Optical storage charging station double-layer multi-target capacity optimization configuration method

Abstract

The invention belongs to the technical field of electric power system planning and operation, and discloses an optical storage charging station double-layer multi-target capacity optimization configuration method, which comprises the following steps: constructing a photovoltaic output scene library through a CNN, carrying out typical scene clustering by utilizing a K-means + + algorithm, predicting an electric vehicle charging load in combination with an LSTM-Transform hybrid model, and generating an optimized charging load curve; a planning-operation double-layer optimization model is established, the outer layer carries out capacity configuration with the minimum whole life cycle cost as the target, the inner layer makes an operation strategy based on the operation economy and reliability target, and dynamic coupling of planning and operation is achieved through bidirectional parameter transmission; a grey wolf algorithm is adopted to optimize capacity configuration parameters, an operation strategy is solved in combination with mixed integer linear programming, and an optimal configuration scheme is obtained through multi-round iterative feedback. According to the method, the relation between planning and operation of the optical storage charging station is considered, the economic benefit is optimal, and the stability and reliability of a power grid are improved.

Description

Technical Field

[0001] This invention belongs to the field of power system planning and operation technology, specifically relating to a two-layer multi-objective capacity optimization configuration method for photovoltaic-storage-charging stations. Background Technology

[0002] With the rapid popularization of electric vehicles and the development of renewable energy technologies, photovoltaic-storage charging stations, as a new type of energy infrastructure, are becoming an indispensable and important component of the power system. By integrating photovoltaic power generation systems, energy storage devices, and electric vehicle charging facilities, photovoltaic-storage charging stations can not only effectively alleviate the charging demand of electric vehicles and realize the self-sufficiency of photovoltaic power generation, but also optimize the local consumption of renewable energy, reduce carbon emissions, and improve the reliability of the power system's stable operation.

[0003] Currently, the planning and configuration of photovoltaic-storage charging stations still faces many technical bottlenecks: the spatiotemporal randomness of charging loads leads to grid voltage fluctuations and equipment overload risks, which are particularly prominent in urban areas; the configuration of charging stations needs to comprehensively consider multiple objectives such as investment and operation and maintenance costs, photovoltaic absorption efficiency, and grid stability. Existing research methods mostly use single-layer models combined with multi-objective optimization algorithms for solving problems, but when planning the capacity of photovoltaic-storage charging stations, they rarely consider the impact of short-term operation on the economics and long-term capacity planning of charging stations. Due to the complex nonlinear coupling relationships between various optimization objectives, traditional single-layer models are difficult to effectively reconcile the conflicts between system economic and technical indicators. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a dual-layer, multi-objective capacity optimization configuration method for photovoltaic-storage charging stations. This method considers the relationship between the planning and operation of photovoltaic-storage charging stations, achieving optimal economic benefits and improving grid stability and reliability.

[0005] The present invention discloses a dual-layer multi-objective capacity optimization configuration method for photovoltaic-storage-charging stations, comprising the following steps:

[0006] Step 1: Construct a photovoltaic power output scenario library using a generative adversarial network (GAN), cluster typical scenarios, and combine the LSTM-Transformer hybrid model to predict electric vehicle charging load, generating an optimized electric vehicle charging load curve.

[0007] Step 2: Establish a two-layer optimization model for planning and operation, including an outer capacity planning model and an inner optimization operation model. The outer capacity planning model configures capacity with the goal of minimizing the total life cycle cost, while the inner optimization operation model formulates operation strategies based on the goals of operational economy and reliability. Dynamic coupling between planning and operation is achieved through bidirectional parameter transmission.

[0008] Step 3: Solve the results obtained in Step 2 to obtain the optimal configuration scheme.

[0009] Furthermore, in step 1, sufficient data reflecting the true characteristics of photovoltaic power output is generated using a generative adversarial network (GAN) to construct a photovoltaic power output database, and typical scenarios are clustered, specifically as follows:

[0010] Step 1-1: Collect time-series random variables {x} at Z time points. 1,1 , ..., x i,t , ..., x m,Z The photovoltaic output curve is constructed by generating photovoltaic power generation scenarios, where m is the number of photovoltaic cells and Z is the number of time points, 1≤i≤m and 1≤t≤Z; photovoltaic power generation scenarios are generated through GAN and a photovoltaic power generation scenario library is constructed.

[0011] Among them, the generator network G and the discriminator network D are important components of generative adversarial networks;

[0012] The loss function of the generator is:

[0013]

[0014] In the formula: z~P z (z) represents the data z obtained by sampling from the generated photovoltaic data; G(z) represents the generated photovoltaic data; D[G(z)] represents the probability that the generated photovoltaic data matches the actual photovoltaic data; E represents the probability distribution of the generated photovoltaic data;

[0015] The loss function of the discriminator is:

[0016]

[0017] In the formula: x ~ Pdata(x) is the data x obtained by sampling real photovoltaic data; D(x) represents the probability that the discriminant data x matches the actual photovoltaic data; D[G(z)] is the probability that the generated photovoltaic data matches the actual photovoltaic data;

[0018] The generator and discriminator are trained with the following objective function:

[0019]

[0020] Steps 1-2: Use the K-means++ algorithm to cluster typical scenarios and generate typical photovoltaic power output scenarios, including:

[0021] Step 1-2-1: Select any photovoltaic output curve from the photovoltaic output scenario library as the cluster center point G1;

[0022] Step 1-2-2: Calculate the distance D from G1 to other photovoltaic output curves in the photovoltaic output scenario library;

[0023] Steps 1-2-3: Randomly select the next cluster center point using a roulette wheel method until K initial points are selected as cluster centers;

[0024] Steps 1-2-4: Calculate the distance between each photovoltaic output curve in the photovoltaic output scenario library and the K cluster centers, and assign it to the cluster corresponding to the nearest cluster center;

[0025] Step 1-2-5: Update the cluster center point G of each cluster according to the distance D. i ;

[0026] Step 1-2-6: Repeat steps 1-2-4 and 1-2-5 until the position of the cluster center point no longer changes.

[0027] Furthermore, based on historical charging load data, the K-means++ clustering method is used to classify charging scenarios. A prediction model is established for each clustered scenario to obtain the corresponding charging load prediction data. The obtained charging load prediction data are then integrated to obtain the electric vehicle charging load prediction curve.

[0028] The LSTM-Transformer hybrid model established in this invention combines the advantages of Transformer and Long Short-Term Memory (LSTM). It accurately captures the short-term temporal features of charging load through the gating mechanism of LSTM, analyzes long-range dependencies across time periods using the multi-head self-attention mechanism of Transformer, and preserves temporal information through positional encoding. A hierarchical feature fusion module achieves dynamic weighted interaction between local details and global trends, thereby obtaining accurate and reliable prediction data.

[0029] The LSTM-Transformer hybrid model includes an encoder and a decoder.

[0030] The encoder is calculated using the following formula:

[0031]

[0032] In the formula: X is the input data at the current time; P is the position-encoded data; X' is the final data input to the encoder; Q is the corresponding data for which attention needs to be calculated; K is the projection vector used for similarity matching with Q; V is the actual information content represented in the input data; A is the attention output data; M is the multi-head attention output data; A' is the normalized value of the output data; W q Here is the corresponding query weight matrix; f is the output value after feedback from the neural network; W k W is the key weight matrix; v W1 and W2 are the weight matrix; W1 and W2 are the trainable and optimizable weights; b1 and b2 are the trainable and optimizable biases.

[0033] After the above calculations, the final output of the encoder is as follows:

[0034] H ec =LayerNorm(A+f) (5)2) Decoder, the calculation formula is as follows:

[0035] H′ t-1 =W·[H ec H t-1 ]+b (6)

[0036] In the formula: W is the trainable and optimizable weight; b is the trainable and optimizable bias; H t–1 H' represents the hidden state of the previous time step; t–1 This is the new hidden state;

[0037] The new hidden state, the current input data, and the cell state C from the previous time step. t–1 The input is given to the current cell of the LSTM, and the calculation formula is as follows:

[0038]

[0039] In the formula: f t The result of the forget gate calculation; i t The result of the input gate calculation; o t The output gate's calculation result; W n (n represents f, i, C, o) represents the weights corresponding to the forget gate, input gate, cell state, and output gate; b n (n represents f, i, C, o) represents the biases corresponding to the forget gate, input gate, cell state, and output gate; x t Let be the input at time t; Candidate values ​​for cell state; C t H represents the cell state at time t; t Let O be the hidden state at time t; t This is the output at time t;

[0040] Finally, the prediction results of each scenario are integrated by a weighted fusion method. The predicted value of each scenario is multiplied by its corresponding probability weight and then summed to obtain the electric vehicle charging load prediction curve.

[0041] Before entering the encoder, the multidimensional input sequence is positionally encoded to provide a temporal reference for the model; after positional encoding, the data enters the attention layer for deep computation and interaction.

[0042] The attention output at each time step is further fed into a feedforward neural network, which consists of two fully connected layers with a non-linear activation function in between for non-linear transformation and mapping of features. Information from each location is then independently mapped non-linearly and linearly by the fully connected layers, resulting in the final attention calculation result. At each time step, the encoder output H... ec The hidden state H of the previous time step of the LSTM t–1 After concatenation, a dimensionality reduction mapping is performed through a linear layer (state aggregation linear layer) to generate a new aggregated hidden state H' with the same dimension as the LSTM hidden state. t–1 New hidden state H' t–1 The input data at the current time step and the cell state C at the previous time step. t-1 The input is fed into the current cell of the LSTM. After precise calculations through the forget gate, input gate, and output gate, the hidden state and cell state continue to be passed to the next time step. Simultaneously, the hidden state is transformed through a linear layer to output the predicted value for the current time step. Employing the LSTM-Transformer hybrid model described in this invention, the Transformer decoder enhances its ability to capture global dependencies in temporal data by incorporating a Long Short-Term Memory (LSTM) mechanism.

[0043] Furthermore, the planning-operation dual-layer optimization model includes an outer-layer capacity planning model and an inner-layer optimization operation model. The outer-layer capacity planning model aims to minimize the total life-cycle cost and plans the capacity of photovoltaic and energy storage in the photovoltaic-storage charging station.

[0044] The objective function of the outer capacity planning model is as follows:

[0045]

[0046] In the formula: F1 is the total life cycle cost of the photovoltaic-storage charging station; The cost of purchasing electricity from the grid under scenario w; Revenue from charging services provided to electric vehicles in scenario w; For electricity sales revenue under scenario w; h w Let v be the probability of scenario w occurring; v be the number of photovoltaic-storage-charging station operating scenarios; C s For investment and maintenance costs, C rep For replacement costs;

[0047]

[0048] In the formula: c t The unit price of electricity at time t; Δt represents the power absorbed from the grid at time t under scenario w; Δt represents the charging / discharging time of the photovoltaic-storage charging station.

[0049] The revenue from electricity sales considering time-of-use pricing is as follows:

[0050]

[0051] In the formula: Let t be the electricity price sold to the grid at time t; Let be the grid power connected to the charging station at time t in scenario w;

[0052]

[0053] Where: δ t EV Considering the time-of-use electricity pricing for electric vehicle charging revenue in scenario w at time t; The charging power of scenario w at time t;

[0054] Investment and operation costs C of photovoltaic-storage charging stations s for:

[0055]

[0056] In the formula: C chu C represents the initial investment cost; PVom,t C represents the photovoltaic operation and maintenance cost in year t. Bom,t Rt represents the energy storage operation and maintenance cost in year t; r0 represents the discount rate; N represents the total lifecycle of the photovoltaic-storage charging station; Ct represents the energy storage operation and maintenance cost in year t ... r0 represents the discount rate; N represents the total lifecycle of the photovoltaic-storage charging station; Ct< / Brep,k Cost of replacing the energy storage system for the kth time; t k N represents the year of the k-th replacement. rep For the number of energy storage replacement cycles;

[0057]

[0058] In the formula: C PV C represents the investment cost of a photovoltaic system. B C represents the investment cost of the energy storage system. C C is the investment cost of charging piles; PS C represents the cost of building a photovoltaic system. BS t1 represents the cost of the energy storage system; t2 represents the operating years of the photovoltaic system; t2 represents the operating years of the energy storage system; Q C The unit price of the charging pile; η C This refers to the number of charging stations;

[0059] The inner-layer optimized operation model aims to maximize the economic benefits of photovoltaic-storage charging stations while minimizing voltage fluctuations in the distribution network. It optimizes the operation of these stations, reducing voltage fluctuations and improving grid reliability. The photovoltaic-storage charging station utilizes photovoltaic power to supply energy storage and electric vehicles. When photovoltaic output is insufficient, energy storage and the grid provide power to electric vehicles. Time-of-use pricing can be considered, storing low-priced grid electricity during off-peak hours and supplying or releasing surplus photovoltaic power during peak hours, thus achieving the goal of buying low and selling high, thereby generating profit.

[0060] The optimization goals are as follows:

[0061] 1) The economic benefits of photovoltaic-storage charging stations are expressed by the following formula:

[0062]

[0063] Considering time-of-use pricing, we have:

[0064]

[0065] 2) The number of electric vehicle users is increasing year by year, leading to excessive pressure on the power grid during peak charging periods. Optimizing the configuration of photovoltaic-storage charging stations can improve grid reliability, reduce grid voltage fluctuations, and improve the power quality of the distribution network. Distribution network voltage fluctuations are expressed by the following formula:

[0066]

[0067] In the formula: M is the total number of system nodes; V j (i) represents the voltage at node j at time i; Let J be the average voltage of node j during time period T;

[0068] By optimizing the inner model, the stability of the power grid can be improved while ensuring economic benefits. Combining equations (14) and (16), the objective function shown in equation (17) is obtained:

[0069]

[0070] Furthermore, the constraints of the planning-running two-level optimization model are:

[0071] 1) Power balance constraints

[0072]

[0073] In the formula: Photovoltaic output at time t under scenario w; For scenario w at time t, the power absorbed by the grid is... Let w be the charging / discharging power of the electric vehicle at time t. The charging / discharging power of the energy storage system under scenario w at time t;

[0074] 2) Energy storage charging / discharging power constraints

[0075]

[0076] In the formula: For the maximum charge / discharge power of the energy storage system; P t B Let t be the charging / discharging power of the energy storage system at time t;

[0077] 3) Energy storage state of charge constraints

[0078]

[0079] In the formula: This represents the minimum state of charge of the energy storage. This represents the maximum state of charge of the energy storage. The state of charge of the stored energy at time t;

[0080] 4) Capacity constraints for photovoltaics and energy storage:

[0081]

[0082] In the formula: W pv This refers to the actual installed capacity of photovoltaic power; W pv,max This represents the maximum installed capacity of photovoltaic power; W B This represents the actual installed capacity of energy storage; W B,max This represents the maximum installed capacity for energy storage.

[0083] Considering the coupling relationship between planning and operation, a two-layer multi-objective optimization model for photovoltaic-storage-charging stations is established based on the above. Each layer has its own objective function and constraints, which interact and influence each other. The outer layer model, under its own constraints, transmits the capacity configuration schemes of photovoltaics and energy storage to the inner layer. The inner layer model, based on specific configuration schemes, performs optimization simulations for each scheme and transmits the optimization results to the outer layer. Finally, through multiple iterations of the inner and outer layers, the optimal capacity configuration scheme for the photovoltaic-storage-charging station can be obtained.

[0084] Furthermore, step 3 specifically involves:

[0085] Step 3-1: Input the typical photovoltaic power output curve, the optimized electric vehicle charging load curve, and the system parameters of the distribution network;

[0086] Step 3-2: The outer layer capacity planning model is based on the gray wolf optimization algorithm. The capacity and power of photovoltaic and energy storage are selected as decision variables. Multiple wolf pack positions are randomly generated. After initializing the outer layer population, the results are passed to the inner layer optimization operation model.

[0087] Step 3-3: The inner-layer optimized operation model uses the Cplex solver to solve the optimal charging and discharging strategy of energy storage based on the photovoltaic and energy storage capacity and power passed from the outer-layer capacity planning model, calculates the economic benefits of the photovoltaic-energy storage charging station, and passes the results to the outer-layer capacity planning model.

[0088] Steps 3-4: The outer capacity planning model receives data from the inner optimization operation model, recalculates the new outer population fitness based on the objective function of the outer capacity planning model, and filters the optimal value of the outer objective function based on the gray wolf optimization algorithm to determine the optimal economic benefits and cost investment.

[0089] Step 3-5: Determine if the calculation results converge; if yes, proceed to step 3-6; otherwise, update the position of the outer wolf pack and proceed to step 3-3.

[0090] Step 6: Output the optimal configuration scheme for the photovoltaic-storage-charging station, and the calculation ends.

[0091] The improved Grey Wolf algorithm optimization is specifically as follows:

[0092] 1) Algorithm initialization and chaotic population generation;

[0093] Algorithm parameters are set, including population size, maximum number of iterations, convergence factor, initial and termination values, adjustment factor, and mutation weight. Then, a chaotic mapping strategy is used to generate the initial population, and the Tent mapping equation is used to generate uniformly distributed initial wolf population positions in the solution space of the decision variables. The Tent mapping equation is expressed as follows:

[0094]

[0095] In the formula: z i The state value of the current iteration step; u is the control parameter, which is set to u = 0.5 to ensure the most uniform distribution of each particle;

[0096] 2) Interaction and fitness evaluation between inner and outer layer models;

[0097] Each individual (i.e., a set of photovoltaic-storage capacity configuration schemes) in the initial population generated in step 1) is passed to the inner optimization operation model; the inner model uses the capacity configuration as a fixed parameter, calculates the optimal operation strategy of the system in a typical operation cycle based on the CPLEX solver, and calculates the objective function value characterizing the economy; after completing the calculation, the inner model returns the objective function value result to the outer algorithm;

[0098] 3) External save file updates and leader wolf selection;

[0099] The outer algorithm receives all objective function values ​​returned by the inner algorithm. Based on the Pareto dominance relationship in multi-objective optimization, it updates the outer archive, adds non-dominated solutions from the current population to the archive, and removes individuals dominated by the new solutions from the archive. If the archive exceeds its capacity, it is pruned according to density indices such as crowding distance to retain solutions with better distribution. Subsequently, from the updated outer archive, three non-dominated solutions are selected to serve as α, β, and γ leader wolves, respectively, to guide the population update in this iteration.

[0100] 4) Population location update and mutation operations;

[0101] Each individual wolf in the pack updates its own position according to the positions of the three selected leader wolves, following the encirclement and hunting mechanism of the standard gray wolf algorithm. After the position is updated, a mutation operation is performed on the new generation of the population. By introducing a combination of standard Cauchy random distribution and the current global optimal solution, the individual positions are perturbed to enhance the global search capability.

[0102] The mutation formula is:

[0103]

[0104] In the formula: X G (t) is the global optimal solution at the current time; J(0,1) is the standard Cauchy random distribution at t=1; λ is the adjustment parameter; η is the mutation weight.

[0105] 5) Terminate the judgment and output the result;

[0106] Determine if the current iteration count has reached the preset maximum value; if not, return to step 2) and continue to the next iteration; if it has reached the maximum value, terminate the calculation.

[0107] The beneficial effects of this invention are as follows: In terms of data processing, the method of this invention fully considers the uncertainties of photovoltaic power output and electric vehicle charging load, and uses two methods to obtain photovoltaic power output and electric vehicle charging load data respectively; in terms of the reliability of planning objectives, by constructing a two-layer model that simultaneously considers long-term planning and short-term operation, it is more in line with reality and the results obtained are more reliable; it considers the relationship between the planning and operation of photovoltaic-storage charging stations, and coordinates the optimization of EV charging behavior and the operation of photovoltaic-storage systems to achieve optimal economic benefits and improve the stability and reliability of the power grid. Attached Figure Description

[0108] Figure 1 This is a flowchart of the method described in this invention;

[0109] Figure 2 This is a schematic diagram of the photovoltaic power output curve;

[0110] Figure 3This is a schematic diagram of the charging load of an electric vehicle;

[0111] Figure 4 This is a schematic diagram of electric vehicle charging load before and after participating in demand response;

[0112] Figure 5 This is a schematic diagram of the power output of each module after the process is completed;

[0113] Figure 6 This is a schematic diagram of the annual average voltage fluctuation. Detailed Implementation

[0114] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0115] like Figure 1 As shown, the present invention provides a dual-layer multi-objective capacity optimization configuration method for photovoltaic-storage-charging stations, comprising the following steps:

[0116] Step 1: Construct a photovoltaic power output scenario library using Generative Adversarial Network (GAN), cluster typical scenarios, and combine it with the LSTM-Transformer hybrid model to predict electric vehicle charging load. Guide orderly charging through the time-of-use pricing mechanism to generate an optimized electric vehicle charging load curve.

[0117] Step 2: Establish a two-layer optimization model for planning and operation, including an outer capacity planning model and an inner optimization operation model. The outer capacity planning model configures capacity with the goal of minimizing the total life cycle cost, while the inner optimization operation model formulates operation strategies based on the goals of operational economy and reliability. Dynamic coupling between planning and operation is achieved through bidirectional parameter transmission.

[0118] Step 3: The outer capacity planning model uses an improved gray wolf algorithm to optimize capacity configuration parameters, and combines mixed integer linear programming to solve the operation strategy. The inner optimization operation model obtains the optimal configuration scheme through multiple rounds of iterative feedback.

[0119] To verify the beneficial effects of this invention, scientific demonstration was conducted through experiments:

[0120] (1) Parameter settings and basic data;

[0121] This invention uses MATLAB / Simulink for simulation. The outer capacity planning model is solved using an improved gray wolf algorithm with a population size of 50 and a maximum iteration count of 100. The inner optimization operation model is solved using the Cplex solver. By selecting typical photovoltaic power generation operating conditions and combining actual photovoltaic power generation data from a photovoltaic-storage charging station, the daily photovoltaic output is predicted, as shown below. Figure 2The photovoltaic power output curve is shown. Assuming the daily charging capacity of the charging station is 100 electric vehicles, and the ratio of private cars, taxis, and buses is 7:2:1, based on the actual situation of a photovoltaic-storage charging station, the following can be obtained: Figure 3 The charging power demand curve for electric vehicles at the charging station is shown. Relevant parameters for the photovoltaic-storage charging station are shown in Table 1.

[0122] Table 1. Relevant parameters of photovoltaic and energy storage charging stations

[0123]

[0124] In this embodiment of the invention, the time-of-use pricing is considered when establishing the dual-layer optimization configuration model for photovoltaic-storage-charging stations. The specific time-of-use pricing is shown in Table 2.

[0125] Table 2 Time-of-use Electricity Prices

[0126]

[0127] (2) Generation of photovoltaic power output and electric vehicle charging power demand;

[0128] Based on actual photovoltaic power generation data from a photovoltaic-storage charging station, the method in step 1 is used to predict daily photovoltaic output. The daily average photovoltaic output, daily photovoltaic output volatility, and daily photovoltaic output distribution skewness are used as aggregated characteristic quantities. The aggregated photovoltaic power generation output indices are as follows:

[0129] Average daily photovoltaic output:

[0130]

[0131] In the formula: P ave For the average daily photovoltaic output; P t The photovoltaic output during the t-th time period; T is a period, taken as 24h.

[0132] The daily photovoltaic power output volatility is:

[0133]

[0134] The skewness of the daily photovoltaic power output distribution is:

[0135]

[0136] The annual photovoltaic (PV) power generation data was generated using a generative adversarial network. Daily average PV output, daily PV output volatility, and daily PV output distribution skewness were used as aggregation features for PV scenario clustering. The K-means++ algorithm was then used to aggregate the PV data into four categories. The corresponding PV output and the probability of occurrence for each scenario are shown below. Figure 2As shown, scenario 1 has a probability of 25.9%, scenario 2 has a probability of 28.9%, scenario 3 has a probability of 29.4%, and scenario 4 has a probability of 16.5%. Using the method in step 1, the electric vehicle charging load is predicted. The prediction results for all scenarios are aggregated to obtain the electric vehicle charging load prediction as shown below. Figure 3 As shown.

[0137] Based on photovoltaic power output forecast data, this invention considers optimizing the electric vehicle (EV) charging load by guiding EVs to participate in demand-side response. It guides EVs to charge in an orderly manner through a time-of-use (TOU) pricing strategy, prioritizing charging during the lowest-priced time periods within each TOU period. Furthermore, it fully considers the economic and comfort aspects of electricity consumption for users, obtaining the EV charging load before and after participating in demand response, as shown below. Figure 4 As shown.

[0138] Depend on Figure 4 It can be seen that after participating in demand response, the electric vehicle charging load is relatively smoother. Compared with the curve before optimization, the optimized curve shows a decrease in load during peak charging demand periods and an increase in load during off-peak charging demand periods, thus playing a role in peak shaving and valley filling.

[0139] The obtained photovoltaic output curve and the predicted electric vehicle charging load are substituted into the photovoltaic-storage charging station system for further optimization. Considering the system investment and operation and maintenance costs and stable operation, a two-layer multi-objective capacity optimization configuration model for the photovoltaic-storage charging station is established.

[0140] (3) Analysis of simulation results;

[0141] To investigate the impact of the two-layer multi-objective model designed in this invention on the optimized configuration results of photovoltaic-storage-charging stations, this invention sets up two schemes, as follows:

[0142] Option 1: Use a single-layer model particle swarm optimization algorithm to solve for the optimal capacity configuration;

[0143] Option 2: Optimize the configuration of the photovoltaic-storage charging station using the method of this invention.

[0144] The optimization results under different schemes are shown in Table 2.

[0145] Table 2 Optimization results under different schemes

[0146]

[0147] According to the comparative analysis of the data in Table 2, Scheme 2 shows significant advantages over Scheme 1 in all aspects. In terms of economics, Scheme 2, through the construction of a planning-operation coordinated optimization system, achieves increased returns while reducing initial investment: the photovoltaic installed capacity is optimized from 430kW to 410kW (a decrease of 4.7%), the energy storage capacity is reduced from 533kWh to 483kWh (a decrease of 9.4%), the total life-cycle cost is reduced by 6.2%, including a 13.8% reduction in electricity purchase cost, while the economic benefit increases to 41.8832 million yuan, a 6.6% increase compared to Scheme 1. Regarding grid adaptability, Scheme 2 exhibits lower voltage fluctuations, a 17% reduction compared to Scheme 1, significantly improving power quality and system stability.

[0148] Simulation results show that, compared to Scheme 1, which only performs static optimization on the capacity configuration of the photovoltaic-energy storage system without considering the impact of real-time system operation, Scheme 2 constructs a two-layer architecture of "planning-operation" collaborative optimization. The outer layer focuses on capacity planning decisions for both photovoltaic and energy storage, while the inner layer optimizes the dynamic charging and discharging strategy of energy storage based on the time-of-use pricing mechanism. This two-layer optimization mechanism effectively solves the problem of collaborative optimization between capacity planning and operation strategies. By responding to time-of-use pricing signals in real time, it achieves peak-valley arbitrage, reducing the cost of purchased electricity while increasing electricity sales revenue, resulting in a significant improvement in overall economic benefits compared to Scheme 1. Through the bidirectional feedback mechanism between the planning and operation layers, under the premise of ensuring power supply reliability, Scheme 2 reduces the installed capacity requirement of the photovoltaic-energy storage system by 18.7% compared to Scheme 1, while also reducing grid voltage fluctuations and significantly improving grid operation stability.

[0149] The optimized configuration of the photovoltaic-storage-charging station includes the output of each module and the grid-side power as follows: Figure 5 As shown. By Figure 5 It can be seen that the optimized photovoltaic-storage charging station effectively reduces the power demand on the power grid. The energy storage system purchases electricity from the grid during off-peak hours and discharges during peak hours, thus improving economic efficiency. Figure 5 The data shows that between 12:00 and 15:00, photovoltaic (PV) output exceeds load demand. During this period, PV provides electricity to energy storage, fulfilling the operational strategy of PV-storage charging stations and improving PV utilization. The coordinated operation of PV, energy storage, and the power grid provides power to electric vehicles, alleviating the problem of high grid pressure during peak hours, improving power quality, and enhancing grid stability.

[0150] Annual average voltage fluctuation Figure 6 As shown. By Figure 6 It can be seen that the method described in this invention can accurately predict the photovoltaic output and electric vehicle charging load, and the photovoltaic-storage charging station can reasonably adjust the energy storage output according to the time-of-use electricity price and the photovoltaic output, which effectively reduces the voltage fluctuation of the power grid and improves the stability and power quality of the power grid.

[0151] The above description is merely a preferred embodiment of the present invention and is not intended to further limit the present invention. All equivalent changes made based on the description and drawings of the present invention are within the protection scope of the present invention.

Claims

1. A method for optimizing the capacity configuration of a photovoltaic-storage-charging station using a dual-layer, multi-objective approach, characterized in that: Includes the following steps: Step 1: Construct a photovoltaic power output scenario library using a generative adversarial network (GAN), cluster typical scenarios, and combine the LSTM-Transformer hybrid model to predict electric vehicle charging load, generating an optimized electric vehicle charging load curve. Step 2: Establish a two-layer optimization model for planning and operation, including an outer capacity planning model and an inner optimization operation model. The outer capacity planning model configures capacity with the goal of minimizing the total life cycle cost, while the inner optimization operation model formulates operation strategies based on the goals of operational economy and reliability. Dynamic coupling between planning and operation is achieved through bidirectional parameter transmission. Step 3: Solve the results obtained in Step 2 to obtain the optimal configuration scheme.

2. The method for optimizing the capacity of a photovoltaic-storage-charging station with a dual-layer multi-objective configuration according to claim 1, characterized in that, In step 1, a photovoltaic power output scenario library is constructed using a generative adversarial network (GAN), and typical scenarios are clustered, specifically as follows: Step 1-1: Collect time-series random variables {x} at Z time points. 1,1 , ..., x i,t , ..., x m,Z The photovoltaic output curve is constructed by generating photovoltaic power generation scenarios, where m is the number of photovoltaic cells and Z is the number of time points, 1≤i≤m and 1≤t≤Z; photovoltaic power generation scenarios are generated through GAN and a photovoltaic power generation scenario library is constructed. Steps 1-2: Use the K-means++ algorithm to cluster typical scenarios and generate typical photovoltaic power output scenarios, including: Step 1-2-1: Select any photovoltaic output curve from the photovoltaic output scenario library as the cluster center point G1; Step 1-2-2: Calculate the distance D from G1 to other photovoltaic output curves in the photovoltaic output scenario library; Steps 1-2-3: Randomly select the next cluster center point using a roulette wheel method until K initial points are selected as cluster centers; Steps 1-2-4: Calculate the distance between each photovoltaic output curve in the photovoltaic output scenario library and the K cluster centers, and assign it to the cluster corresponding to the nearest cluster center; Step 1-2-5: Update the cluster center point G of each cluster according to the distance D. i ; Step 1-2-6: Repeat steps 1-2-4 and 1-2-5 until the position of the cluster center point no longer changes.

3. The method for optimizing the capacity of a photovoltaic-storage-charging station with a dual-layer multi-objective configuration according to claim 1, characterized in that, The LSTM-Transformer hybrid model includes an encoder and a decoder. The encoder is calculated using the following formula: In the formula: X is the input data at the current moment; P is the data after position encoding; X' is the final data input to the encoder; Q represents the data for which attention needs to be calculated; K represents the projection vector used for similarity matching with Q; V represents the actual information content represented in the input data; and A represents the attention output data. M represents the multi-head attention output data; A' represents the normalized value of the output data; W q This is the corresponding query weight matrix; f is the output value after feedback from the neural network; W k W is the key weight matrix; v W1 and W2 are the weight matrix; W1 and W2 are the trainable and optimizable weights; b1 and b2 are the trainable and optimizable biases. After the above calculations, the final output of the encoder is as follows: H ec =LayerNorm(A+f) (5) 2) Decoder, the calculation formula is as follows: H′ t-1 =W·[H ec ,H t-1 ]+b (6) In the formula: W is the trainable and optimizable weight; b is the trainable and optimizable bias; H t–1 H' represents the hidden state of the previous time step; t–1 This is the new hidden state; Combine the new hidden state, the current input data, and the cell state C from the previous time step. t–1 The input is given to the current cell of the LSTM, and the calculation formula is as follows: In the formula: f t The result of the forget gate calculation; i t The result of the input gate calculation; o t The output gate's calculation result; W n (n represents f, i, C, o) represents the weights corresponding to the forget gate, input gate, cell state, and output gate; b n (n represents f, i, C, o) represents the biases corresponding to the forget gate, input gate, cell state, and output gate; x t Let be the input at time t; Candidate values ​​for cell state; C t H represents the cell state at time t; t Let O be the hidden state at time t; t This is the output at time t; Finally, the prediction results of each scenario are integrated by a weighted fusion method. The predicted value of each scenario is multiplied by its corresponding probability weight and then summed to obtain the electric vehicle charging load prediction curve.

4. The method for optimizing the capacity of a photovoltaic-storage-charging station with a dual-layer multi-objective configuration according to claim 1, characterized in that, In step 2, the outer capacity planning model aims to minimize the total lifecycle cost and plans the capacity of photovoltaic and energy storage in the photovoltaic-storage charging station. The objective function is as follows: In the formula: F1 is the total life cycle cost of the photovoltaic-storage charging station; The cost of purchasing electricity from the grid under scenario w; Revenue from charging services provided to electric vehicles in scenario w; For electricity sales revenue under scenario w; h w Let v be the probability of scenario w occurring; v be the number of photovoltaic-storage-charging station operating scenarios; C s For investment and maintenance costs, C rep For replacement costs; In the formula: c t The unit price of electricity at time t; Δt represents the power absorbed from the grid at time t under scenario w; Δt represents the charging / discharging time of the photovoltaic-storage charging station. The revenue from electricity sales considering time-of-use pricing is as follows: In the formula: Let t be the electricity price sold to the grid at time t; Let be the grid power connected to the charging station at time t in scenario w; In the formula: Considering the time-of-use electricity pricing for electric vehicle charging revenue in scenario w at time t; The charging power of scenario w at time t; Investment and operation costs C of photovoltaic-storage charging stations s for: In the formula: C chu C represents the initial investment cost; PVom,n C represents the photovoltaic operation and maintenance cost in year n; Bom,n Rn represents the energy storage operation and maintenance cost in year n; r0 represents the discount rate; N represents the total lifecycle of the photovoltaic-storage charging station; Cn represents the energy storage operation and maintenance cost in year n. Brep,k Cost of replacing the energy storage system for the kth time; t k N represents the year of the k-th replacement. rep For the number of energy storage replacement cycles; In the formula: C PV C represents the investment cost of a photovoltaic system. B C represents the investment cost of the energy storage system. C C is the investment cost of charging piles; PS C represents the cost of building a photovoltaic system. BS t1 represents the cost of the energy storage system; t2 represents the operating years of the photovoltaic system; t2 represents the operating years of the energy storage system; Q C This refers to the unit price of a charging pile. η C This refers to the number of charging stations; The inner-layer optimized operation model aims to maximize the economic benefits of the photovoltaic-storage charging station and minimize the voltage fluctuation of the distribution network. The optimization objectives are as follows: 1) The economic benefits of photovoltaic-storage charging stations are expressed by the following formula: Considering time-of-use pricing, we have: 2) Voltage fluctuations in the distribution network are expressed by the following formula: In the formula: M is the total number of system nodes; V j (i) represents the voltage at node j at time i; Let J be the average voltage of node j during time period T; Combining equations (14) and (16), the objective function shown in equation (17) is obtained:

5. The method for optimizing the capacity of a photovoltaic-storage-charging station with a dual-layer multi-objective configuration according to claim 4, characterized in that, The constraints of the planning-running two-level optimization model are: Outer model constraints include: 1) Power balance constraints: In the formula: Photovoltaic output at time t under scenario w; For scenario w at time t, the power absorbed by the grid is... Let w be the charging / discharging power of the electric vehicle at time t. The charging / discharging power of the energy storage system under scenario w at time t; 2) Capacity constraints of photovoltaics and energy storage: In the formula: W pv This refers to the actual installed capacity of photovoltaic power; W pv,max This represents the maximum installed capacity of photovoltaic power; W B This represents the actual installed capacity of energy storage; W B,max This represents the maximum installed capacity for energy storage. Inner model constraints include: 1) Energy storage charging / discharging power constraints: In the formula: For the maximum charge / discharge power of the energy storage system; P t B Let t be the charging / discharging power of the energy storage system at time t; 2) Energy storage state of charge constraints: In the formula: This represents the minimum state of charge of the energy storage. This represents the maximum state of charge of the energy storage. Let t be the state of charge of the stored energy.

6. The method for optimizing the capacity of a photovoltaic-storage-charging station with a dual-layer multi-objective configuration according to claim 4, characterized in that, Step 3 specifically involves: Step 3-1: Input the typical photovoltaic power output curve, the optimized electric vehicle charging load curve, and the system parameters of the distribution network; Step 3-2: The outer layer capacity planning model is based on the gray wolf optimization algorithm. The capacity and power of photovoltaic and energy storage are selected as decision variables. Multiple wolf pack positions are randomly generated. After initializing the outer layer population, the results are passed to the inner layer optimization operation model. Step 3-3: The inner-layer optimized operation model uses the Cplex solver to solve the optimal charging and discharging strategy of energy storage based on the photovoltaic and energy storage capacity and power passed from the outer-layer capacity planning model, calculates the economic benefits of the photovoltaic-energy storage charging station, and passes the results to the outer-layer capacity planning model. Steps 3-4: The outer capacity planning model receives data from the inner optimization operation model, recalculates the new outer population fitness based on the objective function of the outer capacity planning model, and filters the optimal value of the outer objective function based on the gray wolf optimization algorithm to determine the optimal economic benefits and cost investment. Step 3-5: Determine if the calculation results converge; if yes, proceed to step 3-6; otherwise, update the position of the outer wolf pack and proceed to step 3-3. Step 6: Output the optimal configuration scheme for the photovoltaic-storage-charging station, and the calculation ends.

Images