Power service 5G network quality evaluation method based on power virtual private network

Abstract

The invention relates to an electric power service 5G network quality evaluation method based on an electric power virtual private network. The method comprises the following steps: step 1, constructing an electric power service network architecture based on the 5G electric power virtual private network in a plurality of scenes; step 2, carrying out channel modeling on the network architecture to realize uplink and downlink bidirectional reciprocity channel simulation, and realizing a 5G network accurate performance test through a multiple-input multiple-output air interface test; 3, automatically completing the generation and selection of a network index needing to be evaluated according to an evaluation target by adopting machine learning, further obtaining a network feature weight of the selected network index, and obtaining an influence coefficient capable of representing a network feature direction; step 4, performing relatively comprehensive and accurate evaluation on the 5G network by adopting a panoramic 5G network quality evaluation method according to the evaluation index attributes and the classification characteristics; the method has the advantages that the multi-scene electric power virtual private network is constructed, index generation and selection are automatically completed by adopting machine learning, and comprehensive and accurate evaluation of the 5G network is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of 5G network for electric power business, and specifically relates to a 5G network quality assessment method for electric power business based on electric power virtual private network. Background Art

[0002] The current power grid faces problems such as lack of interaction between sources and loads, reduced security dependence and redundant balancing capabilities, and lack of efficiency-enhancing measures. In order to achieve efficient interaction between "source-grid-load-storage", it is necessary to connect massive power terminals in a safe and reliable way. 5G technology can carry the smooth interconnection of all links of power production, transmission and consumption, strongly support the implementation of new power systems with new energy as the main body, and help the construction of energy Internet to carry massive power services. Due to the wide variety of power business scenarios, it is difficult for 5G networks to maintain and control various power business scenarios. In order to ensure that communication network planning can accurately and efficiently meet the actual needs of power application scenarios, it is necessary to conduct quality assessment of the 5G network for power services. At present, the scale of 5G network users and services is growing in a step-by-step manner, but the comprehensive and objective network evaluation index selection and quality assessment method of 5G networks have not yet been clarified; therefore, it is very necessary to provide a 5G network quality assessment method for power services based on power virtual private networks, which can build a variety of power virtual private networks, use machine learning to automatically complete indicator generation and selection, and achieve comprehensive and accurate evaluation of 5G networks. Summary of the invention

[0003] The purpose of the present invention is to overcome the shortcomings of the prior art and to provide a 5G network quality assessment method for power services based on a power virtual private network, which constructs a variety of power virtual private networks, uses machine learning to automatically complete indicator generation and selection, and realizes a comprehensive and accurate assessment of the 5G network.

[0004] The object of the present invention is achieved by: a method for evaluating the quality of a 5G network for electric power services based on an electric power virtual private network, the method comprising the following steps:

[0005] Step 1: Build a 5G power virtual private network-based power business network architecture in multiple scenarios;

[0006] Step 2: Perform channel modeling on the network architecture to achieve uplink and downlink bidirectional reciprocal channel simulation, and perform precise performance testing of the 5G network through multiple-input multiple-output air interface testing;

[0007] Step 3: Use machine learning to automatically generate and select the required evaluation network indicators according to the evaluation objectives, and further obtain the network feature weights of the selected network indicators and the influence coefficients that can represent the direction of the network features;

[0008] Step 4: Use a panoramic 5G network quality assessment method based on the attributes and classification characteristics of the evaluation indicators to conduct a relatively comprehensive and accurate assessment of the 5G network.

[0009] The 5G power virtual private network in step 1 refers to: in the operator's 5G network, based on hard slices with physical isolation strength and soft slices with logical isolation strength, a dedicated network for the power industry is virtualized in the wireless, bearer, core network and other links; the virtual private network is cross-domain integrated with the power communication private network through a secure access platform to achieve end-to-end power business carrying, high-intensity security isolation and resource management.

[0010] The construction of the power business network architecture based on the 5G power virtual private network in multiple scenarios in step 1 specifically includes: load control business scenarios, distribution automation scenarios, smart inspection scenarios, distributed energy scenarios and collection scenarios.

[0011] The step 2 performs channel modeling on the network architecture, realizes uplink and downlink bidirectional reciprocal channel simulation, and reduces the dimension of the channel model through multi-input multi-output air interface test to realize accurate performance test of 5G network. Specifically, a multi-input multi-output air interface, i.e., MIMOOTA test system, is used to realize MIMOOTA channel modeling. To construct the OTA channel model, it is necessary to restore the delay, power, azimuth, cross-polarization ratio and other information of each cluster in a darkroom, and different antenna probes in the OTA darkroom need to be assigned different weight factors, so as to simulate the directional propagation and angle expansion of the common channel data link, i.e., the CDL cluster, and adopt a pre-fading synthesis PFS probe assignment algorithm that is not affected by the phase fluctuation of the system and has stronger robustness.

[0012] In the PFS algorithm, spatial correlation is selected as an evaluation factor for modeling the spatial characteristics of the channel. Spatial correlation is a statistical measure of the similarity of received signals. AoA is expressed in terms of angle φ. p The plane wave is injected into the antenna array with a spacing of d, and the angle between the antenna normal and the horizontal plane is φ a , φ is the angle of the plane wave relative to the antenna element; Ignoring the antenna polarization, for a pair of antennas u and v, assuming that the terminal antenna is an ideal point source antenna, the continuous power angular spectrum under the channel model can be written as: That is, for the defined continuous PAS, the spatial correlation is a function of the distance and relative position between the pairs of virtual antennas; in the multi-probe anechoic chamber MPAC, the continuous PAS in the channel model is simulated by the signal from the discrete multi-probe antenna, which can be derived from the above formula. The spatial correlation function of the discrete antenna probe in the test area is: Among them, ω k is the weight of the Kth probe; the evaluation factor for measuring the OTA modeling standard is the spatial correlation error, i.e., ρ OTA-ρ Target , one limitation that needs to be noted is that the probe weight must satisfy the normalization condition and cannot be less than 0, so the objective function is: Through this scheme, the weight mapping relationship of each cluster in each probe can be calculated, and the corresponding link can be substituted into the channel formula to restore the channel model defined in the standard.

[0013] The step 3 uses machine learning to automatically complete the generation and selection of the required evaluation network indicators according to the evaluation objectives, and further obtains the network feature weights of the selected network indicators and obtains the influence coefficients that can represent the network feature directions, specifically including the following steps:

[0014] Step 3.1: Machine learning and gradient boosting model;

[0015] Step 3.2: Modeling and implementation.

[0016] The machine learning and gradient boosting model in step 3.1 is as follows: for a sample set R d The i-th network characteristic index x in i , which can be expressed as x i ∈R d , for a multivariate linear model, the predicted value of its objective function index is It can be expressed as: where w i is the learning parameter of the type; for the linear model, the learning parameter Θ can be expressed as: Θ={w j j=1,...,d}, where the i-th network characteristic index j is the model parameter w i number; the objective function can be expressed as: Obj(Θ)=L(Θ)+Ω(Θ)(3), where L(Θ) is the model loss function; Ω(Θ) is the type regularization term; the loss function If the loss number adopts the square loss function, then The XGBoost model is an improved algorithm based on the GBDT principle. XGBoost is based on the decision tree model, and learns tree by tree. It uses a greedy algorithm to iterate successively. Each tree fits the deviation of the previous model. Formula (1) is the predicted value of the objective function indicator. It can be expressed as: Where K is the number of leaf nodes; f k ∈R d ; At this time, the objective function formula (3) can be expressed as: If used represents the predicted value of the objective function indicator of the t-th round of iteration algorithm, is the predicted value of the objective function indicator in the previous round, then It can be expressed as In this way, the above formula can be expressed as: Where C is a constant term; after the objective function is transformed by "Taylor second-order expansion", the new objective function can be expressed as follows: in XGBoost calculates which feature to select as the scoring point based on the gain of the model structure score, and the importance of a feature is the sum of the number of times it appears in all trees. In other words, the more an attribute is used to build decision trees in the model, the higher its importance is relatively. The Gain in the XGBoost model is defined by the following formula: in is the score of the left leaf node; is the score of the right leaf node; The score that can be obtained without segmentation; γ is the complexity cost after the introduction of the new leaf node; where G j , H j Define the sum of the first-order and second-order derivatives of each leaf node respectively, and the scores of the left and right leaf nodes can be respectively represented by G j , H j Find, where G j , H j Calculated by the following formula: The network indicator feature importance FI coefficient value is the feature score of each network feature indicator given by the XGBoost model during the training process, indicating the importance coefficient of the feature of each network indicator to the model training.

[0017] The modeling and implementation in step 3.2 specifically include the following steps:

[0018] Step 3.21: Construct the first network indicator set: Generally speaking, a branch company or an independent network construction unit is selected as the evaluation object; the original network indicator set includes a series of network characteristic indicator names and network characteristic value sets, the number of network characteristic indicators is represented by k, the number of objective function indicators is represented by n, and the number of set samples is represented by m; f x is the name of the network characteristic indicator; i is the name of the objective function; assuming that the evaluation function F(x) = Y, the independent variable x is the network characteristic index; Y is the objective function;

[0019] Step 3.22: Network indicator processing: First, according to the evaluation goal, select an objective function indicator y from the first network indicator set * , forming a second network indicator set; the second network indicator set includes n *network feature indicators and 1 objective function indicator; then use the first feature selection model to apply to the second network indicator set to determine whether the network feature parameters appear close to 0, forming a second network indicator set preferably 1; use the second feature selection model to apply to the network indicator set to determine the FI coefficient value of the importance of the network indicator characteristics in the network indicator set to form a second network indicator set preferably 2 or directly form a third network indicator set, and form a third network indicator set with the second network indicator set preferably 1, and use the FI coefficient value to determine the network indicator weights in the third network indicator set.

[0020] In step 4, a panoramic 5G network quality evaluation method is used according to the evaluation index attributes and classification characteristics to perform a relatively comprehensive and accurate evaluation of the 5G network, specifically including: a static threshold linear algorithm, a single dynamic threshold linear algorithm and a dual dynamic threshold linear algorithm; the static threshold linear algorithm refers to a linear algorithm with fixed upper and lower thresholds, and has a clear indicator quality measurement standard for traditional network quality and performance indicators. The algorithm formula is: subjectto(α+β)=1, where eva_score is the evaluation score, and each specific indicator is calculated according to the formula; ParaValue parameter value is the current status of the indicator, and each specific indicator outputs the latest data according to the statistical period; Baseline basic threshold is the lowest standard, and the basic level of the evaluation area is evaluated. The minimum value of the theoretical minimum standard and the current network basic capability level is adopted. Each specific indicator that does not reach the basic threshold will not score; Challengeline challenge threshold is the highest standard, and the minimum value of the theoretical highest standard and the highest standard that can be achieved by the network is adopted. Each specific indicator that reaches the challenge threshold will get full score full marks The α and β linear control factors are used to control the linear evaluation score base of the indicator and the influence of the indicator on the total score of the overall dimension, respectively. For the indicators determined by the upper and lower thresholds, the focus is on quality stability assessment.

[0021] The single dynamic threshold linear algorithm refers to a linear algorithm in which one of the upper and lower thresholds changes dynamically. For network construction and service benchmarking indicators, it is necessary to comprehensively consider factors such as the overall construction scale and changes in competitors, and adjust the basic threshold in a timely manner. Therefore, a single dynamic evaluation standard measurement method is adopted, and the algorithm formula is: subjectto(α+β)=1, where dy_Baseline dynamic basic threshold value adopts the average level of the benchmark object or the basic capability level of the current network to adaptively reflect the benchmark result; indicators with step-by-step improvement of the basic threshold are generally the embodiment of the quality of phased changes, focusing on the evaluation of quality improvement; the dual dynamic threshold linear algorithm refers to a linear algorithm with dynamic changes in both upper and lower thresholds. For market development and benefit indicators, it is necessary to comprehensively consider regional economy, user groups, consumption capacity, behavioral preferences, and local network carrying capacity. Therefore, the dual dynamic evaluation standard measurement method is adopted, and the algorithm formula is: Among them, dy_Baseline is the dynamic basic threshold value, which is determined by the ranking of the corresponding areas within the accounting period, and adopts the indicator value of the worst ranking area; dy_Challengeline is the dynamic challenge threshold value, which is determined by the ranking of the corresponding areas within the accounting period, and adopts the indicator value of the best ranking area.

[0022] Beneficial effects of the present invention: The present invention is a 5G network quality assessment method for power services based on power virtual private networks. During use, the present invention provides a feasible communication method for the power grid to safely and reliably access massive terminals by constructing virtual private networks in different scenarios, thereby ensuring the stable operation of the power grid in diversified scenarios; the present invention uses machine learning to automatically complete the generation and selection of network indicators required for evaluation according to the evaluation objectives, and further obtains the network feature weights of the selected network indicators, and obtains the influence coefficients that can represent the direction of network features, which can be used for 5G network evaluation, especially for multi-objective and more factor-based quantitative evaluation of networks for various types of network features; the present invention constructs a 5G panoramic picture evaluation system to comprehensively evaluate the 5G network, and according to the formulation of a specific scoring and evaluation system, makes a relatively accurate evaluation of the current network and business development, meeting the needs of 5G power business network operation and maintenance optimization; the present invention has the advantages of constructing power virtual private networks in various scenarios, automatically completing indicator generation and selection using machine learning, and achieving comprehensive and accurate evaluation of 5G networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a load control business scenario architecture diagram of the present invention.

[0024] Figure 2 This is a diagram of the distribution automation scenario architecture of the present invention.

[0025] Figure 3 This is the intelligent inspection scenario architecture diagram of the present invention.

[0026] Figure 4 This is a distributed energy scenario architecture diagram of the present invention.

[0027] Figure 5This is a diagram of the collection scenario architecture of the present invention.

[0028] Figure 6 Schematic diagram of OTA probe configuration of the present invention (θ i is the angular position of the ith probe).

[0029] Figure 7 This is a flow chart of the network indicator processing method of the present invention. DETAILED DESCRIPTION

[0030] The present invention will be further described below in conjunction with the accompanying drawings.

[0031] Example 1

[0032] like Figure 1-7 As shown, a method for evaluating the quality of a 5G network for electric power services based on an electric power virtual private network comprises the following steps:

[0033] Step 1: Build a 5G power virtual private network-based power business network architecture in multiple scenarios;

[0034] Step 2: Perform channel modeling on the network architecture to achieve uplink and downlink bidirectional reciprocal channel simulation, and perform precise performance testing of the 5G network through multiple-input multiple-output air interface testing;

[0035] Step 3: Use machine learning to automatically generate and select the required evaluation network indicators according to the evaluation objectives, and further obtain the network feature weights of the selected network indicators and the influence coefficients that can represent the direction of the network features;

[0036] Step 4: Use a panoramic 5G network quality assessment method based on the attributes and classification characteristics of the evaluation indicators to conduct a relatively comprehensive and accurate assessment of the 5G network.

[0037] The 5G power virtual private network in step 1 refers to: in the operator's 5G network, based on hard slices with physical isolation strength and soft slices with logical isolation strength, a dedicated network for the power industry is virtualized in the wireless, bearer, core network and other links; the virtual private network is cross-domain integrated with the power communication private network through a secure access platform to achieve end-to-end power business carrying, high-intensity security isolation and resource management.

[0038] The construction of the power business network architecture based on the 5G power virtual private network in multiple scenarios in step 1 specifically includes: load control business scenarios, distribution automation scenarios, smart inspection scenarios, distributed energy scenarios and collection scenarios.

[0039] In this embodiment, ① load control business scenario: the load control business aims to gather massive interruptible load resources at all levels of society, so as to realize a power grid in which massive resources are awakened, "source-grid-load-storage" is fully interactive, and safety and efficiency are improved; by deploying an interruptible load regulation function master station on the provincial company side, deploying an interruptible load regulation function substation on the prefecture-level city side, and carrying out the construction of a second-level interruptible load access function on the user side; when there is a large shortage of power supply and demand, a load regulation instruction is issued to the pre-selected interruptible load contracted users to realize second-level load regulation, ensure rapid restoration of power supply and demand balance, and ensure stable operation of the power grid; the overall architecture of the load control business scenario is as follows Figure 1 As shown in the figure, the interruptible load rapid response system adopts a hierarchical structure design. The system is divided into three levels: main station layer, municipal substation layer and user access layer. The three levels realize the interaction of load models and real-time data adjustment strategies, and jointly complete the rapid and accurate removal of interruptible loads when an accident occurs. The communication link between the load adjustment substation and the load control terminal is composed of a power communication private network and an operator's 5G virtual private channel. According to the network structure of the load control system, reasonable network boundaries are divided, and corresponding levels of security protection measures are taken at the boundaries; including: 1) Boundaries of the load adjustment master station and the multi-integrated high-elasticity power grid smart dispatching platform: The provincial load adjustment master station and the provincial multi-integrated high-elasticity power grid smart dispatching platform should deploy power with access control function. 1) The load regulation substation and the provincial load regulation master station have longitudinal boundaries. Gigabit longitudinal encryption authentication devices are deployed in the load regulation master station system and the load regulation substation system respectively to realize business data encryption and decryption and two-way identity authentication functions; 2) The load regulation substation and the provincial load regulation master station have longitudinal boundaries. Gigabit longitudinal encryption authentication devices are deployed in the load regulation master station system and the load regulation substation system respectively to realize business data encryption and decryption and two-way identity authentication functions; 3) The load regulation substation and the security access area boundary: a power-specific isolation device with prevention and control functions should be deployed between the load regulation substation system and the security access area to realize physical isolation and one-way transmission of business messages; 4) The security access area and the interruptible load terminal boundary: the interruptible load terminal accesses the isolation gateway at the security access area boundary through the wireless public network, establishes a dedicated encryption tunnel, and realizes data confidentiality, integrity protection and two-way identity authentication functions;

[0040] ② Distribution automation business scenario: mainly realizes the protection and control of the distribution network, detects the status information of the distribution network line or equipment through the relay protection automatic device, quickly realizes the fault judgment and accurate positioning of the distribution network line section or equipment, and quickly isolates the faulty line section or faulty equipment, and then restores the power supply to the normal area; the overall architecture of the distribution automation scenario is as follows Figure 2As shown, the distribution automation system master station and the distribution automation terminal transmit data through the communication network. When the data transmission adopts wired or power wireless private network communication mode, it is gathered at the substation and transmitted to the distribution automation system master station of the city by backbone communication; if the 5G wireless public network communication mode is adopted, the data is directly gathered at the city company and then transmitted to the distribution automation system master station; a set of wireless security access areas are uniformly deployed to prevent malicious information from invading the master station system from the terminal; at the same time, vertical communication security protection technology is adopted, and the authentication encryption technology based on asymmetric encryption algorithm is used for control instructions and parameter setting instructions, so as to realize two-way identity authentication and data encryption between the distribution network terminal and the master station to ensure the confidentiality and integrity of the message;

[0041] ③ Intelligent inspection business scenario: Intelligent inspection mainly uses high-definition imaging and image recognition processing technology. In specific areas or lines, intelligent inspection equipment conducts autonomous inspections and high-speed transmission of video image information, realizing the "machine-replacing-man" intelligent inspection mode, improving the safe operation level and inspection efficiency of the power grid. The overall architecture of the intelligent inspection scenario is as follows: Figure 3 As shown in the figure, if traditional wired private network communication is adopted, the terminal data passes through the substation business convergence layer, through the municipal backbone communication network to the intelligent robot inspection system centralized control station, and then interacts with the operation safety management system and the unified video system; if a wireless virtual private network is adopted, the terminal data directly interacts with the unified video system through the operator network; different IPs are assigned to different services in intelligent inspection, and reasonable firewall strategies are set; after different services enter the UPF, different communication channels should be used for bearing and transmission. The communication channel can use a dedicated transmission channel or a data communication network VPN channel, and different services are not interoperable; remote control instructions use asymmetric encryption algorithm technology for security protection, realize two-way identity authentication and data encryption between the inspection terminal and the main station, and ensure the confidentiality and integrity of the message;

[0042] ④Distributed energy scenario: The overall architecture of distributed energy scenario is as follows Figure 4As shown in the figure, this structure adopts the traditional wired private network communication method. The access terminal needs to be connected by optical fiber and communicate with the distribution automation system through the traditional power communication private network. If a wireless virtual private network is used, the terminal data directly interacts with the power consumption information collection system (provincial deployment) through the operator network; most distributed source and storage services belong to the power grid management information area. In terms of terminal security protection, a terminal identity management system built on a unified security access gateway and IoT management platform can be used to strengthen terminal authentication capabilities using digital certificates, terminal identification and other authentication methods to ensure the legitimacy of all devices accessing the network and the authenticity of the terminal; in terms of border security protection, hardware firewall security access gateways, security isolation devices, anti-virus and intrusion defense and other comprehensive defense measures are deployed at each network boundary to achieve border isolation and security protection, and prevent cross-regional illegal access and external malicious attacks; for a small number of businesses involving the power grid production control area, their security protection plans should refer to load control and distribution automation businesses;

[0043] ⑤ Collection business scenarios: The overall architecture of collection scenarios is as follows: Figure 5 As shown in the figure, if wireless public network communication means such as 5G are used, the terminal data will interact directly with the provincial power consumption information / distribution master station; since the reliability and real-time requirements of collection services are lower than those of control services, the end-to-end delay does not need to reach the millisecond level. In the early stage of 5G power virtual private network construction, when various cities have not yet built a complete security access platform, business data can be accessed to the security access area on the provincial company side in a centralized manner, and unified management and security protection can be carried out to reduce the difficulty of border security protection. As the access volume increases in the future, the security access platform can be lowered to various cities to disperse the data pressure in the provincial security access area, so as to reduce possible link failure points and improve deterministic delay.

[0044] The present invention is a 5G network quality assessment method for power services based on power virtual private networks. During use, the present invention provides a feasible communication method for the power grid to safely and reliably access massive terminals by constructing virtual private networks in different scenarios, thereby ensuring the stable operation of the diversified power grid scenarios. The 5G power virtual private network of the present invention meets the needs of power applications, and is suitable for intelligent management and control of multi-dimensional scenarios of power grid services through 5G public networks, soft and hard slicing technologies, and power UPF deployment security protection measures, including: 1) Load control: When there is a large shortage of power grid supply and demand, the master station can issue load control instructions to the load terminals that can be interrupted in seconds, so as to achieve 1) Current load control in seconds; 2) Distribution automation: Status detection of distribution network lines or equipment to quickly locate and isolate faults, and restore power supply to normal areas; 3) Intelligent inspection: Through the intelligent inspection mode of "machine instead of people", the safe operation level and inspection efficiency of the power grid are improved; 4) Distributed source storage: Rapid access and application of distributed small hydropower, distributed photovoltaic and distributed energy storage; 5) Collection services: Data collection can be performed on distribution transformers and user sides in minutes; The present invention has the advantages of building a virtual private network for power in various scenarios, using machine learning to automatically complete indicator generation and selection, and realizing comprehensive and accurate evaluation of 5G networks.

[0045] Example 2

[0046] like Figure 1-7 As shown, a method for evaluating the quality of a 5G network for electric power services based on an electric power virtual private network comprises the following steps:

[0047] Step 1: Build a 5G power virtual private network-based power business network architecture in multiple scenarios;

[0048] Step 2: Perform channel modeling on the network architecture to achieve uplink and downlink bidirectional reciprocal channel simulation, and perform precise performance testing of the 5G network through multiple-input multiple-output air interface testing;

[0049] Step 3: Use machine learning to automatically generate and select the required evaluation network indicators according to the evaluation objectives, and further obtain the network feature weights of the selected network indicators and the influence coefficients that can represent the direction of the network features;

[0050] Step 4: Use a panoramic 5G network quality assessment method based on the attributes and classification characteristics of the evaluation indicators to conduct a relatively comprehensive and accurate assessment of the 5G network.

[0051] The step 2 performs channel modeling on the network architecture, realizes uplink and downlink bidirectional reciprocal channel simulation, and reduces the dimension of the channel model through multi-input multi-output air interface test to realize accurate performance test of 5G network. Specifically, a multi-input multi-output air interface, i.e., MIMOOTA test system, is used to realize MIMOOTA channel modeling. To construct the OTA channel model, it is necessary to restore the delay, power, azimuth, cross-polarization ratio and other information of each cluster in a darkroom, and different antenna probes in the OTA darkroom need to be assigned different weight factors, so as to simulate the directional propagation and angle expansion of the common channel data link, i.e., the CDL cluster, and adopt a pre-fading synthesis PFS probe assignment algorithm that is not affected by the phase fluctuation of the system and has stronger robustness.

[0052] In the PFS algorithm, spatial correlation is selected as an evaluation factor for modeling the spatial characteristics of the channel. Spatial correlation is a statistical measure of the similarity of received signals. The schematic diagram of the OTA probe setting is shown in Figure 6 As shown, AoA is an angle of φ p The plane wave is injected into the antenna array with a spacing of d, and the angle between the antenna normal and the horizontal plane is φ a , φ is the angle of the plane wave relative to the antenna element; Ignoring the antenna polarization, for a pair of antennas u and v, assuming that the terminal antenna is an ideal point source antenna, the continuous power angular spectrum under the channel model can be written as: That is, for the defined continuous PAS, the spatial correlation is a function of the distance and relative position (angle) between the pairs of virtual antennas; in the multi-probe anechoic chamber MPAC, the continuous PAS in the channel model is simulated by the signal from the discrete multi-probe antenna, which can be derived from the above formula. The spatial correlation function of the discrete antenna probe in the test area is: Among them, ω k is the weight of the Kth probe; the evaluation factor for measuring the OTA modeling standard is the spatial correlation error, i.e., ρ OTA -ρ Target , one limitation that needs to be noted is that the probe weight must satisfy the normalization condition and cannot be less than 0, so the objective function is: Through this scheme, the weight mapping relationship of each cluster in each probe can be calculated, and the corresponding link can be substituted into the channel formula to restore the channel model defined in the standard.

[0053] The step 3 uses machine learning to automatically complete the generation and selection of the required evaluation network indicators according to the evaluation objectives, and further obtains the network feature weights of the selected network indicators and obtains the influence coefficients that can represent the network feature directions, specifically including the following steps:

[0054] Step 3.1: Machine learning and gradient boosting model;

[0055] Step 3.2: Modeling and implementation.

[0056] The machine learning and gradient boosting model in step 3.1 is as follows: for a sample set R d The i-th network characteristic index x in i , which can be expressed as x i ∈R d , for a multivariate linear model, the predicted value of its objective function index is It can be expressed as: where w i is the learning parameter of the type; for the linear model, the learning parameter Θ can be expressed as: Θ={w j j=1,...,d}, where the i-th network characteristic index j is the model parameter w i number; the objective function can be expressed as: Obj(Θ)=L(Θ)+Ω(Θ)(3), where L(Θ) is the model loss function; Ω(Θ) is the type regularization term; the loss function If the loss number adopts the square loss function, then The XGBoost model is an improved algorithm based on the GBDT principle. XGBoost is based on the decision tree model, and learns tree by tree. It uses a greedy algorithm to iterate successively. Each tree fits the deviation of the previous model. Formula (1) is the predicted value of the objective function indicator. It can be expressed as: Where K is the number of leaf nodes; f k ∈R d ; At this time, the objective function formula (3) can be expressed as: If used represents the predicted value of the objective function indicator of the t-th round of iteration algorithm, is the predicted value of the objective function indicator in the previous round, then It can be expressed as In this way, the above formula can be expressed as: Where C is a constant term; after the objective function is transformed by "Taylor second-order expansion", the new objective function can be expressed as follows: in XGBoost calculates which feature to select as the scoring point based on the gain of the model structure score, and the importance of a feature is the sum of the number of times it appears in all trees. In other words, the more an attribute is used to build decision trees in the model, the higher its importance is relatively. The Gain in the XGBoost model is defined by the following formula: in is the score of the left leaf node; is the score of the right leaf node; The score that can be obtained without segmentation; γ is the complexity cost after the introduction of the new leaf node; where G j , H j Define the sum of the first-order and second-order derivatives of each leaf node respectively, and the scores of the left and right leaf nodes can be respectively represented by G j , H j Find, where G j , H j Calculated by the following formula: The network indicator feature importance FI coefficient value is the feature score of each network feature indicator given by the XGBoost model during the training process, indicating the importance coefficient of the feature of each network indicator to the model training.

[0057] The modeling and implementation in step 3.2 specifically include the following steps:

[0058] Step 3.21: Construct the first network indicator set: Generally speaking, a branch company or an independent network construction unit is selected as the evaluation object; the original network indicator set includes a series of network characteristic indicator names and network characteristic value sets, the number of network characteristic indicators is represented by k, the number of objective function indicators is represented by n, and the number of set samples is represented by m; f x is the name of the network characteristic indicator; i is the name of the objective function; assuming that the evaluation function F(x) = Y, the independent variable x is the network characteristic index; Y is the objective function;

[0059] Step 3.22: Network indicator processing: First, according to the evaluation goal, select an objective function indicator y from the first network indicator set * , forming a second network indicator set; the second network indicator set includes n * network feature indicators and 1 objective function indicator; then use the first feature selection model to apply to the second network indicator set to determine whether the network feature parameters appear close to 0, forming a second network indicator set preferably 1; use the second feature selection model to apply to the network indicator set to determine the FI coefficient value of the importance of the network indicator characteristics in the network indicator set to form a second network indicator set preferably 2 or directly form a third network indicator set (a possible implementation method), and form a third network indicator set with the second network indicator set preferably 1 (another possible implementation method), and use the FI coefficient value to determine the network indicator weights in the third network indicator set.

[0060] In step 4, a panoramic 5G network quality evaluation method is used according to the evaluation index attributes and classification characteristics to perform a relatively comprehensive and accurate evaluation of the 5G network, specifically including: a static threshold linear algorithm, a single dynamic threshold linear algorithm and a dual dynamic threshold linear algorithm; the static threshold linear algorithm refers to a linear algorithm with fixed upper and lower thresholds, and has a clear indicator quality measurement standard for traditional network quality and performance indicators. The algorithm formula is: subjectto(α+β)=1, where eva_score is the evaluation score, and each specific indicator is calculated according to the formula; ParaValue parameter value is the current status of the indicator, and each specific indicator outputs the latest data according to the statistical period; Baseline basic threshold is the lowest standard, and the basic level of the evaluation area is evaluated. The minimum value of the theoretical minimum standard and the current network basic capability level is adopted. Each specific indicator that does not reach the basic threshold will not score; Challengeline challenge threshold is the highest standard, and the minimum value of the theoretical highest standard and the highest standard that can be achieved by the network is adopted. Each specific indicator that reaches the challenge threshold will get full score full marks The α and β linear control factors are used to control the linear evaluation score base of the indicator and the influence of the indicator on the total score of the overall dimension, respectively. For the indicators determined by the upper and lower thresholds, the focus is on quality stability assessment.

[0061] The single dynamic threshold linear algorithm refers to a linear algorithm in which one of the upper and lower thresholds changes dynamically. For network construction and service benchmarking indicators, it is necessary to comprehensively consider factors such as the overall construction scale and changes in competitors, and adjust the basic threshold in a timely manner. Therefore, a single dynamic evaluation standard measurement method is adopted, and the algorithm formula is: subjectto(α+β)=1, where dy_Baseline dynamic basic threshold value adopts the average level of the benchmark object or the basic capability level of the current network to adaptively reflect the benchmark result; indicators with step-by-step improvement of the basic threshold are generally the embodiment of the quality of phased changes, focusing on the evaluation of quality improvement; the dual dynamic threshold linear algorithm refers to a linear algorithm with dynamic changes in both upper and lower thresholds. For market development and benefit indicators, it is necessary to comprehensively consider regional economy, user groups, consumption capacity, behavioral preferences, and local network carrying capacity. Therefore, the dual dynamic evaluation standard measurement method is adopted, and the algorithm formula is: Among them, dy_Baseline is the dynamic basic threshold value, which is determined by the ranking of the corresponding areas within the accounting period, and adopts the indicator value of the worst ranking area; dy_Challengeline is the dynamic challenge threshold value, which is determined by the ranking of the corresponding areas within the accounting period, and adopts the indicator value of the best ranking area.

[0062] The present invention is a 5G network quality assessment method for electric power services based on electric power virtual private networks. During use, the present invention provides a feasible communication method for the power grid to safely and reliably access massive terminals by constructing virtual private networks in different scenarios, thereby ensuring the stable operation of the diversified scenarios of the power grid. The present invention adopts machine learning to automatically complete the generation and selection of the required evaluation network indicators according to the evaluation objectives, and further obtains the network feature weights of the selected network indicators and obtains the influence coefficients that can represent the direction of the network features, which can be used for 5G network evaluation, especially for multi-objective and more factor-analytical quantitative evaluation of the network for various types of network features. The present invention constructs a 5G panoramic picture evaluation system to comprehensively evaluate the 5G network, and makes a relatively accurate evaluation of the current network and business development according to the formulation of a specific scoring and evaluation system, so as to meet the needs of 5G electric power business network operation and maintenance optimization. The present invention has the advantages of constructing electric power virtual private networks in various scenarios, automatically completing indicator generation and selection by machine learning, and realizing comprehensive and accurate evaluation of 5G networks.

Claims

1. A 5G network quality assessment method for electric power services based on electric power virtual private networks, characterized by: The method comprises the following steps: Step 1: Build a 5G power virtual private network-based power business network architecture in multiple scenarios; Step 2: Perform channel modeling on the network architecture to achieve uplink and downlink bidirectional reciprocal channel simulation, and perform precise performance testing of the 5G network through multiple-input multiple-output air interface testing; Step 3: Use machine learning to automatically generate and select the required evaluation network indicators according to the evaluation objectives, and further obtain the network feature weights of the selected network indicators and the influence coefficients that can represent the direction of the network features; Step 4: Use a panoramic 5G network quality assessment method based on the attributes and classification characteristics of the evaluation indicators to conduct a relatively comprehensive and accurate assessment of the 5G network.

2. The method for evaluating the quality of a 5G network for electric power services based on an electric power virtual private network as claimed in claim 1, characterized in that: The 5G power virtual private network in step 1 means: based on hard slices with physical isolation strength and soft slices with logical isolation strength in the operator's 5G network, a dedicated network for the power industry is virtualized in the wireless, bearer, core network and other links; The virtual private network is integrated across domains with the power communication private network through a secure access platform to achieve end-to-end power business carrying, high-intensity security isolation and resource management.

3. The method for evaluating the quality of a 5G network for electric power services based on an electric power virtual private network as claimed in claim 2, characterized in that: The construction of the power business network architecture based on the 5G power virtual private network in multiple scenarios in step 1 specifically includes: load control business scenarios, distribution automation scenarios, smart inspection scenarios, distributed energy scenarios and collection scenarios.

4. The method for evaluating the quality of a 5G network for electric power services based on an electric power virtual private network as claimed in claim 1, characterized in that: The step 2 performs channel modeling on the network architecture, realizes uplink and downlink bidirectional reciprocal channel simulation, and reduces the dimension of the channel model through multi-input multi-output air interface test to realize accurate performance test of 5G network. Specifically, a multi-input multi-output air interface, i.e., MIMO OTA test system, is used to realize MIMO OTA channel modeling. To construct the OTA channel model, it is necessary to restore the delay, power, azimuth, cross-polarization ratio and other information of each cluster in a darkroom, and different weight factors need to be assigned to different antenna probes in the OTA darkroom, so as to simulate the directional propagation and angle expansion of the common channel data link, i.e., CDL cluster, and adopt a pre-fading synthesis PFS probe assignment algorithm that is not affected by the phase fluctuation of the system and has stronger robustness.

5. The method for evaluating the quality of a 5G network for electric power services based on an electric power virtual private network as claimed in claim 4, characterized in that: In the PFS algorithm, spatial correlation is selected as an evaluation factor for modeling the spatial characteristics of the channel. Spatial correlation is a statistical measure of the similarity of received signals. AoA is expressed in terms of angle φ. p The plane wave is injected into the antenna array with a spacing of d, and the angle between the antenna normal and the horizontal plane is φ a , φ is the angle of the plane wave relative to the antenna element; Ignoring the antenna polarization, for a pair of antennas u and v, assuming that the terminal antenna is an ideal point source antenna, the continuous power angular spectrum under the channel model can be written as: That is, for the defined continuous PAS, the spatial correlation is a function of the distance and relative position between the pairs of virtual antennas; in the multi-probe anechoic chamber MPAC, the continuous PAS in the channel model is simulated by the signal from the discrete multi-probe antenna, which can be derived from the above formula. The spatial correlation function of the discrete antenna probe in the test area is: Among them, ω k is the weight of the Kth probe; the evaluation factor for measuring the OTA modeling standard is the spatial correlation error, i.e. |ρ OTA -ρ Target |, one limitation that needs to be noted is that the probe weight must satisfy the normalization condition and cannot be less than 0, so the objective function is: Through this scheme, the weight mapping relationship of each cluster in each probe can be calculated, and the corresponding link can be substituted into the channel formula to restore the channel model defined in the standard.

6. The method for evaluating the quality of a 5G network for electric power services based on an electric power virtual private network as claimed in claim 1, characterized in that: The step 3 uses machine learning to automatically complete the generation and selection of the required evaluation network indicators according to the evaluation objectives, and further obtains the network feature weights of the selected network indicators and obtains the influence coefficients that can represent the network feature directions, specifically including the following steps: Step 3.1: Machine learning and gradient boosting model; Step 3.2: Modeling and implementation.

7. The method for evaluating the quality of a 5G network for electric power services based on an electric power virtual private network as claimed in claim 6, characterized in that: The machine learning and gradient boosting model in step 3.1 is as follows: for a sample set R d The i-th network characteristic index x in i , which can be expressed as x i ∈R d , for a multivariate linear model, the predicted value of its objective function index is It can be expressed as: where w i is the learning parameter of the type; for the linear model, the learning parameter Θ can be expressed as: Θ={w j |j=1,...,d}, where the i-th network characteristic index j is the model parameter w i number; the objective function can be expressed as: Obj(Θ)=L(Θ)+Ω(Θ)(3), where L(Θ) is the model loss function; Ω(Θ) is the type regularization term; the loss function If the loss number adopts the square loss function, then The XGBoost model is an improved algorithm based on the GBDT principle. XGBoost is based on the decision tree model, and learns tree by tree. It uses a greedy algorithm to iterate successively. Each tree fits the deviation of the previous model. Formula (1) is the predicted value of the objective function indicator. It can be expressed as: Where K is the number of leaf nodes; f k ∈R d ; At this time, the objective function formula (3) can be expressed as: If used represents the predicted value of the objective function indicator of the t-th round of iteration algorithm, is the predicted value of the objective function indicator in the previous round, then It can be expressed as In this way, the above formula can be expressed as: Where C is a constant term; after the objective function is transformed by "Taylor second-order expansion", the new objective function can be expressed as follows: in XGBoost calculates which feature to select as the scoring point based on the gain of the model structure score, and the importance of a feature is the sum of the number of times it appears in all trees. In other words, the more an attribute is used to build decision trees in the model, the higher its importance is relatively. The Gain in the XGBoost model is defined by the following formula: in is the score of the left leaf node; is the score of the right leaf node; The score that can be obtained without segmentation; γ is the complexity cost after the introduction of the new leaf node; where G j , H j Define the sum of the first-order and second-order derivatives of each leaf node respectively, and the scores of the left and right leaf nodes can be respectively represented by G j , H j Find, where G j , H j Calculated by the following formula: The network indicator feature importance FI coefficient value is the feature score of each network feature indicator given by the XGBoost model during the training process, indicating the importance coefficient of the feature of each network indicator to the model training.

8. The method for evaluating the quality of a 5G network for electric power services based on an electric power virtual private network as claimed in claim 6, characterized in that: The modeling and implementation in step 3.2 specifically include the following steps: Step 3.21: Construct the first network indicator set: Generally speaking, a branch company or an independent network construction unit is selected as the evaluation object; the original network indicator set includes a series of network characteristic indicator names and network characteristic value sets, the number of network characteristic indicators is represented by k, the number of objective function indicators is represented by n, and the number of set samples is represented by m; f x is the name of the network characteristic indicator; i is the name of the objective function; assuming that the evaluation function F(x) = Y, the independent variable x is the network characteristic index; Y is the objective function; Step 3.22: Network indicator processing: First, according to the evaluation goal, select an objective function indicator y from the first network indicator set * , forming a second network indicator set; the second network indicator set includes n * network feature indicators and 1 objective function indicator; then use the first feature selection model to apply to the second network indicator set to determine whether the network feature parameters appear close to 0, forming a second network indicator set preferably 1; use the second feature selection model to apply to the network indicator set to determine the FI coefficient value of the importance of the network indicator characteristics in the network indicator set to form a second network indicator set preferably 2 or directly form a third network indicator set, and form a third network indicator set with the second network indicator set preferably 1, and use the FI coefficient value to determine the network indicator weights in the third network indicator set.

9. The method for evaluating the quality of a 5G network for electric power services based on an electric power virtual private network as claimed in claim 1, characterized in that: In step 4, a panoramic 5G network quality evaluation method is used according to the evaluation index attributes and classification characteristics to perform a relatively comprehensive and accurate evaluation of the 5G network, specifically including: a static threshold linear algorithm, a single dynamic threshold linear algorithm and a dual dynamic threshold linear algorithm; the static threshold linear algorithm refers to a linear algorithm with fixed upper and lower thresholds, and has a clear indicator quality measurement standard for traditional network quality and performance indicators. The algorithm formula is: subject to(α+β)=1, where eva_score is the evaluation score, and each specific indicator is calculated according to the formula; ParaValue parameter value is the current status of the indicator, and each specific indicator outputs the latest data according to the statistical period; Baseline basic threshold is the lowest standard, and the basic level of the evaluation area is evaluated. The minimum value of the theoretical minimum standard and the current network basic capability level is adopted. Each specific indicator that does not reach the basic threshold will not score; Challengeline challenge threshold is the highest standard, and the minimum value of the theoretical highest standard and the highest standard that can be achieved by the network is adopted. Each specific indicator that reaches the challenge threshold will get full score full marks The α and β linear control factors are used to control the base of the linear evaluation score of the indicator and the influence of the indicator on the total score of the overall dimension, respectively. For the indicators determined by the upper and lower thresholds, the focus is on quality stability assessment.

10. The method for evaluating the quality of a 5G network for electric power services based on an electric power virtual private network as claimed in claim 9, characterized in that: The single dynamic threshold linear algorithm refers to a linear algorithm in which one of the upper and lower thresholds changes dynamically. For network construction and service benchmarking indicators, it is necessary to comprehensively consider factors such as the overall construction scale and changes in competitors, and adjust the basic threshold in a timely manner. Therefore, a single dynamic evaluation standard measurement method is adopted, and the algorithm formula is: subject to (α+β)=1, where dy_Baseline dynamic basic threshold value adopts the average level of the benchmark object or the current network basic capability level to adaptively reflect the benchmark result; for indicators with step-by-step improvement of basic threshold, it is generally a reflection of the quality of phased changes, focusing on the evaluation of quality improvement; the dual dynamic threshold linear algorithm refers to a linear algorithm with dynamic changes in both upper and lower thresholds. For market development and benefit indicators, it is necessary to comprehensively consider regional economy, user groups, consumption capacity, behavioral preferences, and local network carrying capacity. Therefore, the dual dynamic evaluation standard measurement method is adopted, and the algorithm formula is: Among them, dy_Baseline is the dynamic basic threshold value, which is determined by the ranking of the corresponding areas within the accounting period, and adopts the indicator value of the worst ranking area; dy_Challengeline is the dynamic challenge threshold value, which is determined by the ranking of the corresponding areas within the accounting period, and adopts the indicator value of the best ranking area.

Images