Transform adaptive distributed multi-target mantis power system real-time scheduling method, medium and processor

Abstract

The invention provides an electric power system real-time scheduling method of a Transform adaptive distributed multi-target mantis, a medium and a processor. The method comprises the following steps: firstly, establishing a power system objective function based on power generation energy consumption, carbon emission, system network loss and electric energy quality, and converting the power system objective function into a single objective problem by adopting a linear weighting method; a bus tearing method is adopted, the centralized power system is divided into a plurality of areas, a mantis method is used for selection operation, and a new solution set is generated. The solution set and the state variable are input to Transform for training after being subjected to normalization processing and position coding, and future performance of the solution and the state variable is predicted. The prediction solution is then incorporated into the mantis method as a new population or a mutation solution, and at the same time, the mantis method is guided to adaptively adjust parameters, and the search strategy and performance are optimized. According to the method, the problem of distributed multi-target real-time scheduling in the power system can be solved, the function of self-adaptive parameter adjustment is realized, and a better decision scheme is provided.

Description

Technical Field

[0001] The present invention belongs to the field of power system dispatching and new energy dispatching, and relates to a distributed optimization method, a multi-objective optimization method, a Transformer method, a heuristic optimization method and a Transformer adaptive distributed multi-objective optimization method, and is suitable for real-time dispatching of power systems. Background Art

[0002] Traditional centralized power system dispatching methods are mainly based on single-objective and centralized optimization, such as power generation energy consumption, carbon emissions, system network losses and power quality. However, with the increase in the complexity of new energy grids and power systems, single-objective optimization methods and centralized optimization methods can no longer meet the needs of complex power systems. Single-objective optimization methods show obvious limitations when dealing with multi-objective problems and cannot take into account the trade-offs between various objectives at the same time. Centralized optimization methods require a lot of computing resources to calculate large power systems, resulting in reduced overall system efficiency and robustness.

[0003] The multi-objective mantis method is a heuristic method based on multiple objectives. It solves multi-objective optimization problems by simulating the predation behavior of mantises. It has strong search capabilities and good convergence performance. However, the multi-objective mantis method is very sensitive to parameter settings. Different problems require different parameter adjustments, and there is a possibility of falling into a local optimal solution. On large-scale problems, the multi-objective mantis algorithm still requires a lot of computing resources and takes a long time to calculate.

[0004] Therefore, the current centralized power system optimization methods have the problems of single objective, low computational efficiency, sensitive parameter settings of multi-objective methods and easy to fall into local optimal solutions. Summary of the invention

[0005] Based on this, it is necessary to provide a Transformer adaptive distributed multi-objective mantis power system real-time scheduling method, device, computer equipment, computer-readable storage medium and computer program product to address the above technical problems.

[0006] In a first aspect, the present application provides a Transformer adaptive distributed multi-objective mantis power system real-time scheduling method. The method comprises:

[0007] Constructing a mathematical model for real-time dispatch of the power system, the mathematical model including an objective function constructed with power generation energy consumption, carbon emissions, system network loss and power quality as objectives; using a linear weighted method to convert a multi-objective optimization problem into a single-objective optimization problem;

[0008] The busbar tearing method is used to divide the centralized power system into N areas, and the constructed mathematical model is solved by the multi-objective mantis method to obtain the Pareto optimal solution set of multiple objectives;

[0009] Initialize the Transformer network parameters; collect the good solutions in each iteration of the Mantis method and state variables Construct an input data set; train the Transformer network;

[0010] The predicted solution and predicted state variables trained by the Transformer model are input into the Mantis method to obtain the Pareto optimal solution set 2. The solution selected by the decision maker from the Pareto optimal solution set 2 is sent to the distributed generator set to allocate the active power of the generator set and execute the power generation instruction.

[0011] In the second aspect, the present application also provides a Transformer adaptive distributed multi-objective mantis power system real-time dispatching device, characterized in that the device includes:

[0012] A model building module is used to build a mathematical model for real-time dispatch of the power system, wherein the mathematical model includes an objective function built with power generation energy consumption, carbon emissions, system network loss and power quality as objectives; a linear weighted method is used to convert a multi-objective optimization problem into a single-objective optimization problem;

[0013] The Pareto optimal solution set solving module is used to divide the centralized power system into N areas by using the busbar tearing method, and solve the constructed mathematical model to obtain the Pareto optimal solution set of multiple objectives by using the multi-objective mantis method;

[0014] The large model network training module is used to initialize the Transformer network parameters; in each iteration of the Mantis method, the collected solutions are and state variables Construct an input data set; train the Transformer network;

[0015] The large model prediction and scheduling instruction execution module is used to input the prediction solution and prediction state variables trained by the Transformer model into the Mantis method, obtain the Pareto optimal solution set 2, send the solution selected by the decision maker from the Pareto optimal solution set 2 to the distributed generator set, allocate the active power of the generator set, and execute the power generation instruction.

[0016] In a third aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0017] Constructing a mathematical model for real-time dispatch of the power system, the mathematical model including an objective function constructed with power generation energy consumption, carbon emissions, system network loss and power quality as objectives; using a linear weighted method to convert a multi-objective optimization problem into a single-objective optimization problem;

[0018] The busbar tearing method is used to divide the centralized power system into N areas, and the constructed mathematical model is solved by the multi-objective mantis method to obtain the Pareto optimal solution set of multiple objectives;

[0019] Initialize the Transformer network parameters; collect the good solutions in each iteration of the Mantis method and state variables Construct an input data set; train the Transformer network;

[0020] The predicted solution and predicted state variables trained by the Transformer model are input into the Mantis method to obtain the Pareto optimal solution set 2. The solution selected by the decision maker from the Pareto optimal solution set 2 is sent to the distributed generator set to allocate the active power of the generator set and execute the power generation instruction.

[0021] In a fourth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0022] Constructing a mathematical model for real-time dispatch of the power system, the mathematical model including an objective function constructed with power generation energy consumption, carbon emissions, system network loss and power quality as objectives; using a linear weighted method to convert a multi-objective optimization problem into a single-objective optimization problem;

[0023] The busbar tearing method is used to divide the centralized power system into N areas, and the constructed mathematical model is solved by the multi-objective mantis method to obtain the Pareto optimal solution set of multiple objectives;

[0024] Initialize the Transformer network parameters; collect the good solutions in each iteration of the Mantis method and state variables Construct an input data set; train the Transformer network;

[0025] The predicted solution and predicted state variables trained by the Transformer model are input into the Mantis method to obtain the Pareto optimal solution set 2. The solution selected by the decision maker from the Pareto optimal solution set 2 is sent to the distributed generator set to allocate the active power of the generator set and execute the power generation instruction.

[0026] In a fifth aspect, the present application further provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0027] Constructing a mathematical model for real-time dispatch of the power system, the mathematical model including an objective function constructed with power generation energy consumption, carbon emissions, system network loss and power quality as objectives; using a linear weighted method to convert a multi-objective optimization problem into a single-objective optimization problem;

[0028] The busbar tearing method is used to divide the centralized power system into N areas, and the constructed mathematical model is solved by the multi-objective mantis method to obtain the Pareto optimal solution set of multiple objectives;

[0029] Initialize the Transformer network parameters; collect the good solutions in each iteration of the Mantis method and state variables Construct an input data set; train the Transformer network;

[0030] The predicted solution and predicted state variables trained by the Transformer model are input into the Mantis method to obtain the Pareto optimal solution set 2. The solution selected by the decision maker from the Pareto optimal solution set 2 is sent to the distributed generator set to allocate the active power of the generator set and execute the power generation instruction.

[0031] The above-mentioned Transformer adaptive distributed multi-objective mantis power system real-time scheduling method, device, computer equipment, storage medium and computer program product construct a mathematical model for real-time scheduling of the power system, wherein the mathematical model includes an objective function constructed with power generation energy consumption, carbon emissions, system network loss and power quality as objectives; a linear weighted method is used to convert the multi-objective optimization problem into a single-objective optimization problem;

[0032] The busbar tearing method is used to divide the centralized power system into N areas, and the constructed mathematical model is solved by the multi-objective mantis method to obtain the Pareto optimal solution set of multiple objectives;

[0033] Initialize the Transformer network parameters; collect the good solutions in each iteration of the Mantis method and state variables Construct an input data set; train the Transformer network;

[0034] The predicted solution and predicted state variables trained by the Transformer model are input into the Mantis method to obtain the Pareto optimal solution set 2. The solution selected by the decision maker from the Pareto optimal solution set 2 is sent to the distributed generator set, the active power of the generator set is allocated, and the power generation instruction is executed. It can simultaneously consider multiple objective factors such as power generation energy consumption, carbon emissions, system network losses and power quality, and is used to solve the distributed multi-objective real-time scheduling problem in the power system. It has the function of adaptively adjusting parameters and generating high-quality real-time scheduling solutions, which can reduce the power generation energy consumption, carbon emissions, system network losses and power quality of the power system.

[0035] Compared with the prior art, the present invention has the following advantages and effects:

[0036] (1) Compared with the traditional centralized power system, the present invention takes into account the coordinated dispatching and operation of new energy power generation such as wind energy, photovoltaic energy, nuclear energy, biomass energy and pumped storage energy, as well as the dispatching problem of distributed power system.

[0037] (2) The present invention considers the objective functions of power generation energy consumption, carbon emissions, system network loss and power quality. Compared with the existing single-objective optimization method, it considers more objectives and constraints, and obtains the optimal Pareto optimal solution set for decision makers to choose.

[0038] (3) The present invention uses a multi-objective mantis method. Compared with the traditional multi-objective method, the multi-objective mantis method not only has powerful exploration capabilities and high computational efficiency, but also can obtain a smaller Pareto optimal solution set.

[0039] (4) The present invention proposes a distributed multi-objective Mantis method. Compared with the traditional single-objective method, the distributed multi-objective Mantis method considers the multi-objective factors of power generation energy consumption, carbon emissions, system network loss and power quality; compared with the traditional multi-objective method, the distributed multi-objective Mantis method takes into account the distribution and divides the centralized power system into a distributed power system. It has the characteristics of fast calculation efficiency and protection of the privacy of the power system.

[0040] (5) The present invention proposes a Transformer adaptive distributed multi-objective Mantis method. Compared with the multi-objective Mantis method, the Transformer adaptive distributed multi-objective Mantis method can adaptively adjust algorithm parameters and search strategies according to the Transformer network prediction solution, thereby improving optimization performance and computational efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 Schematic diagram of a busbar tearing method in one embodiment.

[0042] Figure 2It is a flowchart of a method for real-time dispatching of a power system using a Transformer adaptive distributed multi-objective mantis in one embodiment.

[0043] Figure 3 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0045] In one embodiment, Figure 1 As shown in FIG. 1 , a schematic diagram of the busbar tearing method is provided. Taking three regions as an example, region A, region B and region C are respectively connected through the busbar , and The busbar tearing method is used to tear each busbar at its midpoint. , , , , and Six virtual buses. After tearing, boundary information is exchanged with each other through virtual buses, which not only realizes information confidentiality, but also realizes distributed multi-region collaborative scheduling, with the characteristics of fast computing efficiency and protection of power system privacy.

[0046] In one embodiment, Figure 2 As shown, a flowchart of a real-time dispatching method for a power system based on a Transformer adaptive distributed multi-objective mantis is provided. The steps of the real-time dispatching method for a power system based on a Transformer adaptive distributed multi-objective mantis are as follows:

[0047] Step (1): constructing a mathematical model for real-time dispatch of the power system, wherein the mathematical model includes an objective function constructed with power generation energy consumption, carbon emissions, system network loss and power quality as targets, and complies with generator active power equality constraints, generator active power inequality constraints, thermal power generator unit ramp constraints, node voltage constraints and power system frequency constraints;

[0048] The objective function of power generation energy consumption is:

[0049] (1)

[0050] in, is the objective function of power generation energy consumption; It is the minimum value of power generation energy consumption; It is Thermal power generating units Energy consumption at each moment; It is Wind turbines Energy consumption at each moment; It is Photovoltaic generator sets Energy consumption at each moment; It is Nuclear power generating units Energy consumption at each moment; It is Biomass power generation units Energy consumption at each moment; It is Tidal power generators Energy consumption at each moment; It is Hydroelectric generating units Energy consumption at each moment; is the total number of thermal power generating units; is the total number of wind turbines; is the total number of photovoltaic generators; is the total number of nuclear power generating units; is the total number of biomass generating units; is the total number of tidal energy generators; is the total number of hydroelectric generating units; is the summation symbol; is the total time of scheduling; It is the serial number of the thermal power generating unit; is the serial number of the wind turbine; It is the serial number of the photovoltaic generator set; It is the serial number of the nuclear power generating unit; It is the serial number of the biomass generator set; is the serial number of the tidal energy generator set; is the serial number of the hydroelectric generator set;

[0051] The energy consumption of the generator is calculated as:

[0052] (2)

[0053] in, It is Thermal power generating units Active power output at all times; It is Wind turbines Active power output at all times; It is Photovoltaic generator sets Active power output at all times; It is Nuclear power generating units Active power output at all times; It is Biomass power generation units Active power output at all times; It is Tidal power generators Active power output at all times; It is Hydroelectric generating units Active power output at all times; is the quadratic coefficient of energy consumption of thermal power generating units; is the first-order coefficient of energy consumption of thermal power generating units; is the constant term coefficient of energy consumption of thermal power generating units; is the energy consumption coefficient of the wind turbine generator set; is the energy consumption coefficient of the photovoltaic generator set; is the energy consumption coefficient of nuclear power generating units; is the energy consumption coefficient of the biomass power generator set; is the energy consumption coefficient of the tidal power generator set; is the energy consumption coefficient of the hydroelectric generator set;

[0054] The generator active power equation constraint is:

[0055] (3)

[0056] The generator active power inequality constraint is:

[0057] (4)

[0058] in, yes Power system load demand during the period; It is the lower limit of the output power of the thermal power generating unit; It is the upper limit of the output power of thermal power generating units; is the lower limit of the wind turbine output power; It is the upper limit of the output power of the wind turbine; It is the lower limit of the output power of the photovoltaic generator set; It is the upper limit of the output power of the photovoltaic generator set; It is the lower limit of the output power of nuclear power generating units; It is the upper limit of the output power of nuclear power generating units; It is the lower limit of the output power of the biomass power generator set; It is the upper limit of the output power of the biomass power generator set; It is the lower limit of the output power of tidal energy generator set; It is the upper limit of the output power of tidal energy generators; It is the lower limit of the output power of the hydroelectric generator set; It is the upper limit of the output power of the hydroelectric generator set;

[0059] Climbing constraints of thermal power generating units:

[0060] (4)

[0061] in, It is Thermal power generating units Active power output at all times; is the upward climbing power of the thermal power generating unit; is the downward climbing power of the thermal power generating unit;

[0062] The objective function of carbon emissions is:

[0063] (5)

[0064] in, is the carbon emission objective function; It is the minimum carbon emission; is the quadratic coefficient of carbon emissions from thermal power generating units; is the first-order coefficient of carbon emissions from thermal power generating units; is the constant coefficient of carbon emissions from thermal power generating units;

[0065] The objective function of the system network loss is:

[0066] (6)

[0067] in, is the objective function of system network loss; It is the minimum value of system network loss; is the number of circuit branches; It is a line The conductance between Is a node The voltage amplitude of Is a node The voltage amplitude; Is a node The phase angle of Is a node The phase angle of Is a node and nodes The voltage phase difference between It belongs to the symbol;

[0068] The node voltage constraints are:

[0069] (7)

[0070] in, Is a node The lower limit of the voltage amplitude; Is a node The upper limit of the voltage amplitude; Is a node The lower limit of the voltage amplitude; Is a node The upper limit of the voltage amplitude;

[0071] The objective function of power quality is:

[0072] (8)

[0073] The power system frequency constraint is:

[0074] (9)

[0075] in, is the objective function of power quality; is the minimum value of power quality; Is a node frequency; is the node reference voltage; is the node reference frequency; is the node voltage deviation weight factor; is the frequency deviation weight factor; Is a node The minimum frequency of Is a node The maximum frequency of

[0076] Step (2): Based on the power generation energy consumption, carbon emissions, system network loss and power quality objectives and constraints in step (1), use the linear weighted method to convert the multi-objective optimization problem into a single-objective optimization problem; the specific steps are:

[0077] Step (2.1): Normalize each power generation energy consumption, carbon emissions, system network loss and power quality objective function:

[0078] (10)

[0079] in, It is A normalized power generation energy consumption, carbon emissions, system network losses and power quality objective function; It is objective function; It is The minimum value of the objective function; It is The maximum value of the objective function;

[0080] Step (2.2): Perform weighted summation on the normalized multi-objective function and convert it into a single objective function:

[0081] (11)

[0082] The weight coefficient satisfies:

[0083] (12)

[0084] in, It is a single objective function after all objectives are weighted; It is The weight of each target;

[0085] Step (3): Divide the centralized power system into N areas. In one embodiment, the centralized power system is divided into three areas. The specific steps are as follows:

[0086] Step (3.1): The centralized power system is divided into area A, area B and area C using the busbar tearing method. The torn busbars act as virtual power generation buses in their respective areas to simulate power exchange between different areas. The torn busbars still comply with power flow balance and constraint conditions.

[0087] Step (3.2): After the busbar is torn, the electrical constraints between the sections are met:

[0088] (13)

[0089] in, and They are the virtual busbars between area A and area B respectively; and They are the virtual busbars between area B and area C respectively; and They are the virtual busbars between area A and area C respectively; and are the direction correction coefficients of the virtual busbars between area A and area B respectively; and are the virtual busbar direction correction coefficients between regions B and C, respectively; and are the virtual bus direction correction coefficients between area A and area C respectively; the virtual bus direction correction coefficients meet the following conditions:

[0090] (14)

[0091] in, It is the current emitted between the areas after tearing; The diagonal elements are The diagonal matrix of ; The diagonal elements are The diagonal matrix of ;

[0092] Step (4): The mathematical models constructed in steps (1), (2) and (3) are used to solve the Pareto optimal solution set of multiple objectives using the multi-objective mantis method, thereby optimizing the objective functions of power generation energy consumption, carbon emissions, system network loss and power quality. The specific steps of the multi-objective mantis method are as follows:

[0093] Step (4.1): Input the generator energy consumption coefficient, carbon emission coefficient, generator power upper and lower limits, node frequency, node voltage parameters and total load value into the multi-objective mantis method;

[0094] Step (4.2): According to step (2), use the linear weighting method to transform the multi-objective optimization problem into a single-objective optimization problem;

[0095] Step (4.3): Set the mantis population to ; Set the maximum number of iterations to ; Set the optimal solution of Mantis to , which is initially a zero vector; set the optimal fitness value of mantis to , initially infinite;

[0096] Step (4.4): Initialize the mantis solution:

[0097] (15)

[0098] in, It is The initial solution of a mantis; is the lower limit of the solution; is the upper limit of the solution; It is a number randomly generated between 0 and 1 based on uniform distribution; according to the solution of initializing mantis, the initial fitness value of the objective function F(x) is calculated ;

[0099] Step (4.5): Determine the initial fitness value Is it less than the optimal fitness value? , if so, the optimal solution for Mantis is set to , the optimal fitness value of mantis is set to ; Otherwise, the optimal solution for mantis is And the optimal fitness value of mantis remain unchanged;

[0100] Step (4.6): The mantis method starts to iterate. In each iteration, the position of the mantis is updated according to the exploration phase and the exploitation phase;

[0101] Step (4.7): Enter the exploration phase; the mantis method simulates the process of mantis catching food, and the exploration phase is divided into the pursuit process and the ambush predation process; the first is the pursuit process, and the pursuit of mantis uses formula (16) to enhance the exploration ability of the mantis method; the mathematical model of the pursuit process is:

[0102] (16)

[0103] in, It is Daizhongdi A praying mantis solution; It is a random number generated by Levy flight; is a random number generated by normal distribution; It is Daidi A praying mantis solution; It is Daidi A praying mantis solution; It is Daidi A praying mantis solution; is a binary vector representing 0 or 1; It is the first random number of the Mantis method; is the second random number of the mantis method; input the updated solution into the objective function F(x) and calculate the fitness value ;

[0104] The second is the ambush predation process; the mantis method simulates the behavior of prey looking for food and falls into the attack range of the ambush predator; the mantis solution is updated through the following mathematical model:

[0105] (17)

[0106] in, It is the third random number of the Mantis method; It is the 4th random number of the Mantis method; It is The optimal solution for the mantis in the generation; is the number of iterations of the Mantis method; cos() is the cosine function; the updated solution is input into the objective function F(x) to calculate the fitness value ;

[0107] Step (4.8): Enter the utilization phase; the mantis method simulates the mantis's behavior of attacking prey that enters the attack range, and introduces a random exchange strategy to prevent the mantis method from falling into the trap of local optimality. The mathematical model of the utilization phase is:

[0108] (18)

[0109] in, It is the 5th random number of the Mantis method; It is the 6th random number of the Mantis method; is the speed at which a mantis attacks its prey; is the natural logarithm; input the updated solution into the objective function F(x) and calculate the fitness value ;

[0110] Step (4.9): Entering the stage of sexual cannibalism;

[0111] To improve the diversity and quality of solutions, the mantis method simulates the sexual cannibalistic behavior of mantises;

[0112] By simulating the female mantis attracting the male mantis to her position and exchanging information between the solutions, the mathematical model of sexual cannibalism is:

[0113] (19)

[0114] After the male mantis is attracted to the position of the female mantis, the male uses the crossover operator in the genetic algorithm to mate with the female and produce new offspring:

[0115] (20)

[0116] in, is the solution of the first dimension of the first mantis; input the updated solution into the objective function F(x) and calculate the fitness value ;

[0117] Step (4.10): Perform non-dominated sorting on the updated solutions, divide each solution set into multiple layers, and calculate the crowding distance between adjacent solutions according to formula (21):

[0118] (twenty one)

[0119] in, is the crowding distance; It is The maximum value of the objective function; It is The minimum value of the objective function; It is The objective function The function value of It is The objective function The function value of

[0120] Step (4.11): Store the non-dominated solutions in the Pareto optimal solution set 1;

[0121] Step (4.12): Determine whether ; is the first generation threshold set; if it is not met, update the number of iterations ,Right now , return to step (4.7) and continue iterating. In each iteration of the Mantis method, the solution and state variables ;

[0122] Step (5): Initialize the Transformer network parameters; collect the good solutions in each iteration of the Mantis method and state variables Constructed into input data set;

[0123] Step (5.1): Normalize the solution and state variables to be between 0 and 1 to facilitate model training; the normalization calculation method is:

[0124] (twenty two)

[0125] in, It is A normalized solution; It is A normalized state variable; is the minimum value of the solution; is the maximum value of the solution; is the maximum value of the state variable; is the minimum value of the state variable;

[0126] Step (5.2): The normalized solution and the normalized state variables Combine to form the input matrix ;

[0127] Step (5.3): Add the positional encoding to the input matrix to obtain sequence information that the model can use; the positional encoding is calculated as:

[0128] (twenty three)

[0129] in, is the position in the positional encoding; is the dimension of Transformer; is in position and dimensions in Sinusoidal position encoding at ; is in position and dimensions in The cosine position code at ; sin() is the sine function;

[0130] Step (5.4): Input matrix Perform position encoding:

[0131] (twenty four)

[0132] in, is the position-encoded input matrix; is the positional encoding;

[0133] Step (5.5): Enter the encoder layer; encode the positional input matrix The encoder uses the multi-head self-attention mechanism to capture the complex relationship between the solution and the fitness value. First, the query matrix is ​​calculated. , key matrix Sum Matrix :

[0134] (25)

[0135] in, is the weight of the query matrix; is the weight of the bond matrix; is the weight of the value matrix; secondly, according to the calculated query matrix , key matrix Sum Matrix Calculate the attention weights:

[0136] (26)

[0137] Among them, Attention( ) is the attention weight; softmax( ) is the softmax activation function; is the length of historical data input to the Transformer model; is the dimension of the key matrix; finally, calculate the multi-head self-attention mechanism:

[0138] (27)

[0139] Among them, MultiHead() is a multi-head self-attention mechanism; Concat() is a Concat concatenation function; is the output weight matrix; It is the first independent attention calculation unit; It is the second independent attention calculation unit; It is An independent attention calculation unit;

[0140] Step (5.6): Further feature extraction is performed on the complex relationship between the solution and fitness value captured by the multi-head self-attention mechanism through a feedforward neural network; first, feedforward processing is performed on each position:

[0141] (28)

[0142] Among them, FNN () is a feed-forward neural network; is the weight matrix of the first linear transformation layer in the feedforward network; is the weight matrix of the second linear transformation layer in the feedforward network; is the bias vector of the first linear transformation layer in the feedforward network; is the bias vector of the second linear transformation layer in the feedforward network; ReLU( ) is the ReLU activation function; secondly, the result after processing the feedforward neural network Add the original input:

[0143] (29)

[0144] in, is the output of the feedforward neural network, i.e. the output of the encoder layer , that is, H= ; LayerNorm() is the layer normalization output function; The feedforward neural network further extracts features from the complex relationship between the solution and fitness value captured by the multi-head self-attention mechanism As the output of the encoder layer ;

[0145] Step (5.7): Enter the decoder layer; convert the output of the encoder layer As the input of the decoder layer; the input of the decoder layer is processed by the self-attention layer to capture the internal relationship in the input sequence. The calculation method of the self-attention layer is;

[0146] (30)

[0147] Among them, SelfAttention() is the input sequence of the decoder layer; is the input sequence of the decoder layer;

[0148] Step (5.8): The decoder layer passes the encoder-decoder attention layer to the output of the encoder layer Combined with the input of the decoder layer to capture the relationship between the two; the encoder-decoder attention layer is calculated as:

[0149] (31)

[0150] Among them, EncDecAttention() is the encoder-decoder attention layer;

[0151] Step (5.9): According to formula (28), the output of the encoder layer captured by the encoder-decoder attention layer is fed through a feedforward neural network Further feature extraction is performed based on the relationship between the input of the decoder layer;

[0152] Step (5.10): Enter the output layer;

[0153] The output of the decoder layer is fed into a linear transform to compute the prediction solution:

[0154] (32)

[0155] in, is the predicted solution; is the predicted state variable; Linear() is the linear transformation function; DecoderOutput is the output of the decoder layer; [ ] represents a matrix;

[0156] Step (5.11): Determine whether the maximum training round of the Transformer network has been reached. If so, output the predicted solution and predicted state variables.

[0157] Otherwise, return to step (5.1) and continue iterating;

[0158] Step (6): Input the prediction solution and prediction state variables trained by the Transformer model into the Mantis method;

[0159] Step (6.1): Calculate the error between the t-th generation solution and the predicted solution, the t-th generation state variable and the predicted variable. The error calculation method is:

[0160] (33)

[0161] in, is the error between the t-th generation solution and the predicted solution; is the error between the state variable and the predictor variable of the tth generation; Expressed as the 1st norm;

[0162] Step (6.2): ​​Adaptively adjust the weights of the solution and state variables based on the error calculation. The adaptive adjustment weight calculation method is:

[0163] (34)

[0164] in, is the adaptive weight of the solution; is the adaptive weight of the state variable; It is to adjust the speed; is the error threshold of the solution; is the error threshold of the state variable;

[0165] Step (6.3): Based on the adjusted weight parameters, adaptively adjust the t-th generation solution and state variables:

[0166] (35)

[0167] in, It is the newly generated A mutation solution; It is the newly generated state variables;

[0168] Step (6.4): Iterate the adaptively adjusted solution and state variables and re-execute steps (4.6) to (4.11);

[0169] Step (6.5): Determine whether iter is greater than or equal to the maximum number of iterations maxiter. If so, output the second set of Pareto optimal solutions for the tth generation; otherwise, collect the tth generation solutions, fitness values ​​and state variables, and re-execute step (5) for the next round of iterations.

[0170] Step (6.6): Obtain Pareto optimal solution set 2, send the solution selected by the decision maker from Pareto optimal solution set 2 to the distributed generator set, allocate the active power of the generator set, execute the power generation instruction, and thus enter the next scheduling.

[0171] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 3As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and the external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be realized through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a Transformer adaptive distributed multi-objective mantis power system real-time scheduling method is realized. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0172] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0173] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

[0174] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0175] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0176] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0177] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0178] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0179] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A Transformer adaptive distributed multi-objective mantis power system real-time scheduling method, characterized in that: The method comprises: Constructing a mathematical model for real-time dispatch of the power system, the mathematical model including an objective function constructed with power generation energy consumption, carbon emissions, system network loss and power quality as objectives; using a linear weighted method to convert a multi-objective optimization problem into a single-objective optimization problem; The busbar tearing method is used to divide the centralized power system into N areas, and the constructed mathematical model is solved by the multi-objective mantis method to obtain the Pareto optimal solution set of multiple objectives; Initialize the Transformer network parameters; collect the good solutions in each iteration of the Mantis method and state variables Construct an input data set; train the Transformer network; The predicted solution and predicted state variables trained by the Transformer model are input into the Mantis method to obtain the Pareto optimal solution set 2. The solution selected by the decision maker from the Pareto optimal solution set 2 is sent to the distributed generator set to allocate the active power of the generator set and execute the power generation instruction.

2. According to the Transformer adaptive distributed multi-objective mantis real-time dispatching method for power system in claim 1, it is characterized by: The mathematical model includes an objective function constructed with power generation energy consumption, carbon emissions, system network losses and power quality as targets, and follows generator active power equality constraints, generator active power inequality constraints, thermal power generator set ramp constraints, node voltage constraints and power system frequency constraints.

3. According to the Transformer adaptive distributed multi-objective mantis real-time dispatching method for power system in claim 1, it is characterized by: The method of using the linear weighted method to convert the multi-objective optimization problem into a single-objective optimization problem is as follows: normalizing each objective function; The normalized multi-objective function is weighted summed and converted into a single objective function.

4. According to the Transformer adaptive distributed multi-objective mantis power system real-time scheduling method of claim 1, it is characterized by: The dividing of the centralized power system into N areas is to divide the centralized power system into N areas by adopting a busbar tearing method; the busbar after tearing still complies with the power flow balance and constraint conditions.

5. According to the Transformer adaptive distributed multi-objective mantis real-time dispatching method for power system in claim 1, it is characterized by: The mathematical model to be constructed uses the multi-objective mantis method to solve the Pareto optimal solution set of multiple objectives: Input the generator energy consumption coefficient, carbon emission coefficient, generator power upper and lower limits, node frequency, node voltage parameters and total load value into the multi-objective mantis method; Use the linear weighting method to transform the multi-objective optimization problem into a single-objective optimization problem; Set the mantis population to ; Set the maximum number of iterations to ; Set the optimal solution of Mantis to , which is initially a zero vector; set the optimal fitness value of mantis to , initially infinite; Initialize the mantis solution; calculate the initial fitness value of the objective function F(x) based on the initialized mantis solution ; Determine the initial fitness value Is it less than the optimal fitness value? , if so, the optimal solution for Mantis is set to , the optimal fitness value of mantis is set to ; Otherwise, the optimal solution for Mantis is And the optimal fitness value of mantis remain unchanged; The mantis method starts to iterate. In each iteration, the mantis's position is updated according to the exploration phase and the exploitation phase to obtain an updated solution. Perform non-dominated sorting on the updated solutions and store the non-dominated solutions in the Pareto optimal solution set 1; Determine whether it is satisfied ; is the first generation threshold set; if it is not met, update the number of iterations ,Right now , continue to iterate, in each iteration of the Mantis method, collect the solution and state variables .

6. According to claim 1, the Transformer adaptive distributed multi-objective mantis power system real-time scheduling method is characterized in that: Initialize the Transformer network parameters, collect the good solutions in each iteration of the Mantis method and state variables Construct the input data set and train the Transformer network as: Normalize the solution and state variables; The normalized solution and the normalized state variables Combine to form the input matrix ; Add positional encoding to the input matrix to obtain sequence information that the model can use; The position-encoded input matrix Passed into the multi-head self-attention mechanism, the encoder uses the multi-head self-attention mechanism to capture the complex relationship between the solution and the fitness value; The complex relationship between the solution and fitness value captured by the multi-head self-attention mechanism is further feature extracted through a feedforward neural network; Enter the decoder layer; the output of the encoder layer As input to the decoder layer; Process the input of the decoder layer through a self-attention layer to capture the internal relations in the input sequence; The decoder layer passes the encoder-decoder attention layer to the output of the encoder layer. Combined with the input of the decoder layer to capture the relationship between the two; The output of the encoder layer is captured by the encoder-decoder attention layer through a feed-forward neural network Further feature extraction is performed based on the relationship between the input of the decoder layer; Enter the output layer; The output of the decoder layer maps the high-dimensional output of the decoder layer to the solution space through linear transformation to generate a specific prediction solution; Determine whether the maximum training round of the Transformer network has been reached. If so, output the predicted solution and predicted state variables; otherwise, continue iterating.

7. According to claim 1, the Transformer adaptive distributed multi-objective mantis power system real-time scheduling method is characterized in that: The prediction solution and prediction state variables trained by the Transformer model are input into the Mantis method to obtain the Pareto optimal solution set 2, and the solution selected by the decision maker from the Pareto optimal solution set 2 is sent to the distributed generator set to allocate the active power of the generator set and execute the power generation instruction as follows: Calculate the errors between the t-th generation solution and the predicted solution, the t-th generation state variables and the predicted variables; Compute solutions and adaptively adjust weights of state variables based on errors; Based on the adjusted weight parameters, the t-th generation solution and state variables are adaptively adjusted; Iterate the adaptively adjusted solution and state variables; Determine whether iter is greater than or equal to the maximum number of iterations maxiter. If so, output the t-th generation Pareto optimal solution set; otherwise, collect the t-th generation solution, fitness value and state variable, and iterate again; The second Pareto optimal solution set is obtained, and the solution selected by the decision maker from the second Pareto optimal solution set is sent to the distributed generator set, the active power of the generator set is allocated, the power generation instruction is executed, and the next scheduling is entered.

8. A Transformer adaptive distributed multi-objective mantis power system real-time dispatching device, characterized in that: The device comprises: A model building module is used to build a mathematical model for real-time dispatch of the power system, wherein the mathematical model includes an objective function built with power generation energy consumption, carbon emissions, system network loss and power quality as objectives; a linear weighted method is used to convert a multi-objective optimization problem into a single-objective optimization problem; The Pareto optimal solution set solving module is used to divide the centralized power system into N areas by using the busbar tearing method, and solve the constructed mathematical model to obtain the Pareto optimal solution set of multiple objectives by using the multi-objective mantis method; The large model network training module is used to initialize the Transformer network parameters; in each iteration of the Mantis method, the collected solutions are and state variables Construct an input data set; train the Transformer network; The large model prediction and scheduling instruction execution module is used to input the prediction solution and prediction state variables trained by the Transformer model into the Mantis method, obtain the Pareto optimal solution set 2, send the solution selected by the decision maker from the Pareto optimal solution set 2 to the distributed generator set, allocate the active power of the generator set, and execute the power generation instruction.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the Transformer adaptive distributed multi-objective mantis power system real-time scheduling method described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the Transformer adaptive distributed multi-objective mantis power system real-time scheduling method described in any one of claims 1 to 7 are implemented.

Images