Multi-park equal carbon emission optimization regulation and control method based on channel attention

Abstract

The invention discloses a multi-park equal carbon emission optimization regulation and control method based on channel attention, and the method comprises the steps: carrying out the numerical corresponding processing of the actual carbon emission of a park and the energy consumption condition of the park, building a specific carbon-energy coupling relation, and achieving the overall optimization of the carbon emission of a corresponding park through the planning of an energy consumption plan; according to the power in-out floating condition introduced by new energy equipment, an algorithm based on channel attention is introduced to integrate data of the new energy equipment, after problems are found, further optimization regulation and control are carried out on each park, and system instability caused by fluctuation is controlled. Through overall scheduling, the park can be in the optimal economical efficiency or stability at the two levels of carbon emission and energy consumption from the beginning of regulation, and the process hardly needs time to be solved relative to regulation, so that the process is most efficient and most important; the second link of regulation and control can solve the problems of instability and uncertainty caused by introduction of a large amount of new energy equipment to the maximum extent, and reliable and stable operation of regulation and control is ensured.

Description

Technical Field

[0001] The present invention relates to a multi-park carbon emission optimization and control method based on channel attention, belonging to the technical field of power system optimization and control. Background Art

[0002] Scheduling technologies based on carbon potential are widely used in the context of dual carbon emissions. They further apportion the responsibility for carbon emissions based on the usage of energy-consuming units. By simulating the flow of electricity between nodes and creating a corresponding carbon potential model, carbon emissions can be measured more precisely. Existing technologies use this model to perform scheduling to ensure that the equivalent carbon emissions of each node are within the target range.

[0003] The park's multi-energy optimization and control technology is based on the mixed application and interactive scenarios of various forms of energy in the park. By controlling the park's various multi-state energy units, it continuously optimizes the current energy consumption of the park to achieve relatively economical and controllable results.

[0004] The defects of existing technologies are pointed out: It is believed that the defects of current technologies mainly exist in two aspects: 1. Insufficient consideration is given to the interactive coupling between carbon and electricity aspects, and one-sided emphasis is placed on the optimization of a single level of carbon emissions or a single level of electricity. There are certain technical gaps in the relevant strategies for the coordinated regulation of carbon targets and electricity targets; 2. The large-scale introduction of new energy equipment in the park has brought about stability problems. The highly variable and time-specific load and power supply changes have brought certain challenges to the existing regulation technology. How to propose a universal regulation strategy to a certain extent for different backgrounds still needs further research. Summary of the Invention

[0005] The present invention provides a multi-park carbon emission optimization and control method based on channel attention, which solves the problems disclosed in the background technology.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0007] A multi-park carbon emission optimization and control method based on channel attention:

[0008] Obtain carbon emissions and energy consumption data for multiple parks over multiple days;

[0009] With the goal of minimizing carbon emissions, we will build control plans for multiple parks that take into account carbon emissions and energy consumption.

[0010] Using the improved D-SENet to capture features of the control plan to obtain a feature set;

[0011] Based on the improved NSGA-III algorithm, the control plan is collaboratively optimized at both the energy consumption and carbon emission levels according to the feature set.

[0012] Furthermore, with the goal of minimizing carbon emissions, methods for constructing a control plan for multiple parks that considers carbon emissions and energy consumption include:

[0013] Assume that the carbon energy function of a single park is: ;

[0014] The slight increase rate indicator of a single park is : ;

[0015] It is believed that when the micro-increase rate indicators of each park are equal, the park group has reached the corresponding carbon saving state:

[0016] ;

[0017] Satisfying the following constraints The energy consumption It is possible to:

[0018] ;

[0019] in, is the carbon emissions of the i-th park, are the polynomial coefficients, obtained from data fitting. is the differential operator, is the carbon emission slight increase rate of the park with serial number j, The carbon emission slight increase rate of the park with serial number k; It shows the proportional relationship between energy consumption and carbon emissions on the grid side.

[0020] Furthermore, the method of using the improved D-SENet to capture features of the control plan to obtain a feature set includes:

[0021] Set control interval constraints;

[0022] Integrate data items including new energy;

[0023] Build interactive channels for different data;

[0024] The data items are reorganized through the interactive channel and integrated to obtain corresponding feature sets.

[0025] Furthermore, the method for setting the control interval constraint is:

[0026] Divide a single day into 96 time periods. , the park number is ,, the actual energy consumption of the park in a single day is expressed as , the actual instantaneous power of the park after regulation is , its range is: .

[0027] Furthermore, the method of integrating data items including new energy is as follows:

[0028] Select all data for several days in a year and 96 time periods within a day. The following data list can be obtained for 300 different time periods in a park within a day. The serial number of the load node in the park ; There is a data matrix for each energy unit of carbon emissions ;Energy input and output have data matrix for each energy unit ;

[0029] ; ;

[0030] Generate the corresponding convolution kernel , Element { }Generated by random numbers satisfying the two-dimensional normal distribution, where Perform convolution transformation:

[0031] ; ;

[0032] in, , ,Convolution transformation is performed on the data collected from each park.

[0033] Furthermore, the method of constructing interactive channels for different data is as follows:

[0034] Construct two fully connected layers for perceptual interaction. The first fully connected layer uses the ReLU function Activation, the second fully connected layer uses the Sigmod function Activation, the weight matrix used for coordination is , , output the corresponding data graph weight value after perceptual interaction, and the value initialization principle in the weight matrix is ​​the same as the convolution kernel; ; Where s is the output channel-specific score after interactive processing.

[0035] Furthermore, the data items are reorganized through the interactive channel and integrated to obtain the corresponding feature set by: The corresponding score Data feature map after multiplication and output scoring :

[0036] ;

[0037] The corresponding data graph The corresponding score Data feature map after multiplication and output scoring :

[0038] ;

[0039] For each park’s energy consumption and carbon emissions on a certain day, a corresponding characteristic graph is generated. ;After several days of operation, the corresponding feature set was integrated.

[0040] Furthermore, the method for collaboratively optimizing the control plan at both energy consumption and carbon emission levels based on the feature set based on the improved NSGA-III algorithm includes:

[0041] For any set of determined feature maps , , Each represents a corresponding park data, and secondary upper-layer optimization is performed based on the data of different parks:

[0042] The control object of the secondary upper-level optimization of the inter-park collaboration is the energy consumption of the park.

[0043] The optimization goal is: ;

[0044] The constraints are:

[0045] ;

[0046] in, is the weight coefficient of the overall energy consumption of the park, is the flat weight coefficient of inter-park power distribution, is the weight coefficient of the park’s energy consumption from the grid side, is the utilization rate coefficient of the park’s new energy equipment, is the coefficient of power interaction between parks, is the weight coefficient of the park’s overall carbon emissions, is the weight coefficient of the park’s planned carbon emissions to actual carbon emissions, For the power interaction of the park from the grid side, The output of new energy in the park. is the power interaction between parks, , The upper and lower limits of the power interaction between the park and the grid side, is the upper limit of power interaction between parks, , The upper and lower limits of new energy equipment output.

[0047] Furthermore, the method for collaboratively optimizing the control plan at both energy consumption and carbon emissions levels based on the feature set based on the improved NSGA-III algorithm also includes:

[0048] Each time, the NSGA-III algorithm is used to solve the optimization problem based on the actual constraints of the previous 15 minutes, the actual output of renewable energy, and the power interaction between the park and the grid. This allows the park to achieve the optimal coordination at the park level in terms of energy consumption and carbon emissions. The solution steps are as follows:

[0049] Step 1: Randomly generate an initial population consisting of multiple individuals, each of which consists of a series of decision variables;

[0050] Step 2: Sort the individuals in the population by non-domination and divide them into different levels. The first level contains the non-dominated solutions on the Pareto front, and the subsequent levels contain the solutions dominated by the previous level. The crowding distance of each solution is calculated to measure its distribution density.

[0051] Step 3: Select solutions from the non-dominated levels to form the next generation population, giving priority to retaining all solutions in the previous levels to maintain population quality; within the same level, select some solutions based on the crowding distance to promote population diversity;

[0052] Step 4: Generate a new generation of solutions using crossover and mutation operations; crossover combines two parent solutions to generate a new child solution, while mutation creates diversity by introducing slight random changes to the individual decision variables;

[0053] Step 5: Merge the newly generated solution with the previous generation population to form a new population of size 2N; if the size of the new population exceeds the predetermined value, use the selection mechanism to filter out the optimal solution;

[0054] Step 6: Check whether the preset termination condition, such as the maximum number of iterations or a specific convergence criterion, has been met. If not, return to step 2 to continue the evolution process. If so, output the results, i.e., a set of high-quality solutions on the Pareto frontier.

[0055] The corresponding Pareto front forms an n-dimensional surface, on which a set of n-dimensional data points that are easiest to implement in principle are selected as the energy consumption plan adjustment for quadratic programming;

[0056] For each set of feature maps and the corresponding set of park data, the corresponding solution strategy set is continuously calculated on a 15-minute time scale within a day, and finally a strategy set corresponding to the feature set is formed.

[0057] Furthermore, the method for collaboratively optimizing the control plan at both energy consumption and carbon emissions levels based on the feature set based on the improved NSGA-III algorithm also includes:

[0058] First, the historical data of each park is obtained to generate the corresponding prediction data. The directly generated data is the feature map generated by the operation in step 2. The new feature map is generated. Should be consistent with all the features in the set The most similar group is selected for matching. The selection basis of feature maps is:

[0059] ;

[0060] Get the corresponding closest feature map and execute the strategy for 96 time periods of 24 hours corresponding to the feature map solution, taking into account that the generated feature map matches all the previously generated feature maps:

[0061] ;

[0062] Where a is the minimum feature ratio median. If it is greater than this value, real-time rolling optimization is performed and the new feature map is placed in the feature library.

[0063] The beneficial effects achieved by the present invention are: 1. The actual carbon emissions of the park and the energy consumption (supply) of the park are numerically processed in correspondence, and a specific carbon-energy coupling relationship is established, so that the overall optimal carbon emissions of the corresponding park can be achieved by planning the energy consumption plan; 2. In response to the power inflow and outflow fluctuations introduced by new energy equipment, the SENet algorithm based on channel attention is introduced to integrate its data and then further optimize and regulate each park after problems are found to control system instability caused by fluctuations. 3. Through overall scheduling, the park can be in the optimal economy or stability at the two levels of carbon emissions and energy consumption from the beginning of regulation. Compared with regulation, this process requires almost no solution time and is the most efficient and most important. 4. The second link of regulation can absorb the instability and uncertainty problems brought about by the large-scale introduction of new energy equipment to the greatest extent, and ensure reliable and smooth operation of regulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 It is a schematic diagram of the process of the present invention;

[0065] Figure 2 Schematic diagram of the interactive channels of different data in the present invention. DETAILED DESCRIPTION

[0066] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention.

[0067] like Figure 1 As shown, the present invention provides a multi-park carbon emission optimization and control method based on channel attention, including the following steps:

[0068] 1. Determine energy consumption planning based on the equal carbon emission power increment rate to achieve primary optimization and control;

[0069] 1.1 Collect data on the park’s carbon emissions and electrical energy input and output over multiple days;

[0070] 1.1.1 Collect data from n parks, where the park number is i=1, 2, 3…n, and the number of the load node in the park Carbon emissions , unit is Kg (absorbed carbon is negative, removed carbon is positive), input and output power , the unit is W (the external supply of electric energy is positive, the internal absorption of electric energy is negative);

[0071] 1.1.2 Collect the time parallel items of the above data, and record the above data items in parallel for 365 days (the day number is t=0,1,2,3…,365). The following data is collected and recorded as , (Carbon emissions of the jth energy unit in the ith park on the tth day of information collection).

[0072] 1.1.3 Data requirements for special energy units: PV units combined with energy use for a certain energy unit make ; Carbon capture power plants are used in conjunction with energy output , build relationships ( ); the energy conversion unit makes ; For traditional energy consumption units from the grid side ,in is the monitored carbon emissions, of which It shows the proportional relationship between energy consumption and carbon emissions on the grid side.

[0073] 1.2 Affine scaling of data items;

[0074] It is hoped that the carbon emission ratio and energy input-output ratio of the same energy unit j can appear to be more real and reliable. At the same time, it is hoped that some energy units with large magnitude can match smaller weight coefficients, and energy units with small magnitude can match larger energy coefficients. The above principles are used to determine , .

[0075] 1.3 Integration of data items and formation of park carbon energy function;

[0076] 1.3.1 Data integration: Obtain the carbon emission data and the total output and input power of a single park per day through affine integration :

[0077] ;

[0078] The data items must not be distorted, so the constraints are:

[0079] ;

[0080] 1.3.2 Formation of the Carbon-Energy Function of a Single Park

[0081] First, assume that the carbon energy function of the park is in the following form:

[0082] ;

[0083] Then, based on the corresponding data obtained above, the least squares fitting is performed to obtain the corresponding carbon energy function of a single park.

[0084] 1.4 Determine the optimal energy use plan

[0085] 1.4.1 Determination of the slight increase rate of carbon emissions in multiple parks

[0086] For the park, the slight increase rate indicator of a single park is ;

[0087] It is necessary to fully consider the scale differences between parks and make additional considerations for the absorption of specific objects, especially in the following situations, such as: ;

[0088] At this time, we should consider giving certain weights to some parks to match them or make additional calculations to avoid the solution from seriously deviating from the actual situation.

[0089] 1.4.2 Determination of the optimal energy use plan for each park:

[0090] Here, it is assumed that when the micro-increase rate indicators of each park are equal, the park group has reached the corresponding carbon saving state, that is: ;

[0091] This actually provides the criteria for the park to regulate carbon emissions. Each park continuously catches up with the above balance state through the conversion of different forms of energy and other energy transmission methods, so as to achieve the goal of minimizing carbon emissions in each park under the operating entity.

[0092] However, due to the large differences in scale, energy unit types, and actual conditions between parks, the functions of each park may vary greatly. Therefore, the energy consumption plan given by the above optimal state may be difficult to achieve or may result in multiple parallel solutions. Therefore, a certain approximate compromise is made to meet the following constraints. The resulting energy consumption plan It is possible to:

[0093] ;

[0094] in, is the carbon emissions of the i-th park, are the polynomial coefficients, obtained from data fitting. is the differential operator, is the carbon emission slight increase rate of the park with serial number j, The carbon emission slight increase rate of the park with serial number k; It shows the proportional relationship between energy consumption and carbon emissions on the grid side.

[0095] In addition, if there is more than one optimal solution energy plan for the above equation , we consider two aspects:

[0096] Eliminate local optimal solutions that do not meet actual requirements, such as generating very small energy consumption plans to avoid carbon emissions , this is a control measure that cannot be implemented and needs to be omitted;

[0097] Based on the historical energy consumption of each park or the load forecast results, the solution that matches the actual value most closely is selected, and at the same time, it is hoped that the carbon emissions generated will be minimized. Therefore, an allowable radius range is set at the average power consumption plan based on the needs of the park, and the optimal solution that generates the smallest carbon emissions is selected.

[0098] 2. Feature set matching park collaborative optimization method based on improved channel attention D-SENet;

[0099] The above steps preliminarily provide plans for each park to regulate from the perspective of carbon emissions and energy consumption. However, considering the uncertainty fluctuations brought by renewable energy output, reasonable daily optimization is still required. However, the optimization calculation process may not be achieved in a short time. Therefore, to solve the above problem, a feature set library is established as the matching object for optimization, in which the improved D-SENet is used for feature capture.

[0100] 2.1 Secondary control interval constraint: In the above we use The only definite energy consumption plan was obtained The energy consumption plan can be slightly smaller or larger to create space for secondary regulation. , the park number is , (assuming the total number of parks is n), the actual energy consumption of the park during the day is expressed as The actual immediate plan of the park after regulation is , its range is:

[0101] ;

[0102] When finer time scale regulation is carried out within the day, if the actual result of the regulation is within this range, it is considered that the secondary regulation as a whole meets the previous day-ahead plan.

[0103] 2.2 Data integration including new energy equipment: Considering that the output of new energy equipment fluctuates violently during the day and there are corresponding internal connections within some energy units, all data within a year (several days) and 96 time periods within a day are selected and listed as follows. The following data list can be obtained for 300 different time periods in a park within a day. The time period is , the park number is , the serial number of the load node in the park For energy input, output and carbon emissions, there are two data matrices for each energy unit: , :

[0104] ; ;

[0105] Generate the corresponding convolution kernel , Element { }Generated by random numbers satisfying the two-dimensional normal distribution, where Perform convolution transformation:

[0106] ; ;

[0107] in, , , then perform the following operations on the data collected from each park below.

[0108] 2.2 Compression processing of data items;

[0109] right The data in the data is compressed accordingly, ignoring some unimportant information and making the final result fully reflect the global information of all energy units in the park. This operation is performed on the information of all parks. Here, only the park with serial number i is used as an example:

[0110] ;

[0111] Therefore, the corresponding data graph weight values ​​for each park are obtained below , .

[0112] 2.3 Interaction between different data channels;

[0113] like Figure 2 As shown, in order to allow the data from different parks that are already connected to interact, I put it into two fully connected layers for perceptual interaction. The first fully connected layer uses the ReLU function Activation, the second fully connected layer uses the Sigmod function Activation (the weight matrix of which is , ), outputs the weight value of the corresponding data graph after perceptual interaction. The value initialization principle in the weight matrix here is the same as the convolution kernel.

[0114] ;

[0115] Where s is the output channel-specific score after interaction processing.

[0116] 2.4 Data reorganization;

[0117] Then the corresponding data graph The corresponding score Data feature map after multiplication and output scoring :

[0118] ;

[0119] The corresponding data graph The corresponding score Data feature map after multiplication and output scoring :

[0120] ;

[0121] The data obtained here is considered to be a feature set that has been processed to fully consider the volatility of new energy equipment and the connections between different parks and different energy units. Figure 2 As a result of this processing, a corresponding characteristic graph is generated for the energy consumption and carbon emissions of each park on a certain day. On this basis, a total of 100 days of operations were carried out and the corresponding feature set was obtained through integration.

[0122] 3. Strategy set driven matching campus collaborative optimization method based on improved NSGA-III algorithm

[0123] For any set of determined feature maps , , Each represents a corresponding type of park data, and each type of park data corresponds to an immediate optimal strategy. At the same time, different parks can also adjust their energy-consuming equipment to respond to park planning and adopt corresponding strategies. At the inter-park level, secondary upper-level optimization is performed based on the data of different parks:

[0124] The main control objects of the secondary upper-level optimization of inter-park collaboration are the energy consumption of the park, the optimization objectives and the constraints:

[0125] ;

[0126] in, is the weight coefficient of the overall energy consumption of the park, is the flat weight coefficient of inter-park power distribution, is the weight coefficient of the park’s energy consumption from the grid side, is the utilization rate coefficient of the park’s new energy equipment, is the coefficient of power interaction between parks, is the weight coefficient of the park’s overall carbon emissions, is the weight coefficient of the park’s planned carbon emissions to actual carbon emissions, For the power interaction of the park from the grid side, The output of new energy in the park. is the power interaction between parks, , The upper and lower limits of the power interaction between the park and the grid side, is the upper limit of power interaction between parks, , The upper and lower limits of new energy equipment output.

[0127] Each time, based on the actual constraints of the previous 15 minutes, the actual output of renewable energy, and the power interaction between the park and the grid, the optimization problem is solved and corresponding arrangements are made. The NSGA-III algorithm is used to solve this problem, enabling the park to achieve the optimal coordination at the park level in terms of energy consumption and carbon emissions. The solution steps are as follows:

[0128] Step 1: Initialize the population;

[0129] An initial population of individuals (i.e., solutions) is randomly generated, each of which consists of a series of decision variables.

[0130] Step 2: Perform non-dominated sorting;

[0131] Individuals in the population are sorted according to their non-dominated relationships and divided into different levels. The first level contains non-dominated solutions on the Pareto front, and subsequent levels contain solutions dominated by the previous level. The crowding distance of each solution is calculated to measure its distribution density.

[0132] Step 3: Execute the selection;

[0133] Solutions are selected from non-dominated levels to form the next generation population, with the priority being to retain all solutions from the previous levels to maintain population quality. Within the same level, partial solutions are selected based on the crowding distance to promote population diversity.

[0134] Step 4: Apply genetic operations;

[0135] The crossover and mutation operations are used to generate a new generation of solutions. The crossover operation combines two parent solutions to generate a new child solution, while mutation creates diversity by introducing slight random changes to the decision variables of the individuals.

[0136] Step 5: Update population status;

[0137] The newly generated solution is merged with the previous generation population to form a new population of size 2N. If the size of the new population exceeds the predetermined value, the selection mechanism is used to filter out the optimal solution.

[0138] Step 6: Repeat the iteration;

[0139] Check whether the preset termination conditions, such as the maximum number of iterations or a specific convergence criterion, have been met. If not, return to step 2 to continue the evolution process; if so, output the results, i.e., a set of high-quality solutions on the Pareto front.

[0140] The corresponding Pareto front forms an n-dimensional surface, on which a set of n-dimensional data points that are easiest to implement in engineering in principle are selected as the energy consumption plan adjustment for quadratic programming.

[0141] Therefore, for each set of feature maps and the corresponding set of park data, the corresponding solution strategy set can be continuously calculated on a 15-minute time scale within a day, and finally a strategy set corresponding to the feature set is formed.

[0142] 3.1 Feature-Strategy Driven Matching:

[0143] First, the historical data of each park can be obtained to generate the corresponding forecast data. The directly generated data is generated by the same operation as the above step 2 to generate the feature map. The new feature map is generated. Should be consistent with all the features in the set Match and select the most similar group. The selection basis of feature map is

[0144] ;

[0145] Get the corresponding closest feature map and execute the full-period strategy (96 time points in 24 hours) for solving the corresponding feature map, taking into account that the generated feature map may match all the previously generated feature maps, that is:

[0146] ;

[0147] Where a is the minimum feature ratio median. If it is greater than this value, real-time rolling optimization is performed and the new feature map is placed in the feature library.

[0148] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

[0149] A computer-readable storage medium storing one or more programs, wherein the one or more programs include instructions, which, when executed by a computing device, enable the computing device to perform a multi-park carbon emission optimization and control method based on channel attention.

[0150] A computing device includes one or more processors, one or more memories, and one or more programs, wherein the one or more programs are stored in the one or more memories and are configured to be executed by the one or more processors, and the one or more programs include instructions for executing a multi-park carbon emission optimization and control method based on channel attention.

[0151] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0152] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts 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, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0153] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0154] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0155] 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 in the scope of the claims of the present invention to be approved.

Claims

1. A multi-park carbon emission optimization and control method based on channel attention, characterized by: Obtain carbon emissions and energy consumption data for multiple parks over multiple days; With the goal of minimizing carbon emissions, we will build control plans for multiple parks that take into account carbon emissions and energy consumption. Using the improved D-SENet to capture features of the control plan to obtain a feature set; Based on the improved NSGA-III algorithm, the control plan is collaboratively optimized at both the energy consumption and carbon emission levels according to the feature set.

2. The multi-park carbon emission optimization and control method based on channel attention according to claim 1 is characterized by: With the goal of minimizing carbon emissions, methods for building multiple park control plans that consider carbon emissions and energy consumption include: Assume that the carbon energy function of a single park is: ; The slight increase rate indicator of a single park is : ; It is believed that when the micro-increase rate indicators of each park are equal, the park group has reached the corresponding carbon saving state: ； Satisfying the following constraints The energy consumption It is possible to: ； in, is the carbon emissions of the i-th park, are the polynomial coefficients, obtained from data fitting. is the differential operator, is the carbon emission slight increase rate of the park with serial number j, The carbon emission slight increase rate of the park with serial number k; It shows the proportional relationship between energy consumption and carbon emissions on the grid side.

3. The multi-park carbon emission optimization and control method based on channel attention according to claim 2 is characterized by: Use the improved D-SENet to capture the features of the control plan. Methods for obtaining feature sets include: Set control interval constraints; Integrate data items including new energy; Build interactive channels for different data; The data items are reorganized through the interactive channel and integrated to obtain corresponding feature sets.

4. The multi-park carbon emission optimization and control method based on channel attention according to claim 3 is characterized by: The method to set the control interval constraint is: Divide a single day into 96 time periods. , the park number is ,, the actual energy consumption of the park in a single day is expressed as , the actual instantaneous power of the park after regulation is , its range is: .

5. The multi-park carbon emission optimization and control method based on channel attention according to claim 4 is characterized by: The method for integrating data items containing new energy is: Select all data for several days in a year and 96 time periods within a day. The following data list can be obtained for 300 different time periods in a park within a day. The serial number of the load node in the park ; There is a data matrix for each energy unit of carbon emissions ;Energy input and output have data matrix for each energy unit ; ； ； Generate the corresponding convolution kernel , Element { }Generated by random numbers satisfying the two-dimensional normal distribution, where Perform convolution transformation: ； ; in, , ,Convolution transformation is performed on the data collected from each park.

6. The multi-park carbon emission optimization and control method based on channel attention according to claim 4 is characterized by: The method of building interactive channels for different data is: Construct two fully connected layers for perceptual interaction. The first fully connected layer uses the ReLU function Activation, the second fully connected layer uses the Sigmod function Activation, the weight matrix used for coordination is , , output the corresponding data graph weight value after perceptual interaction, and the value initialization principle in the weight matrix is ​​the same as the convolution kernel; ; Where s is the output channel-specific score after interactive processing.

7. The multi-park carbon emission optimization and control method based on channel attention according to claim 6 is characterized by: The data items are reorganized through the interactive channel and the method of integrating to obtain the corresponding feature set is as follows: The corresponding score Data feature map after multiplication and output scoring : ； The corresponding data graph The corresponding score Data feature map after multiplication and output scoring : ； For each park’s energy consumption and carbon emissions on a certain day, a corresponding characteristic graph is generated. ;After several days of operation, the corresponding feature set was integrated.

8. The multi-park carbon emission optimization and control method based on channel attention according to claim 7 is characterized by: The method for collaboratively optimizing the control plan at both energy consumption and carbon emissions levels based on the feature set based on the improved NSGA-III algorithm includes: For any set of determined feature maps , , Each represents a corresponding park data, and secondary upper-layer optimization is performed based on the data of different parks: The control object of the secondary upper-level optimization of the inter-park collaboration is the energy consumption of the park. The optimization goal is: ; The constraints are: ； in, is the weight coefficient of the overall energy consumption of the park, is the flat weight coefficient of inter-park power distribution, is the weight coefficient of the park’s energy consumption from the grid side, is the utilization rate coefficient of the park’s new energy equipment, is the coefficient of power interaction between parks, is the weight coefficient of the park’s overall carbon emissions, is the weight coefficient of the park’s planned carbon emissions to actual carbon emissions, For the power interaction of the park from the grid side, The output of new energy in the park. is the power interaction between parks, , The upper and lower limits of the power interaction between the park and the grid side, is the upper limit of power interaction between parks, , The upper and lower limits of new energy equipment output.

9. The multi-park carbon emission optimization and control method based on channel attention according to claim 8 is characterized by: The method for collaboratively optimizing the control plan at both energy consumption and carbon emissions levels based on the feature set based on the improved NSGA-III algorithm also includes: Each time, the NSGA-III algorithm is used to solve the optimization problem based on the actual constraints of the previous 15 minutes, the actual output of renewable energy, and the power interaction between the park and the grid. This allows the park to achieve the optimal coordination at the park level in terms of energy consumption and carbon emissions. The solution steps are as follows: Step 1: Randomly generate an initial population consisting of multiple individuals, each of which consists of a series of decision variables; Step 2: Sort the individuals in the population by non-domination and divide them into different levels. The first level contains the non-dominated solutions on the Pareto front, and the subsequent levels contain the solutions dominated by the previous level. The crowding distance of each solution is calculated to measure its distribution density. Step 3: Select solutions from the non-dominated levels to form the next generation population, giving priority to retaining all solutions in the previous levels to maintain population quality; within the same level, select some solutions based on the crowding distance to promote population diversity; Step 4: Generate a new generation of solutions using crossover and mutation operations; crossover combines two parent solutions to generate a new child solution, while mutation creates diversity by introducing slight random changes to the individual decision variables; Step 5: Merge the newly generated solution with the previous generation population to form a new population of size 2N; if the size of the new population exceeds the predetermined value, use the selection mechanism to filter out the optimal solution; Step 6: Check whether the preset termination condition, such as the maximum number of iterations or a specific convergence criterion, has been met. If not, return to step 2 to continue the evolution process. If so, output the results, i.e., a set of high-quality solutions on the Pareto frontier. The corresponding Pareto front forms an n-dimensional surface, on which a set of n-dimensional data points that are easiest to implement in principle are selected as the energy consumption plan adjustment for quadratic programming; For each set of feature maps and the corresponding set of park data, the corresponding solution strategy set is continuously calculated on a 15-minute time scale within a day, and finally a strategy set corresponding to the feature set is formed.

10. The multi-park carbon emission optimization and control method based on channel attention according to claim 9 is characterized in that: The method for collaboratively optimizing the control plan at both energy consumption and carbon emissions levels based on the feature set based on the improved NSGA-III algorithm also includes: First, the historical data of each park is obtained to generate the corresponding prediction data. The directly generated data is the feature map generated by the operation in step 2. The new feature map is generated. Should be consistent with all the features in the set The most similar group is selected for matching. The selection basis of feature maps is: ； Get the corresponding closest feature map and execute the strategy for 96 time periods of 24 hours corresponding to the feature map solution, taking into account that the generated feature map matches all the previously generated feature maps: ； Where a is the minimum feature ratio median. If it is greater than this value, real-time rolling optimization is performed and the new feature map is placed in the feature library.

Images