Chiller group control method and system based on ESC

Abstract

The application relates to an ESC-based cold water unit group control method and system, which comprises the following steps of acquiring meteorological parameters, air conditioner area equipment heat dissipation, personnel load and the like in real time to perform load prediction; establishing a cold water unit energy efficiency model based on a mechanism; designing a model self-updating device which can process and store operation data; the characteristics of the main machine will change due to aging and the like after the main machine runs for a period of time; a model updating strategy verifies the model precision according to actual operation data, and if the deviation is too large, the model parameters are corrected; a group control strategy establishes a target function and a constraint condition, collects cooling water parameters, outdoor environment wet bulb temperature and the like based on a load prediction result, performs optimization through an escape optimization algorithm ESC, and outputs the number of main machine start-stop units and the load rate of each main machine.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of refrigeration and air conditioning, and particularly relates to a cold water unit group control method and system based on an escape optimization ESC algorithm. BACKGROUND

[0002] With the increasing global energy problems and environmental pressures, building energy consumption has become an important field for reducing carbon emissions and energy waste. In modern buildings, the energy consumption of central air conditioning systems usually accounts for more than 40% of the total building energy consumption, and is one of the key objects in the field of building energy saving. As the core equipment of air conditioning systems, the cold water unit bears the task of refrigeration load, and the efficient operation of the cold water unit is crucial for energy saving.

[0003] Traditional cold water unit load control methods often rely on experience setting or fixed modes, lack flexibility and adaptability, and cannot well cope with load changes and external environmental factors, resulting in serious energy waste. In recent years, with the development of information technology, intelligent control technology has begun to be applied to the management of air conditioning systems in order to achieve more refined and intelligent regulation and control effects.

[0004] Most of the existing cold water unit group control methods use fixed modes or experience settings for load distribution, which have the following technical defects:

[0005] Low prediction accuracy: traditional methods usually use static models (such as linear regression, etc.) to predict load demand, but do not consider the dynamic changes of external environment (such as meteorological factors, personnel activities, etc.) and unit operation state, so the prediction accuracy is low, and the cold load supply is prone to excessive or insufficient.

[0006] Model fixation: many control strategies are based on pre-set models for decision-making, which usually ignore the performance degradation of units after long-term operation or changes in external environment, resulting in model misalignment and inability to adapt to the complexity of long-term operation of cold water units.

[0007] Optimization algorithm limitations: existing optimization algorithms, such as genetic algorithm, particle swarm optimization, etc., may fall into local optimal solution when solving multi-unit load balancing, and cannot quickly converge to global optimal solution, and are difficult to cope with load fluctuations and environmental changes in actual operation. SUMMARY

[0008] The purpose of the present application is to solve the problems of inaccurate load prediction and optimization algorithm prone to local optimal solution in the prior art, and to provide a cold water unit group control method and system based on ESC, which maximizes the overall energy efficiency of the cold water unit group.

[0009] To solve the above technical problems, the technical method adopted by the present application is: the present application discloses a cold water unit group control method based on ESC, comprising the following steps:

[0010] S1. Load prediction: the system collects relevant data affecting the cooling load in real time through sensors, including but not limited to meteorological parameters, air conditioning area equipment heat dissipation, personnel load and fresh air load;

[0011] Load prediction algorithm: based on the neural network model combined with long short-term memory network LSTM and attention mechanism KAN layer, combined with real-time collected meteorological data, equipment heating condition and personnel activity condition, load prediction is carried out; the model can capture the time sequence and dynamic change of the cooling load, and improve the prediction accuracy;

[0012] S2. According to the working characteristics of the cold water unit, an energy efficiency model is constructed to describe the relationship between the load rate and power consumption;

[0013] The model considers the working state of the unit, including the energy efficiency performance of different conditions such as part load and full load;

[0014] The model self-updating device is introduced, which can adjust the parameters of the energy efficiency model in real time according to the operation data of the cold water unit, to ensure the matching of load prediction and actual energy efficiency demand;

[0015] If the model prediction deviation is large, the model parameters are automatically corrected to ensure long-term stable operation;

[0016] S3. Based on the load prediction result, the target function of the cold water unit group control is designed, usually minimizing the total energy consumption of the system;

[0017] The constraint conditions include that the load rate of each cold water unit is within a reasonable range to ensure efficient operation of the system;

[0018] The part load rate PLR of the cold water unit is used as the key variable for optimization, and the ESC algorithm is used for optimization to determine the number of on-off units and the load rate of each cold water unit;

[0019] The goal is to optimize the energy efficiency of the entire cold water unit group and reduce energy waste;

[0020] S4. Application of ESC algorithm

[0021] S41. Initialization: first, initialize the load rate PLR of each cold water unit, and randomly distribute it through the individual initialization mechanism of the ESC algorithm; the load rate of each cold water unit is between 0 and 1, reflecting the load state in the initial stage;

[0022] S42. Fitness calculation: Calculate the fitness of each individual, the fitness function evaluates the load distribution scheme by calculating the total energy consumption of the system; the goal is to minimize the total energy consumption of the system while ensuring that the load demand of each chiller is met;

[0023] S43. According to the ESC algorithm, the load distribution problem is transformed into an individual update problem; individuals are divided into three groups: calm, clustered and panic:

[0024] Calm individuals: individuals rationally approach the system load center and update the load distribution scheme;

[0025] Clustered individuals: individuals are influenced by calm individuals and panic individuals, updating the load distribution scheme;

[0026] Panic individuals: individuals make random adjustments when the load fluctuates sharply, simulating the irrational response of load distribution;

[0027] S44. Panic index and convergence: as the number of iterations increases, the panic index gradually decreases, and the individual behavior gradually stabilizes, eventually converging to the global optimal solution;

[0028] S45. Output optimal solution: through multiple iterations, the final load distribution scheme is obtained, and the optimal number of chiller on-off, load rate and other information are output.

[0029] Further, in the step S3, the partial load ratio of the chiller is:

[0030] Where PLR i is the partial load ratio of the ith chiller, is the actual load of the chiller, is the design load of the chiller;

[0031]

[0032] In the formula: P i is the power consumption of the ith chiller, a, b, c, d are the fitting coefficients of the host-partial load ratio curve.

[0033] Further, in the parallel chiller system, sufficient cooling load should be provided to meet the system cooling load demand; to ensure efficient operation of the chiller, the PLR value must be between 0.3 and 1; when the chiller is off, the PLR is 0; the objective function and the constraint condition are as follows:

[0034]

[0035] In the formula, Q need is the total demand of the system cooling load, Obj is the minimum target power consumption of the parallel chiller system.

[0036] Furthermore, in step S4, the following inputs are set: the number of chillers N; the initial load factor of each chiller PLR0 = [PLR1, PLR2, ..., PLR N ], and the initial value satisfies 0.3≤PLR i ≤1 or 0;

[0037] The COP value of each unit COP=[COP1, COP2, ..., COP N ], maximum number of iterations T max , elite pool size E size ;

[0038] Set output: optimal load distribution plan PLR opt , ensuring ∑Q i PLR i ≥Q need , which corresponds to the load distribution with the lowest fitness.

[0039] Furthermore, in the step S41, the following steps are also included:

[0040] S411. Randomly generate the load rate PLR0 of the chiller, and the initial value satisfies 0.3≤PLR i ≤1 or 0;

[0041] S412. Initialize the population, where each individual in the population represents a load distribution plan for a chiller;

[0042] S413. Set the maximum number of iterations T max .

[0043] Furthermore, in the step S42, the following steps are also included:

[0044] S421. For each individual PLR i , calculate its power consumption P i , and calculate the total power consumption of the system;

[0045] S422. Calculate the degree to which the load distribution meets the cooling load demand and ensure ∑Q i PLR i ≥Q need ;

[0046] S423. Comprehensive fitness:

[0047] Among them, λ is a penalty factor. When the cooling load demand is not met, the fitness will increase the penalty term.

[0048] Furthermore, in step S43, the following steps are also included:

[0049] S431. Calculate the fitness of each individual and rank them in ascending order of fitness, select the optimal individual into the elite pool Elite Pool , size E size ;

[0050] S4322. Iterative process: for each individual, the following updates are made:

[0051] Calm group: the individual rationally moves to the average of the individuals in the elite pool:

[0052]

[0053] Where center is the average of the individual load rates in the elite pool;

[0054] Aggregation group: the individual is updated by the influence of the calm and panic groups:

[0055]

[0056] Where PLR p is a randomly selected individual from the panic group, v h is the velocity vector of the aggregation group;

[0057] Panic group: the individual randomly selects a new load allocation:

[0058]

[0059] Where E j is a randomly selected individual from the elite pool, PLR rand is a randomly selected individual, v p is the velocity vector of the panic group.

[0060] Further, when the number of iterations exceeds T max / 2 all individuals tend to the optimal solution in the elite pool, all individuals are updated to be elite pool members or randomly selected individuals; as the number of iterations t increases, the panic index P(t) gradually decreases, simulating the gradual stabilization of individual behavior: P(t) = P max ·(1-t / T max ).

[0061] The application also discloses an ESC-based chiller group control system, which comprises the steps of the chiller group control method and is optimized and adjusted on the basis of the following modules to ensure efficient operation of the chiller group control system and achieve optimal energy efficiency.

[0062] The modules comprise:

[0063] Data acquisition and monitoring module: for real-time acquisition of meteorological parameters, air conditioning area equipment heat dissipation, personnel load and other data, and transmission of real-time data to the data storage module, providing support for load prediction and energy efficiency model;

[0064] Load prediction module: receiving real-time data from the data acquisition module, using a neural network model combining long short-term memory network LSTM and attention mechanism KAN layer for load prediction, and outputting future load demand;

[0065] Energy efficiency model and self-updating mechanism module: according to the working characteristics of the chiller unit, an energy efficiency model is established to describe the relationship between load rate and power consumption; when there is a large deviation between actual operation data and model prediction, the model parameters are automatically adjusted to ensure long-term stable operation;

[0066] Group control strategy module: based on the load prediction results, a group control strategy is designed, with the objective function being the minimization of the total energy consumption of the system, and the constraint condition being the reasonable range of chiller unit load rate (e.g. 30%-100%); this module uses ESC algorithm to optimize the load distribution and the number of on-off units of the chiller unit to achieve optimal energy efficiency of the system;

[0067] ESC algorithm module: responsible for performing load distribution optimization tasks, simulating individual behavior through ESC algorithm to optimize the load distribution of the chiller unit; this module adjusts the PLR and the number of on-off units of each unit according to the fitness of the chiller unit load distribution, and outputs the optimal solution;

[0068] Output module: according to the optimization results, the load rate and the number of on-off units of each chiller unit are outputted, and the scheme is applied to the chiller unit group control system to ensure the optimal energy efficiency of the system;

[0069] Real-time monitoring and precision evaluation module: for real-time monitoring of the running state of the chiller unit, detecting the precision of the load prediction model and the energy efficiency model, ensuring that the load distribution scheme matches the actual operation demand, and making corrections according to real-time data.

[0070] Further, it further comprises:

[0071] Data processing module for pre-processing chiller unit operation data;

[0072] Data storage module for storing pre-processed operation data;

[0073] Model correction module for correcting the parameters of the chiller unit energy efficiency model according to the pre-processed operation data.

[0074] Beneficial effects:

[0075] This invention provides a chiller cluster control method and system based on the ESC algorithm, addressing existing issues such as inaccurate load forecasting and optimization algorithm limitations. By combining a LSTM and KAN neural network for load forecasting and using the ESC algorithm for load distribution optimization, the system continuously optimizes chiller cluster operation under dynamically changing load demands and environmental conditions, ensuring efficient energy utilization, reducing energy consumption and carbon emissions, and improving building energy efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 Provides an overall control logic diagram of the chiller group control strategy for the embodiment of the present invention;

[0077] Figure 2 This is the overall flow chart of load forecasting in the present invention;

[0078] Figure 3 This is a schematic diagram for establishing and updating the energy efficiency model in the present invention. DETAILED DESCRIPTION

[0079] The present invention will be described in further detail below with reference to the accompanying drawings and specific implementation methods.

[0080] Example 1. Chiller Group Control Method Based on ESC

[0081] like Figure 1-3 As shown, a chiller group control method based on ESC includes the following steps:

[0082] Based on real-time meteorological parameters, heat dissipation of equipment in air-conditioned areas, and personnel load, a prediction algorithm is used to perform load forecasting to obtain future load demand.

[0083] The neural network structure includes LSTM layer, KAN layer and output layer:

[0084] 1) The LSTM layer is responsible for processing the input time series data and capturing the long-term dependencies in the data through its internal memory units and gating mechanisms.

[0085] 2) The attention layer (KAN) receives the output of the LSTM layer and assigns different weights to the outputs of different time steps through the attention mechanism, highlighting the impact of important information on the prediction results.

[0086] 3) The output layer is responsible for converting the features processed by the attention layer into the final prediction results.

[0087] Model training process:

[0088] 1) Data preprocessing

[0089] The original load data is cleaned to remove outliers and missing values.

[0090] Normalize the data to eliminate the impact of different dimensions on model training.

[0091] Divide the training set and test set for model training and evaluation.

[0092] 2) Model training

[0093] The LSTM-KAN model is trained using the training set data, and the model parameters are updated through the back-propagation algorithm and optimizer (such as Adam).

[0094] During the training process, strategies such as early stopping can be used to prevent overfitting.

[0095] 3) Model Evaluation

[0096] The trained LSTM-KAN model is evaluated using the test set data to calculate indicators such as prediction error (such as mean square error (MSE), mean absolute error (MAE), etc.) and accuracy.

[0097] A comparative analysis is conducted with other load forecasting models (such as traditional RNN, LSTM, ARIMA, etc.) to evaluate the performance advantages of the LSTM-KAN model.

[0098] Based on real-time meteorological parameters, heat dissipation of equipment in air-conditioned areas, and personnel load, a prediction algorithm is used to perform load forecasting to obtain future load demand.

[0099] like Figure 2 As shown in Figure 2, the overall load forecast specifically includes:

[0100] 1. Input parameters

[0101] The left side of the figure lists four input parameters: meteorological parameters, equipment heat dissipation, personnel load, and fresh air load. These parameters will serve as input data for the neural network to predict load.

[0102] 2. Data processing and data collection: First, data collection is carried out to collect various load-related data.

[0103] Data processing: Processing the collected data, which may include data cleaning, normalization and other operations to facilitate subsequent analysis.

[0104] Load data true value: The processed data will be used as the true value of the load data for training and validation.

[0105] 3. Sample division Training samples: A portion of the data is divided into training samples for training the neural network.

[0106] Validation samples: Another part of the data is divided into validation samples to verify the performance of the trained neural network.

[0107] 4. Neural Network Training Neural Network Training: Use training samples to train the neural network. The figure shows the structure of a multi-layer neural network, including input layer, hidden layer and output layer.

[0108] RMSE compliance: During the training process, the performance of the model is evaluated by calculating the root mean square error (RMSE) and mean absolute error (MAE). The formula is as follows: When the RMSE reaches a preset standard, the training can be stopped.

[0109] 5. Verify and complete error analysis: Use validation samples to verify the trained model to ensure that the model error is within an acceptable range.

[0110] Training completed: If the error analysis meets the requirements, the training process is complete and the trained model can be used for load forecasting.

[0111] The neural network structure includes LSTM layer, KAN layer and output layer:

[0112] 1) The LSTM layer is responsible for processing the input time series data and capturing the long-term dependencies in the data through its internal memory units and gating mechanisms.

[0113] 2) The attention layer (KAN) receives the output of the LSTM layer and assigns different weights to the outputs of different time steps through the attention mechanism, highlighting the impact of important information on the prediction results.

[0114] 3) The output layer is responsible for converting the features processed by the attention layer into the final prediction results.

[0115] Model training process:

[0116] 1) Data preprocessing

[0117] The original load data is cleaned to remove outliers and missing values.

[0118] Normalize the data to eliminate the impact of different dimensions on model training.

[0119] Divide the training set and test set for model training and evaluation.

[0120] 2) Model training

[0121] The LSTM-KAN model is trained using the training set data, and the model parameters are updated through the back-propagation algorithm and optimizer (such as Adam).

[0122] During the training process, strategies such as early stopping can be used to prevent overfitting.

[0123] 3) Model Evaluation

[0124] The trained LSTM-KAN model is evaluated using the test set data to calculate indicators such as prediction error (such as mean square error MSE, mean absolute error MAE, etc.) and accuracy.

[0125] A comparative analysis is conducted with other load forecasting models (such as traditional RNN, LSTM, ARIMA, etc.) to evaluate the performance advantages of the LSTM-KAN model.

[0126] like Figure 1 The logic shown here assumes a chiller cluster consisting of five parallel chillers. Each chiller has different performance characteristics, and load distribution needs to be adjusted based on real-time meteorological data, load forecasts, cooling water temperature, and outdoor wet-bulb temperature. The goal is to reduce overall energy consumption while ensuring smooth operation of the chiller during load fluctuations.

[0127] initialization:

[0128] The load allocation problem for a chiller cluster is transformed into an N-dimensional vector problem. Assume there are five chillers, and the load factor (PLR) of each chiller is initialized to a random value, for example, [0.5, 0.3, 0.7, 0.6, 0.4].

[0129] Algorithm Description:

[0130] enter:

[0131] The number of chillers N; the initial load factor of each chiller is PLR0 = [PLR1, PLR2, ..., PLR N ], and the initial value satisfies 0.3≤PLR i ≤1 or 0; COP value of each unit COP=[COP1,COP2,...,COP N ], maximum number of iterations T max , elite pool size E size .

[0132] Output:

[0133] Optimal load distribution scheme PLR opt , ensuring ∑Q i PLR i ≥Q need , which corresponds to the load distribution with the lowest fitness.

[0134] step:

[0135] initialization:

[0136] Randomly generate the load rate of the chiller PLR0, the initial value satisfies 0.3≤PLR i ≤1 or 0.

[0137] Initialize the population, each individual in the population represents a load distribution scheme of the chiller.

[0138] Set the maximum number of iterations T max .

[0139] Fitness calculation:

[0140] For each individual PLR i , calculate its power consumption P i , and calculate the total power consumption of the system.

[0141] Calculate the degree to which the load distribution meets the cooling load demand, ensure that ∑Q i ·PLR i ≥Q need .

[0142] Comprehensive fitness:

[0143]

[0144] Where λ is a penalty factor, when the cooling load demand is not met, the fitness will increase the penalty term.

[0145] Population update and behavior simulation:

[0146] Calculate the fitness of each individual and arrange it in ascending order of fitness, select the optimal individual into the elite pool Elite Pool (size E size ).

[0147] Iteration process: for each individual, the following updates are made:

[0148] Cold group: the individual rationally moves to the average value of the individuals in the elite pool:

[0149] Where center is the average value of the load rate of the individuals in the elite pool.

[0150] Aggregation group: the individual is updated under the influence of the cold group and the panic group:

[0151] Where PLR p is an individual randomly selected from the panic group, and v h is the speed vector of the aggregation group.

[0152] Panic group: the individual randomly selects a new load distribution:

[0153]

[0154] Among them, E j For individuals randomly selected from the elite pool, PLR rand is a randomly selected individual, v p is the velocity vector of the panic group.

[0155] Convergence phase:

[0156] When the number of iterations exceeds T max / 2All individuals tend to the optimal solution in the elite pool.

[0157] All individuals are updated to elite pool members or randomly selected individuals.

[0158] Panic Index Adjustment:

[0159] As the number of iterations t increases, the panic index P(t) gradually decreases, and the behavior of the simulated individual gradually stabilizes: P(t) = P max (1-t / T max )

[0160] Output the optimal solution:

[0161] Output optimal load distribution scheme PLR opt , which corresponds to the load distribution with the lowest fitness.

[0162] 3) Actual application results:

[0163] Assume that after several rounds of iterations, the load distribution scheme finally determined by the ESC algorithm may be as follows:

[0164] Unit 1: load factor 0.6

[0165] Unit 2: load factor 0.3

[0166] Unit 3: load factor 0.7

[0167] Unit 4: load factor 0.5

[0168] Unit 5: load factor 0.4

[0169] At this time, the total energy consumption of the system is minimized and the chilled water temperature is within the set range. Through the optimization of the ESC algorithm, the chiller group can be dynamically adjusted according to the external environment, load requirements and internal unit performance, thereby achieving energy efficiency optimization while meeting the chilled water temperature and load requirements.

[0170] When the chiller is in the initial operation stage and there is no historical data for the objective function and constraint condition modeling, in the absence of a chiller part-load rate-energy consumption model, based on experience and a large amount of literature research, the empirical P=f(PLR) formula can be used to calculate the safe operation interval of 30% to 100% of the chiller. The required number of units and unit numbers are calculated in order according to the current required load (which can be the load prediction result or the actual measurement result). The load is distributed to the high-efficiency interval of each opened unit according to the empirical formula P=f(PLR), and the start-up scheme, unit number, number of units, and each load rate are output.

[0171] In the initial operation stage of the chiller, an initial model is established, real-time system operation data are collected and stored based on a data acquisition platform, and the state of the chiller system is monitored regularly to detect the accuracy of the load prediction model and the accuracy of the chiller load distribution model. A reasonable accuracy range is set, and when the accuracy exceeds the allowed range, the load prediction model and the chiller load distribution model are corrected to ensure the stability and efficient operation of the system.

[0172] Example 2. Chiller group control system based on ESC

[0173] The system is based on the above-mentioned ESC chiller group control method and realizes the group control optimization of the chiller. The system combines load prediction, energy efficiency optimization, and dynamic adjustment functions to minimize the total energy consumption of the chiller group and improve the operation efficiency. The specific modules include data acquisition, load prediction, energy efficiency optimization, ESC algorithm implementation, accuracy evaluation, etc., and the load distribution and operation state of the chiller are continuously optimized through a closed-loop mechanism.

[0174] 1. Data acquisition and monitoring module

[0175] This module collects real-time data related to the operation of the chiller through sensors and monitoring devices. The specific data includes but is not limited to the following:

[0176] Meteorological parameters: temperature, humidity, wind speed, and other environmental data that affect the load demand of the chiller.

[0177] Air conditioning area equipment heat dissipation: heat dissipation from air conditioning area equipment, which affects the refrigeration load of the chiller.

[0178] Personnel load situation: the amount of personnel activity directly affects the load demand of the air conditioning area.

[0179] These real-time data will be transmitted to other modules of the system for subsequent load prediction and energy efficiency model correction.

[0180] 2. Data processing module

[0181] This module is mainly responsible for pre-processing the operating data of the water chiller, ensuring the accuracy and integrity of the data. The specific functions include:

[0182] Data cleaning: remove outliers, missing values and noise data to ensure data quality.

[0183] Data standardization: normalize different types of data to eliminate the influence of different dimensions and facilitate subsequent model calculations.

[0184] Feature extraction: extract important features related to load prediction and energy efficiency optimization from raw data, such as device power, load fluctuations, etc.

[0185] The pre-processed data will be passed to the downstream load prediction and energy efficiency model modules.

[0186] 3. Data storage module

[0187] The data storage module is used to store processed data. It saves real-time data and historical data after preprocessing in the database for subsequent analysis and model training. This module has the following characteristics:

[0188] Data storage: store real-time collected and processed sensor data, device operating status, load demand prediction, etc.

[0189] Long-term tracking: able to record long-term operating data of the water chiller and provide data support for energy efficiency model calibration.

[0190] Efficient access: use efficient data access mechanisms to ensure that real-time data can be quickly transmitted to other modules for processing.

[0191] 4. Load prediction module

[0192] The load prediction module is one of the core parts of the entire system. By receiving data from the data acquisition and monitoring module, it uses a neural network model based on LSTM (Long Short-Term Memory) and KAN (Attention Mechanism) layers to perform load prediction. The specific steps include:

[0193] Input data: from real-time collected meteorological parameters, device heat generation, personnel activity, etc.

[0194] LSTM layer: process time series data, learn and capture long-term dependencies of load changes.

[0195] KAN layer: assign different time step weights through attention mechanism, highlight the impact of important information on prediction results.

[0196] Output results: output future load demand prediction as the basis for subsequent group control strategies and energy efficiency optimization.

[0197] This module effectively captures the timing and dynamic changes of the load, providing accurate load prediction.

[0198] 5. Energy efficiency model and self-updating mechanism module

[0199] This module establishes an energy efficiency model for the chiller unit, describing the relationship between the part load ratio (PLR) and power consumption. The energy efficiency model takes into account the unit's energy efficiency performance under different operating conditions (such as partial load and full load). The system corrects the model through real-time operation data to ensure the matching of load prediction and actual energy efficiency requirements.

[0200] Model self-updating mechanism: When there is a large deviation between the operation data of the chiller unit and the prediction results, this mechanism can automatically adjust the model parameters to ensure the continuous accuracy of the model.

[0201] Performance optimization: Dynamic adjustment according to data deviation in actual operation to maintain the stability and efficiency of the system.

[0202] 6. Group control strategy module

[0203] The group control strategy module is responsible for designing the group control strategy of the system based on the load prediction results and energy efficiency optimization goals. The main work of this module includes:

[0204] Optimization objective function: Design and implement the optimization objective function, usually to minimize the total energy consumption of the system:

[0205]

[0206] Where Q need is the total demand for system cooling load, and Obj is the minimum target power consumption of the parallel chiller system.

[0207] Where the part load ratio of the chiller unit is:

[0208]

[0209] Where PLR i is the part load ratio of the i-th chiller, is the actual load of the chiller, is the design load of the chiller.

[0210]

[0211] Where P i is the power consumption of the i-th chiller, and a, b, c, d are the fitting coefficients of the main machine-part load ratio curve.

[0212] In a parallel chiller system, sufficient cooling load should be provided to meet the system cooling load demand. To ensure efficient operation of the chiller, the PLR value must be between 0.3 and 1; when the chiller is off, the PLR is 0.

[0213] Load distribution scheme: According to the group control objective function, adjust the load rate and the number of on-off units of each chiller unit to achieve energy efficiency optimization.

[0214] 7. ESC algorithm module

[0215] The ESC algorithm module is used to perform chiller unit group control optimization. This module uses the ESC (Escape Optimization) algorithm to simulate the behavior of the crowd, and adjusts the load distribution of the chiller unit through mechanisms such as exploration phase, calm group, aggregation group, panic group, etc.:

[0216] Individual initialization: Randomly initialize the load rate PLR of each chiller unit, with a range of 0 to 1.

[0217] Fitness calculation: Evaluate the pros and cons of each chiller unit load distribution scheme through the fitness function, with the goal of minimizing system energy consumption.

[0218] Simulation of group behavior: According to different individual behaviors (calm, aggregation, panic), update the load distribution scheme, and finally converge to the global optimal solution.

[0219] This module optimizes the load distribution of the chiller unit through multiple iterations, ensuring that the chiller unit group is energy efficient.

[0220] 8. Output module

[0221] This module outputs the final chiller unit load distribution scheme based on the optimization results. The output content includes:

[0222] Load rate (PLR) of each chiller unit

[0223] Number of on-off units

[0224] Other necessary operating parameters

[0225] The output scheme will be applied to the chiller unit group control system in real time to ensure that the system always operates efficiently.

[0226] 9. Real-time monitoring and precision evaluation module

[0227] This module is used to monitor the running state of the chiller unit in real time and evaluate the precision of the load prediction and energy efficiency model. Its functions include:

[0228] Precision evaluation: Regularly evaluate the gap between load prediction and actual operation to ensure that the model's prediction results match the actual operation requirements.

[0229] Dynamic correction: Based on the evaluation results, the load forecast model and energy efficiency model are corrected to ensure that the system can adjust its operating strategy in a timely manner according to environmental changes.

[0230] This module ensures the stability and accuracy of the entire system during long-term operation.

[0231] The overall workflow of the system

[0232] Data collection: Real-time collection of environmental data, equipment data, personnel activities and other information.

[0233] Data processing: Cleaning, standardization and feature extraction of collected data.

[0234] Load forecasting: Use a neural network combining LSTM and KAN to predict load.

[0235] Energy efficiency model calibration and optimization: Calibrate and optimize energy efficiency models based on real-time data to ensure that predictions match actual operations.

[0236] Group control strategy design: Adjust the load distribution plan of the chiller according to the load forecast results and optimization objectives.

[0237] ESC algorithm execution: The ESC algorithm simulates individual behavior, optimizes load distribution and ensures optimal energy efficiency.

[0238] Output and application: Apply the optimization results to the chiller group control system to ensure efficient operation of the system.

[0239] The chiller group control system, based on the ESC algorithm, proposed in this paper effectively addresses existing issues such as low load forecasting accuracy and the tendency of optimization algorithms to fall into local optimal solutions. Through the collaborative work of multiple modules, including real-time data acquisition, load forecasting, energy efficiency model adjustment, and ESC algorithm optimization, the system achieves efficient operation of chiller groups, minimizes total energy consumption, and ensures dynamic satisfaction of cooling load requirements. This system has significant application prospects and market value.

[0240] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A chiller group control method based on ESC, characterized in that: The steps include: S1. Load Forecasting: The system uses sensors to collect real-time data related to cooling load, including but not limited to meteorological parameters, heat dissipation of equipment in the air-conditioning area, personnel load, and fresh air load. Load forecasting algorithm: A neural network model combining a long short-term memory (LSTM) network with a KAN layer of attention mechanism performs load forecasting based on real-time meteorological data, equipment heating conditions, and personnel activities. The model can capture the sequential nature and dynamic changes of cooling loads, improving forecast accuracy. S2. Based on the operating characteristics of the chiller, construct an energy efficiency model to describe the relationship between load rate and power consumption; The introduction of a model self-updating device can adjust the parameters of the energy efficiency model in real time based on the operating data of the chiller, ensuring that the load forecast matches the actual energy efficiency requirements; If the model prediction deviation is large, the model parameters will be automatically corrected to ensure long-term stable operation; S3. Based on the load forecast results, design the objective function of the chiller group control to minimize the total energy consumption of the system; Constraints include the load factor of each chiller being within a reasonable range to ensure efficient operation of the system; The partial load rate (PLR) of the chiller is used as the key variable for optimization. The ESC algorithm is used to determine the number of chillers on and off and the load rate of each chiller. The goal is to optimize the energy efficiency of the entire chiller fleet and reduce energy waste; S4. Application of ESC Algorithm S41. Initialization: First, initialize the load factor PLR of each chiller and randomly assign it through the individual initialization mechanism of the ESC algorithm. The load factor of each chiller is between 0 and 1, reflecting the load state in the initial stage. S42. Fitness calculation: Calculate the fitness of each individual. The fitness function evaluates the quality of the load distribution scheme by calculating the total energy consumption of the system. The goal is to minimize the total energy consumption of the system while ensuring that the load requirements of each chiller are met; S43. According to the ESC algorithm, the load distribution problem is transformed into an individual update problem; Individuals are divided into three groups: calm, gathered, and panicked: Calm individuals: Individuals rationally move closer to the system load center and update the load distribution plan; Aggregate individuals: Individuals are influenced by calm individuals and panic individuals to update the load distribution plan; Panic individuals: Individuals make random adjustments when load fluctuates drastically, simulating irrational responses in load distribution; S44. Panic Index and Convergence: As the number of iterations increases, the panic index gradually decreases, individual behaviors gradually stabilize, and eventually converge to the global optimal solution. S45. Output the optimal solution: After multiple iterations, the final load distribution plan is obtained, and the optimal number of chiller units on and off and load rate information is output.

2. The chiller group control method based on ESC according to claim 1, characterized in that: In step S3, the partial load rate of the chiller is: Among them, PLR i is the partial load rate of the i-th chiller, is the actual load of the chiller, is the design load of the chiller; Where: P i is the power consumption of the i-th chiller, a, b, c, d are the fitting coefficients of the host-part load rate curve.

3. The chiller group control method based on ESC according to claim 2, characterized in that: In a parallel chiller system, sufficient cooling load should be provided to meet the system cooling load requirements. To ensure efficient operation of the chiller, the PLR ​​value must be between 0.3 and 1. When the chiller is off, the PLR ​​is 0. The objective function and constraints are as follows: Where Q need is the total cooling load demand of the system, and Obj is the minimum target power consumption of the parallel chiller system.

4. The chiller group control method based on ESC according to claim 3, characterized in that: In step S4, the following inputs are set: the number of chillers N; the initial load factor of each chiller PLR0 = [PLR1, PLR2, ..., PLR N ], and the initial value satisfies 0.3≤PLR i ≤1 or 0; The COP value of each unit COP=[COP1, COP2, ..., COP N ], maximum number of iterations T max , elite pool size E size ; Set output: optimal load distribution plan PLR opt , ensuring ∑Q i PLR i ≥Q need , which corresponds to the load distribution with the lowest fitness.

5. The chiller group control method based on ESC according to claim 4, characterized in that: In the step S41, the following steps are also included: S411. Randomly generate the load rate PLR0 of the chiller, and the initial value satisfies 0.3≤PLR i ≤1 or 0; S412. Initialize the population, where each individual in the population represents a load distribution plan for a chiller; S413. Set the maximum number of iterations T max .

6. The chiller group control method based on ESC according to claim 5, characterized in that: In the step S42, the following steps are also included: S421. For each individual PLR i , calculate its power consumption P i , and calculate the total power consumption of the system; S422. Calculate the degree to which the load distribution meets the cooling load demand and ensure ∑Q i PLR i ≥Q need ; S423. Comprehensive fitness: Among them, λ is a penalty factor. When the cooling load demand is not met, the fitness will increase the penalty term.

7. The chiller group control method based on ESC according to claim 6, characterized in that: In the step S43, the following steps are also included: S431. Calculate the fitness of each individual and sort them in ascending order of fitness, and select the best individual to enter the elite pool Pool , size is E size ; S432. Iterative process: For each individual, perform the following updates: Calm group: Individuals rationally move toward the average of the elite pool: Among them, center is the average value of individual load rates in the elite pool; Gathering groups: Individuals affected by calm and panic groups Update: Among them, PLR p is a randomly selected individual from the panic group, v h is the velocity vector of the gathering group; Panic herd: individuals randomly choose new load distribution: Among them, E j For individuals randomly selected from the elite pool, PLR rand is a randomly selected individual, v p is the velocity vector of the panic group.

8. The chiller group control method based on ESC according to claim 7, characterized in that: When the number of iterations exceeds T max / 2All individuals tend to the optimal solution in the elite pool, and all individuals are updated to members of the elite pool or randomly selected individuals; as the number of iterations t increases, the panic index P(t) gradually decreases, and the behavior of the simulated individuals gradually stabilizes: P(t)=P max ·(1-t / T max )。 9. A chiller group control system based on ESC, characterized in that: The method comprises the steps of the chiller group control method according to any one of claims 1 to 8, and is optimized and adjusted based on the following modules to ensure efficient operation of the chiller group control system and achieve optimal energy efficiency; The modules include: Data acquisition and monitoring module: used to collect meteorological parameters, heat dissipation of air-conditioning area equipment, and personnel load data in real time, and transmit the real-time data to the data storage module to provide support for load forecasting and energy efficiency models; Load forecasting module: Receives real-time data from the data acquisition module, uses a neural network model that combines the long short-term memory network (LSTM) with the attention mechanism (KAN) layer to perform load forecasting and output future load demand; Energy efficiency model and self-update mechanism module: Based on the operating characteristics of the chiller, an energy efficiency model is established to describe the relationship between load rate and power consumption. When there is a large deviation between actual operating data and model predictions, the model parameters are automatically adjusted to ensure long-term stable operation. Group control strategy module: Based on load forecast results, the group control strategy is designed. The objective function is to minimize the total energy consumption of the system, and the constraint condition is the reasonable range of the chiller load rate. This module uses the ESC algorithm to optimize the load distribution and the number of chillers on and off to achieve optimal system energy efficiency. ESC algorithm module: This module is responsible for executing load distribution optimization tasks. It uses the ESC algorithm to simulate individual behavior and optimize the load distribution of the chillers. Based on the adaptability of the chiller load distribution, this module adjusts the PLR ​​and the number of units on and off for each unit and outputs the optimal solution. Output module: Based on the optimization results, it outputs the load rate and number of on / off units of each chiller, and applies the solution to the chiller group control system to ensure the optimal energy efficiency of the system; Real-time monitoring and accuracy assessment module: used to monitor the operating status of the chiller in real time, detect the accuracy of the load forecast model and energy efficiency model, ensure that the load distribution plan matches the actual operating requirements, and make corrections based on real-time data.

10. The chiller group control system based on ESC according to claim 9, characterized in that: Also includes: Data processing module, used for pre-processing chiller operation data; A data storage module, used for storing pre-processed operation data; The model correction module is used to correct the parameters of the chiller energy efficiency model based on the preprocessed operating data.

Images