Comprehensive energy system scheduling method and device based on particle swarm optimization

Abstract

The invention discloses an integrated energy system scheduling method and device based on a particle swarm algorithm, and relates to the technical field of energy and communication systems. According to the method, optimal scheduling can be carried out under a relatively small time scale of the IES, and the uncertainty and challenge of renewable energy power generation and load requirements are solved. While a stepped carbon transaction mechanism is considered, the accuracy and availability of multi-time scale scheduling are improved, and a promising solution is provided for low-carbon economic scheduling. Meanwhile, parameters in an existing multi-time-scale multi-objective optimization low-carbon economic dispatching model solving algorithm are determined through empirical evaluation and need to be manually and repeatedly adjusted, the timeliness of the algorithm is reduced, and the obtained dispatching result has a lifting space. The particle swarm parameter optimization method based on reinforcement learning can objectively give parameter initial values, and the solving speed of a scheduling strategy is improved.

Description

Technical Field

[0001] The present invention discloses a comprehensive energy system scheduling method and device based on a particle swarm optimization algorithm, which relates to the technical field of energy and communication systems. Background Art

[0002] The world is witnessing a strong momentum towards a low-carbon and sustainable energy system, and countries around the world have adopted various policy tools to encourage clean energy technologies and reduce greenhouse gas emissions. In recent years, China has achieved historic achievements in green and low-carbon development. The pace of the green transformation of energy has accelerated. As of the end of June 2024, the installed capacity of renewable energy reached 1.653 billion kilowatts, accounting for 53.8% of the total installed capacity; the industrial structure has been continuously optimized and upgraded, and the world's largest and most complete new energy industrial chain has been built; the resource utilization efficiency has been continuously improved. In 2023, China's energy consumption per unit of GDP and carbon emission intensity decreased by more than 26% and 35% respectively compared with 2012, and the main resource output rate increased by more than 60%. One of the policy tools is the ladder-type carbon tax (LTCT) combined with the carbon quota market, which is introduced to promote the integration of renewable energy and optimize energy utilization. The ladder carbon tax is a mechanism that sets different prices for carbon trading according to different carbon emission intervals. Carbon tax involves the government charging emitters according to the amount of greenhouse gas emissions per ton generated by the emitters, which prompts enterprises and individuals to take measures such as adopting alternative energy sources or implementing new technologies to reduce emissions to avoid additional costs. The integrated energy system (IES) seamlessly integrates renewable energy into the traditional energy structure and is an ideal way to improve energy utilization efficiency and reduce carbon emissions.

[0003] However, in the carbon quota market, due to the volatility and uncertainty of new energy power generation and load, it is currently difficult to accurately predict and control carbon trading costs. Due to the inherent volatility and uncertainty associated with new energy power generation output and load distribution, it is difficult to accurately predict and control carbon trading costs in the multi-time scale scheduling of IES. Since the LTCT mechanism in the carbon quota market represents a relatively macroscopic large time scale concept, there is currently a lack of in-depth research on dynamic adjustment and practical implementation at a smaller time scale. Innovative technologies need to be developed to solve the problems related to IES scheduling. Summary of the Invention

[0004] In view of the problems existing in the prior art, the present invention provides a comprehensive energy system scheduling method and device based on a particle swarm optimization algorithm, and the technical solutions adopted are as follows:

[0005] In a first aspect, a comprehensive energy system scheduling method based on a particle swarm optimization algorithm, the method includes:

[0006] S1. Based on the integrated energy system, collect the output of new energy units, the output of gas turbines, the network load, and the parameters of energy storage devices as the initial system data;

[0007] S2. Based on the initial data, establish a scheduling model for the collaborative optimization of the source, load, and storage in the park;

[0008] S3. Based on the output characteristics of each unit, the load demand, and the peak shaving and valley filling functions of the energy storage device, construct the constraint conditions of the scheduling model;

[0009] S4. Develop a scheduling strategy according to the constraint conditions, solve the scheduling model through the scheduling strategy, and obtain an optimized economic scheduling plan for multi-time scale and multi-objective optimization.

[0010] In some implementation manners, in S1, the new energy machines include photovoltaic generator sets and wind turbine generator sets; the parameters of the energy storage device include the capacity, charge and discharge efficiency, and state of charge of the energy storage device.

[0011] In some implementation manners, in S3, the constraint conditions include wind turbine output constraints, photovoltaic unit output constraints, gas turbine ramp constraints, gas turbine output constraints, energy storage device charge and discharge power constraints, energy storage device capacity constraints, energy storage device state of charge constraints, energy storage device charge and discharge conservation constraints, daily total power conservation constraints of the energy storage device, and interaction power constraints between the park power grid and the external network.

[0012] In some implementation manners, S4 specifically includes:

[0013] S41. Generate an initial particle population according to the system initial data;

[0014] S42. According to the initial particle population, set the initial position of the particles and evaluate the particles;

[0015] S43. Save the evaluated non-dominated solutions to the spare set, solve the initial parameter values through a multi-objective optimization algorithm based on reinforcement learning to improve the particle swarm, and perform dynamic non-linear adjustment;

[0016] S44. According to the best positions of the particles in the population, update the particle positions and calculate the fitness;

[0017] S45. According to the results of the fitness calculation, determine whether the particles need to update their positions and update the best positions of the particles;

[0018] S46. Save the non-dominated solutions to the spare set and generate particles at sparse positions;

[0019] S47. Repeat steps S43 - S46 until the set termination condition is met. Select the individual with the highest fitness value in the final population as the optimal solution, which serves as the optimal scheduling plan for the park.

[0020] In a second aspect, an embodiment of the present invention provides a scheduling device for an integrated energy system based on a particle swarm algorithm. The device includes:

[0021] An initialization module for collecting the output of new energy units, the output of gas turbines, network loads, and energy storage device parameters as system initial data according to the integrated energy system;

[0022] A model establishment module for establishing a scheduling model for the collaborative optimization of source - load - storage in the park based on the initial data;

[0023] A model adjustment module for constructing the constraint conditions of the scheduling model according to the output characteristics of each unit, load demand, and the peak - shaving and valley - filling function of the energy storage device;

[0024] A model optimization module for formulating a scheduling strategy according to the constraint conditions, solving the scheduling model through the scheduling strategy, and obtaining an optimized economic scheduling plan for multi - time - scale and multi - objective optimization.

[0025] In some implementation manners, in the initialization module, the new energy units include photovoltaic power generation units and wind power generation units; the energy storage device parameters include the capacity, charge - discharge efficiency, and state of charge of the energy storage device.

[0026] In some implementation manners, in the model adjustment module, the constraint conditions include wind turbine output constraints, photovoltaic unit output constraints, gas turbine ramp - up constraints, gas turbine output constraints, energy storage device charge - discharge power constraints, energy storage device capacity constraints, energy storage device state - of - charge constraints, energy storage device charge - discharge conservation constraints, daily total power conservation constraints of the energy storage device, and interaction power constraints between the park power grid and the external network.

[0027] In some implementation manners, the model optimization module specifically includes:

[0028] A subgroup unit for generating an initial particle population according to the system initial data;

[0029] An evaluation unit for evaluating the particles by setting the initial positions of the particles according to the initial particle population;

[0030] An adjustment unit for saving the evaluated non - dominated solutions to a backup set, solving the initial parameter values through a multi - objective optimization algorithm for improving the particle swarm based on reinforcement learning, and performing dynamic non - linear adjustment;

[0031] A position unit, configured to update the particle position and calculate the fitness according to the best position of the particles in the population;

[0032] An update unit, configured to determine whether the particle needs to update its position according to the result of the fitness calculation, and update the best position of the particle;

[0033] A generation unit, configured to save the non-dominated solutions to a standby set and generate particles at sparse positions;

[0034] A loop unit, configured to repeat the operations between the adjustment unit and the generation unit until a set termination condition is met, and select the individual with the highest fitness value in the final population as the optimal solution, which is used as the optimal scheduling scheme for the park.

[0035] In a third aspect, an embodiment of the present invention provides an electronic device, including a memory and a processor, where the memory is used to store one or more computer instructions, and when the one or more computer instructions are executed by the processor, the method described in the first aspect above is implemented.

[0036] In a fourth aspect, an embodiment of the present invention provides a computer storage medium, where a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, the method described in the first aspect is implemented.

[0037] Compared with the prior art, the present invention has the following beneficial effects: The method of the present invention can optimize the scheduling for a smaller time scale of IES, effectively solve the uncertainty and challenges of renewable energy generation and load demand. While considering the stepped carbon trading mechanism, it improves the accuracy and availability of multi-time scale scheduling, providing a promising solution for low-carbon economic scheduling. At the same time, the parameters (such as inertia weight, learning factor, etc.) in the current multi-time scale multi-objective optimization low-carbon economic scheduling model solving algorithm are determined by empirical evaluation, which requires manual adjustment repeatedly, reducing the timeliness of the algorithm, and there is room for improvement in the obtained scheduling results. The proposed particle swarm parameter optimization method based on reinforcement learning can objectively give the initial values of the parameters, improving the solution speed of the scheduling strategy. The proposed improved particle swarm algorithm uses an update formula for calculating the inertia weight and learning factor, which can overcome the problem that it is difficult to balance the convergence speed and convergence accuracy and is prone to falling into local optima. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0039] Figure 1 It is a schematic diagram of the physical model of a multi - objective integrated energy system for constructing a ladder - type carbon trading mechanism provided by an embodiment of the present invention.

[0040] Figure 2 It is a framework diagram of the multi - time - scale schedule arrangement of an integrated energy system provided by an embodiment of the present invention;

[0041] Figure 3 It is a flowchart of the implementation of a multi - time - scale scheduling strategy provided by an embodiment of the present invention;

[0042] Figure 4 It is a structural diagram of the action network of a multi - objective optimization algorithm based on the improved particle swarm by reinforcement learning provided by an embodiment of the present invention;

[0043] Figure 5 It is a flowchart of the scheduling algorithm of a multi - objective optimization algorithm based on the improved particle swarm by reinforcement learning provided by an embodiment of the present invention. Detailed implementation manners

[0044] It should be noted that the methods used in the present invention are all conventional methods without special regulations; the raw materials and devices used are all conventional commercially available products without special regulations, and their sources are not specifically limited.

[0045] Embodiment 1:

[0046] Figure 1 It shows a schematic diagram of the physical model of a multi - objective integrated energy system for constructing a ladder - type carbon trading mechanism. As Figure 1 shown, the integrated energy system scheduling method based on the particle swarm algorithm provided in this embodiment includes:

[0047] S1. According to the integrated energy system, collect the output of new - energy units, the output of gas turbines, network loads, and energy - storage device parameters as the initial data of the system;

[0048] S2. According to the initial data, establish a scheduling model for the coordinated optimization of the source - load - storage in the park;

[0049] S3. According to the output characteristics of each unit, load demand, and the peak - shaving and valley - filling functions of the energy - storage device, construct the constraint conditions of the scheduling model;

[0050] S4. According to the constraint conditions, formulate a scheduling strategy, and solve the scheduling model through the scheduling strategy to obtain an optimized economic scheduling plan for multi - time - scale and multi - objective optimization.

[0051] First, as described in S1, the new energy machine includes a photovoltaic power generation unit and a wind power generation unit; the parameters of the energy storage device include the capacity, charge-discharge efficiency, and state of charge of the energy storage device.

[0052] Next, according to S2, the objective function expression with the lowest comprehensive operating cost as the optimization goal is as follows:

[0053] Objective function for day-ahead scheduling:

[0054]

[0055] Objective function for intraday scheduling:

[0056]

[0057] Objective function for real-time scheduling:

[0058]

[0059] In the formula, T 1 , T 2 , T 3 are the cycle numbers in the day-ahead, intraday, and real-time scheduling stages respectively, and β k represents the weight coefficient corresponding to the kth type of polluting gas in the emissions. M represents the total number of polluting gases, and λ i,k is the emission coefficient of the kth gas released from the set Θ. are the electricity purchase and gas purchase price coefficients respectively. are the electricity purchase and gas purchase powers respectively. C LT_in is the stepped carbon trading cost for the day's scheduling, and C LT_rt is the stepped carbon trading cost for real-time scheduling.

[0060] The expression of the stepped carbon trading cost function is as follows:

[0061]

[0062] In the formula, T c is the carbon trading cost coefficient; α is the cost growth coefficient; l is the length of the carbon emission interval.

[0063] Next, according to S3, the constraint conditions include wind turbine output constraint, photovoltaic unit output constraint, gas turbine ramp constraint, gas turbine output constraint, energy storage device charge-discharge power constraint, energy storage device capacity constraint, energy storage device state of charge constraint, energy storage device charge-discharge conservation constraint, energy storage device daily total power conservation constraint, and park power grid and external network interaction power constraint;

[0064] Power balance:

[0065] P PV (t)+PWT (t) + P grid (t) + P DE (t) + P FC (t) + P GT (t) + P BESS (t) = P L (t) + P EL (t) + P EB (t) + P EC (t)

[0066] Wherein, P PV (t), P WT (t), P grid (t), P DE (t), P FC (t), P GT (t), P BESS (t) are respectively the photovoltaic power generation power, wind power generation power, grid power purchase power, diesel power generation power, fuel cell power, gas turbine power, and electrical energy storage power; P L (t), P EV (t), P EL (t), P EB (t), P EC (t) are respectively the load electrical power, electrolyzer electrical power, electric boiler electrical power, and electric refrigeration equipment electrical power.

[0067] Thermal balance:

[0068] P EB (t) + P GB (t) + P HESS (t) = P Heat (t)

[0069] Wherein, P EB (t), P GB (t), P HESS (t) are respectively the thermal power of the electric boiler, the thermal power of the gas boiler, and the thermal energy storage power.

[0070] Cold energy balance:

[0071] P AC (t) + P EC (t) = P Cool (t)

[0072] Wherein, P AC (t), P EC (t) are respectively the absorption chiller power and the electric chiller power; P Cool (t) is the cooling load.

[0073] The formulated scheduling strategy includes: obtaining the system network load power demand P1(t) and the total power generation P2(t) of each unit at the current time period t; if P1(t) > P2(t), query the storage capacity of the energy storage device at the current time period t. If the energy storage is sufficient, all deficits are supplemented by the energy storage device. Otherwise, the energy storage device supplies energy until the lower limit of the capacity, and at the same time purchases electricity from the external power grid to fill the remaining deficit power. If P1(t) < P2(t), query the storage capacity of the energy storage device at the current time period t. If the energy storage device can accommodate all the surplus power, the surplus power is consumed by the energy storage device. Otherwise, the energy storage device stores energy until the upper limit of the capacity, and at the same time sells the remaining surplus power to the external power grid.

[0074] Further, the update formulas for the inertia weight and learning factor are as follows:

[0075] The improved inertia weight and learning factor are as follows:

[0076]

[0077] Make the initial inertia weight become the maximum inertia weight, where m 2 Should be expressed as:

[0078]

[0079] Where m 1 Is the control factor, k 1 And k 2 Are two constants, t is the current iteration, and T is the total number of iterations; And Are the upper and lower limits of the inertia weight, as well as the initial and final values of the inertia weight factor. And Are the upper and lower limits of the individual learning acceleration factor, as well as the initial and final values of c 1 ;

[0080] Further, when updating the velocity and position of each particle, the following update formula is adopted:

[0081]

[0082] In the formula, And Are random numbers between 0 and 1. And Are learning factors. And Are respectively the best position in the current k-th iteration and the best position of the particles in the entire particle swarm. According to the chaotic decreasing inertia weight method, the inertia weight ω k Changes with the iteration index k.

[0083] In some implementations, in S3, the constraint conditions include the output constraint of the wind turbine generator set, the output constraint of the photovoltaic generator set, the ramp constraint of the gas turbine, the output constraint of the gas turbine, the charge-discharge power constraint of the energy storage device, the capacity constraint of the energy storage device, the state of charge constraint of the energy storage device, the charge-discharge conservation constraint of the energy storage device, the daily total power conservation constraint of the energy storage device, and the interactive power constraint between the park power grid and the external network.

[0084] Next, according to S4, a solution algorithm based on the multi-objective optimization scheduling model of the improved particle swarm is used for solution.

[0085] S41, Generate an initial particle population with the input initial data. In the population, each particle individual represents a unique scheduling scheme. These schemes specifically cover the output of the new energy generator sets, the operating power of the gas turbine, the load demand of the park network, and the charge-discharge strategies of the energy storage devices.

[0086] S42, Set the initial positions of the particles and evaluate the particles.

[0087] S43, Save the evaluated non-dominated solutions to the spare set, and use the multi-objective optimization algorithm based on reinforcement learning to improve the particle swarm (MORLPSO) to solve the initial values of parameters such as the inertia weight and learning factor.

[0088] Perform dynamic non-linear adjustment on the inertia weight and learning factor

[0089] S44, Select the best positions of the particles in the population, update the particle positions and calculate the fitness.

[0090] S45, Determine whether the particles need to be updated in position and update the best positions of the particles.

[0091] S46, Save the non-dominated solutions to the spare set and generate particles at sparse positions.

[0092] S47, Repeat S43 to S46 until the set termination condition is met, and select the individual with the highest fitness value in the final population as the optimal solution, that is, obtain the optimal scheduling scheme for the park.

[0093] Among them, in S431, use the particle swarm parameter optimization method based on reinforcement learning to solve the initial values of the parameters, specifically including:

[0094] S431, Random initialization. First, randomly initialize the actor network and the action value network, and copy the target actor network and the target action value network.

[0095] S432, Initialize the buffer R. Initialize the buffer R to save the running state, actions, and rewards.

[0096] S433. Environment optimization, including the initialization of the particle swarm algorithm and the initialization of the objective function to be optimized.

[0097] S434. Obtain the running state parameters from the particle swarm.

[0098] S435. Input the state to obtain an action; input the running state s t into the actor network to get the action a t , and add a certain amount of random noise (the noise is used to help the network explore the policy space and will be removed after training). The random noise follows a normal distribution with a mean of 0 and a variance of 0.5. The formula is: a t = Actor + Ν(0, 0.5)

[0099] S436. Convert the action into parameter adjustment; use an equation to convert the resulting action into the required ω, c 1 and c 2 .

[0100] After training, obtain a trained action network model for subsequent runs.

[0101] In S436, the actions designed in this paper are used to control the operation parameters ω, c 1 and c 2 in the PSO. The parameters to be controlled can be set as needed. The actions output by the action network do not individually control the above parameters of each particle, but divide the particles into five subgroups and generate five sets of parameters to control different groups. We determine the corresponding subgroup according to the index of the particle. For example, if the index of the particle is 7, the corresponding subgroup is 7 % 5 = 2, that is, the subgroup is 2. The indices of the subgroups are 0, 1, 2, 3, and 4. This method ensures that the number of individuals in each subgroup is similar. For the traditional particle swarm algorithm, the resulting action vector is 20 - dimensional and divided into 5 groups, each group targeting a subgroup, as Figure 4 shown. For a subgroup, the action vector is four - dimensional: a[0] to a[3]. The ω, c 1 and c 2 required for each round of the optimization algorithm are generated based on a[0] to a[3] as follows:

[0102]

[0103] where scale is a parameter to help normalize c 1 and c 2 , and this is optional. An example can be seen in Figure 4 . To utilize the performance of the original parameter set, configure the new ω, c 1 , c 2 as follows:

[0104]

[0105] where ω origin , c 1origin and c 2origin represent the original parameters of the algorithm. If there are more parameters to be configured in the algorithm, the number of output parameters can be increased and configured like c 1 or ω.

[0106] S437, iteratively obtain new states and rewards; perform the iteration of PSO using the above parameters to obtain a new reward r t+1 and a new state s t+1 .

[0107] The reward function is used to calculate the reward after performing an action. Its goal is to motivate the particle swarm algorithm to obtain a better gBest(t). Therefore, the reward function is designed as follows:

[0108]

[0109] where gBest(t) is the best solution in the t-th iteration.

[0110] S438, save the state, action, and reward; save the state, action, and reward to buffer R.

[0111] S439, draw data from the buffer experience pool for gradient update; randomly select a batch of experiences from buffer R. Update the weight parameters of the action value network by minimizing the loss function.

[0112] S4310, update the weights of the target actor value network and the target actor network; update the weights of the target action-value network and the target actor network. If the particle swarm has not completed the iteration, increment t by 1 and return the process to S434. If the training is not completed, return the process to S433.

[0113] Among them, the formula for minimizing the loss function is as follows:

[0114] L(θ Q ) = (r(s t , a t ) + γQ'(s t+1 , a t+1 |θ Q′ ) - Q(s t , a t |θ Q )) 2

[0115] The formula for updating the weights of the action value network is as follows:

[0116]

[0117] In the formula, μ is the actor network, Q is the actor value network, and Q' is the actor target value network. θ μ 、θ μ' 、θ Q and θ Q′ are the weights of these neural networks.

[0118] The solution of the model established in the above steps is for the multi-objective optimization problem of multi-dimensional non-linear complex large systems. Traditional multi-objective optimization algorithms are no longer applicable, and the solution is difficult. It is proposed to use the improved MORLPSO algorithm for solution. In the daily scheduling, the comprehensive energy optimization scheduling problem involves high-dimensional non-linearity and multiple constraints. The particle swarm optimization (PSO) algorithm has the advantages of easy implementation, high computational efficiency, and robust optimization ability. Due to the influence of parameters, the inertia weight and learning factor of the traditional particle swarm optimization algorithm are fixed, resulting in a slow convergence speed and easy to fall into local optimum. The proposed algorithm optimizes the inertia weight and learning factor of the particle swarm algorithm, improves the inertia weight and acceleration factor of the traditional multi-objective particle swarm optimization algorithm (MOPSO), solves the problem that the traditional particle swarm algorithm is easy to fall into local optimum, and has the advantages of stronger global search ability, higher accuracy, better flexibility and adaptability. The MOPSO algorithm based on dynamic adaptive variable parameters is used to solve the scheduling model, and its process is as Figure 5 shown, which specifically includes the following steps:

[0119] Step 1: Generate an initial particle population with the input initial data. In the population, each particle represents a unique scheduling scheme. These schemes specifically cover the output of new energy units, the operating power of gas turbines, the load demand of the park network, and the charge and discharge strategies of energy storage devices.

[0120] Step 2: Set the initial position of the particles and evaluate the particles.

[0121] Step 3: Save the evaluated non-dominated solutions to the spare set, and use MORLPSO to solve the initial values of parameters such as inertia weight and learning factor.

[0122] Step 4: Perform dynamic non-linear adjustment on the inertia weight and learning factor

[0123] Step 5: Select the best position of the particles in the population, update the particle positions and calculate the fitness.

[0124] Step 6: Determine whether the particles need to update their positions and update the best positions of the particles.

[0125] Step 7: Save the non-dominated solutions to the spare set and generate particles at sparse positions.

[0126] Step 8: Repeat steps 4 to 7 until the set termination condition is met. Select the individual with the highest fitness value in the final population

[0127] as the optimal solution, i.e., obtain the optimal scheduling plan for the park.

[0128] The traditional multi-objective particle swarm optimization algorithm (MOPSO) is a population-based optimization algorithm mainly used to solve multi-objective optimization problems. It extends the traditional single-objective PSO algorithm. In single-objective PSO, each particle optimizes only one objective function, while in MOPSO, each particle optimizes multiple objective functions. Therefore, each particle maintains multi-dimensional position and velocity vectors for optimization across multiple objective functions. In the traditional multi-objective particle swarm algorithm, each particle is regarded as a candidate solution. The position vector of each particle represents the performance of the candidate solution in each objective function, while the velocity vector represents the direction and speed of the candidate solution's movement. Based on the position and velocity vectors of each particle, the fitness value of each particle in multiple objective functions can be calculated, and the position and velocity of each particle can be updated accordingly. The position and velocity update of each particle are as follows:

[0129]

[0130] where and are random numbers between 0 and 1. and are learning factors. and are the best position in the current k-th iteration and the best position of the particles in the entire particle swarm, respectively. According to the chaotic descent inertia weight method, the inertia weight ω k varies with the iteration index k.

[0131] Due to the influence of parameters, the inertia weight and learning factors in the traditional particle swarm optimization algorithm are fixed, resulting in a slow convergence speed and being prone to falling into local optima. The inertia weight determines the movement speed and direction of the particle in the search space, while the learning factor controls the importance of the particle in individual learning and group learning. The improved particle swarm algorithm adopts adaptive weight adjustment. The inertia weight and learning factor are improved as follows:

[0132]

[0133] Make the initial inertia weight the maximum inertia weight, where m 2 should be expressed as

[0134]

[0135]

[0136] where m 1 is the control factor, k 1 and k 2 are two constants, t is the current iteration, and T is the total number of iterations; and are the upper and lower limits of the inertia weight, as well as the initial and final values of the inertia weight factor. In the early stage of iteration, to prevent the algorithm from falling into a local minimum, the inertia weight factor can be set to a relatively large value, making it less likely to fall into a local minimum and facilitating global search. In the later stage of iteration, the inertia weight factor gradually decreases until it reaches a relatively small value, which is beneficial to the local search and convergence of the algorithm. The inertia weight change usually takes three forms: fixed weight, linearly varying weight, and non-linearly varying weight. The arctangent function is selected to adjust the inertia weight. As the number of iterations increases, the inertia weight gradually decreases from the initial value to the final value, realizing non-linear dynamic adjustment. and are the upper and lower limits of the individual learning acceleration factor, as well as the initial and final values of c 1 ;

[0137] and are the upper and lower limits of the global learning acceleration factor, as well as the final and initial values of c 2 . In the early stage of iteration, the particle velocity needs to increase to avoid falling into a local optimum. A larger c 1 endows the particle with better self-learning ability, while a smaller c 2 endows the particle with poorer swarm learning ability, which is beneficial to global search. In the later stage of iteration, the particle velocity needs to decrease to prevent premature departure from the best region. Using a smaller c 1 and a larger c 2 , the particle has stronger swarm learning ability and poorer self-learning ability, making the algorithm easier to converge. The inertia weight only affects a part of the particle velocity update. Adjusting the inertia weight non-linearly alone cannot fully improve the performance of the MOPSO algorithm. In addition to adjusting the inertia weight, the individual optimal acceleration factor c 1 and the global optimal acceleration factor c 2 also have a significant impact on the performance of the algorithm. Using the exponential function as the basis for the non-linear time-varying dynamic adjustment of c 1 and c 2 to obtain better multi-objective optimization results.

[0138] Example 2:

[0139] Based on Embodiment 1, a physical model of a multi-objective integrated energy system considering a stepped carbon trading mechanism is constructed. The specific implementation is as follows:

[0140] In the integrated energy system scheduling model of the present invention, the electricity, heat, cold, and gas loads of users are mainly considered. The structure includes a combined cooling, heat, and power subsystem composed of a gas turbine (GT), a gas boiler (GB), and an absorption chiller, a power generation subsystem composed of a wind turbine (WT) and a photovoltaic generator (PV), an energy storage subsystem including power and heat storage devices, and an energy storage subsystem composed of a generator and a heat storage device. And an electrolysis process, which mainly involves the production of hydrogen by electrolyzing water through an electrolyzer. In addition, it also includes a process in which hydrogen is used as fuel to generate electricity through a fuel cell.

[0141] First, a carbon quota model is constructed:

[0142] There are three different types of carbon sources in the constructed model: electricity purchased from the grid, gas boilers, and combined cooling, heat, and power systems.

[0143]

[0144] In the formula, E IES , E e,buy , E CCHP and E GB are the carbon emission quotas of the integrated energy system, purchased electricity, combined cooling, heat, and power, and gas boilers respectively; χ e , χ g are the carbon emission quotas per unit coal consumption of coal-fired units and the carbon emission quotas per unit gas consumption of gas units respectively; P e,buy (t) is the purchased electricity at time period t; P CCHP,e (t), P CCHP,h (t), P CCHP,C (t) are the electric power, heat power, and cooling power output by the combined cooling, heat, and power system at time period t respectively. P GB,h (t) is the heat energy output of the gas boiler in time slot t; T is the scheduling period.

[0145] Next, a stepped carbon emission trading model is constructed:

[0146] The actual carbon amount participating in the carbon trading market is equal to the actual carbon emissions minus the carbon quota of the integrated energy system.

[0147] E IES,t = E IES,a - E IES

[0148] In the formula, E IES,t is the actual carbon amount participating in the carbon trading market at time t, and E IES,a is the actual carbon emissions of the integrated energy system, EIES Carbon quota of integrated energy system

[0149] The actual carbon trading cost is represented by the following piecewise function:

[0150]

[0151] In the formula, T c is the carbon trading cost coefficient; α is the cost growth coefficient; l is the length of the carbon emission range.

[0152] Next, construct a multi-objective integrated energy system cost function model for the stepped carbon trading mechanism:

[0153] In day-ahead scheduling, taking the comprehensive cost of the integrated energy system as the objective function, the comprehensive cost mainly includes two categories: system operation and maintenance cost and environmental protection cost. The system operation cost includes the cost related to interaction with the main power grid, the operation and maintenance costs of gas turbines, gas boilers, and diesel generators, as well as their fuel costs and energy storage maintenance costs. The cost of interacting with the main power grid includes two categories: purchase and sale, which can be expressed as follows:

[0154]

[0155] In the formula, C grid (t) is the cost of interacting with the main power grid; C buy (t) is the purchase cost; C sell (t) is the sale cost; c buy (t), c sell (t) are the purchase and sale cost coefficients respectively; P buy (t), P sell (t) are the purchase and sale powers respectively.

[0156] The environmental protection costs generated by gas turbines and diesel generators include two categories: operation and maintenance costs and fuel costs. They are expressed as follows:

[0157]

[0158] In the formula, C GT (t), C GB (t), C DE (t), are the costs of gas turbines, gas boilers, diesel engines, and other equipment respectively; C GT.om (t), C GB.om (t), C DE.om (t), are the maintenance costs of gas turbines, gas boilers, diesel engines, and other equipment respectively; C GT.om (t), C GB.om (t), C DE.om(t) is the fuel cost of gas turbines, gas boilers, and diesel engines.

[0159] The cost C of interacting with the main grid GRID,EN is expressed as follows:

[0160]

[0161] In the formula, C k is the cost factor for treating the k-th type of pollutant, and γ grid,k is the emission amount of the k-th type of pollutant generated during the interaction with the main grid. P grid,buy (t) is the purchased power. Similarly, C GT,EN (t), C GB,EN (t), C DE,EN (t) can be expressed in the same way.

[0162] The pollutant emission cost is as follows:

[0163]

[0164] In the formula, the scheduling period T = 24, and β k represents the weight coefficient corresponding to the k-th type of polluting gas in the emissions. M represents the total number of polluting gases, and also includes the emission coefficient λ of the k-th gas released by the set Θ when purchasing from the main power grid, GT, GB, and FC i,k .

[0165] Next, construct a multi-objective integrated energy system constraint model for the stepped carbon trading mechanism:

[0166] The uncertainty modeling of the photovoltaic output power, wind power output power, and their source loads in the constraint model is determined by the prediction level and its prediction error. The predicted power output of WT is affected by the wind speed and the rated power:

[0167]

[0168] where v in , v out , v rated and v t represent the cut-in wind speed, cut-out wind speed, rated wind speed, and real-time wind speed respectively; P wt,r represents the rated power of WT. The model of the predicted power output of PV is as follows:

[0169]

[0170] where η pv represents the solar radiation efficiency; S pv represents the solar radiation area; θ tis the light intensity at time period t. Considering the volatility and uncertainty of the output power of renewable energy and the load, we introduce the prediction error as a measure of the deviation between the predicted value and the actual value. The actual output powers on both sides of the power supply and the load can be expressed as:

[0171]

[0172] In the formula, represents the set composed of the predicted values of WT, PV, and the load; represents the set composed of the prediction errors of WT, PV, and the load; P wt,t , P pv,t , P load,t respectively represent the actual values of WT, PV, and the load. Their prediction errors all satisfy the following normal distribution:

[0173]

[0174] In the formula, σ t represents the standard deviation of the slot prediction error; ε t represents the ratio of the deviation of the predictions of WT, PV, and the load to the predicted output at time slot t.

[0175] A diesel generator is a power generation device that generates electrical energy by burning diesel. During operation, it will incur fuel costs, maintenance costs, carbon emissions, and other pollutant disposal costs.

[0176]

[0177] In the formula, C DE.f (t), C DE.om (t), C DE.en (t) respectively represent the operation and maintenance costs, fuel costs, and pollutant treatment costs of the diesel engine (DE) at time slot t. P DE (t) is the power generated by the DE at time slot t; K DE.om is the operation and maintenance cost coefficient of the DE; γ de,k is the emission of the k-th type of pollutant generated during the operation of the DE; C k is the cost coefficient for treating the k-th type of pollutant; α, β, γ are the relevant coefficients of the DE, α = 0.00011, β = 0.1801, γ = 6.

[0178] During the power generation process of a gas turbine (GT), it also incurs fuel costs, maintenance costs, and pollutant disposal costs. Its cost is expressed as follows:

[0179]

[0180] In the formula, C GT.f (t), CGT.om (t) and C GT.en (t) are the diesel cost, operation and maintenance cost, and pollutant treatment cost at time period t respectively; LHV is the calorific value unit, representing the net calorific value of natural gas; C gas represents the cost coefficient for treating carbon emissions from natural gas combustion; P GT (t) is the carbon emission generated by DE at time period t; K om is the operation and maintenance cost coefficient of DE; γ gt,k is the emission of the kth type of pollutant generated by DE during operation; C k is the cost coefficient for treating the kth type of pollutant.

[0181] The dynamic process of the charging state of the energy storage device is expressed as follows:

[0182]

[0183] In the formula, SOC(t) is the battery capacity state at time slot t, and P BESS (t) is the charging and discharging power at time slot t. A positive value indicates the charging state, and a negative value indicates the discharging state.

[0184] Power balance:

[0185] P PV (t) + P WT (t) + P grid (t) + P DE (t) + P FC (t) + P GT (t) + P BESS (t) = P L (t) + P EL (t) + P EB (t) + P EC (t)

[0186] In the formula, P PV (t), P WT (t), P grid (t), P DE (t), P FC (t), P GT (t), P BESS (t) are the photovoltaic power generation power, wind power generation power, power grid purchase power, diesel power generation power, fuel cell power, gas turbine power, and electrical energy storage power respectively; P L (t), P EV (t), P EL (t), P EB (t), P EC (t) are the load electrical power, electrolyzer electrical power, electric boiler electrical power, and electric refrigeration equipment electrical power respectively.

[0187] Thermal balance:

[0188] P EB (t) + P GB (t) + P HESS (t) = P Heat (t)

[0189] Wherein, P EB (t), P GB (t), P HESS (t) are respectively the thermal power of the electric boiler, the thermal power of the gas boiler, and the thermal energy storage power.

[0190] Cold energy balance:

[0191] P AC (t) + P EC (t) = P Cool (t)

[0192] Wherein, P AC (t), P EC (t) are respectively the power of the absorption chiller and the power of the electric chiller; P Cool (t) is the cooling load.

[0193] Based on the physical model, the specific implementation method of the multi-time scale stepped carbon trading rolling coordinated scheduling model is as follows:

[0194] This embodiment proposes a modeling method for long-time delay sequences in an intelligent evolution system under multiple time scales. The scheduling strategy parameters for multiple time scales are set as shown in the following table:

[0195]

[0196] The multi-time scale stepped carbon trading rolling coordinated scheduling model is as Figure 2 shown. As the scheduling period shortens, the accuracy of renewable energy output and load forecasting will become higher and higher. Based on the carbon emission results of the superior scheduling, the carbon trading stepped cost of the inferior scheduling is established, and the optimal solution for the stepped carbon trading energy consumption of the inferior scheduling is determined. Figure 2 Shows the multi-time scale rolling scheduling process, and on this basis, a multi-time scale scheduling strategy model is established; Figure 3 Demonstrates the solution process of the scheduling strategy based on different time scales. By introducing the stepped carbon trading model for different time scales, the system can determine the proportion of carbon quotas and update the interval step length in a timely manner to ensure the economic operation of the IES.

[0197] Embodiment 3:

[0198] An embodiment of the present invention provides a comprehensive energy system scheduling device based on a particle swarm algorithm, and the device includes:

[0199] An initialization module, configured to collect the output of new energy units, the output of gas turbines, network loads, and energy storage device parameters as system initial data according to the comprehensive energy system;

[0200] A model establishment module, configured to establish a scheduling model for collaborative optimization of source, load, and storage in a park according to the initial data;

[0201] A model adjustment module, configured to construct the constraint conditions of the scheduling model according to the output characteristics of each unit, load demand, and the peak shaving and valley filling functions of the energy storage device;

[0202] A model optimization module, configured to formulate a scheduling strategy according to the constraint conditions, solve the scheduling model through the scheduling strategy, and obtain an optimized economic scheduling plan for multi-time scale and multi-objective optimization.

[0203] In some implementation manners, in the initialization module, the new energy machines include photovoltaic generator sets and wind turbine generator sets; the energy storage device parameters include the capacity, charge and discharge efficiency, and state of charge of the energy storage device.

[0204] In some implementation manners, in the model adjustment module, the constraint conditions include wind turbine output constraints, photovoltaic unit output constraints, gas turbine ramp constraints, gas turbine output constraints, energy storage device charge and discharge power constraints, energy storage device capacity constraints, energy storage device state of charge constraints, energy storage device charge and discharge conservation constraints, daily total power conservation constraints of the energy storage device, and interaction power constraints between the park power grid and the external network.

[0205] In some implementation manners, the model optimization module specifically includes:

[0206] A subgroup unit, configured to generate an initial particle population according to the system initial data;

[0207] An evaluation unit, configured to evaluate the particles by setting the initial positions of the particles according to the initial particle population;

[0208] An adjustment unit, configured to save the evaluated non-dominated solutions to a backup set, solve the initial values of the parameters through a multi-objective optimization algorithm for improving the particle swarm based on reinforcement learning, and perform dynamic non-linear adjustment;

[0209] A position unit, configured to update the particle positions and calculate the fitness according to the best positions of the particles in the population;

[0210] An update unit, configured to determine whether a particle needs to update its position according to the result of the fitness calculation, and update the best position of the particle;

[0211] A generation unit, configured to save the non-dominated solutions into a standby set and generate particles at sparse positions;

[0212] A loop unit, configured to repeat the operations between the adjustment unit and the generation unit until a set termination condition is met, and select the individual with the highest fitness value in the final population as the optimal solution, which is used as the optimal scheduling plan for the park.

[0213] Embodiment 4:

[0214] This embodiment also provides an electronic device, including a memory and a processor. The memory is used to store one or more computer instructions. Among them, when the one or more computer instructions are executed by the processor, the method of Embodiment 1 is implemented;

[0215] In practical applications, the processor can be implemented by an application specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), a controller, a microcontroller unit (MCU), a microprocessor, or other electronic components, and is used to execute the method in the above embodiments.

[0216] The method implemented in this embodiment is as described in Embodiment 1.

[0217] Embodiment 5:

[0218] This embodiment also provides a computer storage medium. A computer program is stored in the computer-readable storage medium. When the computer program is executed by one or more processors, the method of Embodiment 1 is implemented;

[0219] Among them, the computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0220] The method implemented in this embodiment is as described in Embodiment 1.

[0221] Finally, it should be noted that the above content is only used to illustrate the technical solution of the present invention, rather than a limitation on the protection scope of the present invention. Any simple modification or equivalent replacement of the technical solution of the present invention by those of ordinary skill in the art does not depart from the essence and scope of the technical solution of the present invention.

Claims

1. A comprehensive energy system scheduling method based on particle swarm algorithm, characterized in that: The method comprises: S1, according to the comprehensive energy system, collects the output of new energy units, gas turbine output, network load and energy storage device parameters as the system initial data; S2, establishing a dispatching model for coordinated optimization of sources, loads and storage in the park based on the initial data; S3, constructing the constraint conditions of the scheduling model according to the output characteristics of each unit, load demand and the peak-shaving and valley-filling function of the energy storage device; S4, formulating a dispatching strategy according to the constraint conditions, solving the dispatching model through the dispatching strategy, and obtaining an optimized multi-time scale and multi-objective economic dispatching plan.

2. The method according to claim 1, characterized in that In S1, the new energy machine includes a photovoltaic generator set and a wind generator set; the energy storage device parameters include the capacity, charging and discharging efficiency and charge state of the energy storage device.

3. The method according to claim 1, characterized in that In S3, the constraints include wind turbine output constraints, photovoltaic unit output constraints, gas turbine climbing constraints, gas turbine output constraints, energy storage device charging and discharging power constraints, energy storage device capacity constraints, energy storage device charge state constraints, energy storage device charging and discharging conservation constraints, energy storage device daily total power conservation constraints, and park power grid and external grid interaction power constraints.

4. The method according to claim 1, characterized in that: The S4 specifically includes: S41, generating an initial particle population according to the initial data of the system; S42, according to the initial particle population, setting initial positions of particles and evaluating the particles; S43, saving the evaluated non-inferior solution to a backup set, solving the initial value of the parameter by using a multi-objective optimization algorithm based on a reinforcement learning-improved particle swarm, and performing dynamic nonlinear adjustment; S44, according to the best position of the particle in the group, by updating the particle position and performing fitness calculation; S45, judging whether the particle needs to update its position according to the result of the fitness calculation, and updating the optimal position of the particle; S46, saving the non-inferior solution to a backup set and generating particles at sparse locations; S47, repeat steps S43-S46 until the set termination condition is met, and select the individual with the highest fitness value in the final population as the optimal solution, which serves as the optimal scheduling plan for the park.

5. A comprehensive energy system scheduling device based on particle swarm algorithm, characterized in that: The device comprises: An initialization module is used to collect the output of new energy units, gas turbine output, network load and energy storage device parameters as system initial data according to the comprehensive energy system; A model building module, used to build a dispatching model for coordinated optimization of sources, loads and storage in the park according to the initial data; A model adjustment module is used to construct the constraint conditions of the scheduling model according to the output characteristics of each unit, load demand and the peak-shaving and valley-filling effect of the energy storage device; The model optimization module is used to formulate a scheduling strategy according to the constraints, solve the scheduling model through the scheduling strategy, and obtain an optimized multi-time scale and multi-objective optimized economic scheduling plan.

6. The method according to claim 5, characterized in that In the initialization module, the new energy machine includes a photovoltaic generator set and a wind generator set; the energy storage device parameters include the capacity, charging and discharging efficiency and charge state of the energy storage device.

7. The method according to claim 5, characterized in that In the model adjustment module, the constraints include wind turbine output constraints, photovoltaic unit output constraints, gas turbine climbing constraints, gas turbine output constraints, energy storage device charging and discharging power constraints, energy storage device capacity constraints, energy storage device charge state constraints, energy storage device charging and discharging conservation constraints, energy storage device daily total power conservation constraints, and park power grid and external network interaction power constraints.

8. The method according to claim 5, characterized in that The model optimization module specifically includes: A subgroup unit, used for generating an initial particle population according to the initial data of the system; An evaluation unit, configured to set initial positions of particles and evaluate the particles according to the initial particle population; The adjustment unit is used to save the evaluated non-inferior solution to the backup set, solve the initial value of the parameter through the multi-objective optimization algorithm of the particle swarm improved based on reinforcement learning, and perform dynamic nonlinear adjustment; The position unit is used to update the particle position and calculate the energy fitness according to the optimal position of the particle in the group; An updating unit, used for judging whether the particle needs to be updated in position according to the result of the fitness calculation, and updating the optimal position of the particle; A generation unit, used to save non-inferior solutions to a backup set and generate particles at sparse locations; The loop unit is used to repeat the operations between the adjustment unit and the generation unit until the set termination condition is met, and select the individual with the highest fitness value in the final population as the optimal solution, which serves as the optimal scheduling plan for the park.

9. An electronic device, characterized in that: The invention comprises a memory and a processor, wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions, when executed by the processor, implement the method as described in any one of claims 1 to 4.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it is used to implement the method as described in any one of claims 1 to 4 above.

Images