Particle swarm calculation improvement method for load optimization control of water chilling unit

Abstract

The invention provides a particle swarm calculation improvement method for water chilling unit load optimization control. The particle swarm calculation improvement method comprises the following steps: establishing an energy consumption simulation model, and importing the energy consumption simulation model as a target function into a simulation environment; inputting the initialized population and the parameters into an energy consumption simulation model, calculating a particle fitness value under the scheme, and recording an initial global optimal solution; introducing a self-adaptive inertia weight factor on the basis of the particle swarm algorithm, and dynamically adjusting the inertia weight according to the relationship between a particle fitness value and an average fitness value; an adaptive learning factor strategy is introduced on the basis of the particle swarm optimization; based on a variation mechanism of a genetic algorithm, a variation probability which is reduced along with the number of iterations is designed on each particle dimension; judging whether the current number of iterations is the maximum number of iterations, and if yes, outputting an optimal solution; and if not, continuing iterative calculation. The improved algorithm has higher optimization efficiency in load distribution of the water chilling unit, and 3% of energy consumption can be saved compared with a traditional expert control strategy.

Description

Technical Field

[0001] The present invention belongs to the technical field of building energy system optimization control, and in particular relates to an improved particle swarm computing method for load optimization control of a chiller. Background Art

[0002] Global energy consumption has continued to grow in recent years, and the building industry has become a significant energy consumer. Heating, ventilation, and air conditioning (HVAC) systems account for a significant portion of a building's total energy consumption. As the core component of an HVAC system, the operating efficiency of chillers directly impacts the overall system's energy consumption. Therefore, precisely optimizing chiller load while maintaining indoor comfort can significantly reduce building energy consumption, significantly impacting carbon emissions.

[0003] Currently, optimization methods for building cooling systems fall into two main categories: hardware optimization based on equipment improvements, such as high-efficiency compressors and new refrigerants; and software optimization strategies based on optimization algorithms, such as optimizing load distribution and adjusting operating parameters. Because most building cooling systems have been in operation for many years and hardware retrofits are costly, software optimization methods based on optimization algorithms have become a current research focus. Traditional methods for optimizing load distribution primarily include rule-based expert system control, heuristic algorithms (such as genetic algorithms (GAs) and simulated annealing (SAs), and linear programming. However, in load optimization scenarios targeting complex, multi-extreme-value objective functions, traditional algorithms often lack strong global escape capabilities in the late iteration stages, tending to converge prematurely to suboptimal solutions and making it difficult to obtain a truly optimal load distribution strategy. The particle swarm optimization (PSO) algorithm is widely used for optimization problems due to its advantages of low parameter count and ease of implementation. However, the classic PSO algorithm is also prone to falling into local optima when faced with high-dimensional, multi-extreme-value problems. Therefore, improvements are needed to enhance its search efficiency and robustness. Most existing studies fail to effectively balance global search capabilities and local search accuracy, or the optimization results are subject to specific initial conditions, resulting in unstable optimization effects under different working conditions. Summary of the Invention

[0004] The purpose of the present invention is to provide an improved particle swarm computing method for load optimization control of chillers, aiming to optimize the existing particle swarm algorithm and solve the problems of multiple extreme values and multiple constraints faced by load optimization control of building chillers.

[0005] The present invention is implemented as follows: an improved particle swarm computing method for load optimization control of a chiller, the method comprising the following steps:

[0006] Step S1: Establish an energy consumption simulation model between the load distribution and power consumption of the chiller unit, and import the energy consumption simulation model into the simulation environment as an objective function;

[0007] Step S2: Input the initialized population and parameters into the energy consumption simulation model. The energy consumption simulation model calculates the fitness value of the particles under the allocation scheme according to the scheme and records the initial global optimal solution.

[0008] Step S3: Based on the particle swarm algorithm, an adaptive inertia weight factor is introduced to dynamically adjust the inertia weight according to the relationship between the particle fitness value and the average fitness value;

[0009] Step S4: Based on the particle swarm algorithm, an adaptive learning factor strategy is introduced to enhance the local search capability of the algorithm in the early iteration and the global search capability in the later iteration;

[0010] Step S5: Based on the mutation mechanism of the genetic algorithm, a mutation probability that decreases with the number of iterations is designed in each particle dimension to enhance the global search capability of the multi-extreme solution space and avoid premature convergence;

[0011] Step S6: Determine whether the current number of iterations is the maximum number of iterations. If so, stop the iteration and output the optimal solution; if not, continue the iterative calculation.

[0012] A further technical solution of the present invention is: in step S1, the input of the energy consumption simulation model is the load distribution of each chiller, and the output is the energy consumption per unit time under the distribution strategy.

[0013] A further technical solution of the present invention is: in step S2, when the particle fitness is generally high or is tending to the local optimum, ω should be increased to strengthen the local search; when the particle fitness is dispersed or the average fitness is low, ω should be reduced to strengthen the global search. The specific calculation formula is as follows:

[0014]

[0015] Among them, f represents the current fitness value of the particle, f min is the minimum fitness of the particle swarm, f avg is the average fitness value, ω min is the lower limit of the inertia weight, ω max is the upper limit of the inertia weight.

[0016] A further technical solution of the present invention is: in step S2, the adaptive adjustment strategy dynamically adjusts the search direction and intensity of the algorithm for particles with different fitness levels, so that particles with high fitness tend to search locally and particles with low fitness tend to explore globally.

[0017] A further technical solution of the present invention is: in step S3, let the local learning factor c local and the global learning factor c global Make reverse changes during the iteration process, use a larger c in the early stage localImprove local search depth and use larger c in the later stage local Guide particles to jump out of the local area. The specific calculation formula is as follows:

[0018]

[0019] Among them, maxIter is the maximum number of iterations, iter is the current number of iterations, c min is the lower limit of the learning factor, c max is the upper limit of the learning factor.

[0020] A further technical solution of the present invention is: in step S4, in the multi-extreme solution space, a mutation operation based on a genetic algorithm provides random perturbations to the particles. When the mutation probability is appropriate, the local optimal situation is broken. The specific calculation formula is as follows:

[0021]

[0022] in, is the mutation probability, D is the particle dimension, i is the current iteration number, g max is the maximum number of iterations, n represents the nth dimension of the particle, and m is used to mark the probability of mutation of the nth dimension of the particle in the i-th iteration. When the random number satisfies When the function is called, a mutation operation is performed on the nth dimension of the particle. rand(0,1) refers to a random number uniformly distributed in the interval (0,1). Each time it is called, a floating point number between 0 and 1 is randomly generated.

[0023] A further technical solution of the present invention is to bidirectionally couple the particle swarm algorithm improved through steps S2 to S5 with the energy consumption simulation model, and iteratively output the optimal load distribution strategy for the chiller.

[0024] The beneficial effects of the present invention are as follows: the present invention combines adaptive inertia weights and mutation strategies, thereby enhancing the algorithm's search capability in a multi-extreme space and effectively preventing the particle swarm from falling into a local optimum in the early stages; by using adaptive learning factors, local search and global search are respectively emphasized at different stages of the algorithm, so that the algorithm can efficiently approach the global optimal solution within a limited number of iterations, thereby improving convergence efficiency; in the case of energy consumption optimization of building chillers, the load distribution can be accurately optimized to achieve significant energy-saving effects, and the algorithm has good applicability to large public buildings; and Matlab / Simulink is used for joint simulation and control, which can be easily expanded to more heating, ventilation and air conditioning (HVAC) systems or other complex energy consumption optimization problems. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is the main flow chart of the method of the present invention;

[0026] Figure 2 It is the overall structure diagram of the improved particle swarm algorithm of the present invention;

[0027] Figure 3 The schematic diagram of energy efficiency comparison between IPSO control and expert strategy under different load distribution strategies for chillers is shown below;

[0028] Figure 4 This is a comparison curve of the average particle velocity of the particle swarm optimization algorithm (PSO) and the improved particle swarm optimization algorithm (IPSO) during the iteration process. DETAILED DESCRIPTION

[0029] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments.

[0030] The present invention provides an improved particle swarm computing method for load optimization control of a chiller, the method comprising the following steps:

[0031] Step S1: Establish an energy consumption simulation model between the load distribution and power consumption of the chiller unit, and import the energy consumption simulation model into the simulation environment as an objective function;

[0032] Step S2: Input the initialized population and parameters into the energy consumption simulation model. The energy consumption simulation model calculates the fitness value of the particles under the allocation scheme according to the scheme and records the initial global optimal solution.

[0033] Step S3: Based on the particle swarm algorithm, an adaptive inertia weight factor is introduced to dynamically adjust the inertia weight according to the relationship between the particle fitness value and the average fitness value;

[0034] Step S4: Based on the particle swarm algorithm, an adaptive learning factor strategy is introduced to enhance the local search capability of the algorithm in the early iteration and the global search capability in the later iteration;

[0035] Step S5: Based on the mutation mechanism of the genetic algorithm, a mutation probability that decreases with the number of iterations is designed in each particle dimension to enhance the global search capability of the multi-extreme solution space and avoid premature convergence;

[0036] Step S6: Determine whether the current number of iterations is the maximum number of iterations. If so, stop the iteration and output the optimal solution; if not, continue the iterative calculation.

[0037] This paper designs an improved particle swarm algorithm for multi-unit load optimization. By incorporating adaptive inertia weights, adaptive learning factors, and mutation strategies into the algorithm, it addresses the problem of the search being easily trapped in local optima during chiller load optimization, caused by multiple extreme values in the energy consumption simulation model. This approach offers valuable benefits for energy conservation and efficiency improvement in complex building load environments. Addressing the multi-extremum nature of building chiller load distribution, this paper introduces mutation operations inspired by genetic algorithms to enable the algorithm to promptly escape local optima, thereby expanding the searchable region and improving its ability to find the global optimal solution. To achieve a balance between rapid convergence and global search, the algorithm dynamically adjusts algorithm parameters, including the inertia weight and learning factor, based on the particle fitness distribution. This ensures that high-fitness particles retain their existing search trends while low-fitness particles have opportunities for enhanced exploration. By integrating this improved algorithm into the chiller load optimization framework, it is hoped that the potential for energy savings within the building system will be exploited while maintaining user comfort, improving overall operational efficiency, and reducing operating costs.

[0038] Preferably, in step S1, the input of the energy consumption simulation model is the load distribution of each chiller, and the output is the energy consumption per unit time under the distribution strategy.

[0039] The energy consumption simulation model in step S1 can be established in Simulink according to industry standards and automatically dispatched and called in Matlab. The input of the energy consumption simulation model is the load distribution of each chiller, and the output is the energy consumption per unit time under the distribution strategy. The model has multiple extreme values. To facilitate algorithm calling, this simulation model is encapsulated and interconnected with the particle swarm optimization (IPSO) algorithm through the Simulink interface to realize automatic iterative calling.

[0040] Preferably, in step S2, the inertia weight ω is the core parameter of the particle velocity continuity and needs to be adaptively adjusted according to the fitness distribution of the particle swarm. When the particle fitness is generally high or tending to the local optimum, ω should be increased to strengthen the local search. When the particle fitness is dispersed or the average fitness is low, ω should be reduced to strengthen the global search. The specific calculation formula is as follows:

[0041]

[0042] Among them, f represents the current fitness value of the particle, f min is the minimum fitness of the particle swarm, f avg is the average fitness value, ω min is the lower limit of the inertia weight, ω max is the upper limit of the inertia weight.

[0043] Preferably, in step S2, the adaptive adjustment strategy dynamically adjusts the search direction and intensity of the algorithm for particles with different fitness levels, so that particles with high fitness tend to search locally and particles with low fitness tend to explore globally, thereby improving the overall convergence efficiency.

[0044] Preferably, in step S3, let the local learning factor c local and the global learning factor c global Make reverse changes during the iteration process, use a larger c in the early stage local Improve local search depth and use larger c in the later stage local Guide particles to jump out of the local area. The specific calculation formula is as follows:

[0045]

[0046] Among them, maxIter is the maximum number of iterations, iter is the current number of iterations, c min is the lower limit of the learning factor, c max is the upper limit of the learning factor. The adaptive adjustment strategy dynamically adjusts the capabilities of local search and global search during the iteration process, thereby preventing the algorithm from converging to a suboptimal solution too early.

[0047] Preferably, in step S4, in the multi-extreme solution space, a mutation operation based on a genetic algorithm provides random perturbations for particles. When the mutation probability is appropriate, the local optimal situation is broken. The specific calculation formula is as follows:

[0048]

[0049] in, is the mutation probability, D is the particle dimension, i is the current iteration number, g max is the maximum number of iterations, n represents the nth dimension of the particle (i.e., the nth component), and the particle has multiple components in multidimensional space, n refers to one of the dimensions. m is used to mark the mutation probability of the nth dimension of the particle in the i-th iteration, as a subscript mark of the mutation probability. When the random number satisfies When , a mutation operation is performed on the nth dimension of the particle. rand(0,1) refers to a random number generated with a uniform distribution in the interval (0,1). Each time it is called, it randomly generates a floating-point number between 0 and 1 (excluding 0 and 1). The triggering probability of the mutation mechanism gradually decays as the iteration progresses, thereby maintaining sufficient jump capability in the early stages of the iteration and ensuring stable convergence of the results in the later stages of the iteration. When optimizing the load distribution of a chiller, if the average particle speed drops to zero during the iteration, it means that the algorithm has fallen into a local optimum. The mutation mechanism can increase the particle speed again, thereby maintaining it at a certain level for continuous search. The addition of this strategy can significantly avoid premature convergence in the complex multi-extreme energy consumption optimization space.

[0050] Preferably, the particle swarm algorithm improved through steps S2 to S5 is bidirectionally coupled with the energy consumption simulation model to iteratively output the optimal load distribution strategy for the chiller, and compared and evaluated with the conventional expert control strategy.

[0051] The specific process of the algorithm is as follows: the improved particle swarm optimization algorithm (IPSO) is connected to the chiller model in Simulink. Algorithm initialization and loop iteration: the particle swarm size, maximum number of iterations, and upper and lower limits of the inertia weight and learning factor are set; the particle position (unit load distribution plan) and speed are randomly initialized; based on the distribution plan, the energy consumption model calculates the particle fitness value (i.e., chiller power consumption) under the plan and records the initial global optimal solution. Adaptive parameters and mutation strategy execution: at the beginning of each iteration, the inertia weight is updated according to the particle fitness distribution; the local learning factor and the global learning factor are dynamically adjusted; the mutation probability of the particle is calculated dimension by dimension, and if the random conditions are met, the speed and position of the corresponding dimension are updated; the above process is repeated until the maximum number of iterations or the convergence criterion is reached, and finally the load distribution strategy for each unit with the lowest total energy consumption is output.

[0052] A simulation test was conducted under the cooling demand scenario of a large public building, such as Figure 3 As shown in Figure 2, the energy efficiency ratio under IPSO control is generally higher than that of the common control method and the average control method. The load distribution scheme finally obtained reduces the power consumption by about 3% compared with the expert strategy. Figure 4 As shown in the figure, the particle swarm velocity of the classic PSO algorithm tends to zero after 50 iterations and falls into a local optimum. However, the improved particle swarm algorithm IPSO of the present invention can always maintain a certain speed range to search for more feasible solutions, greatly improving the ability to mine multiple extreme value problems.

[0053] The method of the present invention is described below by means of a specific embodiment.

[0054] In this example, an improved particle swarm optimization (IPSO) algorithm is used to optimize the load distribution of a building's chillers, thereby improving energy efficiency. The following are the specific implementation steps and detailed code analysis.

[0055] 1. Problem Background and Objectives

[0056] Optimization goal: The particle swarm optimization algorithm (PSO) is used to calculate the load distribution ratio of the chiller at each time point. The optimization goal is to maximize the energy efficiency ratio (i.e., save energy consumption).

[0057] Multi-unit load distribution: In this problem, there are three chillers. The key to optimization is to properly distribute the loads of these three units to achieve an overall improvement in energy efficiency.

[0058] 2. Input parameters

[0059] First, you need to load the load data for each time node (for example, read the load demand from an Excel file) and set the main parameters of the algorithm. The detailed code is as follows.

[0060]

[0061] 3. Objective Function

[0062] To maximize the energy efficiency ratio, we define an objective function, calculate_efficiency, to calculate the energy efficiency under the current load distribution. This will serve as the optimization goal. This is achieved using the "chiller load distribution" → "unit energy efficiency ratio" mapping model established in Simulink. The particle swarm generates a load distribution, and the Simulink model returns the energy efficiency ratio under that distribution. The detailed code is as follows.

[0063]

[0064]

[0065] 4. IPSO optimization process

[0066] Next, we will use the improved particle swarm optimization (IPSO) for optimization. The following code implements the initialization of the particle swarm, the iteration process, and the dynamic adjustment of the inertia weight and learning factor.

[0067] 4.1. Initialize particle swarm

[0068] During the initialization phase, the load distribution for each particle is randomly generated and the distribution ratio is ensured to be within the specified range. The detailed code is as follows.

[0069]

[0070] 4.2. Calculate inertia weight

[0071] The adaptive inertia weight is dynamically adjusted according to the relationship between the particle's fitness value and the average fitness value. The detailed code is as follows.

[0072]

[0073] 4.3 Adaptive Learning Factor

[0074] The adjustment of the learning factor changes dynamically according to the iterative process. The detailed code is as follows.

[0075]

[0076] 4.4 Mutation Mechanism

[0077] In each iteration, the mutation probability is dynamically adjusted according to the current number of iterations to help escape the local optimal solution. The detailed code is as follows.

[0078]

[0079]

[0080] 4.5 Fitness calculation and constraint processing

[0081] According to the load distribution of each particle, its energy efficiency is calculated and it is ensured that it meets the load constraints. The detailed code is as follows.

[0082]

[0083] 5. Final optimization results

[0084] After iterative optimization, we obtain the optimal load distribution and energy efficiency ratio at each time node. We save and visualize the results. The detailed code is as follows.

[0085]

[0086]

[0087] 6. Detailed code explanation

[0088] pso_improved: This function implements the improved particle swarm optimization algorithm (IPSO). Among them, the inertia weight (ω) and the learning factor (c local ,c global ) Dynamically adjusts the fitness of each iteration to help the algorithm balance local search and global search at different stages. Mutation mechanism: The probability of mutation operation gradually decreases according to the current number of iterations, thus avoiding premature convergence to the local optimal solution. This operation enhances the search ability by perturbing the position of particles. Constraint processing:

[0089] The apply_boundary_with_load_limit function ensures that the position of each particle (i.e., the load distribution) satisfies the load limit conditions.

[0090] 7. Results display

[0091] After the code is executed, the optimal load distribution ratio for each time node will be obtained, and the energy efficiency ratio trend of each time node will be plotted. By comparing the energy efficiency ratios at different time nodes, the optimization results can be evaluated.

[0092] 8. Conclusion

[0093] By applying this algorithm, we can achieve optimal control of building chiller loads and improve system energy efficiency. By improving the particle swarm optimization algorithm (IPSO) with multiple optimization strategies (such as adaptive inertia weights, learning factors, and mutation mechanisms), we can effectively avoid local optimality and achieve a global optimal solution.

[0094] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An improved particle swarm computing method for load optimization control of a chiller, characterized in that: The method comprises the following steps: Step S1: Establish an energy consumption simulation model between the load distribution and power consumption of the chiller unit, and import the energy consumption simulation model into the simulation environment as an objective function; Step S2: Input the initialized population and parameters into the energy consumption simulation model. The energy consumption simulation model calculates the fitness value of the particles under the allocation scheme according to the scheme and records the initial global optimal solution. Step S3: Based on the particle swarm algorithm, an adaptive inertia weight factor is introduced to dynamically adjust the inertia weight according to the relationship between the particle fitness value and the average fitness value; Step S4: Based on the particle swarm algorithm, an adaptive learning factor strategy is introduced to enhance the local search capability of the algorithm in the early iteration and the global search capability in the later iteration; Step S5: Based on the mutation mechanism of the genetic algorithm, a mutation probability that decreases with the number of iterations is designed in each particle dimension to enhance the global search capability of the multi-extreme solution space and avoid premature convergence; Step S6: Determine whether the current number of iterations is the maximum number of iterations. If so, stop the iteration and output the optimal solution; if not, continue the iterative calculation.

2. The improved particle swarm computing method for load optimization control of a chiller according to claim 1 is characterized in that: In step S1 , the input of the energy consumption simulation model is the load distribution of each chiller, and the output is the energy consumption per unit time under the distribution strategy.

3. The improved particle swarm computing method for load optimization control of a chiller according to claim 1, characterized in that: In step S2, when the particle fitness is generally high or tending to the local optimum, ω should be increased to strengthen the local search. When the particle fitness is dispersed or the average fitness is low, ω should be reduced to strengthen the global search. The specific calculation formula is as follows: Among them, f represents the current fitness value of the particle, f min is the minimum fitness of the particle swarm, f avg is the average fitness value, ω min is the lower limit of the inertia weight, ω max is the upper limit of the inertia weight.

4. The improved particle swarm computing method for load optimization control of a chiller according to claim 1, characterized in that: In step S2, the adaptive adjustment strategy dynamically adjusts the search direction and intensity of the algorithm for particles with different fitness levels, so that particles with high fitness tend to search locally and particles with low fitness tend to explore globally.

5. The improved particle swarm computing method for load optimization control of a chiller according to claim 1, characterized in that: In step S3, let the local learning factor c local and the global learning factor c global Make reverse changes during the iteration process, use a larger c in the early stage local Improve local search depth and use larger c in the later stage local Guide particles to jump out of the local area. The specific calculation formula is as follows: Among them, maxIter is the maximum number of iterations, iter is the current number of iterations, c min is the lower limit of the learning factor, c max is the upper limit of the learning factor.

6. The improved particle swarm computing method for load optimization control of a chiller according to claim 1, characterized in that: In step S4, in the multi-extreme solution space, the mutation operation based on the genetic algorithm provides random perturbations for the particles. When the mutation probability is appropriate, the local optimal situation is broken. The specific calculation formula is as follows: in, is the mutation probability, D is the particle dimension, i is the current iteration number, g max is the maximum number of iterations, n represents the nth dimension of the particle, and m is used to mark the probability of mutation of the nth dimension of the particle in the i-th iteration. When the random number satisfies When the function is called, a mutation operation is performed on the nth dimension of the particle. rand(0,1) refers to a random number uniformly distributed in the interval (0,1). Each time it is called, a floating point number between 0 and 1 is randomly generated.

7. The improved particle swarm computing method for load optimization control of a chiller according to claim 1, characterized in that: The particle swarm algorithm improved by steps S2 to S5 is bidirectionally coupled with the energy consumption simulation model to iteratively output the optimal load distribution strategy for the chiller.

Images