Flexible resource coordination scheduling optimization method for self-adaptive power distribution network fault off-grid operation micro-grid

Abstract

The invention discloses an adaptive power distribution network fault off-grid operation micro-grid flexible resource coordinated scheduling optimization method. The method comprises the following steps of collecting flexible resource operation historical data in a micro-grid and corresponding environment parameters as sample data; establishing an RBF neural network prediction model based on the sample data; building a microgrid flexible resource coordination optimization mathematical model; and proposing a micro-grid flexible resource coordination optimization strategy based on the RBF neural network prediction model and the micro-grid flexible resource coordination optimization mathematical model, and determining a final flexible resource scheduling scheme. Coordination and interaction of flexible resources of the micro-grid in the emergency state of the power distribution network are supported, power failure loss is reduced, and the supporting effect of the micro-grid is brought into full play.

Description

Technical Field

[0001] The present invention relates to the technical field of coordinated optimization scheduling of resources within a microgrid, and particularly to a flexible resource coordinated control optimization method for a microgrid that operates off-grid adaptively during a distribution network fault. Background Art

[0002] Currently, the demand for clean and safe energy in countries around the world is increasing day by day. The proposal of the dual-carbon goal has further promoted the large-scale development of renewable clean energies such as solar energy and wind energy. In the short term, the energy system in China's power system dominated by coal will gradually transform into a new power system dominated by new energy. However, the problems of poor stability and low density of renewable energy pose higher requirements for aspects such as the power quality and safe and stable operation of the power system.

[0003] The concept of a microgrid is proposed to enable the flexible and efficient application of distributed power sources, solve the problem of grid connection of a large number of distributed power sources with diverse forms, and at the same time facilitate the transition of the traditional power grid to a smart grid. However, the inherent complexity of the microgrid poses challenges to existing coordinated control strategies, and the randomness and differences of various renewable energies within the grid are also problems that are difficult to solve with existing prediction methods. A microgrid has two operating modes: off-grid operation and grid-connected operation. When operating in grid-connected mode, the coordinated operation of resources within the grid is uniformly scheduled by the large power grid. When the distribution network is in an emergency state, the microgrid operates off-grid. How to achieve coordinated control of flexible resources within the microgrid under the premise of disconnecting from the large power grid scheduling, so that the microgrid can reduce the loss of load and lower the operating cost while operating safely and stably will be a key research direction. Specifically, how to coordinate the scheduling of various flexible resources within the microgrid according to the output prediction of various flexible resources within the microgrid, combined with the change of load demand, with the goal of the optimal economic operation of the microgrid is the key research direction. The core problems can be summarized into two major problems: one is the output prediction problem of various flexible resources in the microgrid; the other is the problem of collaborative optimization scheduling of various resources to ensure the economic and stable operation of the microgrid. Summary of the Invention

[0004] The purpose of the present invention is to solve the deficiencies of the above-mentioned existing technologies, and thus provide a flexible resource coordinated scheduling optimization method for a microgrid that operates off-grid adaptively during a distribution network fault, reducing fault losses and improving the stable operation level of the distribution network.

[0005] The present invention summarizes the power generation influencing factors by mathematically modeling power generation systems such as photovoltaic and wind power. Based on historical output data and load demand data, an RBF neural network prediction model is established to achieve the prediction of power output and load demand during the expected power restoration time, that is, the time when the microgrid is reconnected to the grid. A coordinated optimization model of flexible resources in the microgrid with the goal of optimal economic operation of the microgrid is built, and a coordinated optimization strategy for flexible resources in the microgrid is proposed. Then, an improved particle swarm intelligent algorithm is used to solve the optimal solution of the model, and the final coordinated scheduling plan of flexible resources in the microgrid is determined. The power outage loss caused by distribution network faults is reduced, the supporting role of the microgrid is fully exerted, and the economic and stable operation of the off-grid microgrid is guaranteed.

[0006] To achieve the above object, the present invention proposes an adaptive coordinated scheduling and optimization method for flexible resources in an off-grid microgrid during distribution network faults, including the following steps:

[0007] (1) Collect the historical operation data of flexible resources within the microgrid and the corresponding environmental parameters as sample data;

[0008] (2) Establish an RBF neural network prediction model based on the sample data;

[0009] (3) Build a coordinated optimization mathematical model for flexible resources in the microgrid;

[0010] (4) Propose a coordinated optimization strategy for flexible resources in the microgrid based on the RBF neural network prediction model and the coordinated optimization mathematical model of flexible resources in the microgrid, and determine the final flexible resource scheduling plan.

[0011] In the step (1), the historical data includes the output of distributed power sources within the microgrid and load value data on multiple typical days and for at least one continuous week, and the environmental parameters include weather, temperature, wind speed, etc. under the corresponding operation data.

[0012] In the step (2), establishing an RBF neural network prediction model based on the sample data is specifically as follows:

[0013] (21) Take a set of sample data, and perform clustering analysis on the data using the K-Means clustering method. The input layer of the corresponding RBF prediction model is environmental parameters such as weather, temperature, and wind speed in the sample data, and the output layer is the power output value or load value under the corresponding input parameters;

[0014] (22) Based on the data clustering results of the K-Means clustering method, determine the centers and widths of the hidden layer units of the RBF neural network prediction model;

[0015] Assume that there are a total of K clustering results in the sample data. Let the nearest center of the kth clustering data group be T k (n); (n = 1, 2,...), that is, the center of the nth iteration of the kth group of clustering data, and its center satisfies:

[0016] C k (S) = min k ||S i -T k (n)|| (k = 1, 2, …) (1)

[0017] Wherein, C k (S) is the minimum distance from the k-th group of clustering data samples to the center, and S i is the i-th sample point of the k-th group of clustering data;

[0018] Judge whether the center converges. If so, determine the center value; otherwise, continue to iterate until convergence. The specific expression of the iteration process is:

[0019] T k (n + 1) = T k (n) + α(S i -T k (n)) (0 < α < 1) (2)

[0020] Wherein, T k (n + 1) is the adjusted center, T k (n) is the center before adjustment, and α is the adjustment coefficient;

[0021] After determining the center value, determine the width according to the Gaussian function. Specifically:

[0022]

[0023] Wherein, c k is the final center of the k-th group of clustering data, and σ k is the width of the k-th group of clustering data;

[0024] The hidden layer unit of the RBF neural network prediction model corresponding to the k-th group of clustering data can be expressed as:

[0025]

[0026] Wherein, is the function of the hidden unit corresponding to the k-th group of clustering data, and is a non-negative nonlinear function that radially symmetrically decays with respect to the center point. ||x|| is the Euclidean norm, and X is the input vector;

[0027] (23) Take another group of sample data and initialize it, input it into the determined hidden layer function, and use the least squares method to train the output layer weights to obtain a more accurate RBF prediction model. The weights are expressed as:

[0028]

[0029] where ω is the weight from the hidden layer to the output layer, h is the number of nodes in the hidden layer, and c max is the maximum distance between the selected centers; x p is the p-th input sample, and c i is the clustering center of the training samples;

[0030] (24) Determine whether the weight error is the smallest. If so, determine the mathematical function in the final RBF neural network prediction model; otherwise, go to step (35) to iterate the weights.

[0031] (25) Update the weights using the stochastic gradient descent method, and calculate the gradient Δl old of the loss function with respect to the weight ω i (ω old ), where η is the learning rate. The updated weight is expressed as:

[0032] ω new = ω old - ηΔl i (ω old ).

[0033] The weight error is calculated as follows:

[0034] Suppose a set of sample data has Q samples. The input of the sample data is x i , and the corresponding output is y i . The output obtained by inputting x i into the RBF neural network prediction model is The weight error e is expressed as:

[0035]

[0036] In step (3), build a coordinated optimization mathematical model for flexible resources in the microgrid, specifically:

[0037] (31) The objective function for building a coordinated optimization mathematical model for flexible resources in the microgrid with the optimal operation of the microgrid system is:

[0038]

[0039] where COE i,t is the generation cost of the i-th power source at time t, PC t is the penalty cost for wind and light curtailment at time t, CR t is the compensation cost for interruptible load at time t, that is, the compensation cost for the load cut due to insufficient power; N is the total number of power sources participating in coordinated dispatch at time t;

[0040] The penalty cost for wind and light curtailment is expressed as:

[0041] PC t = λa ·(p wt-a +p pv-a ) (7)

[0042] In the formula, λ a is the penalty factor for unit curtailment of light and wind power, p wt-a is the wind power curtailment, p pv-a is the light power curtailment;

[0043] The compensation cost of the interruptible load is expressed as:

[0044] CR t =λ b ·p load ·h (8)

[0045] In the formula, λ b is the compensation amount for unit interruptible load, p load is the interruptible load, and h is the interruption time;

[0046] The power generation cost is expressed as:

[0047] COE i,t =CF i,t +IV i,t +OM i,t (9)

[0048] In the formula, CF i,t is the fuel cost of the i-th power source at time t, IV i,t is the depreciation cost of the i-th power source at time t, and OM i,t is the maintenance cost of the i-th power source at time t;

[0049] The fuel cost is expressed as:

[0050]

[0051] In the formula, P i,t is the power generation power of the i-th power source at time t, η is the power generation efficiency of the power source, and C is the power generation fuel cost per unit power.

[0052] The depreciation cost is expressed as:

[0053]

[0054] In the formula, C INS,i is the installation cost of the i-th power source, P r,i is the rated power of the i-th power source, f r,i is the capacity factor of the i-th power source, d is the depreciation rate, and m is the service life of the power source;

[0055] The maintenance cost is expressed as:

[0056] OMi,t = k m,i × p i,t (12)

[0057] Wherein, k m,i is the unit operation and maintenance cost of the i-th power source.

[0058] (32) Determine the constraint conditions of the microgrid flexible resource coordination optimization mathematical model, including the output constraint of the microgrid, the power balance constraint of the microgrid, the power ramp constraint of the power source, and the energy storage constraint;

[0059] 1) Output constraint of the microgrid

[0060] Corresponding to the operation constraints of the power sources in the microgrid, that is, the output power of each power source in the microgrid is within the upper and lower limits of the power of each power source, which is expressed as:

[0061] P i,min ≤ P i,t ≤ P i,max (13)

[0062] Wherein, P i,t is the output power of the i-th power source at time t, P i,min , P i,max are the upper and lower limits of the output power of the i-th power source;

[0063] 2) Power balance constraint of the microgrid

[0064] At all times, the output power in the microgrid should be equal to the load power in the grid. Among them, when the energy storage is charging, it is regarded as the load power, and when it is discharging, it is regarded as the output power. When the output of the microgrid does not meet all the loads, the load power in the grid is the total load power in the grid minus the load power of the shed load, which is expressed as:

[0065]

[0066] Wherein, P Gi,t is the output power of the i-th power source at time t, P Dk,t is the charge and discharge power of the k-th energy storage at time t, negative for charging and positive for discharging, P Lj,t is the load power of the j-th load node at time t, P LOSSj,t is the load power of the shed load of the j-th load node at time t; n represents the total number of power sources in the microgrid,, M represents the total number of energy storages in the microgrid, and m represents the total number of load nodes in the microgrid;

[0067] 3) Ramp constraint of the power sources in the microgrid

[0068] The output power of the conventional power source in the microgrid needs to increase gradually within a certain period of time and cannot reach the maximum value suddenly; at startup, to protect the engine and the system, a reasonable acceleration time and a ramp-up stage are required to gradually reach the maximum speed, which is expressed as:

[0069]

[0070] In the formula, λ Gi,t is the ramp rate of the i-th power source at time t, and P Gi,t , P Gi,t+1 are the output powers of the i-th power source at time t and t + 1, Δt is the time interval between time t and t + 1, and λ minGi,t , λ maxGi,t are the upper and lower limits of the ramp rate of the i-th power source at time t;

[0071] 4) Energy storage constraints include energy storage power constraints, energy storage state of charge constraints, and energy storage capacity constraints;

[0072] The energy storage power constraint means that the energy storage charge and discharge power are within the maximum charge and discharge power range of the energy storage, which is expressed as:

[0073]

[0074] In the formula, P Ck,max is the maximum charging power of the k-th energy storage, and P Dk,max is the maximum discharging power of the k-th energy storage; P Ck,t , P Dk,t represent the charging power and discharging power of the k-th energy storage at time t;

[0075] The energy storage state of charge constraint can prevent overcharging and over-discharging of the energy storage, that is, the energy storage state of charge is within the allowable range, which is expressed as:

[0076]

[0077] In the formula, SOC max , SOC min are the maximum and minimum state of charge of the energy storage, SOC 0 is the initial state of charge of the energy storage system, and SOC k,t represents the state of charge of the k-th energy storage at time t; E k is the capacity of the k-th energy storage;

[0078] The energy storage capacity constraint ensures that the energy storage capacity at all times meets the requirements, which is expressed as:

[0079]

[0080] In the formula, E Max , E Minare the upper and lower limits of the energy storage capacity, Δt is the time interval, and η 1 , η 2 are the discharging and charging efficiencies of the energy storage respectively.

[0081] Step (4) is specifically as follows:

[0082] (41) Divide the off-grid operation time of the microgrid into several moments of equal time length;

[0083] (42) Apply the RBF neural network prediction model to predict the new energy output value and load value in the microgrid at the first moment;

[0084] (43) Based on the prediction results in (42), combined with the adjustable maximum output value of the conventional power source in the microgrid and the maximum discharge of the energy storage, judge whether the total output of all resources at this moment meets the load demand. If yes, go to (44); otherwise, first consider cutting off the interruptible load and unimportant load based on the load priority and load type in the microgrid, output the specific load shedding plan at this moment, that is, the flexible resource coordinated scheduling plan in the microgrid, and go to (46);

[0085] (44) Judge whether the total new energy output meets the total load. If yes, the energy storage in the microgrid is not used as a power source, the energy storage is in the charging mode at this moment, the energy storage is not used as a power source, solve the flexible resource coordinated optimization mathematical model of the microgrid at this time, obtain the specific output of each power source, output the flexible resource coordinated scheduling plan in the microgrid, and go to (46); otherwise, go to (45);

[0086] (45) The energy storage is used as a power source and participates in the optimal scheduling. The energy storage is in the discharging moment at this time. Solve the flexible resource coordinated optimization mathematical model of the microgrid at this time, obtain the specific output of each power source and the energy storage, and output the flexible resource coordinated scheduling plan in the microgrid;

[0087] (46) Select the next moment, judge whether it is the last moment. If yes, go to (47); otherwise, go to (42);

[0088] (47) Output the overall optimal scheduling plan during the off-grid operation time of the microgrid.

[0089] Solve the flexible resource coordinated optimization mathematical model of the microgrid by improving the particle swarm optimization algorithm, specifically as follows:

[0090] (61) Particle initialization

[0091] The dimension of the particles in the particle swarm is determined by the number of power generation units. Two types of particles, namely state optimization particles and power optimization particles, are selected according to the microgrid system structure. Particle initialization is the initialization of the particle swarm and control parameters, including the size, initial position, and initial velocity of the particle swarm. Usually, it is defaulted that the particles are evenly distributed in space, and the initial velocity can be set to 0 or a random number. The initialization formula is:

[0092]

[0093] In the formula, x i (0) and v i (0) represent the initial position and initial velocity of the particle. are the minimum and maximum position values on the dimension where the assumed optimal solution is located. are the minimum and maximum velocity values on the dimension where the assumed optimal solution is located. r j represents a random number in the range of 0 to 1.

[0094] (62) Set the historical optimal position as the current position, and the optimal individual as the current global optimal.

[0095] (63) Fitness value

[0096] Determine the fitness function according to the objective function and calculate the fitness value of each particle:

[0097] G = 1 - F(20)

[0098] In the formula, G represents the fitness function and F is the objective function.

[0099] (54) Judge whether the particle is an optimal particle and mutate the particle:

[0100] For the position of ordinary particles, a random mutation strategy is adopted. The particle swarm will forget its own historical best position and only remember the best position of the population. At the same time, when the particle velocity of ordinary particles becomes zero, its velocity is mutated to obtain a new velocity:

[0101]

[0102] In the formula, represent the position and velocity at the (t + 1)-th moment of the algorithm operation after the mutation of ordinary particles. R 1 and R 2 represent random numbers in the range of [0, 1]. μ represents the mutation direction control parameter. x i (t) and v i (t) are the position and velocity at the t-th moment of the algorithm operation before the particle mutation. Cl is a constant in the range of [0, 1].

[0103] For the optimal particle, its mutation strategy during the movement of the optimal particle enables the optimal particle to continue searching for the optimal value within its nearby area, enhancing the convergence speed and convergence accuracy of the particle swarm algorithm. It is expressed by the formula:

[0104]

[0105] In the formula, represents the position and velocity at the (t + 1)-th moment of the algorithm operation after the optimal particle mutates, and R 3 、R 4 represent random numbers within the range of [0, 1], p gd (t) represents the global optimal particle at the t-th moment of the particle swarm algorithm operation, ξ represents the proportionality factor determining the size of the search area of the optimal particle, and the update formula is:

[0106]

[0107] In the formula, ξ i (t + 1) represents the proportionality factor determining the size of the search area of the optimal particle at the (t + 1)-th moment of the algorithm operation, S represents the number of consecutive successful finds of new optimal positions, F represents the number of consecutive unsuccessful finds of new optimal positions, S C 、F C represent the set values of the number of consecutive successful finds of new optimal positions and the number of consecutive unsuccessful finds of new optimal positions, which can be specifically determined according to the actual situation;

[0108] (65) Particle update

[0109] When the particle swarm algorithm performs particle update, the position and velocity of the i-th particle at the (t + 1)-th moment of the algorithm operation are updated as:

[0110]

[0111] In the formula, x i (t + 1), v i (t + 1) represent the position and velocity of the i-th particle at the (t + 1)-th moment of the algorithm operation, the position and velocity of the i-th particle at the t-th moment of the algorithm operation after mutation, ω represents the inertia weight, p id represents the individual known optimal solution, p gd represents the population known optimal solution, c 1 、c 2 represent the learning factors, i.e., acceleration constants, r 1 、r 2 represent random numbers within the range of 0 to 1;

[0112] (66) Update individual optimal and global optimal

[0113] Update the individual optimal and global optimal of the entire particle swarm according to the particle update after mutation.

[0114] (67) Determine whether the iteration times of the particle swarm reach the maximum iteration times. If so, output the final result; otherwise, go to step (63).

[0115] Compared with the prior art, an adaptive coordinated control optimization method for flexible resources of a microgrid operating off-grid during a distribution network fault provided by the present invention has the following characteristics:

[0116] Regarding the problem of adaptive coordinated optimization of flexible resources of a microgrid operating off-grid during a distribution network fault, according to historical power output data and load demand data, the output power and load demand during the predicted fault recovery time, i.e., the off-grid operation time of the microgrid, are predicted through an RBF neural network prediction model. Furthermore, an objective function for optimal operation and a coordinated optimization model for flexible resources of the microgrid are established. Based on an improved particle swarm optimization algorithm, the optimal value is solved and combined with the coordinated optimization strategy for flexible resources of the microgrid to obtain the final coordinated optimization scheme for flexible resources of the microgrid. While giving full play to the potential of each flexible resource in the microgrid, the operation and power output of various power generation equipment, energy storage, etc. are reasonably configured, the operation cost of the microgrid is reduced, the economy of the microgrid is improved, and the coordination scheme is adjusted in a timely manner according to load changes and output predictions to ensure the economic and stable operation of the microgrid. BRIEF DESCRIPTION OF THE DRAWINGS

[0117] Figure 1 It is a schematic diagram of the construction process of the RBF neural network prediction model of the present invention;

[0118] Figure 2 It is a schematic diagram of the usage process of an adaptive coordinated control optimization method for flexible resources of a microgrid operating off-grid during a distribution network fault of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0119] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0120] Such as Figure 1 , the adaptive coordinated scheduling optimization method for flexible resources of a microgrid operating off-grid during a distribution network fault of the present invention includes the following steps:

[0121] (1) Collect the historical operation data of flexible resources within the microgrid and the corresponding environmental parameters as sample data;

[0122] Historical data includes multiple typical days and data on the output of distributed power sources and load values within the microgrid for at least a continuous week. Environmental parameters include weather, temperature, wind speed, etc. corresponding to the operating data.

[0123] (2) Establish an RBF neural network prediction model based on the sample data. The construction steps are as Figure 2 , specifically:

[0124] (21) Take a set of sample data and perform clustering analysis on the data using the K-Means clustering method. The input layer of the corresponding RBF prediction model is environmental parameters such as weather, temperature, and wind speed in the sample data, and the output layer is the power output value or load value corresponding to the input parameters.

[0125] (22) Based on the data clustering results of the K-Means clustering method, determine the centers and widths of the hidden layer units of the RBF neural network prediction model.

[0126] Assume that there are K clustering results in the sample data. Let the nearest center of the sample in the kth clustering data group be T k (n); (n = 1, 2,...) that is, the center of the kth group of clustering data in the nth iteration, and its center satisfies:

[0127] C k (S) = min k ||S i -T k (n)|| (k = 1, 2,...) (1)

[0128] In the formula, C k (S) is the minimum distance from the sample of the kth group of clustering data to the center, and S i is the ith sample point of the kth group of clustering data;

[0129] Judge whether the center converges. If it converges, determine the center value; otherwise, continue to iterate until convergence. The specific expression of the iteration process is:

[0130] T k (n + 1) = T k (n) + α(S i -T k (n)) (0 < α < 1) (2)

[0131] In the formula, T k (n + 1) is the adjusted center, T k (n) is the center before adjustment, and α is the adjustment coefficient;

[0132] After determining the center value, determine the width according to the Gaussian function, specifically:

[0133]

[0134] where c k is the final center of the k-th group of clustering data, and σ k is the width of the k-th group of clustering data;

[0135] The hidden layer units of the RBF neural network prediction model corresponding to the k-th group of clustering data can be expressed as:

[0136]

[0137] where is the function of the hidden unit corresponding to the k-th group of clustering data, which is a non-negative nonlinear function that radially symmetrically decays with respect to the center point. ||x|| is the Euclidean norm, and X is the input vector;

[0138] (23) Take another set of sample data and initialize it, input it into the determined hidden layer function, and use the least squares method to train the weights of the output layer to obtain a more accurate RBF prediction model. The weights are expressed as:

[0139]

[0140] where ω is the weight from the hidden layer to the output layer, h is the number of nodes in the hidden layer, and c max is the maximum distance between the selected centers; x p is the p-th input sample, and c i is the clustering center of the training samples;

[0141] (24) Judge whether the weight error is the smallest. If so, determine the mathematical function in the final RBF neural network prediction model; otherwise, go to step (33) to iterate the weights.

[0142] (25) Update the weights using the stochastic gradient descent method, and calculate the gradient Δl old of the loss function with respect to the weight ω i (ω old ), where η is the learning rate. The updated weight is expressed as:

[0143] ω new = ω old - ηΔl i (ω old ).

[0144] The weight error, and the specific calculation method is:

[0145] Suppose a set of sample data has Q samples. The input of the sample data is x i , and the corresponding output is y i . The output obtained by inputting x i into the RBF neural network prediction model is The weight error e is expressed as:

[0146]

[0147] (3) Build a mathematical model for coordinated optimization of flexible resources in the microgrid;

[0148] (31) The objective function for building a mathematical model for coordinated optimization of flexible resources in the microgrid with the optimal operation of the microgrid system is:

[0149]

[0150] In the formula, COE i,t is the power generation cost of the i-th power source at time t, PC t is the penalty cost for wind and light curtailment at time t, CR t is the compensation cost of the interruptible load at time t, that is, the compensation cost of the load cut due to insufficient power; N is the total number of power sources participating in coordinated dispatching at time t;

[0151] The penalty cost for wind and light curtailment is expressed as:

[0152] PC t =λ a ·(p wt-a +p pv-a ) (7)

[0153] In the formula, λ a is the penalty factor for unit wind and light curtailment, p wt-a is the wind curtailment volume, p pv-a is the light curtailment volume;

[0154] The compensation cost of the interruptible load is expressed as:

[0155] CR t =λ b ·p load ·h (8)

[0156] In the formula, λ b is the compensation amount for unit interrupted load, p load is the interrupted load, and h is the interruption time;

[0157] The power generation cost is expressed as:

[0158] COE i,t =CF i,t +IV i,t +OM i,t (9)

[0159] In the formula, CF i,t is the fuel cost of the i-th power source at time t, IV i,t is the depreciation cost of the i-th power source at time t, OM i,tThe maintenance cost of the i-th power source at time t;

[0160] The fuel cost, expressed as:

[0161]

[0162] In the formula, P i,t is the power generation power of the i-th power source at time t, η is the power generation efficiency of the power source, and C is the power generation fuel cost per unit power.

[0163] The depreciation cost is expressed as:

[0164]

[0165] In the formula, C INS,i is the installation cost of the i-th power source, P r,i is the rated power of the i-th power source, f r,i is the capacity factor of the i-th power source, d is the depreciation rate, and m is the service life of the power source;

[0166] The maintenance cost is expressed as:

[0167] OM i,t = k m,i × p i,t (12)

[0168] In the formula, k m,i is the unit operation and maintenance cost of the i-th power source.

[0169] (32) Determine the constraint conditions of the coordinated optimization mathematical model of the microgrid flexible resources, including the output constraint of the microgrid, the power balance constraint of the microgrid, the power ramp constraint of the power source, and the energy storage constraint;

[0170] 1) Output constraint of the microgrid

[0171] Corresponding to the operation constraints of the power sources in the microgrid, that is, the output power of each power source in the microgrid is within the upper and lower limits of the power of each power source, expressed as:

[0172] P i,min ≤ P i,t ≤ P i,max (13)

[0173] In the formula, P i,t is the output power of the i-th power source at time t, P i,min 、P i,max are the upper and lower limits of the output power of the i-th power source;

[0174] 2) Power balance constraint of the microgrid

[0175] At every moment, the output power within the microgrid should be equal to the load power within the grid. Among them, when the energy storage is charging, it acts as the load power, and when discharging, it acts as the output power. When the microgrid output cannot meet all the loads, the load power within the grid is the total load power within the grid minus the load power of the shed load, which is expressed as:

[0176]

[0177] In the formula, P Gi,t is the output power of the i-th power source at time t, P Dk,t is the charge and discharge power of the k-th energy storage at time t, negative for charging and positive for discharging, P Lj,t is the load power of the j-th load node at time t, P LOSSj,t is the load power of the shed load of the j-th load node at time t; n represents the total number of power sources within the microgrid, M represents the total number of energy storages within the microgrid, and m represents the total number of load nodes within the microgrid;

[0178] 3) Ramp constraint of power sources within the microgrid

[0179] The output power of conventional power sources within the microgrid needs to increase gradually within a certain time and cannot suddenly reach the maximum value; at startup, to protect the engine and the system, a reasonable acceleration time and ramp stage are required to gradually reach the maximum speed, which is expressed as:

[0180]

[0181] In the formula, λ Gi,t is the ramp rate of the i-th power source at time t, P Gi,t 、P Gi,t+1 are the output powers of the i-th power source at time t and t + 1, Δt is the time interval between time t and t + 1, λ minGi,t 、λ maxGi,t are the upper and lower limits of the ramp rate of the i-th power source at time t;

[0182] 4) Energy storage constraints include energy storage power constraint, energy storage state of charge constraint, and energy storage capacity constraint;

[0183] The energy storage power constraint means that the charge and discharge power of the energy storage is within the maximum charge and discharge power range of the energy storage, which is expressed as:

[0184]

[0185] In the formula, P Ck,max is the maximum charging power of the k-th energy storage, P Dk,max is the maximum discharging power of the k-th energy storage; P Ck,t 、P Dk,t represent the charging power and discharging power of the k-th energy storage at time t;

[0186] The energy storage state of charge constraint can prevent overcharging and over-discharging of the energy storage, that is, the state of charge of the energy storage is within the allowed range, which is expressed as:

[0187]

[0188] In the formula, SOC max and SOC min are the maximum and minimum state of charge of the energy storage, SOC 0 is the initial state of charge of the energy storage system, and SOC k,t represents the state of charge of the k-th energy storage at time t; E k is the capacity of the k-th energy storage;

[0189] The energy storage capacity constraint ensures that the energy storage capacity at all times meets the requirements, which is expressed as:

[0190]

[0191] In the formula, E Max and E Min are the upper and lower limits of the energy storage capacity respectively, Δt is the time interval, η 1 and η 2 are the discharging and charging efficiencies of the energy storage respectively.

[0192] (4) Propose a coordinated optimization strategy for flexible resources in the microgrid based on the RBF neural network prediction model and the coordinated optimization mathematical model of flexible resources in the microgrid, and determine the final flexible resource scheduling plan. Specifically:

[0193] (41) Divide the off-grid operation time of the microgrid into several moments with equal time lengths;

[0194] (42) Apply the RBF neural network prediction model to predict the new energy output value and load value in the microgrid at the first moment;

[0195] (43) Based on the prediction results in (42), combined with the adjustable maximum output value of the conventional power source in the microgrid and the maximum discharge of the energy storage, judge whether the total output of all resources at this moment meets the load demand. If so, go to (44); otherwise, first consider cutting off the interruptible load and unimportant load based on the load priority and load type in the microgrid, output the specific load shedding plan at this moment, that is, the coordinated scheduling plan of flexible resources in the microgrid, and go to (46);

[0196] (44) Judge whether the total new energy output meets the total load. If so, the energy storage in the microgrid is not used as a power source, and the energy storage is in the charging mode at this moment. The energy storage is not used as a power source, solve the coordinated optimization mathematical model of flexible resources in the microgrid at this time, obtain the specific output of each power source, output the coordinated scheduling plan of flexible resources in the microgrid, and go to (46); otherwise, go to (45);

[0197] (45) With energy storage as the power source and participating in the optimal dispatch, at this moment, the energy storage is in the discharging state. Solve the mathematical model for the coordinated optimization of flexible resources in the microgrid at this time to obtain the specific outputs of each power source and the energy storage, and output the coordinated dispatch plan for flexible resources in the microgrid;

[0198] (46) Select the next moment, determine whether it is the last moment. If it is, go to (47); if not, go to (42);

[0199] (47) Output the overall optimal dispatch plan during the off-grid operation time of the microgrid.

[0200] Further, the present invention solves the mathematical model for the coordinated optimization of flexible resources in the microgrid by improving the particle swarm optimization algorithm. Specifically:

[0201] (61) Particle initialization

[0202] The dimension of the particles in the particle swarm is determined by the number of power generation units. According to the microgrid system structure, two types of particles, namely state optimization particles and power optimization particles, are selected correspondingly. Particle initialization is the initialization of the particle swarm and control parameters, including the size of the particle swarm, the initial position, and the initial velocity. Usually, it is default that the particles are evenly distributed in space, and the initial velocity can be set to 0 or a random number. The initialization formula:

[0203]

[0204] In the formula, x i (0), v i (0) represent the initial position and initial velocity of the particle, are the minimum and maximum position values on the dimension where the assumed optimal solution is located, are the minimum and maximum velocity values on the dimension where the assumed optimal solution is located, r j represents a random number in the range of 0 to 1;

[0205] (62) Set the historical optimal position as the current position, and the optimal individual as the current global optimum;

[0206] (63) Fitness value

[0207] Determine the fitness function according to the objective function and calculate the fitness value of each particle:

[0208] G = 1 - F(20)

[0209] In the formula, G represents the fitness function, and F is the objective function;

[0210] (54) Determine whether the particle is an optimal particle and mutate the particle:

[0211] For the positions of ordinary particles, a random mutation strategy is adopted. The particle swarm will forget its own historical best position and only remember the best position of the population. At the same time, when the velocity of an ordinary particle becomes zero, its velocity is mutated to obtain a new velocity:

[0212]

[0213] In the formula, represents the position and velocity at the (t + 1)-th moment of the algorithm operation after the mutation of an ordinary particle. R 1 and R 2 represent random numbers within the range [0, 1]. μ represents the mutation direction control parameter. x i (t) and v i (t) are the position and velocity at the t-th moment of the algorithm operation before the particle mutation. C l is a constant within the range [0, 1];

[0214] For the optimal particle, its mutation strategy during the movement of the optimal particle enables the optimal particle to continue searching for the optimal value within its nearby area, strengthening the convergence speed and convergence accuracy of the particle swarm algorithm. It is expressed by the formula:

[0215]

[0216] In the formula, represents the position and velocity at the (t + 1)-th moment of the algorithm operation after the mutation of the optimal particle. R 3 and R 4 represent random numbers within the range [0, 1]. p gd (t) represents the global optimal particle at the t-th moment of the particle swarm algorithm operation. ξ represents the proportionality factor determining the size of the search area of the optimal particle. The update formula is:

[0217]

[0218] In the formula, ξ i (t + 1) represents the proportionality factor determining the size of the search area of the optimal particle at the (t + 1)-th moment of the algorithm operation. S represents the number of consecutive successful discoveries of new optimal positions. F represents the number of consecutive unsuccessful discoveries of new optimal positions. S C and F C represent the set values of the number of consecutive discoveries of new optimal positions and the number of consecutive unsuccessful discoveries of new optimal positions, which can be specifically determined according to the actual situation;

[0219] (65) Particle update

[0220] When the particle swarm algorithm performs particle update, the position and velocity of the i-th particle at the (t + 1)-th moment of the algorithm operation are updated as:

[0221]

[0222] where x i (t + 1) and v i (t + 1) represent the position and velocity of the i-th particle at the (t + 1)-th moment during the algorithm operation, the position and velocity of the i-th particle at the t-th moment after the mutation of the algorithm, ω represents the inertia weight, p id represents the known optimal solution of the individual, p gd represents the known optimal solution of the population, c 1 and c 2 represent the learning factors, i.e., acceleration constants, r 1 and r 2 represent random numbers in the range of 0 to 1;

[0223] (66) Update the individual optimal and global optimal

[0224] Update the individual optimal and global optimal of the entire particle swarm according to the update of the mutated particles.

[0225] (67) Determine whether the iteration times of the particle swarm reach the maximum iteration times. If so, output the final result; otherwise, go to step (63).

[0226] Common termination conditions can be the maximum iteration times, an acceptable optimal solution, when the iteration exceeds a certain number of times and the algorithm result no longer improves, when the regularization community radius approaches 0, and when the slope of the objective function tends to 0. The present invention uses 40 iteration times as the termination condition. Determine whether the iteration times of the particle swarm reach the termination condition. If so, output the final result; otherwise, go to (63).

Claims

1. A method for optimizing the coordinated dispatch of flexible resources of microgrids with adaptive distribution network failure and off-grid operation, characterized in that: The following steps are involved: (1) Collect historical operation data of flexibility resources in the microgrid and corresponding environmental parameters as sample data; (2) Establish an RBF neural network prediction model based on sample data; (3) Build a mathematical model for the coordinated optimization of microgrid flexible resources; (4) A microgrid flexible resource coordination optimization strategy based on the RBF neural network prediction model and the microgrid flexible resource coordination optimization mathematical model is proposed to determine the final flexible resource scheduling plan.

2. The adaptive distribution network fault off-grid operation microgrid flexible resource coordinated dispatch optimization method according to claim 1 is characterized in that: In step (1), the historical data includes the output and load value data of distributed power sources in the microgrid network for multiple typical days and for at least one week, and the environmental parameters include the weather, temperature, wind speed, etc. under the corresponding operating data.

3. The adaptive distribution network fault off-grid operation microgrid flexible resource coordinated dispatch optimization method according to claim 1 is characterized in that: In step (2), an RBF neural network prediction model is established based on sample data, specifically: (21) Take a set of sample data and use K-Means clustering method to cluster the data. The corresponding RBF prediction model input layer is the environmental parameters such as weather, temperature, wind speed, etc. in the sample data, and the output layer is the power output value or load value under the corresponding input parameters; (22) Based on the K-Means clustering results, determine the center and width of the hidden layer units of the RBF neural network prediction model; Assume that there are K clustering results for the sample data, and let the nearest center of the sample of the kth cluster data group be T k (n); (n=1,2,…) is the center of the kth group of cluster data at the nth iteration, and its center satisfies: C k (S)=min k ||S i -T k (n)|| (k=1,2,…) (1) In the formula, C k (S) is the minimum distance between the k-th group of clustered data samples and the center, Si is the i-th sample point of the k-th group of clustered data; Determine whether the center converges. If so, determine the center value. Otherwise, continue iterating until convergence. The specific expression of the iterative process is: T k (n+1)=T k (n)+α(S i -T k (n)) (0<α<1) (2) Where, T k (n+1) is the adjusted center, T k (n) is the center before adjustment, α is the adjustment coefficient; After determining the center value, the width is determined according to the Gaussian function, specifically: In the formula, c k is the final center of the kth group of cluster data, σ k is the width of the kth group of cluster data; The hidden layer unit of the RBF neural network prediction model corresponding to the kth group of clustering data can be expressed as: In the formula, It is the function of the hidden unit corresponding to the k-th group of cluster data. It is a non-negative nonlinear function with radial symmetric decay about the center point. ||x|| is the Euclidean norm, and X is the input vector. The RBF neural network prediction model consists of k hidden layer units corresponding to k groups of clustered data; (23) Take another set of sample data and initialize it, input it into the determined hidden layer function, and use the least squares method to train the output layer weights to obtain a more accurate RBF prediction model. The weights are expressed as: In the formula, ω is the weight from the hidden layer to the output layer, h is the number of nodes in the hidden layer, c max is the maximum distance between the selected centers; x p is the pth input sample, c i is the cluster center of the training sample; (24) Determine whether the weight error is the minimum or observe the changing trend of the weight error. If the weight error is the minimum or fluctuates very little and tends to a very small threshold under a certain iteration, determine the specific weight of the hidden layer unit of the RBF neural network prediction model, otherwise go to step (25) to iterate the weight; (25) Use the stochastic gradient descent method to update the weights and calculate the loss function with respect to the weight ω before iteration. old The gradient Δl i (ω old ), η is the learning rate. The updated weight is expressed as: oh new =ω old -ηΔl i (oh old )。 4. The adaptive distribution network fault off-grid operation microgrid flexible resource coordinated dispatch optimization method according to claim 3 is characterized in that: Weight error, specifically calculated as: Suppose a set of sample data has a total of Q samples, and the input of the sample data is x i , the corresponding output is y i , x i The output obtained by inputting into the RBF neural network prediction model is The weight error e is expressed as:

5. The adaptive distribution network fault off-grid operation microgrid flexible resource coordinated dispatch optimization method according to claim 1 is characterized in that: In step (3), a mathematical model for the coordinated optimization of microgrid flexible resources is constructed, specifically: (31) The objective function of the mathematical model for the coordinated optimization of microgrid flexible resources is constructed based on the optimal operation of the microgrid system: In the formula, COE i,t is the power generation cost of the ith power source at time t, PC t is the penalty cost of wind and solar power abandonment at time t, CR t is the compensation cost of the interruptible load at time t, i.e., the compensation cost of the load cut due to insufficient power; N is the total number of power sources participating in the coordinated dispatch at time t; The penalty cost for wind and solar curtailment is expressed as: PC t =λ a ·(p wt-a +p pv-a ) (7) In the formula, λ a is the penalty factor for abandoned solar and wind volume, p wt-a is the abandoned air volume, p pv-a is the amount of abandoned light; The compensation cost of interruptible load is expressed as: CR t =λ b ·p load ·h (8) In the formula, λ b is the unit interruption load compensation, p load is the interruption load, h is the interruption time; The cost of electricity generation is expressed as: COE i,t =CF i,t +IV i,t +OM i,t (9) Where CF i,t is the fuel cost of the ith power source at time t, IV i,t is the depreciation cost of the i-th power source at time t, OM i,t is the maintenance cost of the i-th power source at time t; Fuel cost, expressed as: Where P i,t is the power generation of the i-th power source at time t, η is the power generation efficiency of the power source, and C is the power generation fuel cost per unit power. The depreciation cost is expressed as: In the formula, C INS,i is the installation cost of the ith power source, P r,i is the rated power of the ith power supply, f r,i is the capacity factor of the i-th power source, d is the depreciation rate, and m is the service life of the power source; The maintenance cost is expressed as: IF i,t =k m,i ×p i,t (12) In the formula, k m,i is the unit operation and maintenance cost of the ith power supply. (32) Determine the constraints of the microgrid flexible resource coordination optimization mathematical model, including the microgrid output constraint, microgrid power balance constraint, power source power climbing constraint, and energy storage constraint; 1) Output constraints of microgrids The corresponding power supply operation constraints in the microgrid are that the output power of each power supply in the microgrid is within the upper and lower limits of each power supply, which can be expressed as: P i,min ≤P i,t ≤P i,max (13) Where P i,t is the output power of the i-th power source at time t, P i,min , P i,max is the upper and lower limits of the output power of the i-th power supply; 2) Microgrid power balance constraints The output power of the microgrid should be equal to the load power in the grid at all times. The energy storage acts as load power when charging and acts as output power when discharging. When the output of the microgrid does not meet all loads, the load power in the grid is the load power of all loads in the grid minus the load power of the removed load, expressed as: Where P Gi,t is the output power of the i-th power source at time t, P Dk,t is the charge and discharge power of the kth energy storage at time t, charging is negative and discharging is positive, P Lj,t is the load power of the jth load node at time t, P LOSSj,t is the load power of the jth load node when the load is removed at time t; n represents the total number of power sources in the microgrid, M represents the total number of energy storage in the microgrid, and m represents the total number of load nodes in the microgrid; 3) Ramp constraints of power sources in microgrids The output power of conventional power supply in the microgrid needs to increase gradually within a certain period of time and cannot reach the maximum value suddenly. At the start, in order to protect the engine and system, a reasonable acceleration time and climbing stage are required to gradually reach the maximum speed, which can be expressed as: In the formula, λ Gi,t is the ramp rate of the ith power source at time t, P Gi,t , P Gi,t+1 is the output power of the i-th power source at time t and time t+1, Δt is the time interval between time t and time t+1, and λ minGi,t , maxGi,t are the upper and lower limits of the ramp rate of the i-th power source at time t; 4) Energy storage constraints include energy storage power constraints, energy storage charge rate constraints and energy storage capacity constraints; The energy storage power constraint means that the energy storage charging and discharging power is within the maximum charging and discharging power range of the energy storage, which is expressed as: Where P Ck,max is the maximum charging power of the kth energy storage, P Dk,max is the maximum discharge power of the kth energy storage; P Ck,t , P Dk,t represents the charging power and discharging power of the kth energy storage at time t; Energy storage charge constraint can avoid overcharging and over-discharging of energy storage, that is, the energy storage charge rate is within the allowed range, expressed as: Where SOC max , SOC min is the maximum and minimum charge rate of energy storage, SOC0 is the initial charge rate of energy storage system, SOC k,t represents the charge rate at the kth energy storage time t; E k is the kth energy storage capacity; Energy storage capacity constraints ensure that the energy storage capacity meets the requirements at all times, expressed as: In the formula, E Max , E Min are the upper and lower limits of the energy storage capacity, Δt is the time interval, η1 and η2 are the efficiencies of energy storage discharge and charging, respectively.

6. The adaptive distribution network fault off-grid operation microgrid flexible resource coordinated dispatch optimization method according to claim 1 is characterized in that: Step (4) is specifically: (41) Divide the off-grid operation time of the microgrid into several moments of equal time length; (42) The RBF neural network prediction model is used to predict the output value and load value of new energy in the microgrid at the first moment; (43) Based on the prediction result of (42), combined with the adjustable maximum output value of the conventional power source in the microgrid and the maximum discharge of the energy storage, determine whether the total output of all resources at this moment meets the load demand. If so, go to (44). Otherwise, based on the load priority and load type in the microgrid, first consider cutting off the interruptible load and the unimportant load, output the specific load cutting plan at this moment, that is, the flexible resource coordination scheduling plan in the microgrid, and go to (46); (44) Determine whether the total output of new energy sources meets the total load. If yes, the energy storage in the microgrid does not serve as a power source. At this moment, the energy storage is in charging mode and does not serve as a power source. Solve the mathematical model of the coordinated optimization of flexible resources in the microgrid at this time, obtain the specific output of each power source, output the coordinated dispatching plan of flexible resources in the microgrid, and go to (46). Otherwise, go to (45); (45) Energy storage is used as a power source and participates in the optimization scheduling. At this moment, the energy storage is in the discharging moment. The mathematical model of the coordinated optimization of the flexible resources of the microgrid is solved to obtain the specific output of each power source and energy storage, and output the coordinated scheduling plan of the flexible resources in the microgrid; (46) Select the next moment and determine whether it is the last moment. If yes, go to (47); otherwise, go to (42); (47) Output the overall optimal dispatching plan during the off-grid operation time of the microgrid.

7. The adaptive distribution network fault off-grid operation microgrid flexible resource coordinated dispatch optimization method according to claim 6 is characterized in that: The mathematical model of microgrid flexible resource coordination optimization is solved by improving the particle swarm algorithm, specifically: (61) Particle initialization The dimension of particles in the particle swarm is determined by the number of power generation units. According to the microgrid system structure, two types of particles, state optimization and power optimization, are selected. Particle initialization is the initialization of the particle swarm and control parameters, including the size, initial position, and initial velocity of the particle swarm. Usually, the particles are uniformly distributed in the space by default, and the initial velocity can be set to 0 or a random number. The initialization formula is: In the formula, x i (0), v i (0) represents the initial position and initial velocity of the particle, is the minimum and maximum position value on the dimension where the optimal solution is assumed to be located, is the minimum and maximum speed value in the dimension where the optimal solution is assumed to be located, r j Represents a random number in the range of 0 to 1; (62) Set the historical optimal position as the current position, and the optimal individual as the current global optimal; (63) Fitness value Determine the fitness function according to the objective function and calculate the fitness value of each particle: G=1-F (20) In the formula, G represents the fitness function and F is the objective function; (54) Determine whether the particle is the optimal particle and mutate the particle: A random mutation strategy is used for the positions of ordinary particles. The particle swarm will forget its own best historical position and only remember the best position of the population. At the same time, when the particle speed of an ordinary particle becomes zero, its speed will be mutated to obtain a new speed: In the formula, represents the position and speed of the ordinary particle after the algorithm is mutated at time t+1, R1 and R2 represent random numbers in the interval [0,1], μ represents the mutation direction control parameter, x i (t), v i (t) is the position and speed of the particle at time t before the algorithm is run, C l is a constant in the interval [0,1]; For the optimal particle, its mutation strategy makes the optimal particle continue to search for the optimal value in its vicinity during its movement process, so that the convergence speed and convergence accuracy of the particle swarm algorithm are enhanced, which can be expressed as follows: In the formula, represents the position and speed of the optimal particle after the algorithm is mutated at time t+1, R3 and R4 represent random numbers in the interval [0,1], and p gd (t) represents the global optimal particle at the tth moment of the particle swarm algorithm operation, ξ represents the proportional factor that determines the size of the optimal particle search field, and the update formula is: In the formula, ξ i (t+1) represents the proportional factor that determines the size of the optimal particle search area at the t+1 moment of the algorithm. S represents the number of consecutive successful attempts to find a new optimal position. F represents the number of consecutive unsuccessful attempts to find a new optimal position. S C 、F C The setting value representing the number of times of finding a new optimal position continuously and the number of times of failing to find a new optimal position continuously can be specifically formulated according to the actual situation; (65) Particle Update When the particle swarm algorithm updates particles, the position and speed of the i-th particle at the time t+1 of the algorithm operation are updated as follows: In the formula, x i (t+1), v i (t+1) represents the position and velocity of the ith particle at time t+1 when the algorithm runs. The position and velocity of the mutated particle at the tth moment of the algorithm operation. ω represents the inertia weight, p id represents the individual's known optimal solution, p gd represents the known optimal solution of the population, c1 and c2 represent learning factors or acceleration constants, and r1 and r2 represent random numbers in the range of 0 to 1; (66) Update individual optimum and global optimum According to the particle update after particle mutation, update the individual optimum and global optimum of the entire particle swarm; (67) Determine whether the number of iterations of the particle swarm has reached the maximum number of iterations. If so, output the final result; otherwise, go to step (63).

Images