Mobile charging pile control system based on artificial intelligence

Abstract

The invention discloses a mobile charging pile control system based on artificial intelligence, and relates to the technical field of intelligent charging, and the system constructs a real-time power grid load state diagram, combines the operation state data of a fast charging station and power grid sensing information, dynamically models the power grid load distribution and transmission capability, and improves the power grid load distribution and transmission capability. A Gaussian function is introduced into the state diagram to quantify the influence of the fast charging station on the node load, and a traditional power transmission capability model is improved, so that the power grid state expression is more intuitive; a load state diagram is processed through a graph neural network GNN, spatio-temporal evolution characteristics of a load are captured, a load change trend and a regulation level of each node in a future short time period are predicted, a multi-target optimization algorithm is adopted to formulate a preliminary regulation strategy based on a load prediction diagram, and the purposes of relieving a peak-valley difference, reducing energy loss and improving the service capability of a fast charging station are achieved. And an optimization problem is solved through a genetic algorithm, so that the globality and effectiveness of a regulation and control strategy are guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent charging, and particularly to a mobile charging pile control system based on artificial intelligence. Background Art

[0002] With the rapid popularization of electric vehicles, mobile charging piles, as emergency power supply facilities, have become an important part of the charging network.

[0003] Such charging piles can quickly respond to temporary charging demands in transportation hubs, large event venues, etc. However, the control methods of traditional mobile charging piles rely on fixed power output and simple scheduling mechanisms, which are prone to causing instantaneous load impacts on the power grid during peak charging periods, resulting in problems such as voltage dips, frequency fluctuations, and harmonic pollution, significantly increasing the instability risk of the power grid; in addition, traditional methods lack the ability to perceive the overall state of the regional power grid, cannot predict peak power demands, and are difficult to effectively allocate resources to relieve local pressure.

[0004] To address the above problems, some traditional solutions integrate battery energy storage devices in mobile charging piles to charge during low load periods and release energy during high load periods to smooth load fluctuations; some other solutions reduce the simultaneous startup of charging devices through timing control to reduce peak loads; however, these methods are limited by real-time response capabilities and intelligent levels and cannot meet the grid interaction requirements of fast charging stations in complex scenarios. Therefore, there is an urgent need for a control solution for mobile charging piles based on artificial intelligence to solve such problems. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] The present invention provides a mobile charging pile control system based on artificial intelligence to solve

[0007] the problems of insufficient response capabilities of traditional solutions, difficulty in perceiving the power grid state, and inability to dynamically optimize peak loads and harmonic control.

[0008] To solve the above technical problems, the present invention provides the following technical solutions:

[0009] An embodiment of the present invention provides a mobile charging pile control system based on artificial intelligence, which includes,

[0010] a data acquisition module for real-time collecting sensing data and operating state data of the power grid and fast charging stations. The data acquisition module integrates the collected information and constructs a real-time state map of the power grid load, where nodes represent the power grid load distribution and edge weights quantify the power transmission capabilities or flow relationships between nodes;

[0011] The load prediction module predicts the dynamic distribution of the grid load in the short term in the future based on the grid load status map provided by the data acquisition module;

[0012] The load prediction module uses the Graph Neural Network (GNN) model. Through the training of historical load data, it captures the spatio-temporal evolution characteristics of the grid load and outputs a load prediction map;

[0013] The strategy regulation module is used to formulate and optimize the regulation strategy of the fast charging station based on the output of the load prediction module;

[0014] The quality assurance module is used to monitor and optimize the power quality in real time during the operation of the fast charging station;

[0015] The quality assurance module uses the Generative Adversarial Network (GAN) model to monitor the harmonic characteristics injected into the grid by the fast charging station, separate the harmonic components in the power signal, locate the frequency bands with high impact on the power quality, and automatically adjust the PWM control parameters based on the detection results to suppress harmonic currents.

[0016] As a preferred solution of the mobile charging pile control system based on artificial intelligence according to the present invention, wherein: the strategy regulation module constructs a preliminary regulation strategy by using a multi-objective optimization algorithm, and the optimization objectives include: alleviating the peak-valley difference of the grid load,

[0017] reducing the energy loss during power transmission,

[0018] and maximizing the service capacity of the fast charging station.

[0019] As a preferred solution of the mobile charging pile control system based on artificial intelligence according to the present invention, wherein: the control method of the mobile charging pile control system based on artificial intelligence is:

[0020] Step S1: Deploy a sensing network in the power grid and the fast charging station to collect sensing data in real time, and at the same time monitor the operation status of the fast charging station. Combine the sensing data and the operation status data to construct a real-time status map of the grid load, and this status map is represented in a graph structure:

[0021] The nodes describe the load distribution at different positions of the grid,

[0022] The weights of the edges quantify the power transmission capacity or flow relationship between the nodes;

[0023] Step S2: Use the Graph Neural Network (GNN) to process the grid load status map generated in Step S1 to predict the dynamic distribution of the grid load in the short term in the future. The GNN model is trained based on historical load data, and a load map is output through prediction;

[0024] Step S3: Based on the predicted load map output in Step S2, use a multi-objective optimization algorithm to construct a preliminary regulation strategy for the fast charging station;

[0025] In Step S3, the optimization objectives of the multi-objective optimization algorithm include:

[0026] Alleviating the peak-valley difference of the grid load, reducing the energy loss during power transmission, and maximizing the service capacity of the fast charging station;

[0027] Step S4: Based on the preliminary regulation strategy, use the deep reinforcement learning DRL algorithm to continuously collect the sensing data and operating status data of the grid and the fast charging station, autonomously learn the regulation strategies in different scenarios, and dynamically adjust the power distribution of the fast charging station under the condition of load change;

[0028] Step S5: During the operation of the fast charging station, monitor the harmonic characteristics injected into the grid in real time, construct a generative adversarial network GAN model, use the GAN model to separate the harmonic components in the power signal, and locate the frequency bands with high impact on the power quality; based on the detection results, automatically adjust the PWM control parameters of the fast charging station.

[0029] As a preferred solution of the mobile charging pile control system based on artificial intelligence according to the present invention, wherein: the sensing data includes the dynamic operating data of voltage, current and load;

[0030] The operating status of the fast charging station includes charging demand and power output.

[0031] As a preferred solution of the mobile charging pile control system based on artificial intelligence according to the present invention, wherein: the step of combining the sensing data and the operating status data to construct a real-time status map of the grid load is

[0032] Collect the sensing data of the grid nodes and the operating status data of the fast charging station, and represent them as Dt:

[0033] Dt = {V i t, I i t, P i t | i ∈ N}, St = {C j t, O j t | j ∈ M},

[0034] wherein, Dt represents the set of node sensing data at time t, St represents the set of fast charging station status data at time t, V i t represents the voltage of node i at time t, I i t represents the current of node i at time t, P i t represents the load power of node i, C j t represents the charging demand of fast charging station j, O jLet \(t\) denote the power output of fast charging station \(j\), \(N\) and \(M\) represent the set of grid nodes and the set of fast charging stations respectively.

[0035] Calculate the total load of the node by combining the operation data of the grid node and the load impact function of the fast charging station. The calculation formula is:

[0036] \(L\) i \(_t = P\) i \(_t+\sum\) j∈M \(\varphi_{C}\) j \(_{t,O}\) j \(_{t,d}\) ij ,

[0037] where \(L\) i \(_t\) represents the total load of node \(i\) at time \(t\), and \(\varphi_{C}\) j \(_{t,O}\) j \(_{t,d}\) ij is the load impact function calculated based on the charging demand, power output of fast charging station \(j\) and the distance \(d\) to node \(i\) ij .

[0038] It is expressed in the form of a Gaussian function:

[0039]

[0040] where \(O\) j \(_t\) represents the power output of fast charging station \(j\), \(d\) ij represents the distance from fast charging station \(j\) to node \(i\), and \(\sigma\) is the distribution parameter of the Gaussian function;

[0041] Calculate the edge weight between nodes. The calculation formula is:

[0042]

[0043] where \(W\) ij \(_t\) represents the edge weight between node \(i\) and node \(j\), \(R\) ij represents the line resistance between the two nodes, \(\beta\) is the weight adjustment factor, and \(\gamma\) represents the dynamic influence coefficient of load difference on transmission capacity.

[0044] Based on the node load and edge weight, construct the real-time state graph \(G_t\) of the grid load:

[0045] \(G_t=(N, E, L_t, W_t)\),

[0046] where \(G_t\) represents the grid load state graph at time \(t\), \(N\) and \(E\) represent the set of nodes and the set of edges respectively, \(L_t = \{L\) i \(_t|i\in N\}\) represents the set of node loads, and \(W_t=\{W\) ij \(_t|i, j\in E\}\) represents the set of edge weights.

[0047] As a preferred solution of the mobile charging pile control system based on artificial intelligence according to the present invention, wherein: the load graph shows the load change trend of each node within the future time window and its spatial correlation.

[0048] As a preferred solution of the mobile charging pile control system based on artificial intelligence according to the present invention, wherein: the step of using the graph neural network GNN to process the power grid load status graph generated in step S1 and predicting the dynamic distribution of the power grid load in the future short time period is,

[0049] Input the power grid load status graph Gt = N, E, Lt, Wt generated in step S1 into the graph neural network model, and the node feature matrix Xt and the adjacency matrix At are expressed as:

[0050] Xt = {L i t∣i∈N}, At = {W ij t∣i,j∈E},

[0051] wherein, Xt represents the node feature matrix, which contains the load information of the nodes, L i t represents the load of node i at time t, At represents the adjacency matrix, which contains the edge weight information, W ij t represents the edge weight between node i and node j,

[0052] Adopt the graph convolutional neural network GCN as the load prediction model, and the propagation rule of GCN is:

[0053] H l+1 =σD -1 / 2 AD -1 / 2 H l W l ,

[0054] wherein, H l represents the node feature matrix of the l-th layer, H l+1 represents the node feature matrix of the l+1-th layer, A represents the adjacency matrix, D represents the degree matrix of the adjacency matrix, defined as D ii =∑ j A ij , W l represents the trainable weight matrix of the l-th layer, σ represents the activation function, and the ReLU function is adopted, defined as σx = max0,x;

[0055] The output of the GCN model is the load prediction graph at future time t+Δt:

[0056] L pred t+Δt = H L ,

[0057] wherein, L predt + Δt represents the predicted value of the node load at the future time t + Δt, H L represents the node feature matrix of the last layer of the GCN model;

[0058] The model is trained with historical load data, and the mean squared error MSE is used as the loss function:

[0059]

[0060] where, represents the loss value, represents the true load value of node i, represents the predicted load value of node i, and N represents the number of nodes.

[0061] As a preferred solution of the mobile charging pile control system based on artificial intelligence described in the present invention, wherein: for the predicted load map output based on step S2, the steps of constructing the preliminary regulation strategy of the fast charging station by using the multi-objective optimization algorithm are as follows,

[0062] Define the future load prediction map of step S2 as:

[0063] G pred t + Δt = N,E,L pred t + Δt,Wt,

[0064] where, G pred t + Δt represents the predicted load map at the future time t + Δt, L pred t + Δt represents the predicted value of the node load, and Wt represents the set of edge weights;

[0065] The regulation strategy needs to optimize the following objectives simultaneously:

[0066] Alleviate the peak-valley difference of the grid load, and define the peak-valley difference of the load as:

[0067] Obj 1 = maxL pred t + Δt - minL pred t + Δt,

[0068] where, Obj 1 represents the peak-valley difference of the load, maxL pred t + Δt and minL pred t + Δt represent the maximum and minimum load values respectively,

[0069] Reduce the power transmission energy loss, and define the energy loss as:

[0070]

[0071] where, Obj 2 represents the total energy loss, Wij Let \(t\) denote the weight of edge \(i, j\). and denote the predicted load values of node \(i\) and node \(j\) respectively.

[0072] Maximize the service capacity of the fast charging station. The service capacity of the fast charging station is defined as:

[0073] Obj 3 =\(\sum\) j∈M O j _{t + \Delta t}\),

[0074] where \(Obj\) 3 represents the total power output of the fast charging station, and \(O\) j _{t + \Delta t}\) represents the power output of fast charging station \(j\) at time \(t + \Delta t\);

[0075] Adopt the weighted summation method to combine the objective functions into a single optimization objective. The combination formula is:

[0076]

[0077] where represents the comprehensive optimization objective function, and \(\lambda\) 1 , \(\lambda\) 2 , \(\lambda\) 3 are the weight coefficients respectively;

[0078] Based on the optimization objective, use the genetic algorithm GA to solve the optimization problem and output the preliminary regulation strategy of the fast charging station:

[0079]

[0080] where represents the optimal power distribution strategy of fast charging station \(j\).

[0081] As a preferred solution of the mobile charging pile control system based on artificial intelligence described in the present invention, where: the steps of continuously collecting the sensing data and operation status data of the power grid and the fast charging station by using the deep reinforcement learning DRL algorithm and autonomously learning the regulation strategies in different scenarios are as follows.

[0082] Define the system state as \(S\) t , including the power grid load state and the fast charging station operation state:

[0083] S t =\(\{G_t, O_t, D_t\}\),

[0084] where \(G_t\) represents the real-time power grid load state graph, \(O_t = \{O\) j _{t|j \in M}\) represents the power output set of the fast charging station, and \(D_t = \{V\) i _t, I i _t, Pi {ti | i ∈ N} represents the real-time sensing data of nodes,

[0085] Define action A t as the fast charging station power adjustment strategy:

[0086]

[0087] where, represents the new power distribution of fast charging station j at time t,

[0088] Reward function R t is the objective function that comprehensively considers the power grid stability and the efficiency of fast charging stations, expressed as:

[0089]

[0090] where, R t represents the reward value at time t, λ 1 , λ 2 , λ 3 are weight coefficients, maxLt and minLt represent the peak and valley values of the power grid load, and ΔL ij represents the load difference between node i and node j, represents the new power distribution of fast charging station j,

[0091] Adopt the deep Q-network DQN as the reinforcement learning framework, and the Q-function update rule is:

[0092] QS t , A t → QS t , A t + α · R t + γ · maxQS t+1 , A t+1 - QS t , A t ,

[0093] where, QS t , A t represents the Q value of action A in state S t under, α represents the learning rate, R t represents the immediate reward, γ represents the discount factor, and maxQS t , A t+1 , A t+1 represents the optimal Q value of the next state;

[0094] Based on the DQN model, select the optimal action A t , and update the fast charging station power distribution:

[0095]

[0096] Among them, represents the optimal power distribution of the fast charging station j.

[0097] As a preferred solution of the mobile charging pile control system based on artificial intelligence described in the present invention, wherein: the steps of constructing the generative adversarial network GAN model and using the GAN model to separate the harmonic components in the power signal and locate the frequency bands with high impact on power quality are as follows

[0098] The power signal I grid injected into the power grid by the fast charging station is decomposed into fundamental wave and harmonic components:

[0099] I grid t = I base t + I harm t,

[0100] wherein, I grid t represents the total current signal injected into the power grid by the fast charging station, and I base t represents the fundamental wave component I harm t represents the harmonic component;

[0101] The generative adversarial network GAN is used to separate the harmonic signal. The GAN includes a generator G and a discriminator D:

[0102] The generator G inputs the noise z and outputs the estimated harmonic component:

[0103]

[0104] Among them, represents the harmonic signal estimated by the generator, and z represents the random noise,

[0105] The discriminator D is used to distinguish the real signal and the generated signal:

[0106] Dx = σW·x + b,

[0107] wherein, Dx represents the discrimination result, x is the input signal, W and b respectively represent the weight and bias parameters, and σ is the Sigmoid activation function;

[0108] The adversarial loss is used to optimize the GAN:

[0109] L GAN = ElogDI harm t + Elog1 - DGz,

[0110] wherein, L GAN represents the adversarial loss, and E represents the expectation operation;

[0111] The frequency domain analysis of the harmonic signal is carried out based on the fast Fourier transform FFT:

[0112]

[0113] Among them, Ff represents the harmonic intensity corresponding to the frequency f, and the high-impact frequency band f is determined through the harmonic intensity distribution 1 , f 2 ;

[0114] Adjust the PWM pulse width modulation control parameter θ according to the high-impact frequency band PWM , and the optimization formula is:

[0115]

[0116] Among them, represents the optimal PWM control parameter, |Ff| represents the harmonic intensity of the frequency f,

[0117] During the operation of the fast charging station, monitor the harmonic characteristics in real time and adjust the control parameters:

[0118]

[0119] The beneficial effects of the present invention are as follows: The present invention constructs a real-time power grid load status map, combines the operation status data of the fast charging station and the power grid sensing information, dynamically models the power grid load distribution and transmission capacity. A Gaussian function is introduced in the status map to quantify the impact of the fast charging station on the node load, and the traditional power transmission capacity model is improved, making the expression of the power grid status more intuitive; The load status map is processed by the graph neural network GNN to capture the spatio-temporal evolution characteristics of the load and predict the load change trend of each node in the short term in the future. At the regulation level, a multi-objective optimization algorithm is used to formulate a preliminary regulation strategy based on the load prediction map, aiming to alleviate the peak-valley difference, reduce energy loss, and improve the service capacity of the fast charging station. The genetic algorithm is used to solve the optimization problem to ensure the global and effective regulation strategy; Dynamically adjust the power distribution of the fast charging station using real-time data, adapt to the changing scenarios of the power grid load, further improve the coordination between the power grid and the fast charging station. The harmonic signal injected into the power grid by the fast charging station is separated and analyzed through the generative adversarial network GAN, the high-impact frequency band is located in real time, and the PWM control parameter is dynamically adjusted in combination with the frequency domain analysis result to achieve harmonic suppression and power quality optimization.

[0120] The present invention realizes high precision and intelligence in load prediction, power regulation, and harmonic control, effectively alleviates the power grid instability problem caused by peak loads, and improves the service capacity and resource utilization efficiency of the fast charging station. Description of the Drawings

[0121] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0122] Figure 1 It is a schematic framework diagram of the mobile charging pile control system based on artificial intelligence of the present invention. Specific embodiments

[0123] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will describe the specific embodiments of the present invention in detail with reference to the drawings of the specification.

[0124] In the following description, many specific details are set forth to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0125] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that can be included in at least one implementation manner of the present invention. The "in one embodiment" that appears in different places in this specification does not all refer to the same embodiment, nor is it a separate or selectively exclusive embodiment from other embodiments.

[0126] Embodiment 1, referring to Figure 1 , this embodiment provides a mobile charging pile control system based on artificial intelligence, including:

[0127] A data acquisition module for real-time acquisition of sensing data and operating state data of the power grid and fast charging stations. The data acquisition module integrates the acquired information and constructs a real-time state map of the power grid load, where nodes represent the power grid load distribution, and the edge weights quantify the power transmission capacity or flow relationship between nodes;

[0128] A load prediction module for predicting the dynamic distribution of the power grid load in the short term in the future based on the power grid load state map provided by the data acquisition module;

[0129] The load prediction module uses a graph neural network GNN model. Through the training of historical load data, it captures the spatio-temporal evolution characteristics of the power grid load and outputs a load prediction map;

[0130] A strategy regulation module for formulating and optimizing the regulation strategy of the fast charging station based on the output of the load prediction module;

[0131] The strategy control module constructs a preliminary control strategy using a multi-objective optimization algorithm. The optimization objectives include:

[0132] Alleviating the peak-valley difference of the power grid load,

[0133] Reducing the energy loss during power transmission,

[0134] and maximizing the service capacity of the fast charging station;

[0135] The quality assurance module is used to monitor and optimize the power quality in real time during the operation of the fast charging station;

[0136] The quality assurance module uses a generative adversarial network (GAN) model to monitor the harmonic characteristics injected into the power grid by the fast charging station, separate the harmonic components in the power signal, locate the frequency bands with high impact on power quality, and automatically adjust the PWM control parameters based on the detection results to suppress harmonic current.

[0137] This embodiment also provides a control method for the mobile charging pile control system based on artificial intelligence, including:

[0138] Step S1: Deploy a sensing network in the power grid and the fast charging station to collect sensing data in real time, and at the same time monitor the operating status of the fast charging station. Combine the sensing data and the operating status data to construct a real-time status graph of the power grid load, which is represented in a graph structure:

[0139] Nodes describe the load distribution at different positions of the power grid,

[0140] The weight of the edge quantifies the power transmission capacity or flow relationship between nodes;

[0141] The sensing data includes voltage, current, and dynamic operating data of the load;

[0142] The operating status of the fast charging station includes charging demand and power output;

[0143] The steps of combining the sensing data and the operating status data to construct a real-time status graph of the power grid load are as follows:

[0144] Collect the sensing data of the power grid nodes and the operating status data of the fast charging station, and represent them as Dt:

[0145] Dt = {V i t, I i t, P i t | i ∈ N}, St = {C j t, O j t | j ∈ M},

[0146] where Dt represents the set of node sensing data at time t, St represents the set of fast charging station status data at time t, V i$V_{i,t}$ represents the voltage of node $i$ at time $t$, $I$ i $I_{i,t}$ represents the current of node $i$ at time $t$, $P$ i $P_{i,t}$ represents the load power of node $i$ at time $t$, $C$ j $O_{j,t}$ represents the charging demand of fast charging station $j$ at time $t$, $O$ j $N_{j,t}$ represents the power output of fast charging station $j$ at time $t$, $N$ and $M$ respectively represent the set of grid nodes and the set of fast charging stations,

[0147] Combined with the operation data of grid nodes and the load influence function of fast charging stations, calculate the total load of nodes. The calculation formula is:

[0148] $L$ i $_{i,t}=P$ i $_{i,t}+\sum$ j∈M $\varphi_{C}$ j $_{j,t,O}$ j $_{j,t,d}$ ij ,

[0149] where, $L$ i $_{i,t}$ represents the total load of node $i$ at time $t$, $\varphi_{C}$ j $_{j,t,O}$ j $_{j,t,d}$ ij is the load influence function calculated based on the charging demand, power output of fast charging station $j$ and the distance $d$ ij from fast charging station $j$ to node $i$,

[0150] which is represented in the form of a Gaussian function:

[0151]

[0152] where, $O$ j $_{j,t}$ represents the power output of fast charging station $j$, $d$ ij represents the distance from fast charging station $j$ to node $i$, $\sigma$ is the distribution parameter of the Gaussian function;

[0153] Calculate the edge weight between nodes. The calculation formula is:

[0154]

[0155] where, $W$ ij $_{i,j,t}$ represents the edge weight between node $i$ and node $j$, $R$ ij represents the line resistance between the two nodes, $\beta$ is the weight adjustment factor, $\gamma$ represents the dynamic influence coefficient of load difference on transmission capacity,

[0156] Based on node load and edge weight, construct the real-time state graph $G_{t}$ of grid load:

[0157] $G_{t}=(N,E,L_{t},W_{t})$,

[0158] Among them, \(G_t\) represents the power grid load status graph at time \(t\), \(N\) and \(E\) respectively represent the node set and the edge set, \(L_t=\{L_{t|i}\in N\}\) represents the node load set, and \(W_t = \{W_{t|i,j}\in E\}\) represents the edge weight set; i \(t|i\in N\}\) represents the node load set, \(W_t=\{W\) ij \(t|i,j\in E\}\) represents the edge weight set;

[0159] Specifically, a Gaussian function is introduced here to model the impact of fast charging stations on node loads, and the traditional power transmission capacity model is improved to dynamically quantify edge weights; the power grid load is modeled according to the dynamic demand characteristics of fast charging stations, and at the same time, the load distribution and transmission capacity of the power grid are more intuitively expressed in the form of a state graph.

[0160] Step S2, use the graph neural network GNN to process the power grid load status graph generated in step S1, predict the dynamic distribution of the power grid load in the short term in the future, the GNN model is trained based on historical load data, and the load graph is output through prediction;

[0161] The load graph shows the load change trend of each node in the future time window and its spatial correlation;

[0162] The steps of using the graph neural network GNN to process the power grid load status graph generated in step S1 and predicting the dynamic distribution of the power grid load in the short term in the future are as follows:

[0163] Input the power grid load status graph \(G_t = \{N, E, L_t, W_t\}\) generated in step S1 into the graph neural network model. The node feature matrix \(X_t\) and the adjacency matrix \(A_t\) are expressed as:

[0164] \(X_t=\{L\) i \(t|i\in N\}, A_t=\{W\) ij \(t|i,j\in E\}\),

[0165] Among them, \(X_t\) represents the node feature matrix, which contains the load information of the nodes, \(L\) i \(t\) represents the load of node \(i\) at time \(t\), \(A_t\) represents the adjacency matrix, which contains the edge weight information, \(W\) ij \(t\) represents the edge weight between node \(i\) and node \(j\),

[0166] Adopt the graph convolutional neural network GCN as the load prediction model. The propagation rule of GCN is:

[0167] \(H\) l+1 \(^l=\sigma(D^{-\frac{1}{2}}\widetilde{A}D^{-\frac{1}{2}}H^{l - 1}W^l)\) -1 / 2 \(\widetilde{A}=A + I\) -1 / 2 \(H\) l \(^0 = X\) l ;

[0168] Among them, \(H\) l \(^l\) represents the node feature matrix of the \(l\)th layer, \(H\) l+1represents the node feature matrix of the (l + 1)-th layer, A represents the adjacency matrix, D represents the degree matrix of the adjacency matrix, defined as D ii = ∑ j A ij , W l represents the trainable weight matrix of the l-th layer, σ represents the activation function, and the ReLU function is adopted, defined as σ(x) = max(0, x);

[0169] The output of the GCN model is the load prediction graph at future time t + Δt:

[0170] L pred t+Δt = H L ,

[0171] where, L pred t+Δt represents the predicted node load value at future time t + Δt, and H L represents the node feature matrix of the last layer of the GCN model;

[0172] The model is trained with historical load data, and the mean squared error (MSE) is used as the loss function:

[0173]

[0174] where, represents the loss value, represents the true load value of node i, represents the predicted load value of node i, and N represents the number of nodes;

[0175] Specifically, this step uses the graph neural network (GNN) to model the spatio-temporal evolution characteristics of the power grid load. With the node feature matrix and the adjacency matrix as inputs, it predicts the load change trend in the future period. The propagation rule of the GCN model ensures that information can be effectively aggregated in the graph structure. By combining the mean squared error loss to optimize the model parameters, accurate load prediction is carried out.

[0176] Step S3: Based on the predicted load graph output in Step S2, a preliminary regulation strategy for the fast charging station is constructed using a multi-objective optimization algorithm;

[0177] In Step S3, the optimization objectives of the multi-objective optimization algorithm include:

[0178] Mitigating the peak-valley difference of the power grid load, reducing the energy loss during power transmission, and maximizing the service capacity of the fast charging station;

[0179] The steps of constructing a preliminary regulation strategy for the fast charging station using a multi-objective optimization algorithm based on the predicted load graph output in Step S2 are as follows,

[0180] Define the future load prediction graph in Step S2 as:

[0181] G pred t + Δt = N, E, L pred t + Δt, Wt,

[0182] Among them, G pred t + Δt represents the predicted load diagram at future time t + Δt, L pred t + Δt represents the predicted node load value, and Wt represents the set of edge weights;

[0183] The regulation strategy needs to optimize the following objectives simultaneously:

[0184] Mitigate the peak - valley difference of the power grid load. The peak - valley difference of the load is defined as:

[0185] Obj 1 = maxL pred t + Δt - minL pred t + Δt,

[0186] Among them, Obj 1 represents the peak - valley difference of the load, maxL pred t + Δt and minL pred t + Δt represent the maximum and minimum load values respectively,

[0187] Reduce the power transmission energy loss. The energy loss is defined as:

[0188]

[0189] Among them, Obj 2 represents the total energy loss, W ij t represents the weight of edge i, j, and respectively represent the predicted load values of node i and node j,

[0190] Maximize the service capacity of the fast - charging station. The service capacity of the fast - charging station is defined as:

[0191] Obj 3 = ∑ j∈M O j t + Δt,

[0192] Among them, Obj 3 represents the total power output of the fast - charging station, O j t + Δt represents the power output of fast - charging station j at time t + Δt;

[0193] Use the weighted summation method to combine the objective functions into a single optimization objective. The combined formula is:

[0194]

[0195] Among them, denotes the comprehensive optimization objective function, λ 1 , λ 2 , λ 3 are the weight coefficients respectively;

[0196] Based on the optimization objective, use the genetic algorithm GA to solve the optimization problem and output the preliminary regulation strategy of the fast charging station:

[0197]

[0198] where denotes the optimal power distribution strategy of fast charging station j;

[0199] Specifically, through the multi-objective optimization method, construct the preliminary regulation strategy of the fast charging station based on the load prediction map; use the genetic algorithm to solve the non-linear objective function, and maximize the regulation efficiency of the fast charging station on the premise of ensuring the stability of the power grid operation and the quality of service.

[0200] Step S4, based on the preliminary regulation strategy, use the deep reinforcement learning DRL algorithm to continuously collect the sensing data and operation status data of the power grid and the fast charging station, autonomously learn the regulation strategies in different scenarios, and dynamically adjust the power distribution of the fast charging station under the condition of load change;

[0201] The steps of using the deep reinforcement learning DRL algorithm to continuously collect the sensing data and operation status data of the power grid and the fast charging station and autonomously learn the regulation strategies in different scenarios are as follows

[0202] Define the system state as S t , including the power grid load status and the operation status of the fast charging station:

[0203] S t ={Gt, Ot, Dt},

[0204] where Gt represents the real-time load status map of the power grid, Ot={O j t∣j∈M} represents the power output set of the fast charging station, and Dt={V i t, I i t, P i t∣i∈N} represents the real-time sensing data of the node,

[0205] Define the action A t as the power adjustment strategy of the fast charging station:

[0206]

[0207] where represents the new power distribution of fast charging station j at time t,

[0208] The reward function R tThe objective function for integrating power grid stability and fast charging station efficiency is expressed as:

[0209]

[0210] Among them, R t represents the reward value at time t, λ 1 , λ 2 , λ 3 are weight coefficients, maxLt and minLt represent the peak and valley values of the power grid load, and ΔL ij represents the load difference between node i and node j, represents the new power distribution of fast charging station j,

[0211] The deep Q-network DQN is adopted as the reinforcement learning framework, and the Q-function update rule is:

[0212] QS t , A t →QS t , A t +α·R t +γ·maxQS t+1 , A t+1 -QS t , A t .

[0213] Among them, QS t , A t represents the Q value of action A t under state S t , α represents the learning rate, R t represents the immediate reward, γ represents the discount factor, and maxQS t+1 , A t+1 represents the optimal Q value of the next state;

[0214] Based on the DQN model, the optimal action A t is selected to update the power distribution of the fast charging station:

[0215]

[0216] Among them, represents the optimal power distribution of fast charging station j,

[0217] Specifically, the power distribution of the fast charging station is dynamically adjusted through deep reinforcement learning DQN, and an adaptive control strategy for the dynamic change of the power grid load is constructed. The reward function integrates the power grid load balance, energy loss, and the service ability of the fast charging station; combined with the efficient learning ability of DQN, the fast charging station can autonomously optimize the operating parameters in different scenarios, improving the power grid stability and control efficiency.

[0218] Step S5, during the operation of the fast charging station, the harmonic characteristics injected into the power grid are monitored in real time, a Generative Adversarial Network (GAN) model is constructed, the GAN model is used to separate the harmonic components in the power signal, and the frequency band with high impact on power quality is located; based on the detection results, the PWM control parameters of the fast charging station are automatically adjusted.

[0219] The steps of constructing a Generative Adversarial Network (GAN) model and using the GAN model to separate the harmonic components in the power signal and locate the frequency band with high impact on power quality are as follows:

[0220] The power signal I grid t injected by the fast charging station into the power grid is decomposed into fundamental wave and harmonic components:

[0221] I grid t = I base t + I harm t,

[0222] where, I grid t represents the total current signal injected by the fast charging station into the power grid, and I base t represents the fundamental wave component I harm t represents the harmonic component;

[0223] The Generative Adversarial Network (GAN) is used to separate the harmonic signal. The GAN includes a generator G and a discriminator D:

[0224] The generator G takes the input noise z and outputs the estimated harmonic component:

[0225]

[0226] where, represents the harmonic signal estimated by the generator, and z represents the random noise,

[0227] The discriminator Dx is used to distinguish between the real signal and the generated signal:

[0228] Dx = σW·x + b,

[0229] where, Dx represents the discrimination result, x is the input signal, W and b represent the weight and bias parameters respectively, and σ is the Sigmoid activation function;

[0230] The GAN is optimized using the adversarial loss:

[0231] L GAN = ElogDI harm t + Elog1 - DGz,

[0232] where, L GAN represents the adversarial loss, and E represents the expectation operation;

[0233] The frequency domain analysis of the harmonic signal is performed based on the Fast Fourier Transform (FFT).

[0234]

[0235] Among them, Ff represents the harmonic intensity corresponding to the frequency f, and the high-impact frequency band f is determined through the harmonic intensity distribution. 1 , f 2 ;

[0236] Adjust the PWM pulse width modulation control parameter θ according to the high-impact frequency band. PWM , and the optimization formula is:

[0237]

[0238] Among them, represents the optimal PWM control parameter, |Ff| represents the harmonic intensity of the frequency f.

[0239] During the operation of the fast charging station, the harmonic characteristics are monitored in real time and the control parameters are adjusted:

[0240]

[0241] Specifically, a generative adversarial network GAN is constructed here to separate the harmonic signals injected into the power grid by the fast charging station, and the frequency domain analysis is combined to locate the high-impact harmonic frequency bands; by optimizing the PWM control parameters, the influence of harmonic current on the power quality is suppressed in real time. Utilizing the signal generation and separation capabilities of GAN, the problems of harmonic detection and suppression are effectively solved, and the adaptive optimization of harmonic governance in the fast charging station is realized through PWM dynamic regulation, with both real-time performance and high precision.

[0242] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A mobile charging pile control system based on artificial intelligence, characterized in that: include, The data acquisition module is used to collect sensor data and operating status data of the power grid and fast charging stations in real time. The data acquisition module integrates the collected information and constructs a real-time status diagram of the power grid load, where the nodes represent the power grid load distribution and the edge weights quantify the power transmission capacity or flow relationship between the nodes; The load prediction module predicts the dynamic distribution of the power grid load in the short term in the future based on the power grid load status diagram provided by the data acquisition module; The load forecasting module uses the graph neural network (GNN) model to capture the spatiotemporal evolution characteristics of the grid load through training with historical load data and output a load forecasting graph. The strategy control module is used to formulate and optimize the control strategy of the fast charging station based on the output of the load prediction module; The quality assurance module is used to monitor and optimize the power quality in real time during the operation of the fast charging station; The quality assurance module uses the generative adversarial network (GAN) model to monitor the harmonic characteristics of the fast charging station injected into the power grid and separate the harmonic components in the power signal; locate the frequency bands that have a high impact on power quality, and automatically adjust the PWM control parameters based on the detection results to suppress harmonic currents.

2. The mobile charging pile control system based on artificial intelligence as claimed in claim 1, characterized in that: The strategy control module uses a multi-objective optimization algorithm to construct a preliminary control strategy. The optimization objectives include: alleviating the peak-to-valley difference of the power grid load, Reduce energy loss during power transmission, And maximize the service capacity of fast charging stations.

3. The mobile charging pile control system based on artificial intelligence as claimed in claim 2, characterized in that: The control method of the mobile charging pile control system based on artificial intelligence is: Step S1, deploy sensor networks in the power grid and fast charging stations, collect sensor data in real time, monitor the operating status of the fast charging stations, and combine the sensor data with the operating status data to construct a real-time state diagram of the power grid load. The state diagram is represented by a graph structure: Nodes describe the load distribution at different locations in the power grid. The weight of the edge quantifies the power transmission capacity or flow relationship between nodes; Step S2, using the graph neural network GNN to process the power grid load state diagram generated in step S1, and predict the dynamic distribution of the power grid load in the future short period of time. The GNN model is trained based on historical load data and outputs the load diagram through prediction; Step S3, based on the predicted load diagram outputted in step S2, a preliminary control strategy of the fast charging station is constructed using a multi-objective optimization algorithm; In step S3, the optimization objectives of the multi-objective optimization algorithm include: Alleviate the peak-valley difference of power grid load, reduce energy loss during power transmission, and maximize the service capacity of fast charging stations; Step S4: Based on the preliminary control strategy, the deep reinforcement learning (DRL) algorithm is used to continuously collect sensor data and operating status data of the power grid and fast charging stations, autonomously learn the control strategies under different scenarios, and dynamically adjust the power distribution of the fast charging stations when the load changes; Step S5: During the operation of the fast charging station, the harmonic characteristics of the power grid are monitored in real time, a generative adversarial network (GAN) model is constructed, and the GAN model is used to separate the harmonic components in the power signal and locate the frequency band with a high impact on the power quality; based on the detection results, the PWM control parameters of the fast charging station are automatically adjusted.

4. The mobile charging pile control system based on artificial intelligence as claimed in claim 3, characterized in that: The sensing data includes dynamic operation data of voltage, current and load; The operating status of the fast charging station includes charging demand and power output.

5. The mobile charging pile control system based on artificial intelligence as claimed in claim 4, characterized in that: The step of combining the sensor data with the operating status data to construct a real-time status diagram of the power grid load is: The sensor data of the power grid nodes and the operating status data of the fast charging station are collected and expressed as D(t): D(t)={(V i (t),I i (t),P i (t))∣i∈N},S(t)={(C j (t),O j (t))∣j∈M}, Where D(t) represents the node sensor data set at time t, S(t) represents the fast charging station status data set at time t, V i (t) represents the voltage of node i at time t, I i (t) represents the current of node i at time t, P i (t) represents the load power of node i, C j (t) represents the charging demand of fast charging station j, O j (t) represents the power output of fast charging station j, N and M represent the set of grid nodes and the set of fast charging stations, respectively. The total load of the node is calculated by combining the operation data of the grid node and the load impact function of the fast charging station. The calculation formula is: L i (t)=P i (t)+∑ j∈M φ(C j (t),O j (t),d ij ), Among them, L i (t) represents the total load of node i at time t, φ(C j (t),O j (t),d ij ) is based on the charging demand, power output and distance d from fast charging station j to node i. ij Calculated load impact function, It is expressed in the form of Gaussian function: Among them, O j (t) represents the power output of fast charging station j, d ij represents the distance from fast charging station j to node i, σ is the distribution parameter of the Gaussian function; Calculate the edge weight between nodes, the calculation formula is: Among them, W ij (t) represents the edge weight between node i and node j, R ij represents the line resistance between two nodes, β is the weight adjustment factor, and γ represents the dynamic influence coefficient of load difference on transmission capacity. Based on the node load and edge weight, the real-time state graph G(t) of the power grid load is constructed: G(t)=(N,E,L(t),W(t)), Where G(t) represents the grid load state diagram at time t, N and E represent the node set and edge set respectively, and L(t) = {L i (t)|i∈N} represents the node load set, W(t)={W ij (t)|(i,j)∈E} represents the edge weight set.

6. The mobile charging pile control system based on artificial intelligence as claimed in claim 5, characterized in that: The load graph shows the load variation trend and spatial correlation of each node in a future time window.

7. The mobile charging pile control system based on artificial intelligence as claimed in claim 6, characterized in that: The step of using the graph neural network GNN to process the power grid load state diagram generated in step S1 and predicting the dynamic distribution of the power grid load in a short period of time in the future is as follows: The grid load state graph G(t) = (N, E, L(t), W(t)) generated in step S1 is input into the graph neural network model, and the node feature matrix X(t) and adjacency matrix A(t) are expressed as: X(t)={L i (t)∣i∈N},A(t)={W ij (t)∣(i,j)∈E}, Among them, X(t) represents the node feature matrix, which contains the load information of the node, L i (t) represents the load of node i at time t, A(t) represents the adjacency matrix, including edge weight information, and W ij (t) represents the edge weight between node i and node j, The graph convolutional neural network (GCN) is used as the load prediction model. The propagation rule of GCN is: H (l+1) =σ(D -1 / 2 AD -1 / 2 H (l) W (l) ), Among them, H (l) represents the node feature matrix of the lth layer, H (l+1) represents the node feature matrix of the l+1th layer, A represents the adjacency matrix, and D represents the degree matrix of the adjacency matrix, which is defined as D ii =∑ j A ij , W (l) represents the trainable weight matrix of the lth layer, σ represents the activation function, using the ReLU function, defined as σ(x) = max(0,x); The output of the GCN model is the load forecast graph for the future time t+Δt: L pred (t+Δt)=H (L) , Among them, L pred (t+Δt) represents the predicted value of the node load at the future time t+Δt, H (L) Represents the node feature matrix of the last layer of the GCN model; The model is trained by historical load data, and the mean square error (MSE) is used as the loss function: in, represents the loss value, represents the actual load value of node i, represents the predicted load value of node i, and N represents the number of nodes.

8. The mobile charging pile control system based on artificial intelligence as claimed in claim 7, characterized in that: The step of constructing a preliminary control strategy for a fast charging station using a multi-objective optimization algorithm based on the predicted load diagram output in step S2 is: The future load forecast graph of step S2 is defined as: G pred (t+Δt)=(N,E,L pred (t+Δt),W(t)), Among them, G pred (t+Δt) represents the predicted load diagram at future time t+Δt, L pred (t+Δt) represents the node load prediction value, W(t) represents the edge weight set; The control strategy needs to optimize the following objectives simultaneously: To alleviate the peak-to-valley difference of the power grid load, the peak-to-valley difference of the load is defined as: Obj1=max(L pred (t+Δt))-min(L pred (t+Δt)), Among them, Obj1 represents the load peak-to-valley difference, max(L pred (t+Δt)) and min(L pred (t+Δt)) represent the maximum and minimum load values, respectively, Reduce the energy loss of power transmission, and define the energy loss as: Where Obj2 represents the total energy loss, W ij (t) represents the weight of edge (i,j), and They represent the load prediction values ​​of node i and node j respectively, Maximize the fast charging station service capacity, which is defined as: Obj3=∑ j∈M O j (t+Δt), Where Obj3 represents the total power output of the fast charging station, O j (t+Δt) represents the power output of fast charging station j at time t+Δt; The weighted summation method is used to merge the objective functions into a single optimization objective. The combined formula is: F=λ1·Obj1+λ2·Obj2-λ3·Obj3, in, represents the comprehensive optimization objective function, λ1, λ2, λ3 are weight coefficients respectively; Based on the optimization goal, the genetic algorithm GA is used to solve the optimization problem and output the preliminary control strategy of the fast charging station: in, represents the optimal power allocation strategy of fast charging station j.

9. The mobile charging pile control system based on artificial intelligence as claimed in claim 8, characterized in that: The steps of using the deep reinforcement learning DRL algorithm to continuously collect sensor data and operating status data of the power grid and fast charging station and autonomously learn the control strategies in different scenarios are as follows: Define the system state as S t , including the grid load status and fast charging station operation status: S t ={G(t),O(t),D(t)}, Where G(t) represents the real-time load state diagram of the power grid, O(t) = {O j (t)|j∈M} represents the power output set of the fast charging station, D(t)={(V i (t),I i (t),P i (t))|i∈N} represents the real-time sensor data of the node, Define action A t Power adjustment strategy for fast charging stations: in, represents the new power allocation of fast charging station j at time t, Reward function R t is the objective function of comprehensive grid stability and fast charging station efficiency, expressed as: Among them, R t represents the reward value at time t, λ1,λ2,λ3 are weight coefficients, max(L(t)) and min(L(t)) represent the peak and valley values ​​of the grid load, ΔL ij represents the load difference between node i and node j, represents the new power allocation of fast charging station j, The deep Q network DQN is used as the reinforcement learning framework, and the Q function update rule is: Q(S t ,A t )→Q(S t ,A t )+α·[R t +γ·maxQ(S t+1 ,A t+1 )-Q(S t ,A t )], Among them, Q(S t ,A t ) indicates state S t Next action A t Q value, α represents the learning rate, R t represents the immediate reward, γ represents the discount factor, maxQ(S t+1 ,A t+1 ) represents the optimal Q value of the next state; Based on the DQN model, select the optimal action A t , update the fast charging station power distribution: in, represents the optimal power allocation of fast charging station j.

10. The mobile charging pile control system based on artificial intelligence as claimed in claim 9, characterized in that: The steps of constructing a generative adversarial network (GAN) model, using the GAN model to separate harmonic components in the power signal, and locating the frequency band with high impact on power quality are as follows: Power signal I injected into the grid by the fast charging station grid (t) is decomposed into fundamental and harmonic components: I grid (t)=I base (t)+I harm (t), Among them, I grid (t) represents the total current signal injected into the grid by the fast charging station, I base (t) represents the fundamental wave component I harm (t) represents the harmonic component; The generative adversarial network GAN is used to separate the harmonic signal. GAN includes a generator G and a discriminator D: The generator G(z) takes noise z as input and outputs estimated harmonic components: in, represents the harmonic signal estimated by the generator, z represents random noise, The discriminator D(x) is used to distinguish between real signals and generated signals: D(x)=σ(W·x+b), Where D(x) represents the discrimination result, x represents the input signal, W and b represent the weight and bias parameters respectively, and σ represents the Sigmoid activation function; Using adversarial loss to optimize GAN: L GAN =E[logD(I harm (t))]+E[log(1-D(G(z)))], Among them, L GAN represents adversarial loss, E represents expected operation; Frequency domain analysis of harmonic signals based on Fast Fourier Transform FFT: Where F(f) represents the harmonic intensity corresponding to frequency f, and the high-impact frequency band [f1, f2] is determined by the harmonic intensity distribution; Adjust the PWM pulse width modulation control parameter θ according to the high impact frequency band PWM , the optimization formula is: in, represents the optimal PWM control parameter, |F(f)| represents the harmonic intensity of frequency f, During the operation of the fast charging station, the harmonic characteristics are monitored in real time and the control parameters are adjusted:

Images