Intelligent power grid expansion planning and stability prediction method based on 5G and CNN

Abstract

The invention relates to a smart power grid extension planning and stability prediction method based on 5G and CNN, and the method comprises the following steps: 1, building an energy consumption calculation model of a 5G base station: the energy consumption calculation model comprises baseline energy consumption and incremental energy consumption, and the baseline energy consumption is divided into BBU baseline energy consumption and RRU baseline energy consumption; step 2, probability distribution of 5G base station load energy consumption: constructing a probability model of the 5G base station load energy consumption; 3, a power distribution network probability planning model considering the 5G base station load energy consumption probability characteristic is constructed, and line capacity expansion of a power distribution network and capacity expansion of a 110kV transformer substation transformer are planned with the load energy consumption probability characteristic of the base station as the basic condition; step 4, mathematical modeling based on a convolutional neural network: adopting the CNN introducing a nonlinear function to predict the stability of the smart power grid; 5, predicting the stability of the intelligent power grid based on the CNN; the method has the advantages that the energy consumption calculation model is established, the active load probability distribution model is constructed, and the power network stability prediction model is formed.

Description

Technical Field

[0001] The present invention belongs to the technical field of smart grids, and specifically relates to a smart grid expansion planning and stability prediction method based on 5G and CNN. Background Art

[0002] In the future, 5G base stations will be integrated into urban power grids in large numbers. Since the time and purpose of users' use of mobile networks are not fixed, the business traffic generated by their communications will be random and intermittent, which will cause the energy consumption of 5G base stations to fluctuate uncertainly with the access of a large number of users, thus affecting the configuration of the power supply system of 5G base stations. However, current research has not taken into account the impact of the new topological structure of 5G base stations, and has not fully considered the regular characteristics of business traffic in different regions and categories. In addition, with the large-scale construction of 5G base stations, the uncertainty of the capacity configuration of its power supply system will affect the expansion planning of the distribution network, and increase the difficulty of capacity configuration and capacity expansion of electrical components such as main transformers and lines. Therefore, it is necessary to analyze the impact of the topological structure of 5G base stations and the probabilistic characteristics of massive business traffic of users on the energy consumption of 5G base stations. , it is of great significance to study the distribution network planning method taking into account the 5G base station load; in addition, with the research of 5G communication technology and the construction of new infrastructure, smart grids have developed rapidly, which can improve the service quality of power systems to a great extent. At the same time, in the era of big data, the interconnection of all things has led to the access of massive sensors and other access devices in the power network to the power network, which has also brought a large burden to the smart grid. In order to provide users with a steady supply of qualified electricity, the power system is required to have high stability and reliability, and the stability problem of the power network needs to be solved urgently; therefore, it is very necessary to provide a smart grid expansion planning and stability prediction method based on 5G and CNN to establish an energy consumption calculation model, construct an active load probability distribution model, and form a power network stability prediction model. Summary of the invention

[0003] 1. Technical issues

[0004] In view of the above-mentioned existing technical status, this application mainly addresses the following technical problems:

[0005] 1. How to comprehensively analyze the impact of 5G base station topology and the probabilistic characteristics of massive user traffic on the energy consumption of 5G base stations, and study the distribution network planning method taking into account the load of 5G base stations;

[0006] 2. How to ensure the stability and reliability of the power network when a large number of sensors and other access devices are connected to the power network due to the Internet of Things?

[0007] (II) Technical solution

[0008] The purpose of the present invention is to overcome the shortcomings of the prior art and to provide a smart grid expansion planning and stability prediction method based on 5G and CNN, which establishes an energy consumption calculation model, constructs an active load probability distribution model, and forms a power network stability prediction model.

[0009] The object of the present invention is achieved by: a smart grid expansion planning and stability prediction method based on 5G and CNN, the method comprising the following steps:

[0010] Step 1: Establish an energy consumption calculation model for 5G base stations: including baseline energy consumption and incremental energy consumption, where baseline energy consumption is divided into BBU baseline energy consumption and RRU baseline energy consumption;

[0011] Step 2: Probability distribution of 5G base station load energy consumption: Construct a probability model of 5G base station load energy consumption;

[0012] Step 3: Construct a distribution network probabilistic planning model that takes into account the probabilistic characteristics of 5G base station load energy consumption: Based on the probabilistic characteristics of base station load energy consumption, plan the expansion of distribution network lines and 110kV substation transformers;

[0013] Step 4: Mathematical modeling based on convolutional neural network: CNN with nonlinear function is used to predict the stability of smart grid;

[0014] Step 5: CNN-based smart grid stability prediction.

[0015] Furthermore, the baseline energy consumption represents the energy consumption used to send fixed wireless signals when no user accesses the 5G base station, including the reference signal CRS, the primary synchronization signal PSS, the secondary synchronization signal SSS, the physical broadcast channel PBCH, the physical control format indicator channel and the physical hybrid automatic repeat request indicator channel PHICH; the incremental energy consumption represents the energy consumption to meet the uplink and downlink data service traffic requirements when a large number of users access the 5G base station, including the RRU module energy consumption.

[0016] Furthermore, the energy consumption calculation model of the 5G base station in step 1 is specifically as follows: Assuming that the total energy consumption of the 5G base station per hour is E total Indicates that the RRU incremental energy consumption and baseline energy consumption are respectively E RRU,incre and E RRU,base Indicates that the BBU baseline energy consumption is E BBU,base The energy consumption calculation model of 5G base station can be expressed as: E total =E RRU,incre +E RRU,base +E BBU,base .

[0017] Furthermore, the RRU incremental energy consumption is specifically: the incremental energy consumption E of the RRU RRU,incre , the calculation formula is: E RRU,incre =E DLd,incre +E DLs,incre , where E DLd,incre E is the incremental energy consumption of downlink data generated by RRU during a given observation time; DLs,incre It is the incremental energy consumption generated by downlink signaling of RRU within a given observation time.

[0018] Furthermore, the RRU incremental energy consumption is specifically: when the physical resource utilization rate of the base station is zero, the baseline power consumption P of the RRU in_base It can be decomposed into the sum of the baseline power consumption of RRU downlink data, RRU downlink signaling and RRU uplink data. Therefore, the baseline energy consumption of RRU during the observation time is E RRU,base Equal to: E RRU,incre =E DLd,base +E DLs,base +E ULd,base , where E DLd,base E is the baseline energy consumption generated by RRU downlink data per hour; DLs,base is the baseline energy consumption generated by RRU downlink signaling; E ULd,base Baseline energy consumption generated by RRU uplink data.

[0019] Furthermore, the probability model of 5G base station load energy consumption in step 2 is as follows: Assuming that the building complex adopts H i (i∈Φ) indicates that Φ is the set of all types of building clusters within the radiation radius of the base station; the building cluster H i The corresponding population density ρ i The average population density during the observation time is obtained through the historical heat map; the number of mobile phone activations in the building complex satisfies λ i ~N(a λ ,b λ ) is normally distributed; then, the building complex H i The expected total population is E i =ρ i N i , N i For the building complex H i Area; Building complex H i The expected number of mobile phone activations R i =E i λ i Considering that the services provided by mobile phone users can be divided into mobile video services and non-mobile video services, the building group H i The ratio of the number of these two businesses satisfies δ i ~N(a δ ,bδ ), then the building complex H i PDCP layer traffic of mobile video and non-mobile video are respectively equal to: In the formula, the traffic speeds corresponding to the mobile video service and the non-mobile video service are uniformly distributed, where the random variables y1, y2 and w1, w2 are the downlink and uplink data traffic speeds of the user's mobile video service and non-mobile video; z1 and z2 are the total number of resource blocks occupied by the control channels of the mobile video service and the non-mobile video service, respectively;

[0020] Since the ratio of mobile video services to non-mobile video services is δ i , and its cellular Internet access time is also estimated approximately in proportion; thus, the building complex H i Duration of mobile video and non-mobile video traffic within 10 minutes are respectively equal to: In the formula, the random variable t i For the building complex H i The probability density function of the Beta distribution of the downlink data and signaling service duration corresponding to the users in the network is: In the formula, Γ(x) is the Γ function; α and β are the characteristic parameters of the Beta distribution; Substituting the above formula into the energy consumption calculation model, the total energy consumption of the 5G base station within 10 minutes can be obtained: The probability distribution of is:

[0021] Furthermore, the step 3 is specifically as follows: the load energy consumption corresponding to the i-th line of the 5G base station connected to the distribution network is The maximum total load P of the i-th distribution network line L,i ; The remaining capacity of the 110kV substation transformer is S k (k∈Ω), Ω is the set of all transformers in a certain planning area; the allowed capacity of the i-th distribution network line is M i (i∈Λ k ), Λ k is the set of all distribution network lines supplied by transformer k in the station; the variable for the expansion of the distribution network line 0-1 is x, and the variable for the expansion of the 110kV substation transformer 0-1 is x T , then the objective function of the short-term planning model for distribution network reconnection is: minF = B T x1+L T x T, where B is the cost vector for capacity expansion of the distribution network line; L is the cost vector for capacity expansion of the 110kV substation transformer; accordingly, the constraints are the 10kV feeder capacity constraint and the 110kV main transformer capacity constraint, and the 0-1 integer programming model of the distribution network taking into account the probabilistic characteristics of 5G base station energy consumption is:

[0022] Furthermore, the mathematical modeling based on convolutional neural network in step 4 is specifically: the formula of the neuron model is: Among them, w i is the weight of the linear function; b is the bias unit; then, the model needs to be optimized through the loss function, which is defined as follows: Where n is the number of training sets; h (i) is the predicted output; y (i) True value.

[0023] Furthermore, the CNN-based smart grid stability prediction in step 5 includes the following steps:

[0024] Step 5.1: Preprocess the data using standardization: First, calculate the mathematical expectation E(X i ) and standard deviation S i , and then the standardized value is obtained according to the following formula: Among them, X ij is the jth value of the i-th feature of the data; E(X i ) is the mean of the i-th feature; S i is the standard deviation of the i-th feature;

[0025] Step 5.2: Convert the 12 features obtained after the standardized power network dataset into a matrix and input it into the CNN model for feature extraction;

[0026] Step 5.3: Update the weights of the pooling layer in the CNN model;

[0027] Step 5.4: Use multiple optimization algorithms to optimize the CNN model;

[0028] Step 5.5: To avoid significant deviations in the algorithm in early iterations, a deviation correction is added to the formula;

[0029] Step 5.6: After adding bias correction, the CNN model can continuously update the weights and biases in the iterative process to optimize the prediction effect of the model.

[0030] Furthermore, the weight update of the convolutional layer in the CNN model in step 5.2 includes the following steps:

[0031] Step 5.21: First, we need to define the error function, the formula is as follows: Among them, N is the number of samples; K is the category of the sample; represents the nth sample y k The kth dimension of is the nth sample p k The kth dimension of the network output;

[0032] Step 5.22: Next, define the output of the convolutional layer l for: in, represents the jth feature of the i-th layer; f is the activation function of CNN; M j The feature matrix representing the input data; is the convolution kernel that connects the input i-th feature and the output j-th feature in layer l; Then it is the bias parameter corresponding to the j feature in layer l;

[0033] Step 5.23: Error Sensitivity It can be expressed as: in, w is the weight; b is the bias unit; layer l is the convolution layer; l+1 is the pooling layer; up is the upsampling operation that makes the size of the matrix of layer l+1 consistent with that of layer l, β j is the convolution kernel The value in ;

[0034] Step 5.24: Next, find the error function with respect to the bias and convolution kernel Partial derivative of: The above formula is the sum of all error sensitivities in layer l, and (u,v) is the position of each element in the sensitivity matrix.

[0035] (III) Beneficial effects

[0036] 1. Through the analysis of the actual physical structure of 5G base stations and the derivation of the probabilistic characteristics of business traffic, an energy consumption calculation model taking into account the probabilistic characteristics of 5G base station loads was established;

[0037] 2. A distribution network expansion planning method that takes into account the probabilistic characteristics of 5G base station loads is proposed, and the problem of increased difficulty in distribution network expansion planning caused by the uncertainty of 5G base station loads is solved. This is of great significance for reasonable distribution network expansion planning after 5G is widely connected to the distribution network in the future;

[0038] 3. A CNN-based smart grid stability prediction algorithm is proposed. The data generated by the power network is collected, processed by the CNN model, and finally the judgment result of the smart grid stability is output;

[0039] 4. The convolutional neural network-based smart grid stability prediction algorithm of the present invention first processes the data collected from the smart grid, and then inputs it into the CNN model. After continuous iterative calculations, a power network stability prediction model is finally obtained; the algorithm consists of two convolutional layers, two pooling layers and two fully connected layers, and the algorithm has a good accuracy rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 This is a single neuron modeling diagram of the smart grid expansion planning and stability prediction method based on 5G and CNN in the present invention.

[0041] Figure 2 This is a CNN-based smart grid stability prediction modeling diagram of the smart grid expansion planning and stability prediction method based on 5G and CNN in the present invention. DETAILED DESCRIPTION

[0042] The present invention is further described below in conjunction with embodiments and / or drawings.

[0043] Example 1

[0044] like Figure 1-2 As shown, a smart grid expansion planning and stability prediction method based on 5G and CNN includes the following steps:

[0045] Step 1: Establish an energy consumption calculation model for 5G base stations: including baseline energy consumption and incremental energy consumption, where baseline energy consumption is divided into BBU baseline energy consumption and RRU baseline energy consumption;

[0046] In the present invention, the baseline energy consumption represents the energy consumption used to send fixed wireless signals when no user accesses the 5G base station, including the reference signal CRS, the primary synchronization signal PSS, the secondary synchronization signal SSS, the physical broadcast channel PBCH, the physical control format indicator channel and the physical hybrid automatic repeat request indicator channel PHICH; the incremental energy consumption represents the energy consumption for meeting the uplink and downlink data service traffic requirements when a large number of users access the 5G base station, including the RRU module energy consumption.

[0047] The energy consumption calculation model of 5G base stations is as follows: Assume that the total energy consumption of 5G base stations per hour is E total Indicates that the RRU incremental energy consumption and baseline energy consumption are respectively E RRU,incre and E RRU,base Indicates that the BBU baseline energy consumption is E BBU,base The energy consumption calculation model of 5G base station can be expressed as: E total =E RRU,incre +E RRU,base +EBBU,base .

[0048] ① RRU incremental energy consumption: The incremental energy consumption of RRU E RRU,incre , the calculation formula is: E RRU,incre =E DLd,incre +E DLs,incre , where E DLd,incre E is the incremental energy consumption of downlink data generated by RRU during a given observation time; DLs,incre is the incremental energy consumption generated by downlink signaling of the RRU within a given observation time;

[0049] E DLd,incre It can be determined by the proportion of downlink data traffic passing through the packet data convergence protocol layer PDCP to the maximum downlink throughput within a given observation time; E DLs,incre It can be determined by the ratio of the resource blocks required for downlink data and downlink signaling to the total number of resource blocks within a given observation time; thus, E DLd,incre and E DLs,incre The calculation model is: Where P in_max is the maximum power consumption of the RRU equipment; P in_base is the measured baseline power consumption of the RRU; T max_DLd T is the maximum downlink throughput of the PDCP layer for transmission services; OTT_DLd To transmit the downlink data traffic of the PDCP layer of the service; OTT_DLd is the duration of the downlink data transmission service within a given observation time; R DLd is the number of resource blocks occupied by downlink data; R DLs is the number of resource blocks occupied by downlink signaling; R OTT_DLs is the total number of resource blocks occupied by the 5G base station control channel; D OTT_DLs is the signaling duration of the 5G base station within a given observation time;

[0050] According to the above formula, E DLd,incre It's about T OTT_DLd and D OTT_DLd Function of E DLs,incre It's about R OTT_DLs and D OTT_DLs function, so the incremental energy consumption of RRU is E RRU,incre Equal to: E RUU,incre =E DLd,incre (T OTT_DLd ,D OTT_DLd )+E DLs,incre (R OTT_DLs ,D OTT_DLs ).

[0051] ② RRU incremental energy consumption: When the physical resource utilization of the base station is zero, the baseline power consumption of the RRU is P in_base It can be decomposed into the sum of the baseline power consumption of RRU downlink data, RRU downlink signaling and RRU uplink data. Therefore, the baseline energy consumption of RRU during the observation time is E RRU,base Equal to: E RRU,incre =E DLd,base +E DLs,base +E ULd,base , where E DLd,base E is the baseline energy consumption generated by RRU downlink data per hour; DLs,base is the baseline energy consumption generated by RRU downlink signaling; E ULd,base is the baseline energy consumption generated by the RRU uplink data, and the calculation formulas are:

[0052] Where, T(t) = T max_DLd +T max_ULd , where T max_DLd , T max_ULd are the maximum downlink and uplink throughputs of the cell PDCP layer respectively; ρ data is the proportion of uplink and downlink data in the total uplink and downlink resources; ρ DLs U is the proportion of downlink signaling in the total uplink and downlink resources; max is the proportion of fixed signaling overhead in the transmission process to the total power consumption; T OTT_ULd To transmit the uplink data traffic of the PDCP layer of the service; OTT_ULd The duration of uplink data transmission in a unit hour;

[0053] According to the above formula, E DLd,base It's about T OTT_DLd and D OTT_DLd Function of E DLs,base It's about T OTT_DLs and D OTT_DLs Function of E ULd,base It's about T OTT_ULd and D OTT_ULd function; therefore, the baseline energy consumption of RRU E RUU,base Equal to: E RUU,base =E DLd,base (T OTT_DLd ,D OTT_DLd )+E DLs,base (R OTT_DLs ,D OTT_DLs )+E ULs,base (T OTT_ULs ,D OTT_ULs), substituting the incremental energy consumption of RRU and the baseline energy consumption of RRU into the energy consumption calculation model, the total energy consumption E of the base station wireless equipment of the 5G base station within a given observation time can be obtained. total It is closely related to the business type composition and business traffic. Under the same business type, E total It is directly proportional to the PDCP layer downlink data flow, duration, total number of resource blocks occupied by the control channel and signaling duration corresponding to the service.

[0054] Step 2: Probability distribution of 5G base station load energy consumption: Construct a probability model of 5G base station load energy consumption;

[0055] In the present invention, the probability model of 5G base station load energy consumption in step 2 is specifically as follows: Assuming that the building complex adopts H i (i∈Φ) indicates that Φ is the set of all types of building clusters within the radiation radius of the base station; the building cluster H i The corresponding population density ρ i The average population density during the observation time is obtained through the historical heat map; the number of mobile phone activations in the building complex satisfies λ i ~N(a λ ,b λ ) is normally distributed; then, the building complex H i The expected total population is E i =ρ i N i , N i For the building complex H i Area; Building complex H i The expected number of mobile phone activations R i =E i λ i Considering that the services provided by mobile phone users can be divided into mobile video services and non-mobile video services, the building group H i The ratio of the number of these two businesses satisfies δ i ~N(a δ ,b δ ), then the building complex H i PDCP layer traffic of mobile video and non-mobile video are respectively equal to: Where λ is the ratio of the number of activated mobile phones of users in different building groups within the radiation radius of the base station; δ is the ratio of the number of users in different building groups carrying out mobile video services and non-mobile video services; y1 and y2 are the downlink and uplink data traffic speeds of the PDCP layer services of mobile video services, respectively; w1 and w2 are the downlink and uplink data traffic speeds of the PDCP layer services of non-mobile video services, respectively; the traffic speeds corresponding to mobile video services and non-mobile video services are uniformly distributed, among which the random variables y1, y2 and w1, w2 are the downlink and uplink data traffic speeds of users' mobile video services and non-mobile video services, and their probability density function expressions are as follows: z1 and z2 are the total number of resource blocks occupied by the mobile video service and non-mobile video service control channels, respectively, and their values ​​are determined by y1 and y2;

[0056] Since the ratio of mobile video services to non-mobile video services is δ i , and its cellular Internet access time is also estimated approximately in proportion; thus, the building complex H i Duration of mobile video and non-mobile video traffic within 10 minutes are respectively equal to: In the formula, the random variable t i For the building complex H i The probability density function of the Beta distribution of the downlink data and signaling service duration corresponding to the users in the network is: Where Γ(x) is the Γ function; α and β are the characteristic parameters of Beta distribution;

[0057] Substituting the above formula into the energy consumption calculation model, the total energy consumption of the 5G base station within 10 minutes can be obtained. In order to further obtain the load energy consumption of 5G base stations to support the expansion planning of distribution networks, the present invention assumes that the probability distribution of user cellular traffic every 10 minutes in each time period is the same, and the probability distribution of the load energy consumption of 5G base stations is All buildings within the base station radiation radius of approximately 10 minutes H i Average energy consumption of mobile video services and non-mobile video services Right now:

[0058] Step 3: Construct a distribution network probabilistic planning model that takes into account the probabilistic characteristics of 5G base station load energy consumption: Based on the probabilistic characteristics of base station load energy consumption, plan the expansion of distribution network lines and 110kV substation transformers;

[0059] In the present invention, step 3 is specifically as follows: take the load energy consumption corresponding to the ith line of the 5G base station connected to the distribution network as The maximum total load P of the i-th distribution network lineL,i ; The remaining capacity of the 110kV substation transformer is S k (k∈Ω), Ω is the set of all transformers in a certain planning area; the allowed capacity of the i-th distribution network line is M i (i∈Λ k ), Λ k is the set of all distribution network lines supplied by transformer k in the station; the variable for the expansion of the distribution network line 0-1 is x, and the variable for the expansion of the 110kV substation transformer 0-1 is x T , then the objective function of the short-term planning model for distribution network reconnection is: minF = B T x1+L T x T , where B is the cost vector for capacity expansion of distribution network lines; L is the cost vector for capacity expansion of 110kV substation transformers;

[0060] Accordingly, the constraints are 10kV feeder capacity constraint and 110kV main transformer capacity constraint. The specific constraints are expressed as follows:

[0061] ① Capacity constraint of the i-th 10kV line:

[0062] ②110kV transformer remaining capacity constraints: In the formula, S load The maximum load of the station except the 10kV distribution network line;

[0063] In summary, the 0-1 integer programming model of the distribution network taking into account the probabilistic characteristics of 5G base station energy consumption is:

[0064]

[0065] Step 4: Mathematical modeling based on convolutional neural network: CNN with nonlinear function is used to predict the stability of smart grid;

[0066] Step 5: CNN-based smart grid stability prediction.

[0067] The present invention is a smart grid expansion planning and stability prediction method based on 5G and CNN. During use, the present invention first derives an energy consumption calculation model of a 5G base station within a given time according to the communication physical equipment composition of the 5G base station; on this basis, the influence of user distribution of different building complexes, user access characteristics in different building complexes and traffic characteristics of different services on the energy consumption of the 5G base station is analyzed, and an active load probability distribution model of the 5G base station is established; after fully considering the widespread access of 5G base stations to the distribution network, a distribution network expansion planning method taking into account the load probability characteristics of 5G base stations is proposed; the present invention has the advantages of establishing an energy consumption calculation model, constructing an active load probability distribution model, and forming a power network stability prediction model.

[0068] Example 2

[0069] like Figure 1-2 As shown, a smart grid expansion planning and stability prediction method based on 5G and CNN includes the following steps:

[0070] Step 1: Establish an energy consumption calculation model for 5G base stations: including baseline energy consumption and incremental energy consumption, where baseline energy consumption is divided into BBU baseline energy consumption and RRU baseline energy consumption;

[0071] Step 2: Probability distribution of 5G base station load energy consumption: Construct a probability model of 5G base station load energy consumption;

[0072] Step 3: Construct a distribution network probabilistic planning model that takes into account the probabilistic characteristics of 5G base station load energy consumption: Based on the probabilistic characteristics of base station load energy consumption, plan the expansion of distribution network lines and 110kV substation transformers;

[0073] Step 4: Mathematical modeling based on convolutional neural network: CNN with nonlinear function is used to predict the stability of smart grid;

[0074] In the present invention, specifically: the formula of the neuron model is: Among them, w i is the weight of the linear function; b is the bias unit; then, the model needs to be optimized through the loss function, which is defined as follows: Where n is the number of training sets; h (i) is the predicted output; y (i) True value; The CNN-based smart grid stability prediction model automatically extracts 12 features related to grid stability through the convolution layer, and then reduces the dimension of the data through the pooling layer, and finally obtains the prediction result through the fully connected layer and the Sofmax classification function; The Accuracy value that measures the quality of the CNN model ranges from 0 to 1, and the Loss value is generally below 1. The lower the Accuracy value and the higher the Loss value, the worse the prediction result of the model, and vice versa.

[0075] Step 5: CNN-based smart grid stability prediction.

[0076] The present invention comprises the following steps:

[0077] Step 5.1: Preprocess the data using standardization: First, calculate the mathematical expectation E(X i ) and standard deviation S i , and then the standardized value is obtained according to the following formula: Among them, X ijis the jth value of the i-th feature of the data; E(X i ) is the mean of the i-th feature; S i is the standard deviation of the i-th feature; the CNN model automatically extracts 12 features such as the reaction time, standard power, and price coefficient of participants in the power network, and standardizes them. In order to input the data into the CNN model, it is necessary to adjust the data structure and convert it into a feature matrix with the same size as the CNN input matrix;

[0078] Step 5.2: Convert the 12 features obtained after the standardized power network dataset into a matrix and input it into the CNN model for feature extraction;

[0079] The weight update of the convolutional layer in the CNN model includes the following steps:

[0080] Step 5.21: First, we need to define the error function, the formula is as follows: Among them, N is the number of samples; K is the category of the sample (that is, the sample is divided into K categories); represents the nth sample y k The kth dimension of is the nth sample p k The kth dimension of the network output; C in the minimization formula can update the weights of the network, making the output prediction value of the CNN model closer to the true value;

[0081] Step 5.22: Next, define the output of the convolutional layer l for: in, represents the jth feature of the i-th layer; f is the activation function of CNN; M j The feature matrix representing the input data; is the convolution kernel that connects the input i-th feature and the output j-th feature in layer l; Then it is the bias parameter corresponding to the j feature in layer l;

[0082] Step 5.23: Error Sensitivity It can be expressed as: in, w is the weight; b is the bias unit; layer l is the convolution layer; l+1 is the pooling layer; up is the upsampling operation that makes the size of the matrix of layer l+1 consistent with that of layer l, β j is the convolution kernel The value in ;

[0083] Step 5.24: Next, find the error function with respect to the bias and convolution kernel Partial derivative of: The above formula is the sum of all error sensitivities in layer l, and (u,v) is the position of each element in the sensitivity matrix.

[0084] Step 5.3: Update the weights of the pooling layer in the CNN model;

[0085] In the present invention, specifically: ① the output of the first layer of the pooling layer Can be defined as: Among them, f is the activation function; down is the downsampling operation; is bias;

[0086] ② Order Error sensitivity It can be expressed as: Error function C in the pooling layer to bias b j The partial derivative of is equal to the error function E in the convolutional layer with respect to the bias b j The partial derivative of

[0087] In the CNN model, the ReLU function is used as the activation unit. The gradient of the ReLU function is always 1 when x>0, and is always 0 when x<0, which provides a guarantee for the rapid convergence of the CNN model. The definition of the activation unit ReLU function is: f(x)=max(0,x).

[0088] Step 5.4: Use multiple optimization algorithms to optimize the CNN model;

[0089] In the present invention, various optimization algorithms such as Adam, RMSprop, and SGD with momentum (MSGD) are used to optimize the CNN model; wherein ① the SGD algorithm with momentum updates the weight w and bias b. The SGD with momentum will have a jump on the accumulated gradient of the previous step, and then a correction unit is added to correct it. The update formula is as follows: dw =βv dw +(1-β)dw,v db =βv db +(1-β)db,w=w-αv dw , b=b-αv db , where β is a hyperparameter with a value of about 0.9; α is the learning rate with a value of about 0.01.

[0090] ②The update formula of RMSpro optimization algorithm is as follows: dw =βs dw +(1-β)dw 2 ,s db =βs db +(1-β)db 2 , Among them, β is a hyperparameter, with a value of about 0.9; α is the learning rate, with a value of about 0.01; ε is a deviation correction unit, with a value of generally 10 -8 .

[0091] ③Adam combines the Momentum gradient descent method and the RMSprop algorithm. The parameter update formula of the Adam algorithm is as follows: dw =β1v dw +(1-β1)dw,v db =β1v db +(1-β1)db,s dw =β2s dw +(1-β2)dw 2 ,s db =β2s db +(1-β2)db 2 , where β1 and β2 are the exponents of gradient accumulation, which are two hyperparameters. In general, β1 takes the value of 0.9 and β2 takes the value of 0.999; dw and bd are the gradient values ​​of the loss function L with respect to the weight w and the bias b in a batch; v dw 、v db is the accumulated gradient momentum of the algorithm during the first t-1 iterations; similarly, s dw 、s db It is also the accumulated gradient momentum during the first t-1 iterations of the algorithm.

[0092] Step 5.5: To avoid significant deviations in the algorithm in early iterations, a deviation correction is added to the formula;

[0093] In the present invention, the deviation correction is specifically:

[0094] Step 5.6: After adding bias correction, the CNN model can continuously update the weights and biases in the iterative process to optimize the prediction effect of the model.

[0095] In the present invention, the weight and bias update formula is specifically: Among them, α is the learning rate. During the training process, the update speed of the weight can be optimized by fine-tuning the learning rate; ε is a hyperparameter, and its value is usually 10 -8 .

[0096] In the smart grid stability prediction algorithm, the weight w and bias b in the CNN model are continuously updated in the iterative process. The CNN algorithm updates the weight w and bias b through continuous iterations, so that the Loss function approaches a minimum value close to 0. At the same time, the Accuracy function will continue to increase, indicating that the prediction effect of the model is getting better and better in continuous iterations.

[0097] The present invention is a smart grid expansion planning and stability prediction method based on 5G and CNN. During use, the present invention first collects data generated by the power network, then pre-processes the collected data and inputs it into the CNN model. After CNN convolution, pooling, and Softmax function processing and continuous iteration, the smart grid stability prediction model is finally obtained; at the same time, MSGD, RMSprop and Adam are used to optimize and train the model respectively. The model can effectively help the power system to give early warning of unknown problems, reduce safety hazards and avoid the occurrence of power accidents; the present invention has the advantages of establishing an energy consumption calculation model, constructing an active load probability distribution model, and forming a power network stability prediction model.

Claims

1. A smart grid expansion planning and stability prediction method based on 5G and CNN, characterized by: The method comprises the following steps: Step 1: Establish an energy consumption calculation model for 5G base stations: including baseline energy consumption and incremental energy consumption, where baseline energy consumption is divided into BBU baseline energy consumption and RRU baseline energy consumption; Step 2: Probability distribution of 5G base station load energy consumption: Construct a probability model of 5G base station load energy consumption; Step 3: Construct a distribution network probabilistic planning model that takes into account the probabilistic characteristics of 5G base station load energy consumption: Based on the probabilistic characteristics of base station load energy consumption, plan the expansion of distribution network lines and 110kV substation transformers; Step 4: Mathematical modeling based on convolutional neural network: CNN with nonlinear function is used to predict the stability of smart grid; Step 5: CNN-based smart grid stability prediction.

2. The method for smart grid expansion planning and stability prediction based on 5G and CNN as claimed in claim 1, characterized in that: The baseline energy consumption represents the energy consumption used to send fixed wireless signals when no user accesses the 5G base station, including the reference signal CRS, the primary synchronization signal PSS, the secondary synchronization signal SSS, the physical broadcast channel PBCH, the physical control format indicator channel and the physical hybrid automatic repeat request indicator channel PHICH; the incremental energy consumption represents the energy consumption to meet the uplink and downlink data service traffic requirements when a large number of users access the 5G base station, including the RRU module energy consumption.

3. The method for smart grid expansion planning and stability prediction based on 5G and CNN as claimed in claim 2, characterized in that: The energy consumption calculation model of the 5G base station in step 1 is as follows: Assume that the total energy consumption of the 5G base station per hour is E total Indicates that the RRU incremental energy consumption and baseline energy consumption are respectively E RRU,incre and E RRU,base Indicates that the BBU baseline energy consumption is E BBU,base The energy consumption calculation model of 5G base station can be expressed as: E total =E RRU,incre +E RRU,base +E BBU,base .

4. The method for smart grid expansion planning and stability prediction based on 5G and CNN as claimed in claim 3, characterized in that: The RRU incremental energy consumption is specifically: the incremental energy consumption E for RRU RRU,incre , the calculation formula is: E RRU,incre =E DLd,incre +E DLs,incre , where E DLd,incre E is the incremental energy consumption of downlink data generated by RRU during a given observation time; DLs,incre It is the incremental energy consumption generated by downlink signaling of RRU within a given observation time.

5. The method for smart grid expansion planning and stability prediction based on 5G and CNN as claimed in claim 3, characterized in that: The RRU incremental energy consumption is specifically: when the physical resource utilization rate of the base station is zero, the RRU baseline power consumption P in_base It can be decomposed into the sum of the baseline power consumption of RRU downlink data, RRU downlink signaling and RRU uplink data. Therefore, the baseline energy consumption of RRU during the observation time is E RRU,base Equal to: E RRU,incre =E DLd,base +E DLs,base +E ULd,base , where E DLd,base E is the baseline energy consumption generated by RRU downlink data per hour; DLs,base is the baseline energy consumption generated by RRU downlink signaling; E ULd,base Baseline energy consumption generated by RRU uplink data.

6. The method for smart grid expansion planning and stability prediction based on 5G and CNN as claimed in claim 1, characterized in that: The probability model of 5G base station load energy consumption in step 2 is as follows: Assuming that the building complex adopts H i (i∈Φ) indicates that Φ is the set of all types of building clusters within the radiation radius of the base station; the building cluster H i The corresponding population density ρ i The average population density during the observation time is obtained through the historical heat map; the number of mobile phone activations in the building complex satisfies λ i ~N(a λ ,b λ ) is normally distributed; then, the building complex H i The expected total population is E i =ρ i N i , N i For the building complex H i Area; Building complex H i The expected number of mobile phone activations R i =E i λ i Considering that the services provided by mobile phone users can be divided into mobile video services and non-mobile video services, the building group H i The ratio of the number of these two businesses satisfies δ i ~N(a δ ,b δ ), then the building complex H i PDCP layer traffic of mobile video and non-mobile video are respectively equal to: In the formula, the traffic speeds corresponding to the mobile video service and the non-mobile video service are uniformly distributed, where the random variables y1, y2 and w1, w2 are the downlink and uplink data traffic speeds of the user's mobile video service and non-mobile video; z1 and z2 are the total number of resource blocks occupied by the control channels of the mobile video service and the non-mobile video service, respectively; Since the ratio of mobile video services to non-mobile video services is δ i , and its cellular Internet access time is also estimated approximately in proportion; thus, the building complex H i Duration of mobile video and non-mobile video traffic within 10 minutes are respectively equal to: In the formula, the random variable t i For the building complex H i The probability density function of the Beta distribution of the downlink data and signaling service duration corresponding to the user in is: In the formula, Γ(x) is the Γ function; α and β are the characteristic parameters of the Beta distribution; Substituting the above formula into the energy consumption calculation model, the total energy consumption of the 5G base station within 10 minutes can be obtained: The probability distribution of is:

7. The method for smart grid expansion planning and stability prediction based on 5G and CNN as claimed in claim 1, characterized in that: The specific step 3 is: take the load energy consumption corresponding to the ith line of the 5G base station connected to the distribution network as The maximum total load P of the i-th distribution network line L,i ; The remaining capacity of the 110kV substation transformer is S k (k∈Ω), Ω is the set of all transformers in a certain planning area; the allowed capacity of the i-th distribution network line is M i (i∈Λ k ),Λ k is the set of all distribution network lines supplied by transformer k in the station; the variable for the expansion of the distribution network line 0-1 is x, and the variable for the expansion of the 110kV substation transformer 0-1 is x T , then the objective function of the short-term planning model for distribution network reconnection is: minF = B T x1+L T x T , where B is the cost vector for capacity expansion of the distribution network line; L is the cost vector for capacity expansion of the 110kV substation transformer; accordingly, the constraints are the 10kV feeder capacity constraint and the 110kV main transformer capacity constraint, and the 0-1 integer programming model of the distribution network taking into account the probabilistic characteristics of 5G base station energy consumption is:

8. The method for smart grid expansion planning and stability prediction based on 5G and CNN as claimed in claim 1, characterized in that: The mathematical modeling based on convolutional neural network in step 4 is specifically: the formula of the neuron model is: Among them, w i is the weight of the linear function; b is the bias unit; then, the model needs to be optimized through the loss function, which is defined as follows: Where n is the number of training sets; h (i) is the predicted output; y (i) True value.

9. The method for smart grid expansion planning and stability prediction based on 5G and CNN as claimed in claim 1, characterized in that: The CNN-based smart grid stability prediction in step 5 includes the following steps: Step 5.1: Preprocess the data using standardization: First, calculate the mathematical expectation E(X i ) and standard deviation S i , and then the standardized value is obtained according to the following formula: Among them, X ij is the jth value of the i-th feature of the data; E(X i ) is the mean of the i-th feature; S i is the standard deviation of the i-th feature; Step 5.2: Convert the 12 features obtained after the standardized power network dataset into a matrix and input it into the CNN model for feature extraction; Step 5.3: Update the weights of the pooling layer in the CNN model; Step 5.4: Use multiple optimization algorithms to optimize the CNN model; Step 5.5: To avoid significant deviations in the algorithm in early iterations, a deviation correction is added to the formula; Step 5.6: After adding bias correction, the CNN model can continuously update the weights and biases in the iterative process to optimize the prediction effect of the model.

10. The method for smart grid expansion planning and stability prediction based on 5G and CNN as claimed in claim 9, characterized in that: The weight update of the convolutional layer in the CNN model in step 5.2 includes the following steps: Step 5.21: First, we need to define the error function, the formula is as follows: Among them, N is the number of samples; K is the category of the sample; represents the nth sample y k The kth dimension of is the nth sample p k The kth dimension of the network output; Step 5.22: Next, define the output of the convolutional layer l for: in, represents the jth feature of the i-th layer; f is the activation function of CNN; M j The feature matrix representing the input data; is the convolution kernel that connects the input i-th feature and the output j-th feature in layer l; Then it is the bias parameter corresponding to the j feature in layer l; Step 5.23: Error Sensitivity It can be expressed as: in, w is the weight; b is the bias unit; layer l is the convolution layer; l+1 is the pooling layer; up is the upsampling operation that makes the size of the matrix of layer l+1 consistent with that of layer l, β j is the convolution kernel The value in ; Step 5.24: Next, find the error function with respect to the bias and convolution kernel Partial derivative of: The above formula is the sum of all error sensitivities in layer l, and (u,v) is the position of each element in the sensitivity matrix.

Images