An energy internet optimization reconstruction method and system based on a graph convolution network

Abstract

The application discloses an energy internet optimization reconstruction method and system based on a graph convolution network, first constructs an energy internet optimization model under a large power shortage based on an objective function of minimizing an important load outage rate, then constructs an economic cost evaluation network based on a graph convolution neural network algorithm, taking the load outage rate as an economic evaluation index, and on the basis, constructs a new energy internet optimization reconstruction model considering the load outage rate index constraint, adopts a water circulation algorithm to solve the constructed energy internet optimization reconstruction model, and obtains an optimal network topology, so that active support and emergency frequency control of the energy internet are realized, and the safety and economy of the energy internet system operation are ensured.

Description

Technical Field

[0001] This invention relates to the field of energy internet optimization and reconstruction and emergency frequency regulation technology, and in particular to an energy internet optimization and reconstruction method and system based on graph convolutional networks. Background Technology

[0002] Distributed renewable energy generation has been integrated into the grid at a high rate, causing traditional distribution networks to gradually evolve into new forms of "active" distribution networks. The continuous integration of large-scale distributed renewable energy into the distribution network has led to a decline in the capacity of traditional generating units in the main power grid, resulting in a shortage of grid regulation resources and reserve capacity, and insufficient support for the stable operation of the distribution system. Simultaneously, the coupling and interdependence of various energy sources, such as electricity, heat, cooling, and gas, are deepening in all aspects of energy production, transmission, conversion, consumption, and storage, forming an energy internet with a new type of power system as its core. As the coupling and interconnection of various energy supply networks, such as power distribution, gas supply, and heating, deepens within the energy internet, the security threats faced by the distribution network are becoming increasingly diversified. To ensure uninterrupted power supply to critical users in the energy internet, a network optimization and reconfiguration method can be adopted. This method improves power flow by changing the state of switches on feeders in the network. Therefore, when the controllable resource regulation capacity is insufficient, emergency frequency control of the energy internet can be achieved based on network topology reconfiguration.

[0003] Traditional heuristic algorithms, when solving large-scale energy internet reconfiguration problems, require numerous iterations within the search space to find the optimal solution, making them prone to getting trapped in local optima and time-consuming. Therefore, traditional methods have limitations when dealing with urgent frequency control problems. Summary of the Invention

[0004] Purpose of the invention: The purpose of this invention is to provide an energy internet optimization and reconfiguration method and system based on graph convolutional networks, which improves the computational speed and global convergence capability of reconfiguration optimization, and can achieve rapid active power support while ensuring optimal economic efficiency when dealing with the requirements of short time scales for large power deficits.

[0005] Technical solution: The energy internet optimization and reconstruction method based on graph convolutional networks provided by this invention includes the following steps:

[0006] S1: Based on the objective function of minimizing the power failure rate of critical loads, construct an energy internet optimization model under high power deficit conditions;

[0007] S2: Construct an economic cost assessment network based on graph convolutional neural network algorithm, using the power failure rate of critical loads as an economic evaluation index;

[0008] S3: Based on the energy internet optimization model and economic cost assessment network, construct an energy internet optimization and reconfiguration model that considers the constraint of the power failure rate of important loads;

[0009] S4: The optimal solution is obtained by using the water cycle algorithm to solve the energy internet optimization and reconstruction model, thus completing the network optimization and reconstruction of the energy internet.

[0010] Furthermore, in step S1, the energy internet optimization model formula is as follows:

[0011]

[0012] Where F is the power failure rate of critical loads, ε is the controllable proportion of critical loads, and ω i εω represents the proportion of significant load reduction at node i. i P Li This represents the amount of significant load that can be reduced at node i based on ε.

[0013] Furthermore, in step S1, the energy internet optimization model has the following constraints:

[0014] 1.1) Power balance constraints:

[0015]

[0016]

[0017]

[0018] Among them, P DGi Let Q be the active power output of the distributed power source at node i. DGi P represents the reactive power output of the distributed power source at node i; Li Q represents the active power consumed by the load at node i. Li G represents the reactive power consumed by the load at node i; ij For the conductance of branch ij, B ij For the susceptance of branch ij, θ ij Let be the phase angle difference between the voltages of node i and node j, and m be the number of branches connected to the node.

[0019] 1.2) Network radiation constraint:

[0020]

[0021] Among them, l ij Let N be the radiation of branch ij, N be the number of network nodes, and E be the set of network branches;

[0022] 1.3) Voltage upper and lower limit constraints:

[0023] U i,min ≤U i ≤U i,max

[0024] Among them, U i Let U be the voltage at node i. i,min U is the lower limit of the voltage at node i. i,max Let i be the upper bound of node i;

[0025] 1.4) Branch capacity constraints:

[0026] S ijmin ≤S ij ≤S ijmax

[0027] Among them, S ij S is the capacity of branch ij. ijmin S is the lower limit of the capacity of branch ij. ijmax This represents the upper limit of the capacity of branch ij;

[0028] 1.5) DG power constraint:

[0029]

[0030] Among them, P DGmin P is the lower limit of the active power output of the distributed power source at node i. DGmax Q represents the upper limit of the active power output by the distributed power source at node i. DGmin Q is the lower limit of the reactive power output of the distributed power source at node i. DGmax This represents the upper limit of the reactive power output of the distributed power source at node i.

[0031] Furthermore, step S2 includes the following steps:

[0032] 2.1) Taking the critical load power failure rate as an economic evaluation indicator, defined as LPSP, then:

[0033]

[0034] in:

[0035] F(t) = Load(t) - E Max (t)

[0036] E Max (t)=E PV (t)+E WT (t)+η(E BESS (t)-E BESS,min )

[0037] In the formula, F(t) is the minimum total power consumption at which the load is cut off; Load(t) is the total power consumption of the load; EMax (t) represents the maximum power generation of the distributed power source; E PV (t) represents photovoltaic power generation; E WT (t) represents the wind turbine's power generation; E BESS (t) represents the energy state of the electric vehicle battery; E BESS,min η represents the minimum battery capacity; η is the charge / discharge efficiency.

[0038] 2.2) Based on graph convolutional neural networks, establish the mapping relationship between network topology and load reduction, and obtain the important load loss rate indicators of each network topology.

[0039] Furthermore, in step S2, the training process of the graph convolutional neural network includes the following steps:

[0040] 2.2.1) Normalization is used to map each component of the node feature to the interval [0,1]. Assuming the node feature vector is x, then:

[0041]

[0042] Where, x j For the feature components of the i-th node before normalization, x represents the feature component of the i-th node after normalization, j∈[1,M]; min x is the minimum value among all components of the node feature. max The maximum value among all components of the node feature;

[0043] The processed data samples are randomly divided into training set, validation set and test set according to the proportion. The training set and validation set are used to determine the various hyperparameters and weights of GCN, and the test set is used to evaluate the performance of the trained GCN model.

[0044] 2.2.2) Initialize the processed data, including initializing weights and biases;

[0045] 2.2.3) Train the GCN model, calculate the loss function between the predicted value and the true value through forward propagation, and update the weights and biases of the GCN using the backpropagation algorithm. After multiple iterations, if the termination condition is met, save the GCN model.

[0046] 2.2.4) Input the test set data into the saved GCN model to obtain the important load power failure rate indicators for each network topology.

[0047] Furthermore, in step S3, based on the energy internet optimization model formula and corresponding constraints, the following steps are set: To determine the maximum permissible load reduction, an energy internet optimization and reconfiguration model is constructed.

[0048] Furthermore, the water cycle algorithm in step S4 includes the following steps:

[0049] 4.1) The rainfall process generates an initial population, which is then initialized and classified as stream, river, or ocean.

[0050] 4.2) Evaluate the fitness of the initialized population based on the cost function;

[0051] 4.3) Determine the intensity of raindrops flowing towards rivers and oceans during the rainfall process, and update the locations of streams and rivers;

[0052] 4.4) Determine if the evaporation conditions are met; if they are, new rainfall will occur.

[0053] 4.5) Determine if the algorithm meets the termination condition. If it does, output the result as the optimal solution; otherwise, repeat steps 4.3) to 4.4).

[0054] This invention provides an energy internet optimization and reconstruction system based on graph convolutional networks, comprising a module for constructing an energy internet optimization model, a module for constructing an economic cost assessment network, a module for constructing an energy internet optimization and reconstruction model, and a water cycle algorithm module.

[0055] The module for constructing an energy internet optimization model is used to build an energy internet optimization model under high power deficit based on the objective function of minimizing the power failure rate of important loads.

[0056] An economic cost assessment network module is constructed to build an economic cost assessment network based on the graph convolutional neural network algorithm, using the power failure rate of critical loads as an economic evaluation index.

[0057] The module for constructing an energy internet optimization and reconfiguration model is used to construct an energy internet optimization and reconfiguration model that considers the constraints of the power failure rate of important loads based on the energy internet optimization model and the economic cost assessment network.

[0058] The water cycle algorithm module is used to solve the energy internet optimization and reconstruction model using the water cycle algorithm to obtain the optimal solution, thereby completing the network optimization and reconstruction of the energy internet.

[0059] Beneficial effects: Compared with the prior art, the significant feature of this invention is that it adopts an energy internet optimization and reconstruction method based on graph convolutional networks. Based on the constraints of economic indicators, it effectively reduces the number of candidate topologies, realizes the calculation of power flow of all selected topologies in a short time, and obtains the optimal solution. While completing the energy internet network optimization and reconstruction, it improves the calculation speed and global convergence capability of reconstruction optimization. When dealing with the requirements of short time scales for large power deficits, it can achieve rapid active power support while ensuring optimal economy, thus ensuring the safe and economical operation of the system. Attached Figure Description

[0060] Figure 1 This is a flowchart illustrating the present invention. Detailed Implementation

[0061] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0062] Example 1

[0063] This invention provides an energy internet optimization and reconstruction method based on graph convolutional networks. Please refer to [link to relevant documentation]. Figure 1 As shown, it includes the following steps:

[0064] S1: Based on the objective function of minimizing the power failure rate of critical loads, construct an energy internet optimization model under high power deficit conditions.

[0065] The formula for the energy internet optimization model is as follows:

[0066]

[0067] Where F is the power failure rate of critical loads, ε is the controllable proportion of critical loads, and ω i εω represents the proportion of significant load reduction at node i. i P Li This represents the amount of significant load that can be reduced at node i based on ε.

[0068] The energy internet optimization model has the following constraints:

[0069] 1.1) Power balance constraints:

[0070]

[0071]

[0072]

[0073] Among them, P DGi Let Q be the active power output of the distributed power source at node i. DGi P represents the reactive power output of the distributed power source at node i; Li Q represents the active power consumed by the load at node i. Li G represents the reactive power consumed by the load at node i; ij For the conductance of branch ij, B ij For the susceptance of branch ij, θ ij Let be the phase angle difference between the voltages of node i and node j, and m be the number of branches connected to the node.

[0074] 1.2) Network radiation constraint:

[0075]

[0076] Among them, l ij Let N be the radiation of branch ij, N be the number of network nodes, and E be the set of network branches;

[0077] 1.3) Voltage upper and lower limit constraints:

[0078]

[0079] Among them, U i Let U be the voltage at node i. i,min U is the lower limit of the voltage at node i. i,max Let i be the upper bound of node i;

[0080] 1.4) Branch capacity constraints:

[0081] S ijmin ≤S ij ≤S ijmax (7)

[0082] Among them, S ij S is the capacity of branch ij. ijmin S is the lower limit of the capacity of branch ij. ijmax This represents the upper limit of the capacity of branch ij;

[0083] 1.5) DG power constraint:

[0084]

[0085] Among them, P DGmin P is the lower limit of the active power output of the distributed power source at node i. DGmax Q represents the upper limit of the active power output by the distributed power source at node i. DGmin Q is the lower limit of the reactive power output of the distributed power source at node i. DGmax This represents the upper limit of the reactive power output of the distributed power source at node i.

[0086] By minimizing the power failure rate of critical loads as the objective function, and by proposing general inequality constraints that consider power balance, voltage, branch capacity, and DG output power, the traditional energy internet optimization and reconfiguration problem can be solved, enabling active power support under power deficit conditions.

[0087] S2: Construct an economic cost assessment network based on graph convolutional neural network algorithm, using the power failure rate of critical loads as an economic evaluation index.

[0088] 2.1) To quantify the economics of the energy internet, the critical load power failure rate is used as an economic evaluation index, defined as LPSP. The smaller the LPSP value, the better the economics of the energy internet system. Therefore:

[0089]

[0090] in:

[0091] F(t) = Load(t) - E Max (t) (10)

[0092] E Max (t)=E PV (t)+E WT (t)+η(E BESS (t)-E BESS,min (11)

[0093] In the formula, F(t) is the minimum total power consumption at which the load is cut off; Load(t) is the total power consumption of the load; E Max (t) represents the maximum power generation of the distributed power source; E PV (t) represents photovoltaic power generation; E WT (t) represents the wind turbine's power generation; E BESS (t) represents the energy state of the electric vehicle battery; E BESS,min η represents the minimum battery capacity; η represents the charge / discharge efficiency.

[0094] 2.2) Based on the load and distributed power output in the energy internet, a mapping relationship between network topology and load reduction is established based on graph convolutional neural network to obtain the load loss rate index of each network topology.

[0095] The input to the graph convolutional neural network includes: active and reactive power data and power regulation values ​​of each node in the energy internet, as well as the node adjacency matrix A. The specific operational expression of the graph convolutional layer is:

[0096] H l =σ(L sym H l-1 W l (12)

[0097] Among them, H l Let H0 be the hidden feature matrix output by the l-th graph convolutional layer, where H0 = X; L sym Let be the symmetric normalized Laplace matrix of the graph. in I N It is an N×N identity matrix. The adjacency matrix of the graph after adding self-joins, where D is the degree matrix of the graph; W lLet be the trainable weight matrix of the l-th layer of the GCN; σ(·) is the activation function.

[0098] The training process of a graph convolutional neural network includes the following steps:

[0099] 2.2.1) Normalization is used to map each component of the node feature to the interval [0,1]. Assuming the node feature vector is x, then:

[0100]

[0101] Where, x j For the feature components of the i-th node before normalization, x represents the feature component of the i-th node after normalization, j∈[1,M]; min x is the minimum value among all components of the node feature. max The maximum value among all components of the node feature;

[0102] The processed data samples are randomly divided into training, validation, and test sets according to a certain ratio. The training and validation sets are used to determine the various hyperparameters and weights of GCN, while the test set is used to evaluate the performance of the trained GCN model.

[0103] 2.2.2) Initialize the processed data, including initializing weights and biases.

[0104] 2.2.3) Train the GCN model, and calculate the loss function between the predicted and actual values ​​through forward propagation. The mean squared error is... The weights and biases of the GCN are updated using the backpropagation algorithm. After multiple iterations, the termination condition is met, which is that the number of iterations or the loss function value is less than a certain threshold. At this point, it indicates that the GCN has learned the mapping pattern between the data, and the GNC model is saved.

[0105] 2.2.4) Input the test set data into the saved GCN model to obtain the important load power failure rate indicators for each network topology.

[0106] S3: Based on the energy internet optimization model and economic cost assessment network, construct an energy internet optimization and reconfiguration model that considers the constraint of the power failure rate of important loads.

[0107] Based on the energy internet optimization model formula and corresponding constraints, the following is set: The maximum allowable load reduction; by considering the constraint of the power failure rate of important loads, the number of candidate topologies can be effectively reduced.

[0108] Since the evaluation network based on graph convolutional neural networks is trained offline and makes decisions online, the trained graph convolutional neural network can calculate the corresponding economic indicators of all topologies in a very short time. Based on this, the optimal solution is obtained through various constraints, so as to realize more rapid and effective active power support for the energy internet.

[0109] S4: The optimal solution is obtained by using the water cycle algorithm to solve the energy internet optimization and reconstruction model, thus completing the network optimization and reconstruction of the energy internet. This water cycle algorithm includes the following steps:

[0110] 4.1) The rainfall process generates an initial population, which is then initialized and classified as stream, river, or ocean.

[0111] The initial population is represented as N. pop ×D matrix

[0112]

[0113] Where, N pop D is the randomly generated population size, and D is the number of control variables.

[0114] 4.2) The fitness of the initialized population is evaluated based on the cost function, that is, each population is evaluated according to its cost function.

[0115] F i =f(x) i,1 ,x i,2 ,x i,3 ,...,x i,D (15)

[0116] Among them, F i Let be the fitness of the i-th population, i = 1, 2, ..., N. pop .

[0117] 4.3) Determine the intensity of raindrops flowing towards rivers and oceans during the rainfall process, and update the locations of streams and rivers.

[0118] Runoff processes renew streams and rivers, causing them to flow into the ocean. The flow rates of streams into rivers and the ocean are defined by equations (16) and (17), and the flow rates of rivers into the ocean are given by equation (18).

[0119]

[0120]

[0121]

[0122] Where iter is the number of iterations, rand is a random number between 0 and 1, and C is the position update coefficient between 1 and 2; Formulas (16), (17), and (18) show that if the position of the stream is better than that of the river, the positions will be switched accordingly, and the same process will be repeated for the ocean and the river.

[0123] 4.4) Determine whether the evaporation conditions are met. If the evaporation conditions are met, new rainfall will form.

[0124] To avoid local convergence of WCA, the concept of evaporation is considered, where water evaporated from seawater returns to the landscape via rainfall. This rainwater forms a new stream, which then flows back to a river or ocean. This evaluation process ensures that the algorithm avoids getting trapped in local minima. The triggering condition for the evaporation process is:

[0125]

[0126]

[0127] Where, d max It is a parameter that is close to 0 and decreases with the number of iterations.

[0128] When rainfall conditions are met, a rainfall process will occur, resulting in new individuals. There are two types of rainfall:

[0129] a: Randomly generate new individuals in the problem space to increase population diversity.

[0130]

[0131] b: Rainfall near the ocean, seeking optimization near the optimal value.

[0132]

[0133] Where UB is the upper limit of the space and LB is the lower limit of the space; μ is the sea area search range. The smaller the value of μ, the smaller the search range. It is usually taken as 0.1.

[0134] 4.5) Determine whether the algorithm meets the termination condition, which is reaching the number of iterations or the loss function value is less than a certain threshold. If it is met, the output result is taken as the optimal solution; otherwise, repeat steps 4.3) to 4.4) to obtain the optimal solution based on the optimization objective, and reconstruct the energy internet based on the obtained optimal topology.

[0135] Example 2

[0136] Corresponding to the energy internet optimization and reconstruction method based on graph convolutional networks in Embodiment 1, this embodiment provides an energy internet optimization and reconstruction system based on graph convolutional networks. Please refer to [link to embodiment 1]. Figure 1As shown, it includes a module for constructing an energy internet optimization model, a module for constructing an economic cost assessment network, a module for constructing an energy internet optimization and reconfiguration model, and a water cycle algorithm module;

[0137] The module for constructing an energy internet optimization model is used to build an energy internet optimization model under high power deficit based on the objective function of minimizing the power failure rate of important loads.

[0138] The formula for the energy internet optimization model is as follows:

[0139]

[0140] Where F is the power failure rate of critical loads, ε is the controllable proportion of critical loads, and ω i εω represents the proportion of significant load reduction at node i. i P Li This represents the amount of significant load that can be reduced at node i based on ε.

[0141] The energy internet optimization model has the following constraints:

[0142] 1.1) Power balance constraints:

[0143]

[0144]

[0145]

[0146] Among them, P DGi Let Q be the active power output of the distributed power source at node i. DGi P represents the reactive power output of the distributed power source at node i; Li Q represents the active power consumed by the load at node i. Li G represents the reactive power consumed by the load at node i; ij For the conductance of branch ij, B ij For the susceptance of branch ij, θ ij Let be the phase angle difference between the voltages of node i and node j, and m be the number of branches connected to the node.

[0147] 1.2) Network radiation constraint:

[0148]

[0149] Among them, l ij Let N be the radiation of branch ij, N be the number of network nodes, and E be the set of network branches;

[0150] 1.3) Voltage upper and lower limit constraints:

[0151]

[0152] Among them, U i Let U be the voltage at node i. i,min U is the lower limit of the voltage at node i. i,max Let i be the upper bound of node i;

[0153] 1.4) Branch capacity constraints:

[0154] S ijmin ≤S ij ≤S ijmax (7)

[0155] Among them, S ij S is the capacity of branch ij. ijmin S is the lower limit of the capacity of branch ij. ijmax This represents the upper limit of the capacity of branch ij;

[0156] 1.5) DG power constraint:

[0157]

[0158] Among them, P DGmin P is the lower limit of the active power output of the distributed power source at node i. DGmax Q represents the upper limit of the active power output by the distributed power source at node i. DGmin Q is the lower limit of the reactive power output of the distributed power source at node i. DGmax This represents the upper limit of the reactive power output of the distributed power source at node i.

[0159] By minimizing the power failure rate of critical loads as the objective function, and by proposing general inequality constraints that consider power balance, voltage, branch capacity, and DG output power, the traditional energy internet optimization and reconfiguration problem can be solved, enabling active power support under power deficit conditions.

[0160] An economic cost assessment network module is constructed to build an economic cost assessment network based on the graph convolutional neural network algorithm, using the power failure rate of critical loads as the economic evaluation index.

[0161] The economic cost assessment network module includes a unit for defining economic evaluation indicators and a graph convolution training unit.

[0162] The economic evaluation index unit is used to quantify the economics of the energy internet. The power failure rate of critical loads is used as the economic evaluation index, defined as LPSP. The smaller the LPSP value, the better the economics of the energy internet system.

[0163]

[0164] in:

[0165] F(t) = Load(t) - E Max (t) (10)

[0166] E Max (t)=E PV (t)+E WT (t)+η(E BESS (t)-E BESS,min (11)

[0167] In the formula, F(t) is the minimum total power consumption at which the load is cut off; Load(t) is the total power consumption of the load; E Max (t) represents the maximum power generation of the distributed power source; E PV (t) represents photovoltaic power generation; E WT (t) represents the wind turbine's power generation; E BESS (t) represents the energy state of the electric vehicle battery; E BESS,min η represents the minimum battery capacity; η represents the charge / discharge efficiency.

[0168] The graph convolutional training unit is used to establish a mapping relationship between network topology and load reduction based on the load and distributed power output in the energy internet optimization, and to obtain the load loss rate index of each network topology.

[0169] The input to the graph convolutional neural network includes: active and reactive power data and power regulation values ​​of each node in the energy internet, as well as the node adjacency matrix A. The specific operational expression of the graph convolutional layer is:

[0170] H l =σ(L sym H l-1 W l (12)

[0171] Among them, H l Let H0 be the hidden feature matrix output by the l-th graph convolutional layer, where H0 = X; L sym Let be the symmetric normalized Laplace matrix of the graph. in I N It is an N×N identity matrix. The adjacency matrix of the graph after adding self-joins, where D is the degree matrix of the graph; W l Let be the trainable weight matrix of the l-th layer of the GCN; σ(·) is the activation function.

[0172] The graph convolutional training unit specifically includes a mapping part, an initialization part, a GCN training part, and an input part;

[0173] The mapping part is used to normalize the node features to map each component to the interval [0,1]. Assuming the node feature vector is x, then:

[0174]

[0175] Where, x j For the feature components of the i-th node before normalization, x represents the feature component of the i-th node after normalization, j∈[1,M]; min x is the minimum value among all components of the node feature. max The maximum value among all components of the node feature;

[0176] The processed data samples are randomly divided into training, validation, and test sets according to a certain ratio. The training and validation sets are used to determine the various hyperparameters and weights of GCN, while the test set is used to evaluate the performance of the trained GCN model.

[0177] The initialization section is used to initialize the processed data, including initializing weights and biases.

[0178] The GCN training part is used to train the GCN model. Through forward propagation, the loss function between the predicted and actual values ​​is calculated, with a mean squared error of... The weights and biases of the GCN are updated using the backpropagation algorithm. After multiple iterations, the termination condition is met, which is that the number of iterations or the loss function value is less than a certain threshold. At this point, it indicates that the GCN has learned the mapping pattern between the data, and the GNC model is saved.

[0179] The input section is used to input the test set data into the saved GCN model to obtain important load power failure rate indicators for each network topology.

[0180] In the Energy Internet Optimization and Reconfiguration Model module, an Energy Internet Optimization and Reconfiguration Model is constructed based on the Energy Internet Optimization Model and the Economic Cost Assessment Network, taking into account the constraint of the power failure rate of important loads.

[0181] Based on the energy internet optimization model formula and constraints, the following is set: The maximum allowable load reduction; by considering the constraint of the power failure rate of important loads, the number of candidate topologies can be effectively reduced.

[0182] Since the evaluation network based on graph convolutional neural networks is trained offline and makes decisions online, the trained graph convolutional neural network can calculate the corresponding economic indicators of all topologies in a very short time. Based on this, the optimal solution is obtained through various constraints, so as to realize more rapid and effective active power support for the energy internet.

[0183] The water cycle algorithm module is used to solve the energy internet optimization and reconstruction model using the water cycle algorithm to obtain the optimal solution, thereby completing the network optimization and reconstruction of the energy internet.

[0184] The water cycle algorithm includes an initialization unit, a fitness evaluation unit, an update unit, an evaporation detection unit, and an algorithm detection unit.

[0185] The initialization unit is used to generate an initial population during the rainfall process and initializes the initial population, which is classified as stream, river, or ocean.

[0186] The initial population is represented as N. pop ×D matrix

[0187]

[0188] Where, N pop D is the randomly generated population size, and D is the number of control variables.

[0189] The fitness evaluation unit is used to evaluate the fitness of the initialized population based on the cost function, that is, to evaluate each population according to its cost function.

[0190] F i =f(x) i,1 ,x i,2 ,x i,3 ,...,x i,D (15)

[0191] Among them, F i Let be the fitness of the i-th population, i = 1, 2, ..., N. pop .

[0192] The update unit is used to determine the intensity of raindrops flowing towards rivers and oceans during the rainfall process, and to update the location of streams and rivers.

[0193] Runoff processes renew streams and rivers, causing them to flow into the ocean. The flow rates of streams into rivers and the ocean are defined by equations (16) and (17), and the flow rates of rivers into the ocean are given by equation (18).

[0194]

[0195]

[0196]

[0197] Where iter is the number of iterations, rand is a random number between 0 and 1, and C is the position update coefficient between 1 and 2; Formulas (16), (17), and (18) show that if the position of the stream is better than that of the river, the positions will be switched accordingly, and the same process will be repeated for the ocean and the river.

[0198] The evaporation unit is used to determine whether the evaporation conditions are met. If the evaporation conditions are met, new rainfall will form.

[0199] To avoid local convergence of WCA, the concept of evaporation is considered, where water evaporated from seawater returns to the landscape via rainfall. This rainwater forms a new stream, which then flows back to a river or ocean. This evaluation process ensures that the algorithm avoids getting trapped in local minima. The triggering condition for the evaporation process is:

[0200]

[0201]

[0202] Where, d max It is a parameter that is close to 0 and decreases with the number of iterations.

[0203] When rainfall conditions are met, a rainfall process will occur, resulting in new individuals. There are two types of rainfall:

[0204] a: Randomly generate new individuals in the problem space to increase population diversity.

[0205]

[0206] b: Rainfall near the ocean, seeking optimization near the optimal value.

[0207]

[0208] Where UB is the upper limit of the space and LB is the lower limit of the space; μ is the sea area search range. The smaller the value of μ, the smaller the search range. It is usually taken as 0.1.

[0209] The algorithm judgment unit is used to determine whether the algorithm meets the termination condition, which is to reach the number of iterations or the loss function value is less than a certain threshold. If the condition is met, the output result is taken as the optimal solution; otherwise, the update unit is repeatedly executed to the evaporation judgment unit, thereby obtaining the optimal solution based on the optimization objective, and the energy Internet is optimized and reconstructed according to the obtained optimal topology.

[0210] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0211] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0212] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0213] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0214] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of the claims of the present invention pending approval.

Claims

1. A method for optimizing and reconstructing an energy internet based on graph convolutional networks, characterized in that, Includes the following steps: S1: Based on the objective function of minimizing the power failure rate of critical loads, construct an optimization model for the energy internet under high power deficit conditions; S2: Construct an economic cost assessment network based on a graph convolutional neural network algorithm, using the power outage rate of critical loads as an economic evaluation index; including the following steps: 2.1) The power failure rate of critical loads is used as an economic evaluation indicator, which is defined as follows: ,but: in: In the formula, The minimum total power required for load shedding; This represents the total power consumption of the load. This represents the maximum power generation capacity of the distributed power source. Photovoltaic power generation; For wind turbine power generation; The energy state of electric vehicle batteries; This is the minimum battery capacity. For charge and discharge efficiency; 2.2) Based on graph convolutional neural networks, establish the mapping relationship between network topology and load reduction, and obtain the important load loss rate indicators for each network topology; S3: Based on the energy internet optimization model and economic cost assessment network, construct an energy internet optimization and reconfiguration model that considers the constraint of the power failure rate of important loads; S4: The optimal solution is obtained by using the water cycle algorithm to solve the energy internet optimization and reconstruction model, thus completing the network optimization and reconstruction of the energy internet.

2. The energy internet optimization and reconstruction method based on graph convolutional networks according to claim 1, characterized in that, In step S1, the formula for the energy internet optimization model is as follows: in, For critical load power loss rate, For the controllable proportion of critical loads, For nodes The proportion of critical load reduction at the location, Indicates at node According to Significant loads that can be reduced; This represents the number of network nodes.

3. The energy internet optimization and reconstruction method based on graph convolutional networks according to claim 2, characterized in that, In step S1, the energy internet optimization model has the following constraints: 1.1) Power balance constraints: in, For nodes The active power output of the distributed power source. For nodes The reactive power output of the distributed power source; For nodes The active power consumed by the load. For nodes The reactive power consumed by the load; branch road electrical conductivity, branch road susceptivity, For nodes and nodes The phase angle difference of the voltage, It is the number of branches connected to the node; 1.2) Network radiation constraint: in, branch road radiation, The number of network nodes. A collection of network branches; 1.3) Voltage upper and lower limit constraints: in, For nodes voltage, For nodes The lower limit of voltage, For nodes The upper limit; 1.4) Branch capacity constraints: in, branch road capacity, branch road Lower limit of capacity branch road Maximum capacity; 1.5) DG power constraint: in, For nodes The lower limit of the active power output of the distributed power source. For nodes The upper limit of the active power output of the distributed power source. For nodes The lower limit of reactive power output by distributed power sources. For nodes The upper limit of reactive power output by a distributed power source.

4. The energy internet optimization and reconstruction method based on graph convolutional networks according to claim 3, characterized in that, In step S2, the training process of the graph convolutional neural network includes the following steps: 2.2.1) Normalization is used to map each component of the node feature to the interval [0,1]. Assume the node feature vector is... ,but: in, For standardization The feature components of each node For the standardized first The feature components of each node, ; It is the minimum value among all components of the node feature. The maximum value among all components of the node feature; The processed data samples are randomly divided into training set, validation set and test set according to the proportion. The training set and validation set are used to determine the various hyperparameters and weights of GCN, and the test set is used to evaluate the performance of the trained GCN model. 2.2.2) Initialize the processed data, including initializing weights and biases; 2.2.3) Train the GCN model, calculate the loss function between the predicted value and the true value through forward propagation, and update the weights and biases of the GCN using the backpropagation algorithm. After multiple iterations, if the termination condition is met, save the GCN model. 2.2.4) Input the test set data into the saved GCN model to obtain the important load power failure rate indicators for each network topology.

5. The energy internet optimization and reconstruction method based on graph convolutional networks according to claim 4, characterized in that, In step S3, based on the energy internet optimization model formula and constraints, the following steps are set: , To determine the maximum permissible load reduction, an energy internet optimization and reconfiguration model is constructed.

6. The energy internet optimization and reconstruction method based on graph convolutional networks according to claim 5, characterized in that, The water cycle algorithm in step S4 includes the following steps: 4.1) The rainfall process generates an initial population, which is then initialized and classified as stream, river, or ocean. 4.2) Evaluate the fitness of the initialized population based on the cost function; 4.3) Determine the intensity of raindrops flowing towards rivers and oceans during the rainfall process, and update the locations of streams and rivers; 4.4) Determine if the evaporation conditions are met; if so, new rainfall will occur. 4.5) Determine if the algorithm meets the termination condition. If it does, output the result as the optimal solution; otherwise, repeat steps 4.3) to 4.

4.

7. An energy internet optimization and reconfiguration system for executing the energy internet optimization and reconfiguration method based on graph convolutional networks as described in any one of claims 1-6, characterized in that, It includes modules for building an energy internet optimization model, building an economic cost assessment network, building an energy internet optimization and reconfiguration model, and a water cycle algorithm module; The module for constructing an energy internet optimization model is used to build an energy internet optimization model under high power deficit based on the objective function of minimizing the power failure rate of important loads. An economic cost assessment network module is constructed to build an economic cost assessment network based on the graph convolutional neural network algorithm, using the power failure rate of critical loads as an economic evaluation index. The module for constructing an energy internet optimization and reconfiguration model is used to construct an energy internet optimization and reconfiguration model that considers the constraints of the power failure rate of important loads based on the energy internet optimization model and the economic cost assessment network. The water cycle algorithm module is used to solve the energy internet optimization and reconstruction model using the water cycle algorithm to obtain the optimal solution, thereby completing the network optimization and reconstruction of the energy internet.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images