An active energy-saving optimization control method for fresh air handling units based on intelligent optimization algorithm

Abstract

The application discloses a fresh air handling unit active energy-saving optimization control method based on an intelligent optimization algorithm, and comprises the following steps: (1) mode judgment; (2) precooling treatment in a precooling mode; (3) preheating treatment in a preheating mode; (4) isenthalpic humidification treatment; (5) recooling section treatment; (6) reheating section treatment; and (7) constraint treatment through the intelligent optimization algorithm to obtain valve opening degree. The application uses the intelligent optimization algorithm, takes the supply air temperature and the supply air dew point as constraint conditions, and takes the cold and hot water energy consumption cost as an optimization target, realizes the optimization search of the opening degree of each coil adjusting valve, reduces the cold and hot water energy consumption cost of the fresh air handling unit under the premise of meeting the supply air temperature and the supply air dew point, and realizes the active energy-saving optimization control of the fresh air handling unit according to the real-time change of external air conditions.

Description

Technical Field

[0001] This invention relates to a proactive energy-saving optimization control method for fresh air handling units based on machine learning and intelligent optimization algorithms. Background Technology

[0002] Fresh air handling units play a crucial role in factory air conditioning systems. Their primary function is to process outside air to meet specific temperature and humidity requirements before delivering it indoors, ensuring the indoor temperature and humidity environment conforms to the needs of production processes. To achieve this, the fresh air handling unit regulates air temperature and humidity through preheating, precooling, recooling, and reheating processes, utilizing four chilled / hot water coils and a humidifier. The heating and cooling capacity of the four chilled / hot water coils is controlled by electrically operated regulating valves on the coils. Currently, PID control is widely used to control the electrically operated regulating valves of the fresh air handling unit. Specifically, sensors measure the deviations between the pre-treated air temperature (or enthalpy), the supply air temperature, and the supply air dew point from the set values. A PID algorithm is then applied to calculate and output the corresponding adjustment amount for the regulating valves on the coils, thereby controlling the valve opening to achieve the set supply air temperature and dew point temperature.

[0003] However, there are relatively few energy-saving control methods currently available for fresh air handling units. Referring to energy-saving control methods for similar equipment such as central air conditioning systems, they can be divided into PID-based energy-saving control methods and model-predictive energy-saving control methods. PID-based energy-saving control methods optimize commonly used PID control methods, aiming to improve the stability and adaptability of the PID algorithm to avoid energy losses caused by lag or fluctuations in the control process. For example, a fuzzy PID-based energy-saving control method can be used to optimize various parameters in PID control.

[0004] In contrast, model-based predictive energy-saving control methods focus on establishing a mathematical model of the control system to predict the control results under different control parameters, thereby selecting control parameters that meet control requirements and are relatively energy-efficient. Current model-based predictive energy-saving control methods employ numerous system modeling and energy-saving optimization techniques. One common approach is to directly use historical system operating data, using control parameters as input variables and control results as output variables to establish a model that reflects the relationship between control parameters and control results.

[0005] In summary, while PID control is widely used in the field of fresh air handling units, research on energy-saving control methods is relatively limited. PID-based energy-saving control methods aim to optimize the performance of conventional PID control, while model-predictive energy-saving control methods focus on establishing system models to predict control results under different control parameters, thereby achieving more energy-efficient control.

[0006] Commonly used PID control methods and PID-based energy-saving control methods typically do not require system modeling of the fresh air handling unit. Essentially, they compare the actual values ​​of target control variables such as pre-processed air temperature or enthalpy, supply air temperature, and supply air dew point with their setpoints. The PID algorithm then calculates and outputs the corresponding control quantity, thereby achieving stable control of the supply air temperature and dew point. Commonly used PID control methods and PID-based energy-saving control methods have the following drawbacks:

[0007] 1. Passive response and lag. This type of method adjusts the controlled variable based on feedback from the control results, thus exhibiting a lag in response to changes in external atmospheric conditions. The system cannot proactively adjust to real-time changes in external atmospheric conditions, potentially leading to instability in the control process or wasted energy.

[0008] 2. Coupling effects are difficult to resolve. In multi-parameter systems such as fresh air handling units, PID control methods struggle to handle the coupling relationships between variables effectively. Especially when multiple PID controllers are connected in series, parameter adjustments become complex, potentially leading to energy waste, such as over-humidification followed by dehumidification.

[0009] 3. Lack of evaluation of energy-saving effect. PID-based energy-saving control methods mainly achieve energy-saving effects by optimizing PID control parameters, but lack quantitative evaluation and basis for energy-saving effects, making it difficult to measure the actual effect of optimization.

[0010] Currently, system modeling methods in model-based predictive energy-saving control primarily focus on single-data-driven system modeling based on historical operating data. This involves directly establishing a mapping relationship between control parameters and control results based on historical data, predicting energy consumption based on these parameters, and selecting the most energy-efficient control parameters. However, model-based predictive energy-saving control methods currently suffer from the following drawbacks:

[0011] 1. Lack of full-cycle system modeling. Existing methods, based on historical operating data, often focus only on single data-driven models, failing to comprehensively describe the energy conversion of the fresh air handling unit's air treatment process, and failing to correlate control parameters with energy supply and demand, resulting in insufficient model transparency and accuracy.

[0012] 2. Low optimization search efficiency. When dealing with a large number of control parameter combinations, existing model-based energy-saving control methods lack efficient optimization search methods, resulting in insufficient optimization quality and efficiency.

[0013] In summary, both commonly used PID control methods and model-based predictive energy-saving control methods have limitations in existing technologies. The lag and coupling issues of PID control can lead to poor control performance, while model-based predictive energy-saving control methods suffer from incomplete modeling processes and inefficient optimization search methods. These drawbacks can result in problems such as control instability, energy waste, and unclear optimization effects. Summary of the Invention

[0014] Purpose of the Invention: The purpose of this invention is to provide an active energy-saving optimization control method for fresh air handling units based on intelligent optimization algorithms. This method applies intelligent optimization algorithms, using supply air temperature and supply air dew point as constraints and chilled / hot water energy costs as optimization objectives, to achieve optimized searching of the opening degree of regulating valves of each coil. Under the premise of meeting the supply air temperature and supply air dew point, it reduces the chilled / hot water energy costs of the fresh air handling unit and realizes active energy-saving optimization control of the fresh air handling unit based on real-time changes in outside air conditions.

[0015] Technical solution: The active energy-saving optimization control method for fresh air handling units based on intelligent optimization algorithms described in this invention includes the following steps:

[0016] Step 1, Mode Determination: Obtain outside air temperature, humidity, and air velocity, and pre-set a threshold; determine whether the outside air temperature is greater than the threshold. If it is, proceed to Step 2, Pre-cooling mode; otherwise, proceed to Step 3, Pre-heating mode.

[0017] Step 2, Pre-cooling process in pre-cooling mode: Use a machine learning prediction model to obtain the pre-cooling section energy under the current individual's pre-cooling opening, temperature, humidity, and airflow rate. Then, use the pre-cooling section mechanism model to obtain the new enthalpy value after pre-cooling under the current temperature, humidity, airflow rate, and pre-cooling section energy.

[0018] Step 3, Preheating treatment in preheating mode: Use a machine learning prediction model to obtain the preheating section energy under the current individual's preheating opening, temperature, humidity, and air flow rate. Then, use the preheating section mechanism model to obtain the new enthalpy value after preheating under the current temperature, humidity, air flow rate, and preheating section energy.

[0019] Step 4, Isoenthalpic Humidification Treatment: After precooling or preheating, the process enters the isoenthalpic humidification stage. The mechanism model is used directly to obtain the temperature and moisture content after humidification at the current enthalpy value.

[0020] Step 5, Recooling stage treatment: After isenthalpic humidification, the recooling stage begins. First, the recooling stage prediction model is used to obtain the recooling stage energy under the conditions of individual recooling opening, temperature, humidity, and air velocity after humidification. Then, the recooling stage mechanism model is used to obtain the humidity and temperature after recooling under the current humidity, enthalpy, air velocity, and recooling stage energy.

[0021] Step 6, Reheat Section Processing: First, use the reheat section prediction model to obtain the reheat section energy under the conditions of individual reheat valve opening and temperature, humidity, and air velocity after recooling. Then, use the reheat section mechanism model to obtain the reheat temperature under the current temperature, humidity, air velocity, and reheat section energy after recooling.

[0022] Step 7: Obtain the valve opening by performing constraint processing through intelligent optimization algorithms.

[0023] Furthermore, the specific steps in step 7 for obtaining the valve opening degree through constraint processing using an intelligent optimization algorithm are as follows:

[0024] Step 7.1: Initialize a population with N individuals;

[0025] Step 7.2: Pre-set the number of iterations for the population, use variable T as a counter, and continuously increase it as the population iterates. Determine whether the number of iterations for the population can be terminated when it reaches the preset value. If it can, then end; otherwise, proceed to step 7.3.

[0026] Step 7.3, Tournament Selection;

[0027] Step 7.4, Directional Crossing;

[0028] Step 7.5, Directed Mutation;

[0029] Step 7.6: Evaluate the sub-solution;

[0030] Step 7.7: Combine the parent population and the offspring population to obtain a population of size 2N;

[0031] Step 7.8: Use tournament selection to select N individuals from the population to enter the next generation;

[0032] Step 7.9: Iterate continuously using the method in Step 7.8 until the termination condition is met, find the best individual in the population, and output the optimal solution and the target value;

[0033] Step 7.10: Control the valve opening by using the output optimal solution and target value.

[0034] Furthermore, the specific steps for initializing a population of N individuals in step 7.1 are as follows:

[0035] Step 7.1.1: Use real number encoding: In real number encoding, each digit represents a different digit value; it is necessary to first determine the actual meaning represented by each gene on the chromosome, and then convert these genes into decimal numbers; based on real number encoding, go through various stages of the genetic algorithm, such as selection, crossover, and mutation;

[0036] Step 7.1.2: Determine the attributes and range of each individual: Each individual should have at least 3 attributes, namely the valve opening degree of precooling, recooling and reheating in the precooling mode, or the valve opening degree of preheating, recooling and reheating in the preheating mode; the value range of each valve opening degree is determined according to the actual situation, and is 0 to 100.

[0037] Step 7.1.3: Generate random individuals: Design the number of individuals N to be 100; for each individual, generate random real values ​​between 0 and 100 for its three attributes, namely valve opening under preheating, recooling, and reheating; use a random number generator to generate these values ​​and save the results in the population list; adjust the range of random number generation according to the distribution of valve opening in historical data. For example, if the distribution of precooling valves is concentrated in the range of 60~90, then random individuals only need to be generated in the range of 50~100.

[0038] Step 7.1.3, Assigning Fitness: Fitness assignment refers to calculating an index to measure the quality of an individual. This index is usually determined based on the characteristics of the problem and the optimization objective. In the case of real number encoding, the fitness function is defined according to the specific problem. For example, the target value that satisfies the constraints is used as the fitness assignment index. That is, by judging the target value (energy consumption) that satisfies the constraints of supply air temperature and dew point temperature, the smaller the target value when the constraints are satisfied, the better the fitness of the individual. If the constraints are not satisfied, only the degree of constraint violation is calculated. The smaller the degree of constraint violation, the better the fitness. Then, based on the fitness of the individuals, evolutionary operators such as selection, crossover, and mutation are used to adjust the population to achieve optimal search.

[0039] Furthermore, the specific steps for tournament selection in step 7.3 are as follows:

[0040] Step 7.3.1: The tournament selection mechanism simulates the competitive process in a tournament. In this selection method, a certain number of individuals are randomly selected for comparison, and then the individual with the best fitness is selected as the parent individual.

[0041] Step 7.3.2: Determine the tournament size, defining the number of individuals compared in each round as 3.

[0042] Step 7.3.3: Repeat the tournament until a sufficient number of individuals are selected:

[0043] A. Randomly select the number of individuals for the tournament scale;

[0044] B. Select the individual with the best fitness from these individuals as the winner.

[0045] Furthermore, the specific steps of the directional crossing in step 7.4 are as follows:

[0046] Step 7.4.1: Set the current population size N to 100, and the decision variable d to 3, i.e., the opening degree of the pre-cooling, re-cooling, and reheating valves; the directional crossover operator performs the directional crossover operation based on the tournament selection results; the directional crossover operator mainly has four parameters, namely the crossover rate. Variable direction crossover rate directional probability Multiplication factors ;

[0047] Step 7.4.2: Assume there are two unequal parent generations. and Crossing is allowed, where j ranges from 1 to d. and These are the optimal and average solutions for the current population, respectively, and the multiplication factor for the unknown parameters in the calculation formula. The directional probability is uniformly set to 0.95. The setting rule is: whenever it is found that the current best fitness value is an improvement compared to the previous generation, It will be assigned a value equal to 0.75; otherwise, its value is 0.5.

[0048] Now, create child individuals according to different situations;

[0049] if Greater than or equal to Then, use the following formulas to create two child solutions respectively. and ;

[0050]

[0051]

[0052]

[0053]

[0054]

[0055]

[0056] Here, val is based on two parent individuals. and The function of the difference between them and These are the upper and lower bounds of the j-th variable, respectively; the Val function measures the similarity between two parent individuals; if the two individuals are significantly different, Val is smaller, and if the difference is small, Val is larger; r3 and r4 are two distinct random numbers generated in the range (0,1). It is a multiplicative factor, therefore It is a parameter based on random numbers, which is related to... Multiplication factor related; based on val and Update child values and This method combines information from two parents; if the difference between the two parents is large (val is small), the offspring is more influenced by the average of the two parents; if the difference between the two parents is small (val is large), the offspring is more influenced by the difference between the two parents. These rules are based on the random number r4 and the directional probability. The update method is determined so that offspring are influenced by their parents while introducing a certain degree of randomness. This method helps maintain the diversity of the population and guides the generation of offspring based on the differences and directional probabilities of the parents.

[0057] if Less than When the child solution is updated, the update rule is as follows:

[0058]

[0059]

[0060]

[0061]

[0062]

[0063]

[0064] If the two parent generations to be crossed have equal values, and When the child solution is obtained, the following formula is used:

[0065]

[0066]

[0067]

[0068]

[0069]

[0070] .

[0071] In summary, directed crossover generates offspring solutions based on the differences between parent solutions, which helps maintain diversity in the population and avoids getting trapped in local optima. By adjusting the mutation coefficient, the degree of variation in offspring solutions can be controlled to some extent, thereby enabling a wider exploration in the solution space. Directed crossover provides a relatively effective way to combine information from parent solutions, making the generated offspring solutions more likely to inherit excellent characteristics and produce better changes in the search space.

[0072] Furthermore, the specific steps of the directional mutation in step 7.5 are as follows:

[0073] Set a population size of N=100 and a number of decision variables of d=3 to optimize a problem. During the optimization process, the mutation operator executes in a variable direction. and Represents a parent solution and its mutated solutions, where i varies from 1 to N and j varies from 1 to d; generate a random number in the range [0,1], if this random number is less than or equal to the mutation probability. Then parent-generation resolution is allowed. Participate in mutation operations; if parent solutions are allowed to participate in mutation, a mutated solution is created under the guidance of directional information; otherwise, the value of the parent solution remains unchanged; in practical use, settings are... =1 / d; Depending on the situation, the rules for calculating the mutated solutions are as follows:

[0074] Directional mutation operators require directional information, so necessary information needs to be collected before use. Information collection involves comparing the parent's solution with the optimal solution obtained from the current population. When comparing, when found The value is greater than or equal to When a mutation solution is created, the following equation is used:

[0075] ,

[0076] ,

[0077]

[0078] and These are two intermediate parameters used to determine... r and r2 are in the range (0,1), and r . and It is the upper and lower bounds of the j-th variable. The directional mutation operator, typically set to between 0.5 and 1, represents the probability of mutation. It considers the relationship between the parent solution and the optimal solution in the population, along with randomness, to generate mutated solutions. By considering the relationship between the optimal solution and the parent solution, the mutation direction is more likely to point towards potentially advantageous directions in the global search space. Using random numbers to control the mutation probability helps control the frequency of mutation operations to some extent, avoiding excessive interference with the optimization process. Introducing information about the optimal solution makes mutation operations more inclined towards directions currently considered superior. This design helps maintain population diversity and explores the solution space more effectively in the global search, increasing the likelihood of finding the global optimum. Overall, this design combines randomness and information about the optimal solution to perform mutation operations in a more guided manner. This approach aims to make mutated solutions more likely to point towards potentially better solutions, improving the algorithm's potential to find the global optimum.

[0079] when and Slightly worse than the mutated solution It can be obtained using the following formula.

[0080]

[0081] Furthermore, the specific steps for evaluating the sub-solution in step 7.6 are as follows:

[0082] Step 7.6.1, Objective Function Evaluation: Substitute the sub-solution into the objective function and calculate its corresponding objective value; the sub-solution stores the valve opening values ​​for precooling or preheating, recooling, and reheating. Substitute these values ​​into the energy consumption calculation formula. The lower the energy consumption, the better. The quality of the currently generated sub-solution is determined by the energy consumption value.

[0083] Step 7.6.2, Constraint Condition Verification: Substitute the sub-solution into the constraint conditions to verify whether the solution meets the constraint conditions. Specifically, substitute the valve opening values ​​of precooling or preheating, recooling, and reheating in the individual into the precooling or preheating section, the isenthalpic humidification section, the recooling section, and the reheating section to obtain the supply air temperature and supply air dew point temperature. Then, perform error calculation, that is, make the difference between the obtained supply air temperature and dew point temperature and the set supply air temperature and dew point temperature. If the absolute value is less than 0.1, it is considered that the control requirements are met and the constraint conditions are satisfied.

[0084] Step 7.6.3, Feasibility Assessment: If all constraints are met, the current solution is considered feasible; otherwise, it is considered infeasible.

[0085] Step 7.6.4: Record the objective function value, constraints, and other relevant information of the sub-solution so that this information can be used for selection and optimization during algorithm iteration.

[0086] Furthermore, the specific steps in step 7.7 to combine the parent population and the offspring population to obtain a population of size 2N are as follows:

[0087] Step 7.7.1: Use direct merging to directly merge the parent and offspring populations to form a large set;

[0088] Step 7.7.2: Deduplicatize the merged population; sort all individuals in the population according to their fitness values ​​using a sorting method; in the sorted solution set, traverse and check whether adjacent solutions are the same. If they are the same, keep one and delete the duplicate solutions.

[0089] Step 7.7.2: Replenish the population size; After removing duplicate individuals from the population, if the population size is less than 2N, use the directional crossover method in step 7.4.1 to generate new offspring to replenish the population size until the population size reaches 2N.

[0090] Furthermore, the specific steps in step 7.8 of using tournament selection to select N individuals from the population to enter the next generation are as follows:

[0091] Step 7.8.1: Determine the scale of the tournament: Determine the number of individuals participating in the competition in each round of the tournament selection, which is set to 3 in this patent.

[0092] Step 7.8.2: Conduct multiple rounds of tournaments; from a population of 2N individuals, conduct multiple rounds of tournament selection, randomly selecting 3 individuals from the population in each round.

[0093] Step 7.8.3: Compare competing individuals; compare the three competing individuals and determine the best-performing individual based on the objective value and constraints, identifying it as the winner. Specifically, substitute the solutions from each individual into the constraints to see if they are satisfied. If the constraints are satisfied, substitute them into the objective function to obtain the corresponding objective value for that individual. The minimum objective value that satisfies the constraints is the optimal solution, and this is the winner. If the constraints are not satisfied, calculate the degree of constraint violation, and the individual with the lowest degree of violation is the winner.

[0094] Step 7.8.4: Repeat the tournament selection process multiple times until N winners are selected.

[0095] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0096] (1) By using directional crossover and directional mutation operators, the information of the current population (the best individual in the current population) serves as a directional guide. Therefore, during the search process, the mutation operator can be used to point to the region where there is a better solution.

[0097] (2) The directional crossover and directional mutation operators used in this invention do not rely entirely on the current population for mutation, but introduce the value of a random number r, so that the mutation probability of the mutation operator depends on r. Therefore, it is possible to perform centralized search or diverse search, thereby improving the diversity of the population.

[0098] (3) This invention improves the algorithm's exploration capability in the early stages of evolution by using the intermediate parameter val, that is, it can help find new solutions or explore new regions of the solution space; in the later stages, it can improve the utilization capability, that is, it can use the information already available to find a good solution. In the early stages of algorithm evolution, more emphasis is placed on exploration, because the location of the optimal solution may not be clear at this time; while in the later stages, more emphasis is placed on utilization, because a better solution may have been found, and it is necessary to make more in-depth use of this information to find a better solution.

[0099] (4) The directional crossover method used in this invention is a crossover operator centered on the parent individual, which can generate offspring solutions using directional information. It also utilizes directional probability (…). This parameter can balance population diversity and selection pressure. Attached Figure Description

[0100] Figure 1 This is a flowchart of the present invention;

[0101] Figure 2 for Figure 1 The flowchart of the intelligent optimization algorithm. Detailed Implementation

[0102] like Figure 1 As shown, a method for active energy-saving optimization control of a fresh air handling unit based on intelligent optimization algorithms includes the following steps:

[0103] Step 1: Pattern determination;

[0104] It acquires outside air temperature, humidity, and air velocity, and presets a threshold value.

[0105] Determine if the outside air temperature is greater than the threshold. If it is, proceed to step 2, pre-cooling mode; otherwise, proceed to step 3, pre-heating mode.

[0106] Step 2, Pre-cooling process in pre-cooling mode: Use a machine learning prediction model to obtain the pre-cooling section energy under the current individual's pre-cooling opening, temperature, humidity, and air velocity conditions. Then, use the pre-cooling section mechanism model to obtain the new enthalpy value after pre-cooling under the current temperature, humidity, air velocity, and pre-cooling energy.

[0107] Step 3: Preheating treatment in preheating mode; use a machine learning prediction model to obtain the preheating section energy under the current individual's preheating opening, temperature, humidity, and air velocity conditions, and then use the preheating section mechanism model to obtain the new enthalpy value after preheating under the current temperature, humidity, air velocity, and precooling energy.

[0108] Step 4, isenthalpic humidification treatment: After precooling or preheating, the isenthalpic humidification stage is entered. The mechanism model is used directly to obtain the temperature and moisture content after humidification at the current enthalpy value.

[0109] Step 5, Recooling Stage Processing: After isenthalpic humidification, the recooling stage begins. First, the recooling stage prediction model is used to obtain the individual recooling opening, temperature after humidification, moisture content, and air velocity to obtain the energy of the recooling stage. Then, the recooling stage mechanism model is used to obtain the moisture content and temperature after recooling based on the current moisture content, enthalpy, air velocity, and recooling energy.

[0110] Step 6, Reheat Section Processing: First, use the reheat section prediction model to obtain the reheat valve opening and temperature, moisture content, and air velocity of each individual to obtain the reheat section energy; then use the reheat section mechanism model to obtain the temperature, moisture content, air velocity, and reheat section energy after recooling to obtain the temperature after reheating.

[0111] Step 7: Obtain the valve opening by performing constraint processing through intelligent optimization algorithms.

[0112] like Figure 2 As shown, the specific steps in step 7 of obtaining the valve opening degree through constraint processing using an intelligent optimization algorithm are as follows:

[0113] Step 7.1: Initialize a population with N individuals;

[0114] Step 7.2: Pre-set the number of iterations for the population, use variable T as a counter, and continuously increase it as the population iterates. Determine whether the number of iterations for the population can be terminated when it reaches the preset value. If it can, then end; otherwise, proceed to step 7.3.

[0115] Step 7.3, Tournament Selection;

[0116] Step 7.4, Directional Crossing;

[0117] Step 7.5, Directed Mutation;

[0118] Step 7.6: Evaluate the sub-solution;

[0119] Step 7.7: Combine the parent population and the offspring population to obtain a population of size 2N;

[0120] Step 7.8: Use tournament selection to select N individuals from the population to enter the next generation;

[0121] Step 7.9: Iterate continuously using the method in Step 7.8 until the termination condition is met, find the best individual in the population, and output the optimal solution and the target value;

[0122] Step 7.10: Control the valve opening by using the output optimal solution and target value.

[0123] Furthermore, the specific steps for initializing a population of N individuals in step 7.1 are as follows:

[0124] Step 7.1.1: Use real number encoding: In real number encoding, each digit represents a different digit value; it is necessary to first determine the actual meaning represented by each gene on the chromosome, and then convert these genes into decimal numbers; based on real number encoding, go through various stages of the genetic algorithm, such as selection, crossover, and mutation;

[0125] Step 7.1.2: Determine the attributes and range of each individual: Each individual should have at least 3 attributes, namely the valve opening degree of precooling, recooling and reheating in the precooling mode, or the valve opening degree of preheating, recooling and reheating in the preheating mode; the value range of each valve opening degree is determined according to the actual situation, and is 0 to 100.

[0126] Step 7.1.3: Generate random individuals: Design the number of individuals N to be 100; for each individual, generate random real values ​​between 0 and 100 for its three attributes, namely valve opening under preheating, recooling, and reheating; use a random number generator to generate these values ​​and save the results in the population list; adjust the range of random number generation according to the distribution of valve opening in historical data. For example, if the distribution of precooling valves is concentrated in the range of 60~90, then random individuals only need to be generated in the range of 50~100.

[0127] Step 7.1.3, Assigning Fitness: Fitness assignment refers to calculating an index to measure the quality of an individual. This index is usually determined based on the characteristics of the problem and the optimization objective. In the case of real number encoding, the fitness function is defined according to the specific problem. For example, the target value that satisfies the constraints is used as the fitness assignment index. That is, by judging the target value (energy consumption) that satisfies the constraints of supply air temperature and dew point temperature, the smaller the target value when the constraints are satisfied, the better the fitness of the individual. If the constraints are not satisfied, only the degree of constraint violation is calculated. The smaller the degree of constraint violation, the better the fitness. Then, based on the fitness of the individuals, evolutionary operators such as selection, crossover, and mutation are used to adjust the population to achieve optimal search.

[0128] Furthermore, the specific steps for tournament selection in step 7.3 are as follows:

[0129] Step 7.3.1: The tournament selection mechanism simulates the competitive process in a tournament. In this selection method, a certain number of individuals are randomly selected for comparison, and then the individual with the best fitness is selected as the parent individual.

[0130] Step 7.3.2: Determine the tournament size, defining the number of individuals compared in each round as 3.

[0131] Step 7.3.3: Repeat the tournament until a sufficient number of individuals are selected:

[0132] A. Randomly select the number of individuals for the tournament scale;

[0133] B. Select the individual with the best fitness from these individuals as the winner.

[0134] Furthermore, the specific steps of the directional crossing in step 7.4 are as follows:

[0135] Step 7.4.1: Set the current population size N to 100, and the decision variable d to 3, i.e., the opening degree of the pre-cooling, re-cooling, and reheating valves; the directional crossover operator performs the directional crossover operation based on the tournament selection results; the directional crossover operator mainly has four parameters, namely the crossover rate. Variable direction crossover rate directional probability Multiplication factors ;

[0136] Step 7.4.2: Assume there are two unequal parent generations. and Crossing is allowed, where j ranges from 1 to d. and These are the optimal and average solutions for the current population, respectively, and the multiplication factor for the unknown parameters in the calculation formula. The directional probability is uniformly set to 0.95. The setting rule is: whenever it is found that the current best fitness value is an improvement compared to the previous generation, It will be assigned a value equal to 0.75; otherwise, its value is 0.5.

[0137] Now, create child individuals according to different situations;

[0138] if Greater than or equal to Then, use the following formulas to create two child solutions respectively. and ;

[0139]

[0140]

[0141]

[0142]

[0143]

[0144]

[0145] Here, val is based on two parent individuals. and The function of the difference between them and These are the upper and lower bounds of the j-th variable, respectively; the Val function measures the similarity between two parent individuals; if the two individuals are significantly different, Val is smaller, and if the difference is small, Val is larger; r3 and r4 are two distinct random numbers generated in the range (0,1). It is a multiplicative factor, therefore It is a parameter based on random numbers, which is related to... Multiplication factor related; based on val and Update child values and This method combines information from two parents; if the difference between the two parents is large (val is small), the offspring is more influenced by the average of the two parents; if the difference between the two parents is small (val is large), the offspring is more influenced by the difference between the two parents. These rules are based on the random number r4 and the directional probability. The update method is determined so that offspring are influenced by their parents while introducing a certain degree of randomness. This method helps maintain the diversity of the population and guides the generation of offspring based on the differences and directional probabilities of the parents.

[0146] if Less than When the child solution is updated, the update rule is as follows:

[0147]

[0148]

[0149]

[0150]

[0151]

[0152]

[0153] If the two parent generations to be crossed have equal values, and When the child solution is obtained, the following formula is used:

[0154]

[0155]

[0156]

[0157]

[0158]

[0159] .

[0160] In summary, directed crossover generates offspring solutions based on the differences between parent solutions, which helps maintain diversity in the population and avoids getting trapped in local optima. By adjusting the mutation coefficient, the degree of variation in offspring solutions can be controlled to some extent, thereby enabling a wider exploration in the solution space. Directed crossover provides a relatively effective way to combine information from parent solutions, making the generated offspring solutions more likely to inherit excellent characteristics and produce better changes in the search space.

[0161] Furthermore, the specific steps of the directional mutation in step 7.5 are as follows:

[0162] Set a population size of N=100 and a number of decision variables of d=3 to optimize a problem. During the optimization process, the mutation operator executes in a variable direction. and Represents a parent solution and its mutated solutions, where i varies from 1 to N and j varies from 1 to d; generate a random number in the range [0,1], if this random number is less than or equal to the mutation probability. Then parent-generation resolution is allowed. Participate in mutation operations; if parent solutions are allowed to participate in mutation, a mutated solution is created under the guidance of directional information; otherwise, the value of the parent solution remains unchanged; in practical use, settings are... =1 / d; Depending on the situation, the rules for calculating the mutated solutions are as follows:

[0163] Directional mutation operators require directional information, so necessary information needs to be collected before use. Information collection involves comparing the parent's solution with the optimal solution obtained from the current population. When comparing, when found The value is greater than or equal to When a mutation solution is created, the following equation is used:

[0164] ,

[0165] ,

[0166]

[0167] and These are two intermediate parameters used to determine... r and r2 are in the range (0,1), and r . and It is the upper and lower bounds of the j-th variable. The directional mutation operator, typically set to between 0.5 and 1, represents the probability of mutation. It considers the relationship between the parent solution and the optimal solution in the population, along with randomness, to generate mutated solutions. By considering the relationship between the optimal solution and the parent solution, the mutation direction is more likely to point towards potentially advantageous directions in the global search space. Using random numbers to control the mutation probability helps control the frequency of mutation operations to some extent, avoiding excessive interference with the optimization process. Introducing information about the optimal solution makes mutation operations more inclined towards directions currently considered superior. This design helps maintain population diversity and explores the solution space more effectively in the global search, increasing the likelihood of finding the global optimum. Overall, this design combines randomness and information about the optimal solution to perform mutation operations in a more guided manner. This approach aims to make mutated solutions more likely to point towards potentially better solutions, improving the algorithm's potential to find the global optimum.

[0168] when and Slightly worse than the mutated solution It can be obtained using the following formula.

[0169]

[0170] Furthermore, the specific steps for evaluating the sub-solution in step 7.6 are as follows:

[0171] Step 7.6.1, Objective Function Evaluation: Substitute the sub-solution into the objective function and calculate its corresponding objective value; the sub-solution stores the valve opening values ​​for precooling or preheating, recooling, and reheating. Substitute these values ​​into the energy consumption calculation formula. The lower the energy consumption, the better. The quality of the currently generated sub-solution is determined by the energy consumption value.

[0172] Step 7.6.2, Constraint Condition Verification: Substitute the sub-solution into the constraint conditions to verify whether the solution meets the constraint conditions. Specifically, substitute the valve opening values ​​of precooling or preheating, recooling, and reheating in the individual into the precooling or preheating section, the isenthalpic humidification section, the recooling section, and the reheating section to obtain the supply air temperature and supply air dew point temperature. Then, perform error calculation, that is, make the difference between the obtained supply air temperature and dew point temperature and the set supply air temperature and dew point temperature. If the absolute value is less than 0.1, it is considered that the control requirements are met and the constraint conditions are satisfied.

[0173] Step 7.6.3, Feasibility Assessment: If all constraints are met, the current solution is considered feasible; otherwise, it is considered infeasible.

[0174] Step 7.6.4: Record the objective function value, constraints, and other relevant information of the sub-solution so that this information can be used for selection and optimization during algorithm iteration.

[0175] Furthermore, the specific steps in step 7.7 to combine the parent population and the offspring population to obtain a population of size 2N are as follows:

[0176] Step 7.7.1: Use direct merging to directly merge the parent and offspring populations to form a large set;

[0177] Step 7.7.2: Deduplicatize the merged population; sort all individuals in the population according to their fitness values ​​using a sorting method; in the sorted solution set, traverse and check whether adjacent solutions are the same. If they are the same, keep one and delete the duplicate solutions.

[0178] Step 7.7.2: Replenish the population size; After removing duplicate individuals from the population, if the population size is less than 2N, use the directional crossover method in step 7.4.1 to generate new offspring to replenish the population size until the population size reaches 2N.

[0179] Furthermore, the specific steps in step 7.8 of using tournament selection to select N individuals from the population to enter the next generation are as follows:

[0180] Step 7.8.1: Determine the scale of the tournament: Determine the number of individuals participating in the competition in each round of the tournament selection, which is set to 3 in this patent.

[0181] Step 7.8.2: Conduct multiple rounds of tournaments; from a population of 2N individuals, conduct multiple rounds of tournament selection, randomly selecting 3 individuals from the population in each round.

[0182] Step 7.8.3: Compare competing individuals; compare the three competing individuals and determine the best-performing individual based on the objective value and constraints, identifying it as the winner. Specifically, substitute the solutions from each individual into the constraints to see if they are satisfied. If the constraints are satisfied, substitute them into the objective function to obtain the corresponding objective value for that individual. The minimum objective value that satisfies the constraints is the optimal solution, and this is the winner. If the constraints are not satisfied, calculate the degree of constraint violation, and the individual with the lowest degree of violation is the winner.

[0183] Step 7.8.4: Repeat the tournament selection process multiple times until N winners are selected.

[0184] Specific embodiments of the present invention are as follows:

[0185] Example 1:

[0186] Table 1 Given Working Condition 1

[0187]

[0188] Table 2 Optimization results for given working condition 1

[0189]

[0190] Using given working condition 1 as input to the intelligent optimization algorithm, the final optimization results are shown in Table 2. The mathematical expression of the mechanism model is as follows:

[0191] m = 43kg / s

[0192] T 外气 = 34.41℃

[0193] Td = 23.88℃

[0194] T 送风 = 22.12℃

[0195] Td 送风 = 9.15℃

[0196] Q 预冷 (99.13%) represents the energy value predicted by the prediction model when the precooling pipe valve opening is 99.13%.

[0197] Q 再冷 (59.0%) represents the energy value predicted by the prediction model when the recooling pipe valve opening is 59.0%.

[0198] Q 再热 (40.0%) represents the energy value predicted by the prediction model when the reheat pipe valve opening is 40.0%.

[0199] (1) Pre-cooling stage:

[0200] (1)

[0201] (2)

[0202] (3)

[0203] (4)

[0204] Combining the above four equations, we get , , .

[0205] (2) Humidification stage:

[0206] (1)

[0207] (2)

[0208] (3)

[0209] (4)

[0210] Combining the above four equations, we get , ,

[0211] (3) Recooling stage:

[0212] (1)

[0213] (2)

[0214] (3)

[0215] (4)

[0216] Combining the above four equations, we get , , ,in The corresponding dew point temperature Td = 9.15℃, which means that the dew point temperature requirement for delivery has been met.

[0217] (4) Reheating stage:

[0218] (1)

[0219] (2)

[0220] Combining the above two equations, we get ℃, meeting the air supply temperature requirements.

[0221] Example 2:

[0222] Table 3 Given Working Condition 2

[0223]

[0224] Table 4 shows the optimization results for given working condition 2.

[0225]

[0226] Using given working condition 1 as input to the intelligent optimization algorithm, the final optimization results are shown in Table 4. The mathematical expression of the mechanism model is as follows:

[0227] m = 43kg / s

[0228] T 外气 = 23.46℃

[0229] Td = 20.87℃

[0230] T 送风 = 22.44℃

[0231] Td 送风 = 9.25℃

[0232] Q 预冷 (72.63%) represents the energy value predicted by the prediction model when the precooling pipe valve opening is 72.63%.

[0233] Q 再冷 (33.84%) represents the energy value predicted by the prediction model when the recooling pipe valve opening is 33.84%.

[0234] Q 再热 (42.76%) represents the energy value predicted by the prediction model when the reheat pipe valve opening is 42.76%.

[0235] (1) Pre-cooling stage:

[0236] (1)

[0237] (2)

[0238] (3)

[0239] (4)

[0240] Combining the above four equations, we get , , .

[0241] (2) Humidification stage:

[0242] (1)

[0243] (2)

[0244] (3)

[0245] (4)

[0246] Combining the above four equations, we get , ,

[0247] (3) Recooling stage:

[0248] (1)

[0249] (2)

[0250] (3)

[0251] (4)

[0252] Combining the above four equations, we get , , ,in The corresponding dew point temperature Td = 9.25℃, which means that the dew point temperature requirement for delivery has been met.

[0253] (4) Reheating stage:

[0254] (1)

[0255] (2)

[0256] Combining the above two equations, we get ℃, meeting the air supply temperature requirements.

Claims

1. A method for active energy-saving optimization control of fresh air handling units based on intelligent optimization algorithms, characterized in that, Includes the following steps: Step 1, Mode Determination: Obtain outside air temperature, humidity, and air velocity, and pre-set a threshold; determine whether the outside air temperature is greater than the threshold. If it is, proceed to Step 2, Pre-cooling mode; otherwise, proceed to Step 3, Pre-heating mode. Step 2, Pre-cooling process in pre-cooling mode: Use a machine learning prediction model to obtain the pre-cooling section energy under the current individual's pre-cooling opening, temperature, humidity, and airflow rate. Then, use the pre-cooling section mechanism model to obtain the new enthalpy value after pre-cooling under the current temperature, humidity, airflow rate, and pre-cooling section energy. Step 3, Preheating treatment in preheating mode: Use a machine learning prediction model to obtain the preheating section energy under the current individual's preheating opening, temperature, humidity, and air flow rate. Then, use the preheating section mechanism model to obtain the new enthalpy value after preheating under the current temperature, humidity, air flow rate, and preheating section energy. Step 4, Isoenthalpic Humidification Treatment: After precooling or preheating, the process enters the isoenthalpic humidification stage. The mechanism model is used directly to obtain the temperature and moisture content after humidification at the current enthalpy value. Step 5, Recooling stage treatment: After isenthalpic humidification, the recooling stage begins. First, the recooling stage prediction model is used to obtain the recooling stage energy under the conditions of individual recooling opening, temperature, humidity, and air velocity after humidification. Then, the recooling stage mechanism model is used to obtain the humidity and temperature after recooling under the current humidity, enthalpy, air velocity, and recooling stage energy. Step 6, Reheat Section Processing: First, use the reheat section prediction model to obtain the reheat section energy under the conditions of individual reheat valve opening and temperature, humidity, and air velocity after recooling. Then, use the reheat section mechanism model to obtain the reheat temperature under the current temperature, humidity, air velocity, and reheat section energy after recooling. Step 7: Obtain the valve opening by performing constraint processing through intelligent optimization algorithms; The specific steps in step 7, where the valve opening is obtained by performing constraint processing using an intelligent optimization algorithm, are as follows: Step 7.1: Initialize a population with N individuals; Step 7.2: Pre-set the number of iterations for the population, use variable T as a counter, and continuously increase it as the population iterates. Determine whether the number of iterations for the population can be terminated when it reaches the preset value. If it can, then end; otherwise, proceed to step 7.

3. Step 7.3, Tournament Selection; Step 7.4, Directional Crossing; Step 7.5, Directed Mutation; Step 7.6: Evaluate the sub-solution; Step 7.7: Combine the parent population and the offspring population to obtain a population of size 2N; Step 7.8: Use tournament selection to select N individuals from the population to enter the next generation; Step 7.9: Iterate continuously using the method in Step 7.8 until the termination condition is met, find the best individual in the population, and output the optimal solution and the target value; Step 7.10: Control the valve opening using the output optimal solution and target value; The specific steps for initializing a population of N individuals in step 7.1 are as follows: Step 7.1.1: Use real number encoding: In real number encoding, each digit represents a different digit value; it is necessary to first determine the actual meaning represented by each gene on the chromosome, and then convert these genes into decimal numbers; based on real number encoding, go through each stage of the genetic algorithm; Step 7.1.2: Determine the attributes and range of each individual: Each individual has at least 3 attributes, namely the valve opening degree of precooling, recooling and reheating in the precooling mode, or the valve opening degree of preheating, recooling and reheating in the preheating mode; the value range of each valve opening degree is determined according to the actual situation, and is 0 to 100. Step 7.1.3: Generate random individuals: Design the number of individuals N to be 100; for each individual, generate random real values ​​between 0 and 100 for its three attributes, namely valve opening under preheating, recooling, and reheating conditions; use a random number generator to generate these values ​​and save the results in the population list; adjust the range of random number generation according to the distribution of valve opening in historical data. Step 7.1.3: Assign fitness: Fitness assignment refers to calculating an index to measure the quality of an individual. This index is determined based on the characteristics of the problem and the optimization objective. Based on the fitness of the individuals, evolutionary operators such as selection, crossover, and mutation are used to adjust the population to achieve optimal search. The specific steps for evaluating the sub-solution in step 7.6 are as follows: Step 7.6.1, Objective Function Evaluation: Substitute the sub-solution into the objective function and calculate its corresponding objective value; Step 7.6.2, Constraint Condition Verification: Substitute the sub-solutions into the constraint conditions to verify whether the solutions meet the constraint conditions. Specifically, the valve opening values ​​for precooling or preheating, recooling, and reheating in the individual are gradually substituted into the precooling or preheating treatment, isenthalpic humidification treatment, recooling section treatment, and reheating section treatment to obtain the supply air temperature and supply air dew point temperature. Then, error calculation is performed, that is, the difference between the obtained supply air temperature and supply air dew point temperature and the set supply air temperature and supply air dew point temperature. If the absolute value is less than 0.1, it is considered that the control requirements are met and the constraint conditions are satisfied. Step 7.6.3, Feasibility Assessment: If all constraints are met, the current solution is considered feasible; otherwise, it is considered infeasible. Step 7.6.4: Record the objective function value, constraints, and other relevant information of the sub-solution so that this information can be used for selection and optimization during algorithm iteration.

2. The active energy-saving optimization control method for fresh air handling units based on intelligent optimization algorithms according to claim 1, characterized in that, The specific steps for tournament selection in step 7.3 are as follows: Step 7.3.1: The tournament selection mechanism simulates the competitive process in a tournament. In this selection method, a certain number of individuals are randomly selected for comparison, and then the individual with the best fitness is selected as the parent individual. Step 7.3.2: Determine the tournament size, defining the number of individuals compared in each round as 3; Step 7.3.3: Repeat the tournament until a sufficient number of individuals are selected: A. Randomly select the number of individuals for the tournament scale; B. Select the individual with the best fitness from these individuals as the winner.

3. The active energy-saving optimization control method for fresh air handling units based on intelligent optimization algorithms according to claim 1, characterized in that, The specific steps in step 7.7 to combine the parent and offspring populations to obtain a population of size 2N are as follows: Step 7.7.1: Use direct merging to directly merge the parent and offspring populations to form a large set; Step 7.7.2: Deduplicatize the merged population; sort all individuals in the population according to their fitness values ​​using a sorting method; in the sorted solution set, traverse and check whether adjacent solutions are the same. If they are the same, keep one and delete the duplicate solutions. Step 7.7.2: Replenish the population size; After removing duplicate individuals from the population, if the population size is less than 2N, use the directional crossover method in step 7.4.1 to generate new offspring to replenish the population size until the population size reaches 2N.

4. The active energy-saving optimization control method for fresh air handling units based on intelligent optimization algorithms according to claim 1, characterized in that, The specific steps in step 7.8, which uses tournament selection to select N individuals from the population to enter the next generation, are as follows: Step 7.8.1: Determine the size of the tournament: Determine the number of individuals participating in the competition in each round of the tournament selection; Step 7.8.2: Conduct multiple rounds of tournament selection; From a population of 2N individuals, conduct multiple rounds of tournament selection, randomly selecting 3 individuals from the population in each round; Step 7.8.3: Compare competing individuals; compare the three competing individuals and determine the best-performing individual based on the target value and constraints, and designate it as the winner; Step 7.8.4: Repeat the tournament selection process multiple times until N winners are selected.

Images