CN119130033A

CN119130033A - Optimal dispatching method, system, equipment and medium for 5G base station and active distribution network

Info

Publication number: CN119130033A
Application number: CN202411189277.3A
Authority: CN
Inventors: 程元; 饶尧; 丁胜; 白新雷; 肖申; 刘政; 宋河; 魏浩; 刘紫懿; 叶子麒; 黄津明
Original assignee: Wuhan Energy Efficiency Evaluation Co Ltd Of State Grid Electric Power Research Institute; State Grid Electric Power Research Institute; Marketing Service Center of State Grid Hebei Electric Power Co Ltd; State Grid Corp of China SGCC
Current assignee: Wuhan Energy Efficiency Evaluation Co Ltd Of State Grid Electric Power Research Institute; State Grid Electric Power Research Institute; Marketing Service Center of State Grid Hebei Electric Power Co Ltd; State Grid Corp of China SGCC
Priority date: 2024-08-28
Filing date: 2024-08-28
Publication date: 2024-12-13

Abstract

The invention provides an optimized scheduling method, system, equipment and medium for a 5G base station and an active power distribution network, which are characterized in that an improved space-time diagram convolutional neural network is established, a diagram convolution method is combined with a gating circulation unit to deeply mine the load flexible adjustment potential of the 5G base station, the space relevance among different stations is automatically learned based on historical data through an adjacency matrix self-adaptive correction method, the utilization value of key data is improved, the prediction precision is improved, the real-time minimum reserve capacity of the 5G base station energy storage is calculated based on the prediction results of the 5G base station load rate, the 5G base station power and the photovoltaic output, an objective function is established with the optimal operation cost as a target, the problem of solving the objective function is decomposed into a main problem and a sub problem by utilizing an improved list generation algorithm to solve, the scheduling plan of the 5G base station and the active power distribution network is obtained, the space-time relevance of different 5G base station groups is guaranteed to be efficiently extracted, and the economy of each base station group is effectively improved.

Description

Optimal scheduling method, system, equipment and medium for 5G base station and active power distribution network

Technical Field

The invention belongs to the technical field of energy management, and relates to an optimal scheduling method, an optimal scheduling system, optimal scheduling equipment and optimal scheduling media.

Background

The energy storage schedulable capacity of a single macro base station of the 5G base station is 4-16kWh, the maximum adjustable power is 2-4kW/h, huge adjustable resources are contained in massive 5G base stations, the optimal scheduling of the 5G base stations and an active power distribution network is a key link of energy conservation and consumption reduction of novel infrastructure, and the 5G base stations can effectively support the safe operation of a power grid through interaction and energy conservation potential excavation with the power grid, so that mutual benefits and win-win of a communication carrier and the power grid are realized. With the construction and development of 5G base stations, the number of 5G base stations rises year by year to form massive distributed flexible resources, however, in the optimal scheduling of the 5G base stations and an active power distribution network, the 5G base stations have the problems of insufficient flexible load adjustment potential excavation and insufficient cooperative interaction capability with a power grid, and if a centralized scheduling method is adopted, the problems of user privacy protection and huge calculation load are faced.

Disclosure of Invention

In order to solve the problems of insufficient flexible load adjustment potential mining, insufficient cooperative interaction capability with a power grid and the problem of user privacy protection and huge calculation load faced by adopting a centralized scheduling method in the background technology, the invention provides an optimal scheduling method, system, equipment and medium for a 5G base station and an active power distribution network.

The method of the invention comprises the following steps:

Based on original input data of optimized scheduling of a 5G base station and an active power distribution network, combining a graph convolution and gating circulation unit to establish a space-time graph convolution neural network, and learning the relation of each node in a graph convolution layer graph structure in the space-time graph convolution neural network by using historical operation data of the 5G base station and the active power distribution network, and adaptively correcting adjacent matrix parameters of a graph convolution layer in the space-time graph convolution neural network to obtain an improved space-time graph convolution neural network;

calculating weights of different input features in original input data of optimal scheduling of the 5G base station and the active power distribution network by using a soft attention mechanism and a scaling point scoring function, and obtaining corrected input features;

inputting the corrected input characteristics into an improved space-time diagram convolutional neural network, and outputting to obtain a 5G base station predicted load rate, 5G base station predicted power and photovoltaic predicted output;

according to the 5G base station predicted power and the standby time of the 5G base station, calculating to obtain the energy storage real-time minimum standby capacity of the 5G base station;

according to the operation conditions of the 5G base station and the active power distribution network, establishing an objective function and an operation constraint condition which aim at optimizing the operation cost;

And solving an objective function through a column constraint generation algorithm based on the 5G base station energy storage real-time minimum reserve capacity, the 5G base station predicted load rate, the photovoltaic predicted output and the operation constraint condition to obtain a scheduling plan of the 5G base station and the active power distribution network, and optimally scheduling the 5G base station and the active power distribution network according to the scheduling plan.

Further, the original input data of the 5G base station and the active power distribution network optimized schedule comprises operation data of the 5G base station, geographic distribution data of the 5G base station, meteorological data of the 5G base station and historical operation data of the 5G base station and the active power distribution network;

The space-time diagram convolutional neural network model is represented as a topological graph g= (V, E, a), where V represents a point set, E represents an edge set, a is an adjacency matrix of the topological graph G, and the convolution range is a neighborhood B of node V _i:

B(v_i)＝{v_j|d(v_j,v_i)≤D}(1),

Wherein D (v _j,v_i) represents the shortest distance from the graph node v _i to v _j, D is the maximum convolution neighborhood, and when D=1, the size of the sampling area is the first-order neighborhood of the node v _i;

The forward propagation of the graph convolutional layer is as follows:

wherein, reLU refers to a modified linear unit activation function, X is a feature matrix of a topological graph G; is an adjacency matrix containing autocorrelation, i.e A is the adjacency matrix of the topology graph G, and I is the identity matrix; Is that The ith row and jth column elements in the table; Is a normalized adjacent matrix, sigma (·) is a sigmoid function, W ₀ is a weight matrix from an upper layer to a picture volume layer, W ₁ is a weight matrix from a picture volume layer to a lower layer, Is a node degree matrix;

Combining a graph convolution unit with a gating circulation unit, introducing a graph convolution algorithm into the gating circulation unit to form a space-time graph convolution unit, forming a space-time graph convolution neural network through stacking of a plurality of space-time graph convolution units, capturing space-time dynamic changes of nodes in a graph, and deeply learning and analyzing space-time graph data, wherein the forward propagation formula is shown in the following formulas (4) - (7):

u_t＝σ(W_u[f(A,X_t),h_t-1]+b_u) (4),

r_t＝σ(W_r[f(A,X_t),h_t-1]+b_r) (5),

c_t＝tanh(W_c[f(A,X_t),(r_t*h_t-1)]+b_c) (6),

h_t＝u_t*h_t-1+(1-u_t)*c_t (7),

Wherein u _t、r_t、c_t is a unit internal parameter, which corresponds to the states of an update gate, a reset gate and a current unit at the time t respectively, h _t is a unit output at the time t, W _u and b _u are a weight matrix and a bias matrix of the update gate respectively, W _r and b _r are a weight matrix and a bias matrix of the reset gate, f (A, X _t) is a graph convolution operation input to a model at the time t, h _t-1 is a hidden state at the time t-1 at the last time, W _c is a weight matrix for calculating a candidate state c _t, and b _c is a bias vector associated with the candidate state c _t;

The method for adaptively correcting the adjacent matrix parameters of the graph convolution layer in the space-time graph convolution neural network comprises the following steps:

For the feature matrix x= (X ₁,x₂,…x_N)^T, define a _mn as the parameter of the m-th row and n-th column of the adjacency matrix, and the expression a _mn is:

Wherein x _m and x _n represent characteristics of a node m and a node N, ω is a weight vector which can be learned in the adjacency matrix parameter, and N is the total number of first-order neighbors of the node m;

Equation (8) uses ReLU units to limit the values of the adjacency matrix to non-negative numbers and normalizes a _mn using the form of softmax;

The weight parameters of a _mn are learned using the loss function L _graph of the following formula:

in the formula, lambda is a penalty coefficient of a regular term, F is a norm of the regular term, and when F-2 regularization is adopted, the formula is as follows:

The total loss function L _loss of the adaptive neural network is shown as follows:

L_loss＝L_mae+L_graph (11),

Where L _mae is the loss function of the original neural network, and the average absolute error is used.

Further, the method for obtaining the corrected input features comprises the following steps:

When the period of the input feature is T, the feature weight vector α _T is:

Wherein X is an original input feature, Q is a query vector of an attention mechanism, namely a predicted target value corresponding to time, D _T is the total length of the input feature, alpha _T is calculated by a soft attention mechanism, and the input feature is normalized by using a scaling dot product scoring function to ensure that weights of different features are reasonably adjusted so as to optimize the input feature of a model;

And carrying out standardized operation on all the input features to obtain corrected input features, wherein the expression is as follows:

Wherein X _ref is the corrected input feature, X _mean is the average value of each feature component of the input X, X _std is the standard deviation of each feature component of the input X;

The input characteristic is standardized according to the formula (13) to generate a corrected input characteristic data set X _ref which is the basis of subsequent prediction, the corrected input characteristic data set X _ref is input into an improved space-time diagram convolutional neural network model, and the 5G base station load rate, power and photovoltaic output in different time periods are predicted by utilizing multi-layer convolutional operation and space-time dependency analysis;

The 5G base station predicted load rate λ _5G is represented by the following formula:

λ_5G＝{λ₁(t),λ₂(t),…,λ_N(t)} (14),

Wherein lambda _N (t) represents the load rate of the Nth 5G base station at the moment of time t;

The 5G base station predicted power P _5G is expressed as:

P_5G＝{P₁(t),P₂(t),…,P_N(t)} (15),

wherein, P _N (t) represents the power of the Nth 5G base station at the moment of time t;

The photovoltaic predicted force P _PV is represented by the formula:

P_PV＝{P₁(t),P₂(t),…,P_M(t)} (16),

where P _M (t) represents the power of the mth 5G base station at time t.

Further, in the calculation of the 5G base station energy storage real-time minimum standby capacity, the 5G base station predicted power of different areas obtained in the formula (15) is combined with the 5G base station minimum standby power time t _s to obtain the 5G base station minimum standby capacity at the moment tThe following formula is shown:

Where P _5G is the 5G base station predicted power.

Further, the expression of the objective function is:

In the formula, Representing short-time load electricity-shortage cost; representing the energy storage charging and discharging cost of the base station; Representing the cost of the photovoltaic output;

The specific expression is shown as follows:

Wherein c _LD、c_PV respectively represents load electricity-shortage unit cost and photovoltaic unit power generation cost, c _5G、c_SH respectively represents 5G base station energy storage unit charge-discharge cost and loss cost, P _t ^ch、P_t ^dis respectively represents t-time charge-discharge power, P _t ^PV represents photovoltaic output, eta _5G represents G base station energy storage charge-discharge efficiency, and P _t ^ΔLD represents load electricity-shortage power;

The operation constraint conditions comprise energy storage charging and discharging constraint, photovoltaic output constraint and power balance constraint of the 5G base station;

the expression of the energy storage charging and discharging constraint of the 5G base station is as follows:

Wherein, P _t ^ch and P _t ^dis respectively represent the charging power and the discharging power at the time t, AndRespectively representing the maximum charging power and the maximum discharging power at the time t, E ^5G represents the energy storage energy of the 5G base station,AndRespectively representing the minimum value and the maximum value of the energy storage energy of the 5G base station;

The expression of the photovoltaic output constraint is as follows:

Wherein P _t ^PV represents a photovoltaic output, Representing the maximum output of the photovoltaic;

The expression of the power balance constraint is:

P_t ^ΔLD＝P_t ^ch+P_t ^LD-P_t ^PV-P_t ^dis (22),

Wherein, P _t ^LD represents the load power of the distribution network at the time t, P _t ^ch and P _t ^dis represent the charging power and the discharging power at the time t respectively, and P _t ^PV represents the photovoltaic output.

Further, the solving method of the objective function is as follows:

the uncertainty of the photovoltaic and load is represented by the set Ω, as shown in the following formula:

In the formula, Actual output of the photovoltaic system is time t; Predicting the power of the photovoltaic system at the moment of time t; Representing the maximum value of the output uncertainty range of the photovoltaic system; Representing the actual load demand of the load at the moment t; Representing a maximum value of the load power uncertainty range; a predicted load demand representing a load at time t;

The formulas (18) - (22) are expressed as an uncertain two-stage robust optimization model by using a compact matrix equation, wherein the first stage is to minimize the running cost of the system, and the second stage is to obtain the worst running scene under the minimized cost, and the specific expression is as follows:

Wherein c ^T x represents an optimal objective function of running cost, d ^T y represents worst running cost when a load loses electricity, u represents uncertain variables including photovoltaic and load power, max represents worst scene under the highest cost in optimal scheduling, H, F, G respectively represents three different coefficient matrixes, and y|representscorresponding state variables or decision variables in a column constraint generation algorithm;

Based on a 5G base station energy storage real-time minimum standby capacity model, a 5G base station prediction load rate, a photovoltaic prediction output and operation constraint conditions, solving an uncertain two-stage robust optimization model by using a column constraint generation algorithm, firstly decoupling an objective function of the robust optimization model, and decomposing the objective function into a main problem and a sub-problem;

The main problems are as follows:

where m represents the number of iterations and y ^m represents the solution for m iterations of the sub-problem.

The sub-problems are:

by giving the (x, u) value, the transformation minimization problem is a deterministic linear optimization problem, expressed as follows:

Wherein d and y represent column vectors, A ₁、A₂、A₃ represents a matrix of m ₁、m₂、m₃ columns and n rows, and b ₁、b₂、b₃ represents m ₁、m₂、m₃ -dimensional column vectors;

the min model of the inner layer in the sub-problems of the robust optimization model is converted into a max model by using a dual theory and is combined with the max of the outer layer, and the conversion mode is as follows:

If the original problems are:

The dual form is:

Wherein z ₁、z₂、z₃ represents that m ₁、m₂、m₃ is a row vector;

according to the principle of dual theory, the dual form is:

and (3) obtaining a scheduling plan of the 5G base station and the active power distribution network based on formulas (24) - (30), wherein the scheduling plan comprises a scheduling plan of the 5G base station and a photovoltaic output plan.

The invention provides an optimized dispatching system of a 5G base station and an active power distribution network, which comprises an improved space-time diagram convolutional neural network building module, an input characteristic correcting module, a prediction result calculating module, a 5G base station energy storage real-time minimum standby capacity calculating module, an objective function and constraint building module and an optimized dispatching module.

The improved space-time diagram convolutional neural network establishing module is used for establishing a space-time diagram convolutional neural network by combining a diagram convolution and gating circulating unit based on original input data of optimized scheduling of a 5G base station and an active power distribution network, and learning the relation of each node in a diagram convolution layer diagram structure in the space-time diagram convolutional neural network by utilizing historical operation data of the 5G base station and the active power distribution network, and adaptively correcting adjacent matrix parameters of a diagram convolution layer in the space-time diagram convolutional neural network to obtain the improved space-time diagram convolutional neural network.

And the input characteristic correction module is used for calculating weights of different input characteristics in the original input data of the 5G base station and the active power distribution network optimized scheduling by utilizing a soft attention mechanism and a scaling dot product scoring function, and obtaining corrected input characteristics.

And the prediction result calculation module is used for inputting the corrected input characteristics into the improved space-time diagram convolutional neural network, and outputting to obtain the 5G base station prediction load rate, the 5G base station prediction power and the photovoltaic prediction output.

And the 5G base station energy storage real-time minimum standby capacity calculation module is used for calculating and obtaining the 5G base station energy storage real-time minimum standby capacity according to the 5G base station predicted power and the standby time of the 5G base station.

The objective function and constraint establishing module is used for establishing an objective function and an operation constraint condition which aim at optimizing the operation cost according to the operation conditions of the 5G base station and the active power distribution network.

The optimal scheduling module is used for solving an objective function through a column constraint generation algorithm based on the 5G base station energy storage real-time minimum reserve capacity, the 5G base station predicted load rate, the photovoltaic predicted output and the operation constraint condition to obtain a scheduling plan of the 5G base station and the active power distribution network, and performing optimal scheduling on the 5G base station and the active power distribution network according to the scheduling plan.

The invention also provides electronic equipment, which comprises a memory and a processor, wherein the memory and the processor are in communication connection, the memory stores computer instructions, and the processor realizes the optimal scheduling method of the 5G base station and the active power distribution network by executing the computer instructions.

The invention also provides a computer readable storage medium, which stores a computer program, and is characterized in that the optimal scheduling method of the 5G base station and the active power distribution network is realized when the computer program is executed by a processor.

Compared with the prior art, the improved space-time diagram convolutional neural network is established, a graph rolling method is combined with a gating circulation unit, the advantages of graph rolling and space feature extraction and the space-time dependence of photovoltaic output of a 5G base station and the mining of time feature of the gating circulation unit are considered, the load flexible adjustment potential of the 5G base station is deeply mined, the space relevance among different stations is automatically learned based on historical data through an adjacent matrix self-adaptive correction method, the input data is weighted through a soft attention mechanism and a scaling point integral scoring function, the utilization value of key data is improved, the prediction precision is improved, the energy storage real-time minimum reserve capacity of the 5G base station is calculated based on the prediction result of the 5G base station load rate, the 5G base station power and the photovoltaic output, an objective function is established with optimal operation cost as a target, the problem of solving the objective function is decomposed into a main problem and a sub problem by applying an improved column constraint generation algorithm, the scheduling plan of the 5G base station and the active power distribution network is obtained, the space-time relevance of different 5G base station groups is effectively extracted, and the economical efficiency of each base station group is effectively improved. The invention improves the flexible load adjustment potential of the 5G base station, solves the problems of user privacy protection and huge calculation burden faced by adopting a centralized scheduling method, and greatly improves the 5G base station load rate, the 5G base station power and photovoltaic output prediction accuracy, thereby effectively evaluating the real-time schedulable capacity of each base station energy storage group, realizing the efficient friendly interaction between the 5G base station and the active power distribution network and reducing the operation cost of the 5G base station and the active power distribution network.

Drawings

FIG. 1 is a flow chart of the method of the present invention.

Fig. 2 is a schematic diagram of the architecture of a modified space-time convolutional neural network.

Fig. 3 is a system architecture diagram of the present invention.

Fig. 4 is a schematic diagram of a certain 5G base station and an active power distribution network in embodiment 5.

Fig. 5 is an adjacency matrix for different approaches.

FIG. 6 shows the prediction results of different methods.

Fig. 7 is a diagram of schedulable capacity for different area 5G base station load rate predictions and stored energy.

Fig. 8 is a schematic diagram of cluster division of a distribution network 5G base station and a photovoltaic.

Fig. 9 shows the power loss variation of the distribution network load.

FIG. 10 is a scheduling plan for each cluster.

FIG. 11 is a graph showing the comparison of the overall running costs of the system in the present invention with the random optimization method and the conventional robust optimization method.

Detailed Description

In order to make the technical problems, technical schemes and beneficial effects to be solved more clear, the application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the application.

Example 1

The flow chart of the optimal scheduling method for the 5G base station and the active power distribution network is shown in figure 1, and the specific steps are as follows.

Firstly, based on original input data of optimized scheduling of a 5G base station and an active power distribution network, combining a graph convolution and gating circulation unit to establish a space-time graph convolution neural network, and learning the relation of each node in a graph convolution layer graph structure in the space-time graph convolution neural network by using historical operation data of the 5G base station and the active power distribution network, and adaptively correcting adjacent matrix parameters of a graph convolution layer in the space-time graph convolution neural network to obtain an improved space-time graph convolution neural network;

specifically, the original input data of the 5G base station and the active power distribution network optimized schedule comprises operation data of the 5G base station, geographic distribution data of the 5G base station, meteorological data of the 5G base station and historical operation data of the 5G base station and the active power distribution network.

The architecture diagram of the improved space-time diagram convolutional neural network is shown in fig. 2, in the improved space-time diagram convolutional neural network, compared with the traditional convolutional neural network, the space-time diagram convolutional neural network generalizes the convolutional network to an unstructured feature space, and diagram structure data with any shape can be processed. The space-time diagram convolutional neural network model is represented as a topological graph g= (V, E, a), where V represents a point set, E represents an edge set, a is an adjacency matrix of the topological graph G, and the convolution range is a neighborhood B of node V _i:

B(v_i)＝{v_j|d(v_j,v_i)≤D}(1),

At this time, the forward propagation of the graph roll stack is as follows:

wherein, reLU refers to a modified linear unit activation function, X is a feature matrix of a topological graph G; is an adjacency matrix containing autocorrelation, i.e A is the adjacency matrix of the topology graph G, and I is the identity matrix; Is that The ith row and jth column elements in the table; Is a normalized adjacent matrix, sigma (·) is a sigmoid function, W ₀ is a weight matrix from an upper layer to a picture volume layer, W ₁ is a weight matrix from a picture volume layer to a lower layer, Is a node degree matrix.

And converting the historical load rate and the geographic distribution characteristics of the 5G base station into node parameters of the graph by adopting a graph convolution method, and learning hidden spatial correlations of power timing curves and photovoltaic output curves of different 5G base stations based on a graph convolution layer. In order to further extract the time sequence characteristics of the 5G base station load rate and the photovoltaic output curve, a graph convolution and gating circulation unit is combined, a graph convolution algorithm is introduced into the gating circulation unit to form a space-time graph convolution unit, a space-time graph convolution neural network is formed by stacking a plurality of space-time graph convolution units, the space-time dynamic change of nodes in a graph is captured, the space-time graph data is deeply learned and analyzed, and the forward propagation formula is shown in the following formulas (4) - (7):

u_t＝σ(W_u[f(A,X_t),h_t-1]+b_u) (4),r_t＝σ(W_r[f(A,X_t),h_t-1]+b_r) (5),

c_t＝tanh(W_c[f(A,X_t),(r_t*h_t-1)]+b_c) (6),h_t＝u_t*h_t-1+(1-u_t)*c_t (7),

Where u _t、r_t、c_t is a unit internal parameter, which corresponds to the states of the update gate, the reset gate, and the current unit at time t respectively, h _t is a unit output at time t, W _u and b _u are a weight matrix and a bias matrix of the update gate respectively, W _r and b _r are a weight matrix and a bias matrix of the reset gate, f (a, X _t) is a convolution operation of a model input at time t, h _t-1 is a hidden state at time t-1 last time, W _c is a weight matrix for calculating candidate state c _t, and b _c is a bias vector associated with candidate state c _t.

The network unit of the space-time diagram convolutional neural network can input diagram structure data, extract the input characteristic space relevance through a diagram convolution method, further extract the input characteristic time relevance based on the gating cycle unit structure, send the extracted characteristics into the full-connection layer and the output layer, then output 5G base station load rate and photovoltaic output prediction results, and obtain 5G base station power prediction results based on the 5G base station load rate.

The adjacent matrix parameters of the graph convolution layer represent the spatial correlation of different prediction sites, in the prediction of the load rate of multiple 5G base stations, the power of the 5G base stations and the photovoltaic output, the adjacent matrix parameters are adaptively corrected for more accurately learning the spatial relationship between different base stations and the power output time sequence curves of the photovoltaic power station, and in the training process, the relationship of each node in the graph structure is learned by utilizing massive historical data samples, so that the adjacent matrix parameters are adaptively corrected.

Specifically, the method for adaptively correcting the adjacency matrix parameters of the graph convolution layer in the space-time graph convolution neural network comprises the following steps:

Wherein x _m and x _n represent characteristics of a node m and a node N, ω is a weight vector which can be learned in the adjacency matrix parameter, and N is a total number of first-order neighbors of the node m.

Equation (8) uses ReLU units to limit the values of the adjacency matrix to non-negative numbers and normalizes a _mn by row using the form of softmax;

When the difference between the output time sequence curves of the photovoltaic power stations represented by x _m and x _n is larger, the smaller the relation of the output time sequence curves in space is, the smaller the value of the adjacency matrix A _mn is, in addition, a regularization term is added in the loss function, the 0 weight solution is avoided, and the total loss function L _loss of the adaptive neural network is shown as the following formula:

L_loss＝L_mae+L_graph (11),

And then, calculating weights of different input features in the original input features by using a soft attention mechanism and a scaling dot product scoring function to obtain corrected input features.

Specifically, the method for obtaining the corrected input features comprises the following steps:

When the period of the input feature is T, the feature weight vector α _T is:

the 5G base station predicted load rate is expressed as:

λ_5G＝{λ₁(t),λ₂(t),…,λ_N(t)} (14),

The 5G base station predicted power is expressed as:

P_5G＝{P₁(t),P₂(t),…,P_N(t)} (15),

The photovoltaic predicted output is expressed as:

P_PV＝{P₁(t),P₂(t),…,P_M(t)} (16),

where P _M (t) represents the power of the mth 5G base station at time t.

And then, calculating to obtain the energy storage real-time minimum standby capacity of the 5G base station according to the 5G base station predicted power and the standby time of the 5G base station.

The energy storage minimum standby capacity of the 5G base station is related to the power consumption and the minimum standby time of the base station equipment, and for the power consumption of the base station equipment, a macro base station of 3 source antenna units (ACTIVE ANTENNA units, AAU) and 1 baseband processing unit (BBU) devices is typically configured. Base station power consumption includes two parts, namely dynamic power consumption and static power consumption, and single base station communication equipment power consumption can be expressed as: the power at the time of the base station t is represented; Representing static power, i.e. power amplifier power in AAU, BBU power, and transmission device power; Represents the maximum dynamic power, i.e., the power of the radio unit in the AUU, which is related to the base station communication load rate lambda _t.

Specifically, combining the 5G base station predicted power of different areas obtained in the formula (15) with the minimum standby power time t _s of the 5G base station to obtain the minimum standby capacity of the 5G base station at the moment tThe following formula is shown:

Where P _5G is the 5G base station predicted power.

And then, according to the operation conditions of the 5G base station and the active power distribution network, establishing an objective function and an operation constraint condition which aim at optimizing the operation cost.

Specifically, the expression of the objective function is:

The specific expression is shown as follows:

In the formula, c _LD、c_PV respectively represents the load electricity-shortage unit cost and the photovoltaic unit power generation cost, c _5G、c_SH respectively represents the 5G base station energy storage unit charge-discharge cost and the loss cost, P _t ^ch、P_t ^dis respectively represents the t-time charge-discharge power, P _t ^PV represents the photovoltaic output, eta _5G represents the G base station energy storage charge-discharge efficiency, and P _t ^ΔLD represents the load electricity-shortage power.

The operation constraint conditions comprise energy storage charge and discharge constraint, photovoltaic output constraint and power balance constraint of the 5G base station;

The expression of the photovoltaic output constraint is:

Wherein P _t ^PV represents a photovoltaic output, Indicating the maximum output of the photovoltaic.

The power balance constraint is expressed as:

P_t ^ΔLD＝P_t ^ch+P_t ^LD-P_t ^PV-P_t ^dis (22),

And finally, solving an objective function through a column constraint generation algorithm based on the 5G base station energy storage real-time minimum reserve capacity, the 5G base station predicted load rate, the photovoltaic predicted output and the operation constraint condition to obtain a scheduling plan of the 5G base station and the active power distribution network, and optimally scheduling the 5G base station and the active power distribution network according to the scheduling plan.

Specifically, the solving method of the objective function is as follows:

In the formula, Actual output of the photovoltaic system is time t; Predicting the power of the photovoltaic system at the moment of time t; Representing the maximum value of the output uncertainty range of the photovoltaic system; Representing the actual load demand of the load at the moment t; Representing a maximum value of the load power uncertainty range; the predicted load demand for the load at time t is indicated.

In the formula, c ^T x represents an optimal objective function of the running cost, d ^T y represents the worst running cost when the load loses electricity, u represents an uncertain variable comprising photovoltaic and load power, max represents the worst scene under the highest cost in the optimal scheduling, H, F, G respectively represents three different coefficient matrixes, and y| represents a corresponding state variable or decision variable in a column constraint generation algorithm.

The main problems are as follows:

The sub-problems are:

If the original problems are:

The dual form is:

according to the principle of dual theory, the dual form is:

And (3) obtaining a scheduling plan of the 5G base station and the active power distribution network based on the formulas (24) - (30), wherein the scheduling plan comprises a scheduling plan of the 5G base station and a photovoltaic output plan.

Example 2

The architecture diagram of the optimal scheduling system for the 5G base station and the active power distribution network is shown in fig. 3, and the optimal scheduling system is composed of an improved space-time diagram convolutional neural network building module, an input characteristic correcting module, a prediction result calculating module, a 5G base station energy storage real-time minimum standby capacity calculating module, an objective function and constraint building module and an optimal scheduling module.

The improved space-time diagram convolutional neural network establishing module is used for establishing a space-time diagram convolutional neural network by combining the diagram convolution and gating circulating unit based on original input data of optimized scheduling of the 5G base station and the active power distribution network, and learning the relation of each node in the diagram convolution layer diagram structure in the space-time diagram convolutional neural network by utilizing historical operation data of the 5G base station and the active power distribution network, and adaptively correcting adjacent matrix parameters of a diagram convolution layer in the space-time diagram convolutional neural network to obtain the improved space-time diagram convolutional neural network.

B(v_i)＝{v_j|d(v_j,v_i)≤D}(1),

At this time, the forward propagation of the graph roll stack is as follows:

wherein, reLU refers to a modified linear unit activation function, X is a feature matrix of a topological graph G; is an adjacency matrix containing autocorrelation, i.e A is the adjacency matrix of the topology graph G, and I is the identity matrix; Is that

The ith row and jth column elements in the table; Is a normalized adjacent matrix, sigma (·) is a sigmoid function, W ₀ is a weight matrix from an upper layer to a picture volume layer, W ₁ is a weight matrix from a picture volume layer to a lower layer, Is a node degree matrix.

u_t＝σ(W_u[f(A,X_t),h_t-1]+b_u) (4),r_t＝σ(W_r[f(A,X_t),h_t-1]+b_r) (5),

c_t＝tanh(W_c[f(A,X_t),(r_t*h_t-1)]+b_c) (6),h_t＝u_t*h_t-1+(1-u_t)*c_t (7),

L_loss＝L_mae+L_graph (11),

And the input characteristic correction module is used for calculating weights of different input characteristics in the original input data of the 5G base station and the active power distribution network optimized scheduling by utilizing a soft attention mechanism and a scaling dot product scoring function to obtain corrected input characteristics.

When the period of the input feature is T, the feature weight vector α _T is:

the 5G base station predicted load rate is expressed as:

λ_5G＝{λ₁(t),λ₂(t),…,λ_N(t)} (14),

The 5G base station predicted power is expressed as:

P_5G＝{P₁(t),P₂(t),…,P_N(t)} (15),

The photovoltaic predicted output is expressed as:

P_PV＝{P₁(t),P₂(t),…,P_M(t)} (16),

where P _M (t) represents the power of the mth 5G base station at time t.

The energy storage minimum standby capacity of the 5G base station is related to the power consumption and the minimum standby time of the base station equipment, and for the power consumption of the base station equipment, a macro base station of 3 source antenna units (ACTIVE ANTENNA units, AAU) and 1 baseband processing unit (BBU) devices is typically configured. Base station power consumption includes two parts, namely dynamic power consumption and static power consumption, and single base station communication equipment power consumption can be expressed as: P _J ^5G represents static power, namely power amplifier power in AAU, BBU power and transmission equipment power; Represents the maximum dynamic power, i.e., the power of the radio unit in the AUU, which is related to the base station communication load rate lambda _t.

Where P _5G is the 5G base station predicted power.

And the objective function and constraint establishing module is used for establishing an objective function and an operation constraint condition which aim at optimizing the operation cost according to the operation conditions of the 5G base station and the active power distribution network.

Specifically, the expression of the objective function is:

In the formula, Representing short-time load electricity-shortage cost; representing the energy storage charging and discharging cost of the base station; The specific expression of the photovoltaic output cost is shown as follows:

The expression of the photovoltaic output constraint is as follows:

Wherein P _t ^PV represents a photovoltaic output, Indicating the maximum output of the photovoltaic

The expression of the power balance constraint is:

P_t ^ΔLD＝P_t ^ch+P_t ^LD-P_t ^PV-P_t ^dis (22),

And the optimal scheduling module is used for solving the objective function through a column constraint generation algorithm based on the 5G base station energy storage real-time minimum standby capacity, the 5G base station predicted load rate, the photovoltaic predicted output and the operation constraint condition to obtain a scheduling plan of the 5G base station and the active power distribution network, and optimally scheduling the 5G base station and the active power distribution network according to the scheduling plan.

Specifically, the solving method of the objective function is as follows:

The main problems are as follows:

The sub-problems are:

If the original problems are:

The dual form is:

according to the principle of dual theory, the dual form is:

Example 3

The electronic device comprises a memory and a processor, wherein the memory and the processor are in communication connection, the memory stores computer instructions, and the processor executes the computer instructions to realize the optimal scheduling method of the 5G base station and the active power distribution network in the embodiment 1 and the optimal scheduling system of the 5G base station and the active power distribution network in the embodiment 2.

Example 4

A computer readable storage medium storing a computer program which when executed by a processor implements the optimized scheduling method of the 5G base station and the active power distribution network described in embodiment 1 and the optimized scheduling system of the 5G base station and the active power distribution network described in embodiment 2.

Example 5

In order to better show the technical effect of the invention, the traditional method and the invention are adopted for comparison analysis simultaneously aiming at the optimal scheduling of a certain 5G base station and an active power distribution network. In the analysis process, a Python compiling environment is used, a model is built on PyTorch, and the improved IEEE33 node is adopted for simulation verification.

The schematic diagrams of a certain 5G base station and an active power distribution network are shown in fig. 4, the active power distribution network is hereinafter simply called as a power distribution network, 5G base stations containing energy storage are set up at 2-5 and 10-21 nodes, 5 base stations are set up under each node, and the equipment parameters of photovoltaic power station 4,5G base stations with the access capacity of 0.5MW are added into both node 8 and node 11 as shown in the following table 1.

Table 15 g base station apparatus parameters

The training set was randomly selected from 80% of the days of input data, and the rest of the data were test sets. In the improved space-time diagram convolutional neural network of the embodiment, the space-time diagram convolutional layer adopts a hidden layer, the number of units is 64, the full-connection layer contains 32 units, and a dropout layer with the rejection rate of 0.2 is added after the full-connection layer. The Adam algorithm is used as a training optimization algorithm, the initial learning rate is 0.001, the prediction time step length is 1h, the data resolution is 5min, and the model review period is 24h.

The root mean square error RMSE and mean absolute error MAE based on the prediction result and the true value evaluate the prediction result expression as follows:

Where T is the total length of the evaluation period, y _t is the true value, Is a model predictive value.

In order to better reflect the effect of the method on the aspects of the space characteristics of the base station and the photovoltaic, the present two traditional methods are adopted to test on a sample of a cloud day of 7 months in a test set, and two space-time diagram convolutional neural networks for fixing the parameters of an adjacent matrix are selected to compare. The two traditional methods are respectively that a first method directly limits the parameters of a base station and a light Fu Linjie matrix within a certain distance to be 1, other parameters of an adjacent matrix to be 0, namely a distance method, and a second method constructs the adjacent matrix based on a K nearest neighbor algorithm, namely a K nearest neighbor method.

The adjacency matrix of different methods is shown in fig. 5, wherein the thickness degree of the connecting line represents the strength of the spatial characteristic coupling relation between the 5G base station and the photovoltaic, and the strength of the coupling relation is mainly represented by the size of the adjacency matrix. Compared with the adjacency matrix based on distance, the adjacency weight calculated by the K nearest neighbor method considers the relation between the historical photovoltaic output and weather data among the stations, has larger change in the adjacency weight parameter structure and higher prediction accuracy, and the adjacency matrix self-adaptive correction method provided by the invention directly learns the influence of different adjacency matrix parameters on a prediction target from the historical data, can describe the relation among the photovoltaic stations more accurately, and further reduces the prediction error.

The prediction results are shown in fig. 6, RMSE and MAE on the full test set are shown in the following table 2, and as can be obtained from the prediction comparison results of fig. 6 and table 2, the prediction results of the present invention are closest to the true values, and the prediction accuracy is highest.

Table 2 prediction error under different methods

Fig. 7 shows the schedulable capacity of the 5G base station load rate prediction and energy storage in different areas, and the reserve capacity of each base station is different mainly because of the difference between the load characteristic of the base station and the number of hours of energy storage and standby, and fig. 7 shows that the standby time of the base station accessed by the node 13 is longer than that of the base station accessed by the node 5, so that the minimum reserve capacity is obviously increased, and the schedulable capacity interval is obviously reduced.

According to the adjacency relationship between the node accessed base station and the photovoltaic obtained by the invention, 5G base station+photovoltaic cluster division is carried out on the distribution network, and a schematic diagram is shown in figure 8. By considering the space-time connection relation between the 5G base station and the photovoltaic, the energy storage capacity and the photovoltaic adjustable capacity of the 5G base station are aggregated, so that the load power failure among distribution network clusters and among distribution network clusters can be effectively supported, and the distribution network load power failure change situation is shown in figure 9.

Fig. 10 shows a scheduling plan obtained by each cluster based on the present invention, and as can be seen from fig. 10, in a period of 00:00-05:00, the load demand is small, the load loss condition is almost no, the photovoltaic does not work, the electricity price is low in the period, the 5G base station energy storage device is charged, and the clusters purchase electricity from the distribution network to maintain the load demand in each cluster. During the period 07:00-18:00, the photovoltaic output of the cluster 1 and the cluster 2 is increased to meet the requirement of short-time high-capacity power loss in the load peak period, the energy storage equipment also starts to release power, and the power is increased to be sold out in the higher stage of the electricity price, so that economic benefits are obtained. And in the period of 21:00-23:00, the energy storage equipment of each cluster is charged, the load electricity shortage condition is slowed down in the period, and the electric quantity obtained by each cluster from the distribution network is reduced.

Fig. 11 illustrates the comparison of the random optimization method (Stochastic optimization, SO), the conventional robust optimization method (Robust optimization, RO) and the overall system operating cost under the present invention. From fig. 11, it can be seen that the overall trend is that the system according to the present invention has a smaller running cost, especially in the 14:00, 19:00, 22:00 periods, and the RO method and the SO method have a higher running cost, mainly because the aggregate capacity of the cluster itself is not accurately estimated, SO that the running cost is increased.

It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein. The solution in the embodiment of the application can be implemented in various computer languages, such as Java, C++, python, javaScript, etc., which are object-oriented programming languages.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

While preferred embodiments of the present application have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. It is therefore intended that the following claims be interpreted as including the preferred embodiments and all such alterations and modifications as fall within the scope of the application.

It will be apparent to those skilled in the art that various modifications and variations can be made to the present application without departing from the spirit or scope of the application. Thus, it is intended that the present application also include such modifications and alterations insofar as they come within the scope of the appended claims or the equivalents thereof.

Claims

The optimal scheduling method for the 1.5G base station and the active power distribution network is characterized by comprising the following steps:

Based on original input data of optimized scheduling of a 5G base station and an active power distribution network, combining a graph convolution and gating circulation unit to establish a space-time graph convolution neural network, and learning the relation of each node in a graph convolution layer graph structure in the space-time graph convolution neural network by using historical operation data of the 5G base station and the active power distribution network, and adaptively correcting adjacent matrix parameters of a graph convolution layer in the space-time graph convolution neural network to obtain an improved space-time graph convolution neural network;

calculating weights of different input features in original input data of optimal scheduling of the 5G base station and the active power distribution network by using a soft attention mechanism and a scaling point scoring function, and obtaining corrected input features;

inputting the corrected input characteristics into an improved space-time diagram convolutional neural network, and outputting to obtain a 5G base station predicted load rate, 5G base station predicted power and photovoltaic predicted output;

according to the 5G base station predicted power and the standby time of the 5G base station, calculating to obtain the energy storage real-time minimum standby capacity of the 5G base station;

according to the operation conditions of the 5G base station and the active power distribution network, establishing an objective function and an operation constraint condition which aim at optimizing the operation cost;

And solving an objective function through a column constraint generation algorithm based on the 5G base station energy storage real-time minimum reserve capacity, the 5G base station predicted load rate, the photovoltaic predicted output and the operation constraint condition to obtain a scheduling plan of the 5G base station and the active power distribution network, and optimally scheduling the 5G base station and the active power distribution network according to the scheduling plan.
2. The optimal scheduling method for the 5G base station and the active power distribution network according to claim 1, wherein in the improved space-time diagram convolutional neural network, the original input data of the optimal scheduling for the 5G base station and the active power distribution network comprises operation data of the 5G base station, geographic distribution data of the 5G base station, meteorological data of the 5G base station and historical operation data of the 5G base station and the active power distribution network;

The space-time diagram convolutional neural network model is represented as a topological graph g= (V, E, a), where V represents a point set, E represents an edge set, a is an adjacency matrix of the topological graph G, and the convolution range is a neighborhood B of node V _i:

B(v_i)＝{v_j|d(v_j,v_i)≤D}(1),

Wherein D (v _j,v_i) represents the shortest distance from the graph node v _i to v _j, D is the maximum convolution neighborhood, and when D=1, the size of the sampling area is the first-order neighborhood of the node v _i;

The forward propagation of the graph convolutional layer is as follows:

wherein, reLU refers to a modified linear unit activation function, X is a feature matrix of a topological graph G; is an adjacency matrix containing autocorrelation, i.e A is the adjacency matrix of the topology graph G, and I is the identity matrix; Is that

The ith row and jth column elements in the table; Is a normalized adjacent matrix, sigma (·) is a sigmoid function, W ₀ is a weight matrix from an upper layer to a picture volume layer, W ₁ is a weight matrix from a picture volume layer to a lower layer, Is a node degree matrix;

Combining a graph convolution unit with a gating circulation unit, introducing a graph convolution algorithm into the gating circulation unit to form a space-time graph convolution unit, forming a space-time graph convolution neural network through stacking of a plurality of space-time graph convolution units, capturing space-time dynamic changes of nodes in a graph, and deeply learning and analyzing space-time graph data, wherein the forward propagation formula is shown in the following formulas (4) - (7):

u_t＝σ(W_u[f(A,X_t),h_t-1]+b_u) (4),

r_t＝σ(W_r[f(A,X_t),h_t-1]+b_r) (5),

c_t＝tanh(W_c[f(A,X_t),(r_t*h_t-1)]+b_c) (6),

h_t＝u_t*h_t-1+(1-u_t)*c_t (7),

Wherein u _t、r_t、c_t is a unit internal parameter, which corresponds to the states of an update gate, a reset gate and a current unit at the time t respectively, h _t is a unit output at the time t, W _u and b _u are a weight matrix and a bias matrix of the update gate respectively, W _r and b _r are a weight matrix and a bias matrix of the reset gate, f (A, X _t) is a graph convolution operation input to a model at the time t, h _t-1 is a hidden state at the time t-1 at the last time, W _c is a weight matrix for calculating a candidate state c _t, and b _c is a bias vector associated with the candidate state c _t;

The method for adaptively correcting the adjacent matrix parameters of the graph convolution layer in the space-time graph convolution neural network comprises the following steps:

For the feature matrix x= (X ₁,x₂,…x_N)^T, define a _mn as the parameter of the m-th row and n-th column of the adjacency matrix, and the expression a _mn is:

Wherein x _m and x _n represent characteristics of a node m and a node N, ω is a weight vector which can be learned in the adjacency matrix parameter, and N is the total number of first-order neighbors of the node m;

Equation (8) uses ReLU units to limit the values of the adjacency matrix to non-negative numbers and normalizes a _mn using the form of softmax;

The weight parameters of a _mn are learned using the loss function L _graph of the following formula:

in the formula, lambda is a penalty coefficient of a regular term, F is a norm of the regular term, and when F-2 regularization is adopted, the formula is as follows:

The total loss function L _loss of the adaptive neural network is shown as follows:

L_loss＝L_mae+L_graph (11),

Where L _mae is the loss function of the original neural network, and the average absolute error is used.
3. The optimal scheduling method for the 5G base station and the active power distribution network according to claim 1, wherein the corrected input characteristics are obtained by the following steps:

When the period of the input feature is T, the feature weight vector α _T is:

Wherein X is an original input feature, Q is a query vector of an attention mechanism, namely a predicted target value corresponding to time, D _T is the total length of the input feature, alpha _T is calculated by a soft attention mechanism, and the input feature is normalized by using a scaling dot product scoring function to ensure that weights of different features are reasonably adjusted so as to optimize the input feature of a model;

And carrying out standardized operation on all the input features to obtain corrected input features, wherein the expression is as follows:

Wherein X _ref is the corrected input feature, X _mean is the average value of each feature component of the input X, X _std is the standard deviation of each feature component of the input X;

The input characteristic is standardized according to the formula (13) to generate a corrected input characteristic data set X _ref, the corrected input characteristic data set X _ref is input into an improved space-time diagram convolutional neural network model, and the 5G base station load rate, the power and the photovoltaic output in different time periods are predicted by utilizing multi-layer convolution operation and space-time dependency analysis;

The 5G base station predicted load rate λ _5G is represented by the following formula:

λ_5G＝{λ₁(t),λ₂(t),…,λ_N(t)} (14),

Wherein lambda _N (t) represents the load rate of the Nth 5G base station at the moment of time t;

The 5G base station predicted power P _5G is expressed as:

P_5G＝{P₁(t),P₂(t),…,P_N(t)} (15),

wherein, P _N (t) represents the power of the Nth 5G base station at the moment of time t;

The photovoltaic predicted force P _PV is represented by the formula:

P_PV＝{P₁(t),P₂(t),…,P_M(t)} (16),

where P _M (t) represents the power of the mth 5G base station at time t.
4. The optimal scheduling method of the 5G base station and the active power distribution network according to claim 3, wherein in the calculation of the 5G base station energy storage real-time minimum reserve capacity, 5G base station predicted power of different areas obtained in the formula (15) is combined with 5G base station minimum reserve power time t _s to obtain the 5G base station minimum reserve capacity at the time tThe following formula is shown:

Where P _5G is the 5G base station predicted power.
5. The optimal scheduling method for the 5G base station and the active power distribution network according to claim 4, wherein the expression of the objective function is:

In the formula, Representing short-time load electricity-shortage cost; representing the energy storage charging and discharging cost of the base station; Representing the cost of the photovoltaic output;

The specific expression is shown as follows:

Wherein c _LD、c_PV represents the load electricity-deficiency unit cost and the photovoltaic unit power generation cost respectively, and c _5G、c_SH represents the charging and discharging cost and the loss cost of the 5G base station energy storage unit respectively; respectively representing the charge and discharge power at the time t; η _5G represents the energy storage charging and discharging efficiency of the G base station; indicating the lack of power to the load;

The operation constraint conditions comprise energy storage charging and discharging constraint, photovoltaic output constraint and power balance constraint of the 5G base station;

the expression of the energy storage charging and discharging constraint of the 5G base station is as follows:

In the formula, AndThe charging power and the discharging power at time t are respectively indicated,AndRespectively representing the maximum charging power and the maximum discharging power at the time t, E ^5G represents the energy storage energy of the 5G base station,AndRespectively representing the minimum value and the maximum value of the energy storage energy of the 5G base station;

The expression of the photovoltaic output constraint is as follows:

In the formula, The photovoltaic output is indicated as being a function of the voltage,Representing the maximum output of the photovoltaic;

The expression of the power balance constraint is:

In the formula, The load power of the distribution network at the moment t is represented,AndThe charging power and the discharging power at time t are respectively indicated,Representing the photovoltaic output.
6. The optimal scheduling method for the 5G base station and the active power distribution network according to claim 5, wherein the solving method for the objective function is as follows:

the uncertainty of the photovoltaic and load is represented by the set Ω, as shown in the following formula:

In the formula, Actual output of the photovoltaic system is time t; Predicting the power of the photovoltaic system at the moment of time t; Representing the maximum value of the output uncertainty range of the photovoltaic system; Representing the actual load demand of the load at the moment t; Representing a maximum value of the load power uncertainty range; a predicted load demand representing a load at time t;

The formulas (18) - (22) are expressed as an uncertain two-stage robust optimization model by using a compact matrix equation, wherein the first stage is to minimize the running cost of the system, and the second stage is to obtain the worst running scene under the minimized cost, and the specific expression is as follows:

Wherein c ^T x represents an optimal objective function of the running cost, d ^T y represents the worst running cost when the load loses electricity, u represents an uncertain variable including photovoltaic and load power, max represents the worst scene under the highest cost in the optimal scheduling, and H, F, G respectively represents three different coefficient matrixes;

Based on a 5G base station energy storage real-time minimum standby capacity model, a 5G base station prediction load rate, a photovoltaic prediction output and operation constraint conditions, solving an uncertain two-stage robust optimization model by using a column constraint generation algorithm, firstly decoupling an objective function of the robust optimization model, and decomposing the objective function into a main problem and a sub-problem;

The main problems are as follows:

where m represents the number of iterations and y ^m represents the solution for m iterations of the sub-problem.

The sub-problems are:

by giving the (x, u) value, the transformation minimization problem is a deterministic linear optimization problem, expressed as follows:

Wherein d and y represent column vectors, A ₁、A₂、A₃ represents a matrix of m ₁、m₂、m₃ columns and n rows, and b ₁、b₂、b₃ represents m ₁、m₂、m₃ -dimensional column vectors;

the min model of the inner layer in the sub-problems of the robust optimization model is converted into a max model by using a dual theory and is combined with the max of the outer layer, and the conversion mode is as follows:

If the original problems are:

The dual form is:

Wherein z ₁、z₂、z₃ represents that m ₁、m₂、m₃ is a row vector;

according to the principle of dual theory, the dual form is:

and (3) obtaining a scheduling plan of the 5G base station and the active power distribution network based on formulas (24) - (30), wherein the scheduling plan comprises a scheduling plan of the 5G base station and a photovoltaic output plan.
The optimized dispatching system of the 7.5G base station and the active power distribution network is characterized by comprising an improved space-time diagram convolutional neural network building module, an input characteristic correcting module, a prediction result calculating module, a 5G base station energy storage real-time minimum standby capacity calculating module, an objective function and constraint building module and an optimized dispatching module;

The improved space-time diagram convolutional neural network establishing module is used for establishing a space-time diagram convolutional neural network by combining a diagram convolution and gating circulating unit based on original input data of optimized scheduling of a 5G base station and an active power distribution network, and learning the relation of each node in a diagram convolution layer diagram structure in the space-time diagram convolutional neural network by utilizing historical operation data of the 5G base station and the active power distribution network, and adaptively correcting adjacent matrix parameters of a diagram convolution layer in the space-time diagram convolutional neural network to obtain the improved space-time diagram convolutional neural network;

The input characteristic correction module is used for calculating weights of different input characteristics in the original input data of the 5G base station and the active power distribution network optimized scheduling by utilizing a soft attention mechanism and a scaling dot product scoring function to obtain corrected input characteristics;

The prediction result calculation module is used for inputting the corrected input characteristics into the improved space-time diagram convolutional neural network, and outputting to obtain a 5G base station prediction load rate, 5G base station prediction power and photovoltaic prediction output;

The 5G base station energy storage real-time minimum standby capacity calculation module is used for calculating and obtaining the 5G base station energy storage real-time minimum standby capacity according to the 5G base station predicted power and the standby time of the 5G base station;

the objective function and constraint establishing module is used for establishing an objective function and an operation constraint condition which aim at optimizing the operation cost according to the operation conditions of the 5G base station and the active power distribution network;

The optimal scheduling module is used for solving an objective function through a column constraint generation algorithm based on the 5G base station energy storage real-time minimum reserve capacity, the 5G base station predicted load rate, the photovoltaic predicted output and the operation constraint condition to obtain a scheduling plan of the 5G base station and the active power distribution network, and performing optimal scheduling on the 5G base station and the active power distribution network according to the scheduling plan.
8. The optimal scheduling system for the 5G base station and the active power distribution network according to claim 7, wherein in the improved space-time diagram convolutional neural network building module, the original input data of the optimal scheduling for the 5G base station and the active power distribution network comprises operation data of the 5G base station, geographic distribution data of the 5G base station, meteorological data of the 5G base station and historical operation data of the 5G base station and the active power distribution network;

The space-time diagram convolutional neural network model is represented as a topological graph g= (V, E, a), where V represents a point set, E represents an edge set, a is an adjacency matrix of the topological graph G, and the convolution range is a neighborhood B of node V _i:

B(v_i)＝{v_j|d(v_j,v_i)≤D} (1),

Wherein D (v _j,v_i) represents the shortest distance from the graph node v _i to v _j, D is the maximum convolution neighborhood, and when D=1, the size of the sampling area is the first-order neighborhood of the node v _i;

At this time, the forward propagation of the graph roll stack is as follows:

wherein, reLU refers to a modified linear unit activation function, X is a feature matrix of a topological graph G; is an adjacency matrix containing autocorrelation, i.e A is the adjacency matrix of the topology graph G, and I is the identity matrix; Is that The ith row and jth column element in the graph, sigma (·) is a sigmoid function, W ₀ is a weight matrix from the upper layer to the graph volume layer, W ₁ is a weight matrix from the graph volume layer to the lower layer,Is a node degree matrix;

Combining a graph convolution unit with a gating circulation unit, introducing a graph convolution algorithm into the gating circulation unit to form a space-time graph convolution unit, forming a space-time graph convolution neural network through stacking of a plurality of space-time graph convolution units, capturing space-time dynamic changes of nodes in a graph, and deeply learning and analyzing space-time graph data, wherein the forward propagation formula is shown in the following formulas (4) - (7):

u_t＝σ(W_u[f(A,X_t),h_t-1]+b_u) (4),

r_t＝σ(W_r[f(A,X_t),h_t-1]+b_r) (5),

c_t＝tanh(W_c[f(A,X_t),(r_t*h_t-1)]+b_c) (6),

h_t＝u_t*h_t-1+(1-u_t)*c_t (7),

Wherein u _t、r_t、c_t is a unit internal parameter, which corresponds to the states of an update gate, a reset gate and a current unit at the time t respectively, h _t is a unit output at the time t, W _u and b _u are a weight matrix and a bias matrix of the update gate respectively, W _r and b _r are a weight matrix and a bias matrix of the reset gate, f (A, X _t) is a graph convolution operation input to a model at the time t, h _t-1 is a hidden state at the time t-1 at the last time, W _c is a weight matrix for calculating a candidate state c _t, and b _c is a bias vector associated with the candidate state c _t;

The method for adaptively correcting the adjacent matrix parameters of the graph convolution layer in the space-time graph convolution neural network comprises the following steps:

For the feature matrix x= (X ₁,x₂,…x_N)^T, define a _mn as the parameter of the m-th row and n-th column of the adjacency matrix, and the expression a _mn is:

Wherein x _m and x _n represent characteristics of a node m and a node N, ω is a weight vector which can be learned in the adjacency matrix parameter, and N is the total number of first-order neighbors of the node m;

Equation (8) uses ReLU units to limit the values of the adjacency matrix to non-negative numbers and normalizes a _mn using the form of softmax;

The weight parameters of a _mn are learned using the loss function L _graph of the following formula:

in the formula, lambda is a penalty coefficient of a regular term, F is a norm of the regular term, and when F-2 regularization is adopted, the formula is as follows:

The total loss function L _loss of the adaptive neural network is shown as follows:

L_loss＝L_mae+L_graph (11),

Where L _mae is the loss function of the original neural network, and the average absolute error is used.
9. The optimal scheduling system for the 5G base station and the active power distribution network according to claim 7, wherein the input characteristic correction module is characterized in that the corrected input characteristic obtaining method comprises the following steps:

When the period of the input feature is T, the feature weight vector α _T is:

Wherein X is an original input feature, Q is a query vector of an attention mechanism, namely a predicted target value corresponding to time, D _T is the total length of the input feature, alpha _T is calculated by a soft attention mechanism, and the input feature is normalized by using a scaling dot product scoring function to ensure that weights of different features are reasonably adjusted so as to optimize the input feature of a model;

And carrying out standardized operation on all the input features to obtain corrected input features, wherein the expression is as follows:

Wherein X _ref is the corrected input feature, X _mean is the average value of each feature component of the input X, X _std is the standard deviation of each feature component of the input X;

The input characteristic is standardized according to the formula (13) to generate a corrected input characteristic data set X _ref which is the basis of subsequent prediction, the corrected input characteristic data set X _ref is input into an improved space-time diagram convolutional neural network model, and the 5G base station load rate, power and photovoltaic output in different time periods are predicted by utilizing multi-layer convolutional operation and space-time dependency analysis;

The 5G base station predicted load rate λ _5G is represented by the following formula:

λ_5G＝{λ₁(t),λ₂(t),…,λ_N(t)} (14),

Wherein lambda _N (t) represents the load rate of the Nth 5G base station at the moment of time t;

The 5G base station predicted power P _5G is expressed as:

P_5G＝{P₁(t),P₂(t),…,P_N(t)} (15),

wherein, P _N (t) represents the power of the Nth 5G base station at the moment of time t;

The photovoltaic predicted force P _PV is represented by the formula:

P_PV＝{P₁(t),P₂(t),…,P_M(t)} (16),

where P _M (t) represents the power of the mth 5G base station at time t.
10. The optimal scheduling system for the 5G base station and the active power distribution network according to claim 9, wherein in the 5G base station energy storage real-time minimum reserve capacity calculation module, 5G base station predicted power of different areas obtained in the formula (15) is combined with 5G base station minimum reserve power time t _s to obtain the minimum reserve capacity of the 5G base station at the time tThe following formula is shown:

Where P _5G is the 5G base station predicted power.
11. The optimal scheduling system for the 5G base station and the active power distribution network according to claim 10, wherein in the objective function and constraint building module, the expression of the objective function is:

In the formula, Representing short-time load electricity-shortage cost; representing the energy storage charging and discharging cost of the base station; The specific expression of the photovoltaic output cost is shown as follows:

Wherein c _LD、c_PV represents the load electricity-deficiency unit cost and the photovoltaic unit power generation cost respectively, and c _5G、c_SH represents the charging and discharging cost and the loss cost of the 5G base station energy storage unit respectively; respectively representing the charge and discharge power at the time t; η _5G represents the energy storage charging and discharging efficiency of the G base station; indicating the lack of power to the load;

The operation constraint conditions comprise energy storage charging and discharging constraint, photovoltaic output constraint and power balance constraint of the 5G base station;

the expression of the energy storage charging and discharging constraint of the 5G base station is as follows:

In the formula, AndThe charging power and the discharging power at time t are respectively indicated,AndRespectively representing the maximum charging power and the maximum discharging power at the time t, E ^5G represents the energy storage energy of the 5G base station,AndRespectively representing the minimum value and the maximum value of the energy storage energy of the 5G base station;

The expression of the photovoltaic output constraint is as follows:

In the formula, The photovoltaic output is indicated as being a function of the voltage,Representing the maximum output of the photovoltaic;

The expression of the power balance constraint is:

In the formula, The load power of the distribution network at the moment t is represented,AndThe charging power and the discharging power at time t are respectively indicated,Representing the photovoltaic output.
12. The optimal scheduling system for the 5G base station and the active power distribution network according to claim 11, wherein in the optimal scheduling module, the solving method of the objective function is as follows:

the uncertainty of the photovoltaic and load is represented by the set Ω, as shown in the following formula:

In the formula, Actual output of the photovoltaic system is time t; Predicting the power of the photovoltaic system at the moment of time t; Representing the maximum value of the output uncertainty range of the photovoltaic system; Representing the actual load demand of the load at the moment t; Representing a maximum value of the load power uncertainty range; a predicted load demand representing a load at time t;

The formulas (18) - (22) are expressed as an uncertain two-stage robust optimization model by using a compact matrix equation, wherein the first stage is to minimize the running cost of the system, and the second stage is to obtain the worst running scene under the minimized cost, and the specific expression is as follows:

Wherein c ^T x represents an optimal objective function of the running cost, d ^T y represents the worst running cost when the load loses electricity, u represents an uncertain variable including photovoltaic and load power, max represents the worst scene under the highest cost in the optimal scheduling, and H, F, G respectively represents three different coefficient matrixes;

Based on a 5G base station energy storage real-time minimum standby capacity model, a 5G base station prediction load rate, a photovoltaic prediction output and operation constraint conditions, solving an uncertain two-stage robust optimization model by using a column constraint generation algorithm, firstly decoupling an objective function of the robust optimization model, and decomposing the objective function into a main problem and a sub-problem;

The main problems are as follows:

where m represents the number of iterations and y ^m represents the solution for m iterations of the sub-problem.

The sub-problems are:

by giving the (x, u) value, the transformation minimization problem is a deterministic linear optimization problem, expressed as follows:

Wherein d and y represent column vectors, A ₁、A₂、A₃ represents a matrix of m ₁、m₂、m₃ columns and n rows, and b ₁、b₂、b₃ represents m ₁、m₂、m₃ -dimensional column vectors;

the min model of the inner layer in the sub-problems of the robust optimization model is converted into a max model by using a dual theory and is combined with the max of the outer layer, and the conversion mode is as follows:

If the original problems are:

The dual form is:

Wherein z ₁、z₂、z₃ represents that m ₁、m₂、m₃ is a row vector;

according to the principle of dual theory, the dual form is:

and (3) obtaining a scheduling plan of the 5G base station and the active power distribution network based on formulas (24) - (30), wherein the scheduling plan comprises a scheduling plan of the 5G base station and a photovoltaic output plan.
13. An electronic device, comprising a memory and a processor, wherein the memory and the processor are in communication connection with each other, the memory stores computer instructions, and the processor implements the optimized scheduling method of the 5G base station and the active power distribution network according to any one of claims 1-6 by executing the computer instructions.
14. A computer readable storage medium storing a computer program, wherein the computer program when executed by a processor implements the optimized scheduling method of a 5G base station and an active power distribution network according to any one of claims 1-6.