Optimization Methods for Distributed Photovoltaic Microgrid Systems Considering Demand-Side Response and Energy Storage

Abstract

This invention discloses a method for optimizing distributed photovoltaic (PV) microgrid systems, belonging to the field of PV microgrid control technology, specifically considering demand-side response and energy storage. The method includes: Step A: establishing a PV day-ahead output prediction model; Step B: establishing an energy storage and load model; Step C: establishing a PV grid integration capacity assessment and optimization model. The day-ahead PV output is predicted using the PSO-SVM regression prediction method. Addressing the severe problem of PV curtailment, the method considers the impact of energy storage and demand response on PV grid integration rate. Starting from constraints such as no node voltage exceeding limits, no power flow overload, and no energy storage exceeding limits, the optimization objective is established by considering the maximum PV grid integration capacity and the minimum network active power loss, forming a distributed PV grid integration capacity assessment and optimization model that comprehensively considers the PV system's grid integration capacity and economic efficiency. Simulation results show that this model can effectively reduce the PV power plant curtailment rate, providing feasibility guidance for the application of PV in the region.

Description

Technical Field

[0001] This invention belongs to the field of photovoltaic microgrid control technology, and specifically relates to an optimization method for distributed photovoltaic microgrid systems that considers demand-side response and energy storage. Background Technology

[0002] With increasing energy shortages and environmental degradation, the development and application of renewable energy have become a focal point of research. Photovoltaic power generation, as the most abundant and geographically less restricted form of new energy, has developed particularly rapidly. However, due to the uncertainty and intermittency of photovoltaic power generation, its integration into the power grid affects voltage and frequency, making it difficult for the grid to meet power quality requirements. Therefore, large-scale photovoltaic integration poses numerous challenges to the safe and stable operation of the distribution network. When the grid cannot absorb large-scale photovoltaic grid connection, significant curtailment of solar power inevitably occurs. Therefore, current research on promoting photovoltaic grid connection and absorption has extremely important application value.

[0003] Energy storage units can store electrical energy when there is a surplus and supply it when there is a shortage. Microgrids can utilize demand response to regulate user electricity consumption through electricity market pricing. Energy storage and demand response, as two regulation methods, can alter the photovoltaic absorption capacity within a microgrid, and studying and analyzing their specific impacts is of great practical significance.

[0004] Therefore, there is an urgent need for an optimization method for distributed photovoltaic microgrid systems that considers demand-side response and energy storage. This method should establish an optimization scheme for distributed photovoltaic microgrid systems that considers demand-side response and energy storage, propose an optimization scheme for photovoltaic microgrid system absorption, fully consider energy storage characteristics and load characteristics, and comprehensively consider the photovoltaic absorption capacity of the distribution network and the economic efficiency of distribution network operation. Summary of the Invention

[0005] The purpose of this invention is to provide an optimization method for distributed photovoltaic microgrid systems that considers demand-side response and energy storage, characterized by the following steps:

[0006] Step A: Establish a photovoltaic day-ahead output prediction model, use a photovoltaic power prediction method based on particle swarm optimization and improved support vector machine, and process the statistical time data of photovoltaic output through direct prediction method based on the analysis of meteorological influencing factors of photovoltaic power generation in physics, and correct the prediction results under different weather conditions by statistically analyzing the historical data patterns.

[0007] Step B: Establish an energy storage and load model, which includes an energy storage model and a load absorption energy consumption model. The energy storage model suppresses the reverse load characteristics of new energy power generation while achieving peak shaving and valley filling. The load absorption energy consumption model adjusts the operating time of the equipment according to the photovoltaic output characteristics, transferring the time-shiftable load during peak electricity consumption periods to the off-peak electricity consumption periods or the peak photovoltaic output periods, and performs energy consumption analysis based on the proportion of different load categories.

[0008] Step C: Establish an optimization model for the photovoltaic microgrid system with energy storage and demand response. The sum of distributed photovoltaic access at each node is used to measure the photovoltaic absorption capacity of the distribution network. The two objective functions of maximizing photovoltaic absorption capacity and minimizing active power loss are weighted using a random weighting method and combined into an optimization objective. Combining power flow constraints, node voltage deviation constraints, distributed photovoltaic power generation energy constraints, and energy upper limit constraints of the energy storage system, the photovoltaic absorption capacity and economy of the distribution network are considered to achieve the optimization of the distributed photovoltaic microgrid system.

[0009] Step A, establishing the photovoltaic day-ahead power output prediction model, includes the following steps:

[0010] Step A1: Based on the least squares support vector machine, the inequality constraints in the optimization formula are converted into equality constraints; based on the data sample set, the regression algorithm of the least squares support vector machine is used to project the input data to a high-dimensional feature space using the nonlinear mapping φ(x), so that the function value y corresponding to sample x can be approximately represented by f(x);

[0011] y=f(x)=ωφ(x)+b (1)

[0012] In the formula, ω is the weight vector in the feature space, and b is the bias.

[0013] The data sample set is: D = {(x i ,y i Let D = |(i=1,2,...,l)}, where: D represents the data sample set, l is the number of samples in the sample set, and x is the number of samples in the sample set. i For the i-th sample data, y i The function value corresponding to the i-th sample data;

[0014] The loss function of the photovoltaic day-ahead power output prediction model is defined as:

[0015]

[0016] In the formula, ε is the insensitivity coefficient, e is the loss function, and f(x) i ) represents the function value corresponding to the i-th sample;

[0017] The least squares support vector regression is expressed as an optimization problem for equation (2):

[0018]

[0019] In the formula, γ is the penalty factor, J(ω,e) is the least squares support vector, and φ(x) is the least squares support vector. i ) is x i nonlinear mapping, e i The sample error calculated by equation (2), y i This represents the function value corresponding to the i-th sample data;

[0020] Based on equation (3), the Lagrange function is established, namely:

[0021]

[0022] In the formula, a i Let L be a Lagrange multiplier, L be a Lagrange function, a denote the set of Lagrange multipliers, and e be the Lagrange multiplier. i The sample error calculated by equation (2);

[0023] The linear equation obtained based on the KKT conditions is:

[0024]

[0025] In the formula, I is an l×l identity matrix, γ -1 The expression for the remainder is: where b is the reciprocal of the penalty factor, b is the bias, and the rest of the expression is:

[0026]

[0027] In the formula, e l Ω is the identity matrix. i,j For x i With x j The product of nonlinear mappings, where a is the Lagrange multiplier matrix, y is the sample function value matrix, and K(x) is the product of nonlinear mappings. i ,x j ) is the kernel function, and T represents transpose;

[0028] Solving the linear equations in equation (6) yields the values ​​of a and b, and the regression fitting function is obtained. The regression fitting function is:

[0029]

[0030] In the formula, K(x,x) i ) is the kernel function, and the kernel function adopts the radial basis function, which is:

[0031]

[0032] In the formula, σ is the width of the radial basis kernel function;

[0033] Step A2: Perform optimization based on the particle swarm optimization algorithm and output the optimization results, which include the optimal penalty parameter c and the optimal kernel function parameter g;

[0034] Suppose that in a D-dimensional search space, there is a population M = (M1, M2, ..., Mn) consisting of n particles. n ), where the position of the i-th particle is X. i =(X i1 ,X i2 ,...,X iD ) T The speed is V i =(V i1 V i2 ,...,V iD ) T The individual extreme value is P i =(P i1 ,P i2 ,...,P iD ) T The global extremum of the population is P. g =(P g1 ,P g2 ,...,P gD ) T ;

[0035] The evolutionary mode of the particle swarm optimization algorithm is as follows:

[0036] Particle swarm optimization evolutionary model = particle historical inertia + particle self-awareness + particle swarm awareness

[0037] The update formulas for velocity and position during particle iterative optimization are expressed as follows:

[0038] V id (k+1)=ωV id (k)+c1r1[P id (k)-X id (k)]+c2r2[P gd (k)-X id (k)] (9)

[0039] X id (k+1)=X id (k)+V id (k+1) (10)

[0040] In the formula, V id (k) represents the current particle's optimization velocity, P id (k) represents the optimal position of the current particle, X id (k) Current particle position, Pgd (k) represents the current swarm optimal position of the particle, ω is the inertia factor, mainly used to balance the global search and local search capabilities; d = 1, 2, ..., n, where d represents the number of iterations; i = 1, 2, ..., D, where i represents the i-th particle; k is the generation number; c1 and c2 are acceleration factors, which are non-negative, where c1 represents the degree to which the particle inherits its own optimal position, and c2 represents the degree to which the particle inherits the global optimal position; r1 and r2 are random numbers with a value space of [0, 1].

[0041] In each iteration, each particle determines its adaptive value based on the objective function, thus determining the optimal position P of the current particle. id (k) and the optimal position of the group P gd (k), update the velocity and position of each particle through equation (9) and equation (10); when the number of iterations reaches the set value or the result meets the accuracy requirements, the optimization ends, otherwise the iteration continues.

[0042] The photovoltaic power generation prediction method based on the particle swarm optimization algorithm and the improved support vector machine in step A includes the following steps:

[0043] Import historical photovoltaic data;

[0044] Optimization is performed using the particle swarm optimization algorithm;

[0045] Determine whether the optimization criteria are met;

[0046] If the optimization criteria are not met, update the historical best position of each particle; update the global best position of the population.

[0047] If the optimization conditions are met, the optimization results are output, including the optimal penalty parameter c and the optimal kernel function parameter g.

[0048] The optimization results are imported into support vector regression for training and prediction.

[0049] Calculate the mean square error and coefficient of determination;

[0050] Output predicted power.

[0051] Step B, establishing the energy storage and load model, includes the following steps:

[0052] Step B1: The energy storage model is based on the power supply and load characteristics of the energy storage system. During off-peak hours, it can store renewable energy as a load; during peak hours, it releases the stored energy to alleviate the power supply pressure on the power system; charging and discharging cannot occur simultaneously within the same time period, and the charging power... and discharge power The following constraints must be satisfied at time t:

[0053]

[0054] When the output of distributed photovoltaic power exceeds the load When the output of distributed photovoltaic power is less than the load

[0055] Step B2: The load absorption energy consumption model calculates the photovoltaic load absorption within the domain. Through the coordination of time-shiftable and fixed loads, energy consumption analysis is performed based on the proportion of different load categories. The energy consumption calculation method for load absorption is as follows:

[0056] C ec =p c ∫t c L c (t)dt+E c (12)

[0057] In the formula, C ec energy consumption for load absorption, p c Time-of-use pricing has three time periods: peak, off-peak, and valley. c For load operating time; L c E represents the load power. c Operating costs.

[0058] The method for calculating the photovoltaic absorption capacity in step C is as follows:

[0059]

[0060] In the formula, f1 represents the photovoltaic absorption capacity; N PV P represents the total number of photovoltaic nodes connected to the system. PV,m Let m be the amount of photovoltaic power connected to the m-th node.

[0061] The method for calculating active power loss in step C is as follows:

[0062]

[0063] In the formula, f2 is the active power loss, {N B} represents the set of all nodes in the system; P ij Q ij These represent the active power and reactive power flowing through branch ij, respectively; U i R is the voltage value at node i; ij Let be the line resistance value between nodes i and j.

[0064] The functional expression for the random weighting method in step C is:

[0065] maxf=ω1f1-ω2f2 (15)

[0066] In the formula, f is the overall objective function, ω1 and ω2 are random numbers, and ω1+ω2=1, ω1≤1, ω2≤1.

[0067] The beneficial effects of this invention are as follows:

[0068] This invention establishes an optimization scheme for a distributed photovoltaic (PV) microgrid system considering demand-side response and energy storage. First, the day-ahead PV output is predicted using a particle swarm optimization for support vector machine (PSO-SVM) regression prediction method. Addressing the severe problem of PV curtailment, a distributed PV curtailment optimization model is established, based on constraints such as no node voltage exceeding limits, no power flow overload, and no energy storage exceeding limits, aiming for maximum PV absorption capacity and optimal economic efficiency. Simulation analysis is used to analyze the impact of energy storage systems and demand-side response on the PV absorption rate. Simulation results show that the model can effectively reduce the curtailment rate of PV power plants, providing feasibility guidance for PV applications within the region.

[0069] This invention utilizes intelligent algorithms to predict day-ahead photovoltaic (PV) output, fully considering energy storage characteristics and load characteristics. To comprehensively consider the PV absorption capacity of the distribution network and the economic efficiency of its operation, and with the dual objectives of maximizing the system's PV absorption capacity and minimizing operating costs, a PV absorption capacity optimization model is established by combining equality and inequality constraints, and an optimization scheme for PV microgrid system absorption is proposed. Attached Figure Description

[0070] Figure 1 This is a flowchart of the optimization method for distributed photovoltaic microgrid systems that considers demand-side response and energy storage, as described in this invention.

[0071] Figure 2 Flowchart of distributed photovoltaic power output prediction method;

[0072] Figure 3 This is a diagram of the node structure of the simulation system;

[0073] Figure 4 This is a schematic diagram of the particle swarm fitness change curve;

[0074] Figure 5 This is a comparison chart of the predicted results and actual results for the test set.

[0075] Figure 6 The diagram shows the photovoltaic output value and typical daily load curves, where (a) is the curve without energy storage and (b) is the curve with energy storage.

[0076] Figure 7The diagram shows the output curves of four photovoltaic power plants, where (a) is the output curve of photovoltaic power plant 2, (b) is the output curve of photovoltaic power plant 3, (c) is the output curve of photovoltaic power plant 4, and (d) is the output curve of photovoltaic power plant 8.

[0077] Figure 8 A schematic diagram of the energy change curves of the energy storage system set up at the four photovoltaic nodes;

[0078] Figure 9 A schematic diagram of the output curves of photovoltaic system, energy storage system, and upstream bus. Detailed Implementation

[0079] This invention provides an optimization method for distributed photovoltaic microgrid systems that considers demand-side response and energy storage. The invention will be further described in detail below with reference to the accompanying drawings.

[0080] This invention proposes an optimization scheme for a distributed photovoltaic microgrid system that considers energy storage systems and demand-side response. It utilizes artificial intelligence algorithms to predict the day-ahead output of photovoltaic power and establishes a photovoltaic day-ahead output prediction model, an energy storage and load model, and an optimization model for the photovoltaic microgrid system that considers energy storage systems and demand response.

[0081] like Figure 1 The embodiment of the present invention shown discloses an optimization method for a distributed photovoltaic microgrid system that considers demand-side response and energy storage, including the following steps:

[0082] Step A: Establish a photovoltaic day-ahead output prediction model, use a photovoltaic power prediction method based on particle swarm optimization and improved support vector machine, and process the statistical time data of photovoltaic output through direct prediction method based on the analysis of meteorological influencing factors of photovoltaic power generation in physics, and correct the prediction results under different weather conditions by statistically analyzing the historical data patterns.

[0083] In this embodiment, a photovoltaic (PV) day-ahead power output prediction model is established. A PV power prediction method based on particle swarm optimization for support vector machine (PSO-SVM) is used. This method analyzes the PV power generation system from a physics perspective. Based on the analysis of meteorological influencing factors of PV power generation, and combined with direct prediction methods, the statistical time data of PV power output is processed. The prediction results under different external weather conditions are corrected by analyzing the statistical patterns of historical data, reducing the dependence of PV power generation prediction on meteorological data and improving the prediction efficiency and accuracy. The specific process is as follows: Figure 2 As shown.

[0084] The photovoltaic power generation prediction method based on the particle swarm optimization algorithm and the improved support vector machine in step A includes the following steps:

[0085] Import historical photovoltaic data;

[0086] Optimization is performed using the particle swarm optimization algorithm;

[0087] Determine whether the optimization criteria are met;

[0088] If the optimization criteria are not met, update the historical best position of each particle; update the global best position of the population.

[0089] If the optimization conditions are met, the optimization results are output, including the optimal penalty parameter c and the optimal kernel function parameter g.

[0090] The optimization results are imported into support vector regression for training and prediction.

[0091] Calculate the mean square error and coefficient of determination;

[0092] Output predicted power.

[0093] Step A, establishing the photovoltaic day-ahead power output prediction model, includes the following steps:

[0094] Step A1: Based on the least squares support vector machine, the inequality constraints in the optimization formula are converted into equality constraints; based on the data sample set, the regression algorithm of the least squares support vector machine is used to project the input data to a high-dimensional feature space using the nonlinear mapping φ(x), so that the function value y corresponding to the sample x can be approximately represented by f(x).

[0095] In this embodiment, the least squares support vector machine transforms the inequality constraints in the optimization formula into equality constraints, thereby significantly reducing computational complexity and improving the efficiency of solving the regression prediction function. The regression algorithm based on the least squares support vector machine is simply called Least Squares Support Vector Regression (LS-SVR). For the data sample set D = {(x...} i ,y i )|(i=1,2,...,l)} where: D represents the data sample set, l is the number of samples in the sample set, x i For the i-th sample data, y i Let f(x) be the function value corresponding to the i-th sample data. The support vector machine uses a nonlinear mapping φ(x) to project the input data into a high-dimensional feature space, so that the function value y corresponding to sample x can be approximately represented by f(x).

[0096] y=f(x)=ωφ(x)+b (1)

[0097] In the formula, ω is the weight vector in the feature space, and b is the bias.

[0098] The loss function of the photovoltaic day-ahead power output prediction model is defined as:

[0099]

[0100] In the formula, ε is the insensitivity coefficient, e is the loss function, and f(x) i ) represents the function value corresponding to the i-th sample.

[0101] Therefore, LS-SVR can be formulated as an optimization problem for the above functions.

[0102] The least squares support vector regression is expressed as an optimization problem for equation (2):

[0103]

[0104] In the formula, γ is the penalty factor, and taking an appropriate value can prevent the model from overfitting or underfitting. J(ω,e) is the least squares support vector, and φ(x) is the least squares support vector. i ) is x i nonlinear mapping, e i The sample error calculated by equation (2), y i This represents the function value corresponding to the i-th sample data;

[0105] The equality constraints in the optimization formula are shown in equation (3).

[0106] Therefore, the Lagrange function can be established from equation (3), that is:

[0107]

[0108] In the formula, a i Let L be a Lagrange multiplier, L be a Lagrange function, a denote the set of Lagrange multipliers, and e be the Lagrange multiplier. i The sample error is calculated using equation (2).

[0109] The linear equation obtained from the KKT (Ksrush-Kuhn-Tucker) conditions is as follows:

[0110]

[0111] In the formula, I is an l×l identity matrix, γ -1 The expression for the remainder is: where b is the reciprocal of the penalty factor, b is the bias, and the rest of the expression is:

[0112]

[0113] In the formula, e lΩ is the identity matrix. i,j For x i With x j The product of nonlinear mappings, where a is the Lagrange multiplier matrix, y is the sample function value matrix, and K(x) is the product of nonlinear mappings. i ,x j ) is the kernel function, and T represents transpose;

[0114] Solving the linear equations in equation (6) yields the values ​​of a and b, and provides the regression fitting function.

[0115] The regression fitting function is:

[0116]

[0117] In the formula, K(x,x) i ) is the kernel function, using the radial basis function (RBF) kernel function, and its expression is:

[0118]

[0119] In the formula, σ is the width of the radial basis kernel function.

[0120] Step A2: Perform optimization based on the particle swarm optimization algorithm and output the optimization results, which include the optimal penalty parameter c and the optimal kernel function parameter g;

[0121] In this embodiment, the particle swarm optimization algorithm is used to select parameters. The following describes its basic algorithm.

[0122] Suppose that in a D-dimensional search space, there is a population M = (M1, M2, ..., Mn) consisting of n particles. n ), where the position of the i-th particle is X. i =(X i1 ,X i2 ,...,X iD ) T The speed is V i =(V i1 V i2 ,...,V iD ) T The individual extreme value is P i =(P i1 ,P i2 ,...,P iD ) T The global extremum of the population is P. g =(P g1 ,P g2 ,...,P gD ) T ;

[0123] The speed formula for the particle swarm optimization algorithm consists of three parts, and is usually expressed as follows:

[0124] The evolutionary mode of the particle swarm optimization algorithm is as follows:

[0125] Particle swarm optimization evolutionary model = particle historical inertia + particle self-awareness + particle swarm awareness

[0126] Therefore, the update formulas for velocity and position during the particle iterative optimization process are expressed as follows:

[0127] V id (k+1)=ωV id (k)+c1r1[P id (k)-X id (k)]+c2r2[P gd (k)-X id (k)] (9)

[0128] X id (k+1)=X id (k)+V id (k+1) (10)

[0129] In the formula, V id (k) represents the current particle's optimization velocity, P id (k) represents the optimal position of the current particle, X id (k) Current particle position, P gd (k) represents the current swarm optimal position of the particle, ω is the inertia factor, mainly used to balance the global search and local search capabilities; d = 1, 2, ..., n, where d represents the number of iterations; i = 1, 2, ..., D, where i represents the i-th particle; k is the generation number; c1 and c2 are acceleration factors, which are non-negative, where c1 represents the degree to which the particle inherits its own optimal position, and c2 represents the degree to which the particle inherits the global optimal position; r1 and r2 are random numbers with a value space of [0, 1].

[0130] Each particle needs to determine its adaptive value based on the objective function during each iteration, thereby determining the optimal position P of the current particle. id (k) and the optimal position of the group P gd (k), the velocity and position of each particle are updated by equations (9) and (10). When the number of iterations reaches the set value or the result meets the accuracy requirements, the optimization ends; otherwise, the iteration continues.

[0131] Step B: Establish an energy storage and load model, which includes an energy storage model and a load absorption energy consumption model. The energy storage model suppresses the reverse load characteristics of new energy power generation while achieving peak shaving and valley filling. The load absorption energy consumption model adjusts the operating time of the equipment according to the photovoltaic output characteristics, transferring the time-shiftable load during peak electricity consumption periods to the off-peak electricity consumption periods or the peak photovoltaic output periods, and performs energy consumption analysis based on the proportion of different load categories.

[0132] In this embodiment, an energy storage and load model is established. The energy storage model is built by fully considering the characteristics of energy storage. The operating time of the equipment is adjusted according to the photovoltaic output characteristics, shifting time-shiftable loads from peak electricity consumption periods to off-peak periods or peak photovoltaic output periods. Energy consumption analysis is performed based on the proportion of different load categories, and a calculation model for load absorption and energy consumption is established.

[0133] Step B1: The energy storage model is based on the power supply characteristics and load characteristics of the energy storage system. During off-peak hours, it can store new energy in the form of load; during peak hours, it releases the stored electricity to alleviate the power supply pressure on the power system.

[0134] In this embodiment, the energy storage system has flexible power throughput characteristics, which can effectively suppress the reverse load characteristics of renewable energy generation while effectively shaving peak loads and filling valleys, promoting the consumption of renewable energy and ensuring the safety and stability of the main grid. In addition to power supply characteristics, the energy storage system also has load characteristics: it can store renewable energy in the form of load during periods of low electricity demand; during peak electricity demand periods, it can release the electricity stored during off-peak hours to alleviate the power supply pressure on the power system.

[0135] Charging and discharging cannot occur simultaneously within the same time period, i.e., charging power... and discharge power The following conditions must be met at time t:

[0136]

[0137] When the output of distributed photovoltaic power exceeds the load When the output of distributed photovoltaic power is less than the load

[0138] Step B2: The load absorption energy consumption model calculates the photovoltaic load absorption within the domain. By coordinating time-shiftable loads and fixed loads, energy consumption analysis is performed based on the proportion of different load categories.

[0139] In this embodiment, the photovoltaic load absorption within the domain is mainly achieved through the coordination of time-shiftable loads and fixed loads. The operating time of the equipment is adjusted according to the photovoltaic output characteristics to transfer the time-shiftable loads during peak electricity consumption periods to the off-peak electricity consumption periods or the peak photovoltaic output periods.

[0140] Based on the proportion of different load categories, energy consumption analysis is performed, and the energy consumption for load absorption is calculated as follows:

[0141] C ec =p c ∫t c L c (t)dt+E c (12)

[0142] In the formula, p c Time-of-use pricing has three time periods: peak, off-peak, and valley. c For load operating time; L c E represents the load power. c Operating costs.

[0143] Step C: Establish an optimization model for the photovoltaic microgrid system with energy storage and demand response. The sum of distributed photovoltaic access at each node is used to measure the photovoltaic absorption capacity of the distribution network. The two objective functions of maximizing photovoltaic absorption capacity and minimizing active power loss are weighted using a random weighting method and combined into an optimization objective. Combining power flow constraints, node voltage deviation constraints, distributed photovoltaic power generation energy constraints, and energy upper limit constraints of the energy storage system, the photovoltaic absorption capacity and economy of the distribution network are considered to achieve the optimization of the distributed photovoltaic microgrid system.

[0144] In this embodiment, an optimization model for an energy storage system and a demand-response photovoltaic microgrid system is established. The sum of the distributed photovoltaic access volume of each node is used to measure the photovoltaic absorption capacity of the distribution network. The maximum absorption capacity and the minimum active power loss are weighted and combined into an optimization objective. The photovoltaic absorption capacity and economic efficiency of the distribution network are considered in combination with power flow constraints, node voltage deviation constraints, distributed photovoltaic power generation energy constraints, and energy upper limit constraints of the energy storage system.

[0145] The sum of the distributed photovoltaic (PV) grid connection amounts at each node is taken as the PV absorption capacity of the distribution network. The calculation method for the PV absorption capacity in step C is as follows:

[0146]

[0147] In the formula, N PV P represents the total number of photovoltaic nodes connected to the system. PV,m Let m be the amount of photovoltaic power connected to the m-th node.

[0148] After photovoltaic grid connection, the calculation method for active power loss in step C is as follows:

[0149]

[0150] In the formula, f2 is the active power loss, {N B} represents the set of all nodes in the system; P ijQ ij These represent the active power and reactive power flowing through branch ij, respectively; U i R is the voltage value at node i; ij Let be the line resistance value between nodes i and j.

[0151] To comprehensively consider the photovoltaic absorption capacity of the distribution network and the economic efficiency of its operation, in this embodiment, the functional expression of the random weighting method in step C is:

[0152] maxf=ω1f1-ω2f2 (15)

[0153] In the formula, f is the overall objective function, ω1 and ω2 are random numbers, and ω1+ω2=1, ω1≤1, ω2≤1.

[0154] To enable those skilled in the art to better understand the present invention and its advantages over the prior art, the applicant provides further explanation in conjunction with specific embodiments.

[0155] 1. Simulation system node structure diagram, particle swarm fitness change curve, and photovoltaic prediction result curve.

[0156] The node structure of the simulation system is as follows Figure 3 As shown, power generation was predicted using open photovoltaic data from Guoneng Rixin. A total of 66,860 data sets were collected from April 1, 2016 to April 30, 2018, with a sampling interval of 15 minutes. 50,000 data sets were randomly selected as the training set, and 50 sets were randomly selected from the remaining samples for experimental comparison.

[0157] In optimizing the parameters of a support vector machine using the particle swarm optimization algorithm, the maximum number of iterations is 100, the number of particles n = 50, the acceleration factors c1 = 1.2, c2 = 1.2, the search range of the penalty parameter c is [0.001, 50], and the search range of the kernel function parameter g is [0.001, 20]. The particle swarm fitness (root mean square error) changes as follows: Figure 4 As shown, the optimal penalty parameter c is 15.8334 and the kernel function parameter g is 0.05591.

[0158] The relative standard deviation (RSD) measures the repeatability of regression predictions, that is, the consistency between independent predictions. The formula is:

[0159]

[0160] In the formula, f RSD x represents the relative standard deviation. i For predicted values, is the predicted average, and n is the number of predictions.

[0161] The root mean square error (MSE) is used to evaluate the accuracy of regression predictions, that is, the degree of agreement between the predicted results and the actual values. The calculation formula is:

[0162]

[0163] In the formula, f MSE The root mean square error, x represents the true value of photovoltaic power output. i This is the predicted value for photovoltaic power output. The operating results are as follows: Figure 5 As shown in the figure, the prediction results of the support vector machine regression test set show that the correlation coefficient is 0.90375 and the root mean square error is 0.69491, which provides a high level of accuracy in predicting photovoltaic power output.

[0164] 2. Analysis of Photovoltaic Absorption Capacity

[0165] To study the impact of energy storage systems on microgrid photovoltaic (PV) consumption, two scenarios were established for comparative optimization analysis: one without energy storage and the other with energy storage. PV output and typical daily load curves are shown below. Figure 6 As shown, (a) is a schematic diagram of the curve without energy storage, and (b) is a schematic diagram of the curve with energy storage.

[0166] from Figure 6 It is clear from this that when energy storage is not added to the power distribution system, such as Figure 6 As shown in (a), there is a significant mismatch between photovoltaic (PV) output and load demand. PV output exceeds load between 9:00 and 17:00, with peak load around 20:00, and PV output drops to zero after 20:00. Adding an energy storage system, as... Figure 6 As shown in (b), the photovoltaic power and load are basically matched between 9 and 17 hours, and the photovoltaic output is absorbed to a greater extent. The photovoltaic absorption rate of the four photovoltaic power plants is compared in Table 1.

[0167] Table 1 Comparison of photovoltaic grid integration rates before and after optimization

[0168]

[0169] Comparing the two scenarios, it can be seen that the introduction of an energy storage system increased the microgrid's photovoltaic (PV) absorption rate from 45.0454% to 78.1356%, significantly improving the PV absorption capacity. The output of the four PV power plants is as follows: Figure 7 As shown, (a) is the output curve of photovoltaic power station 2, (b) is the output curve of photovoltaic power station 3, (c) is the output curve of photovoltaic power station 4, and (d) is the output curve of photovoltaic power station 8.

[0170] The energy changes of the energy storage system installed at the four photovoltaic nodes are as follows: Figure 8As shown, from 9:00 to 17:00, the photovoltaic output can fully supply the system load, and the output of the photovoltaic nodes is in a state of oversupply. At this time, the energy storage system can store new energy in the form of load. Around 20:00, due to the lack of sunlight, the photovoltaic output is zero, and the output of the photovoltaic nodes cannot meet the system load demand, and the system output is in a state of undersupply. At this time, the energy storage system can release the stored electrical energy to alleviate the power supply pressure of the power system. The energy storage system exhibits power supply characteristics.

[0171] Photovoltaic output, energy storage system output, and upstream bus power, such as Figure 9 As shown, the entire distribution network system balances the active load through the combined output of photovoltaic (PV) power and energy storage system. Starting at 10:00 AM, the sum of the PV power output and energy storage system output in the distribution system can fully meet the active load demand. The system does not need to obtain active power from the upstream bus, which theoretically enables on-site control of distributed PV power consumption. The local control layer works quickly, maximizing the PV power consumption rate while minimizing operating costs.

Claims

1. An optimization method for a distributed photovoltaic microgrid system considering demand-side response and energy storage, characterized in that, Includes the following steps: Step A: Establish a photovoltaic day-ahead output prediction model, use a photovoltaic power prediction method based on particle swarm optimization and improved support vector machine, and process the statistical time data of photovoltaic output through direct prediction method based on the analysis of meteorological influencing factors of photovoltaic power generation in physics, and correct the prediction results under different weather conditions by statistically analyzing the historical data patterns. Step B: Establish an energy storage and load model, which includes an energy storage model and a load absorption energy consumption model. The energy storage model suppresses the reverse load characteristics of new energy power generation while achieving peak shaving and valley filling. The load absorption energy consumption model adjusts the operating time of the equipment according to the photovoltaic output characteristics, transferring the time-shiftable load during peak electricity consumption periods to the off-peak electricity consumption periods or the peak photovoltaic output periods, and performs energy consumption analysis based on the proportion of different load categories. Step C: Establish an optimization model for the photovoltaic microgrid system with energy storage and demand response. The sum of distributed photovoltaic access at each node is used to measure the photovoltaic absorption capacity of the distribution network. The two objective functions of maximizing photovoltaic absorption capacity and minimizing active power loss are weighted using a random weighting method and combined into an optimization objective. Combining power flow constraints, node voltage deviation constraints, distributed photovoltaic power generation energy constraints, and energy upper limit constraints of the energy storage system, the photovoltaic absorption capacity and economy of the distribution network are considered to achieve the optimization of the distributed photovoltaic microgrid system.

2. The optimization method for distributed photovoltaic microgrid systems considering demand-side response and energy storage according to claim 1, characterized in that, Step A, establishing the photovoltaic day-ahead power output prediction model, includes the following steps: Step A1: Based on the least squares support vector machine, the inequality constraints in the optimization formula are converted into equality constraints; based on the data sample set, the regression algorithm of the least squares support vector machine is used to project the input data to a high-dimensional feature space using the nonlinear mapping φ(x), so that the function value y corresponding to sample x can be approximately represented by f(x); y=f(x)=ωφ(x)+b (1) In the formula, ω is the weight vector in the feature space, and b is the bias. The data sample set is: D = {(x i ,y i Let D = |(i=1,2,...,l)}, where: D represents the data sample set, l is the number of samples in the sample set, and x is the number of samples in the sample set. i For the i-th sample data, y i The function value corresponding to the i-th sample data; The loss function of the photovoltaic day-ahead power output prediction model is defined as: In the formula, ε is the insensitivity coefficient, e is the loss function, and f(x) i ) represents the function value corresponding to the i-th sample; The least squares support vector regression is expressed as an optimization problem for equation (2): In the formula, γ is the penalty factor, J(ω,e) is the least squares support vector, and φ(x) is the least squares support vector. i ) is x i nonlinear mapping, e i The sample error calculated by equation (2), y i This represents the function value corresponding to the i-th sample data; Based on equation (3), the Lagrange function is established, namely: In the formula, a i Let L be a Lagrange multiplier, L be a Lagrange function, a denote the set of Lagrange multipliers, and e be the Lagrange multiplier. i The sample error calculated by equation (2); The linear equation obtained based on the KKT conditions is: In the formula, I is an l×l identity matrix, γ -1 The expression for the remainder is: where b is the reciprocal of the penalty factor, b is the bias, and the rest of the expression is: In the formula, e l Ω is the identity matrix. i,j For x i With x j The product of nonlinear mappings, where a is the Lagrange multiplier matrix, y is the sample function value matrix, and K(x) is the product of nonlinear mappings. i ,x j ) represents the kernel function, and T represents the transpose; Solving the linear equations in equation (6) yields the values ​​of a and b, and the regression fitting function is obtained. The regression fitting function is: In the formula, K(x,x) i ) is the kernel function, and the kernel function adopts the radial basis function, which is: In the formula, σ is the width of the radial basis kernel function; Step A2: Perform optimization based on the particle swarm optimization algorithm and output the optimization results, which include the optimal penalty parameter c and the optimal kernel function parameter g; Suppose that in a D-dimensional search space, there is a population M = (M1, M2, ..., Mn) consisting of n particles. n ), where the position of the i-th particle is X. i =(X i1 ,X i2 ,...,X iD ) T The speed is V i =(V i1 V i2 ,...,V iD ) T The individual extreme value is P i =(P i1 ,P i2 ,...,P iD ) T The global extremum of the population is P. g =(P g1 ,P g2 ,...,P gD ) T ; The evolutionary mode of the particle swarm optimization algorithm is as follows: Particle swarm optimization evolutionary model = particle historical inertia + particle self-awareness + particle swarm awareness The update formulas for velocity and position during particle iterative optimization are expressed as follows: V id (k+1)=ωV id (k)+c1r1[P id (k)-X id (k)]+c2r2[P gd (k)-X id (k)] (9) X id (k+1)=X id (k)+V id (k+1) (10) In the formula, V id (k) represents the current particle's optimization velocity, P id (k) represents the optimal position of the current particle, X id (k) Current particle position, P gd (k) represents the current swarm optimal position of the particle, ω is the inertia factor, mainly used to balance the global search and local search capabilities; d = 1, 2, ..., n, where d represents the number of iterations; i = 1, 2, ..., D, where i represents the i-th particle; k is the generation number; c1 and c2 are acceleration factors, which are non-negative, where c1 represents the degree to which the particle inherits its own optimal position, and c2 represents the degree to which the particle inherits the global optimal position; r1 and r2 are random numbers with a value space of [0, 1]. In each iteration, each particle determines its adaptive value based on the objective function, thus determining the optimal position P of the current particle. id (k) and the optimal position of the group P gd (k), update the velocity and position of each particle through equation (9) and equation (10); when the number of iterations reaches the set value or the result meets the accuracy requirements, the optimization ends, otherwise the iteration continues.

3. The optimization method for distributed photovoltaic microgrid systems considering demand-side response and energy storage according to claim 1, characterized in that, The photovoltaic power generation prediction method based on the particle swarm optimization algorithm and the improved support vector machine in step A includes the following steps: Import historical photovoltaic data; Optimization is achieved using the particle swarm optimization algorithm; Determine whether the optimization criteria are met; If the optimization criteria are not met, update the historical best position of each particle; update the global best position of the population. If the optimization conditions are met, the optimization results are output, including the optimal penalty parameter c and the optimal kernel function parameter g. The optimization results are imported into support vector regression for training and prediction. Calculate the mean square error and coefficient of determination; Output predicted power.

4. The optimization method for distributed photovoltaic microgrid systems considering demand-side response and energy storage according to claim 1, characterized in that, Step B, establishing the energy storage and load model, includes the following steps: Step B1: The energy storage model is based on the power supply and load characteristics of the energy storage system. During off-peak hours, it can store renewable energy as a load; during peak hours, it releases the stored energy to alleviate the power supply pressure on the power system; charging and discharging cannot occur simultaneously within the same time period, and the charging power... and discharge power The following constraints must be satisfied at time t: When the output of distributed photovoltaic power exceeds the load When the output of distributed photovoltaic power is less than the load Step B2: The load absorption energy consumption model calculates the photovoltaic load absorption within the domain. Through the coordination of time-shiftable and fixed loads, energy consumption analysis is performed based on the proportion of different load categories. The energy consumption calculation method for load absorption is as follows: C ec =p c ∫t c L c (t)dt+E c (12) In the formula, C ec energy consumption for load absorption, p c Time-of-use pricing has three time periods: peak, off-peak, and valley. c For load operating time; L c E represents the load power. c Operating costs.

5. The optimization method for a distributed photovoltaic microgrid system considering demand-side response and energy storage according to claim 1, characterized in that, The method for calculating the photovoltaic absorption capacity in step C is as follows: In the formula, f1 represents the photovoltaic absorption capacity; N PV P represents the total number of photovoltaic nodes connected to the system. PV,m Let m be the amount of photovoltaic power connected to the m-th node.

6. The optimization method for a distributed photovoltaic microgrid system considering demand-side response and energy storage according to claim 1, characterized in that, The method for calculating active power loss in step C is as follows: In the formula, f2 is the active power loss, {N B } represents the set of all nodes in the system; P ij Q ij These represent the active power and reactive power flowing through branch ij, respectively; U i R is the voltage value at node i; ij Let be the line resistance value between nodes i and j.

7. The optimization method for a distributed photovoltaic microgrid system considering demand-side response and energy storage according to claim 1, characterized in that, The functional expression for the random weighting method in step C is: maxf=ω1f1-ω2f2 (15) In the formula, f is the overall objective function, ω1 and ω2 are random numbers, and ω1+ω2=1, ω1≤1, ω2≤1.

Images