calculated; The element DLP(i, j) in the distance matrix DLP=1-S′(i, j); S′(i, j) is the element of the smoothed similarity matrix S′; Find the element DLP(i, j) corresponding to the minimum distance in the distance matrix DLP. At this time, i and j correspond to a separate cluster (i) and {j} respectively. Merge clusters (i) and {j} into a new cluster rt, and delete the rows and columns involving clusters (i) and {j} from DLP. Calculate the distance between the new cluster rt and other clusters x Where n_p is the number of elements in the new cluster rt, and n_x is the number of elements in the other cluster x; Add the distance Dis(r, x) to the distance matrix DLP, and build a hierarchical tree according to the order of merging; The arithmetic mean of the distance between elements in a cluster is used as the inter-cluster distance value, and the inter-cluster distance values ​​are arranged in the order of merging to construct a curve, where the horizontal axis of the curve is the number of clusters and the vertical axis of the curve is the corresponding merged inter-cluster distance value; two straight lines are fitted on the curve, and the intersection of the two straight lines is the inflection point; the merged inter-cluster distance value at the inflection point is used as the cutting threshold; The hierarchical tree is cut at the cutting threshold to form the final clustering cluster. For each wireless access point i, within its cluster; according to the value of the smoothed similarity matrix S′, the first K wireless access points most similar to i are selected.

7. The optical network wireless controller system according to claim 6, characterized in that: The construction method of the spatiotemporal sample points includes: For each suspicious outlier point l, a monitoring neighborhood sphere area is constructed with l as the center and max(L) as the radius, namely the monitoring neighborhood; within the monitoring neighborhood, historical performance data within a fixed time is collected, and the multi-dimensional performance indicator data at each time point t is taken as a sample point X(t); For each sample point X(t), a spatiotemporal feature vector is extracted. The spatiotemporal feature vector is obtained by concatenating the time features and space features. The time features include the periodic features of the hourly, periodic and holiday sample points X(t). The space features include the environmental features of the sample point X(t). The sample point X(t) endowed with the spatiotemporal feature vector is used as the spatiotemporal sample point.

8. The optical network wireless controller system according to claim 7, characterized in that: The method for performing outlier detection includes: All spatiotemporal sample points constitute a spatiotemporal sample point set; define the key parameters of the isolation forest, including the number of trees ntr and the subsampling size sam; For each tree, sam spatiotemporal sample points are randomly selected from the spatiotemporal sample point set as the training set; For each spatiotemporal sample point y in the training set, randomly select a feature v of one dimension from the spatiotemporal feature vector of y, and randomly generate a segmentation value u within its value range for the selected feature v; According to the relationship between the value of y on feature v and the segmentation value u, y is assigned to the left child node or the right child node, and the left child node and the right child node are repeatedly assigned until all the spatiotemporal sample points in the training set are isolated as separate nodes; that is, an isolation path is constructed for each spatiotemporal sample point y; Define the length of the isolation path of the spatiotemporal sample point y, pa(y), as the number of splits required from the root node to y; record the length of the isolation path of each spatiotemporal sample point, recorded as the path length; normalize the path length to the range [0, 1]; that is, construct an isolation tree; repeatedly construct ntr isolation trees; For each spatiotemporal sample point y, calculate its average path length CX(y) on all isolation trees; Define the outlier function, the expression of the outlier function is: Among them, CX(ntr) is the average path length of ntr isolation trees, out(y) is the outlier score of spatiotemporal sample point y; For a suspected outlier point l, calculate the mean μ′(l) and standard deviation σ′(l) of the outlier scores of all spatiotemporal sample points in its monitoring neighborhood; Then the local outlier factor of the suspected outlier point l is Among them, max_out_l is the maximum outlier score of the spatiotemporal sample points in the monitoring neighborhood of the suspected outlier point l; The method for determining the abnormal wireless access point is: presetting an outlier threshold, and if the local outlier factor lie(l) of the suspected outlier point l is greater than the outlier threshold, then determining the suspected outlier point l as an abnormal wireless access point.

9. The optical network wireless controller system according to claim 8, characterized in that: The method of optimizing and adjusting the parameters of the abnormal wireless access point includes: Establish a remote management session, connect to the abnormal wireless access point, obtain the current configuration parameters and performance data of the abnormal wireless access point, and analyze the root cause of the problem of the abnormal wireless access point; The root causes of the problem include improper parameter configuration, changes in the RF environment, network congestion, and malicious attacks; Determine the parameters that need to be optimized and adjusted based on the root cause of the problem; send new optimized parameter configurations to abnormal wireless access points; Abnormal wireless access points access the core network through the PON network, and manage and optimize the uplink bandwidth and latency of abnormal wireless access points; identify the bottleneck of PON network resources based on the root cause of the problem, which may be insufficient uplink bandwidth resources, too high PON network latency, or too high PON network packet loss rate; If the identified bottleneck is insufficient uplink bandwidth resources, the uplink bandwidth share of the abnormal wireless access point is increased and the bandwidth of other wireless access points is reduced; If the identified bottleneck is that the PON network latency is too high, the scheduling and caching strategies of the PON network are optimized for the uplink traffic of the abnormal wireless access point; If the identified bottleneck is that the packet loss rate of the PON network is too high, the abnormal wireless access point will be migrated to other PON network resources.

Images