Central air-conditioning system energy-saving optimization method based on pump machine independent control strategy

Abstract

The invention discloses a central air-conditioning system energy-saving optimization method and system based on a pump machine independent control strategy, and the method comprises the steps: collecting operation data, constructing a load prediction model, and predicting the cooling and heating load of a building at the next moment; dividing the load into a plurality of categories according to the predicted load in combination with a preset classification rule; aiming at each category, with the minimum total energy consumption or maximum energy efficiency ratio of the system as a target, optimizing the operation frequency and the number of the cold and hot water pumps under the condition of meeting thermal comfort and equipment safety constraints to obtain corresponding optimization parameters; and calling the corresponding category optimization parameters, and implementing frequency conversion control on the cold and hot water pump. The load prediction accuracy is improved through the BP neural network, and the energy waste is avoided and the energy efficiency is improved in combination with load classification and pre-optimization. Independent pump control is implemented to realize accurate'quantity adjustment ', and mass-quantity coupling control can be expanded. An accurate mechanism model is not needed, the applicability is high, the energy-saving effect is remarkable, and the application prospect is good.

Description

Technical Field

[0001] This invention belongs to the field of HVAC energy-saving control technology, specifically relating to an energy-saving optimization method and system for central air conditioning systems based on an independent pump control strategy. Background Technology

[0002] Central air conditioning systems, as the main energy-consuming equipment in large public buildings and commercial complexes, typically account for 40% to 60% of the building's total energy consumption. Among them, the energy consumption of the hot and cold water circulation system (including chilled water pumps and cooling water pumps) can account for 20% to 30% of the total energy consumption of the air conditioning system. Traditional water pump control often adopts fixed-frequency operation or simple variable frequency control based on pressure difference / temperature. This results in low adjustment accuracy, slow response, and difficulty in adapting to dynamic changes in building load, leading to common operating phenomena such as "high flow rate and small temperature difference," causing significant energy waste.

[0003] To improve the energy efficiency of air conditioning systems, researchers have proposed various optimization control strategies in recent years, such as model predictive control (MPC), data-driven optimization, and intelligent algorithm optimization. Patent CN114580254B discloses a method and system for regulating indoor temperature in buildings based on model predictive control. First, an indoor temperature prediction model is established to predict room temperature. The method uses feedback control of indoor temperature based on system temperature or pressure differences to reduce air conditioning system energy consumption while improving indoor thermal comfort, exhibiting good control quality for both indoor temperature and cooling capacity. Some solutions achieve system-level collaborative optimization by predicting building load and adjusting the operating status of the main unit and water pumps in advance. However, these methods often rely on precise system mechanism models, resulting in complex modeling, difficult parameter tuning, and large online computational loads, making them difficult to promote and apply in practical engineering. Furthermore, most control strategies treat water pumps as auxiliary equipment for passive regulation, lacking refined modeling and active optimization of the independent operating characteristics of the water system.

[0004] In recent years, with the development of variable frequency technology and intelligent control algorithms, pump-independent control has gradually gained attention. This strategy emphasizes decoupling hot and cold water pumps from the main unit's integrated control, independently adjusting their operating frequency and the number of pumps starting and stopping based on actual hydraulic demand to achieve "on-demand water supply." Combined with load forecasting and optimization algorithms, optimal pump operating parameters can be dynamically matched under different operating conditions, improving system flexibility and energy efficiency. However, achieving low-computational-cost, highly adaptable pump-independent optimization control while ensuring thermal comfort and equipment safety remains a key technological challenge.

[0005] Current energy-saving control methods for central air conditioning systems generally suffer from problems such as insufficient load forecasting accuracy, poor real-time performance of optimization strategies, and weak adaptability of control strategies to complex operating conditions. In particular, under low load or variable operating conditions, pump over-operation or lag in regulation can easily occur, leading to a decrease in system energy efficiency. Furthermore, the lack of a classification optimization mechanism for different load characteristics makes it difficult to balance energy saving and operational stability. Therefore, there is an urgent need for an independent pump control method to achieve energy-efficient and optimized operation of central air conditioning systems within a given operating range. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide an energy-saving optimization method for central air conditioning systems based on an independent pump control strategy. This method aims to solve the technical problems caused by the difficulty of traditional control strategies in balancing the independent adjustment needs of water pumps with the overall energy efficiency optimization of the system, resulting in high energy consumption, slow response, and poor control stability under variable load conditions.

[0007] The objective of this invention is achieved through the following technical solution: An energy-saving optimization method for a central air conditioning system based on an independent pump control strategy includes the following steps: Collect operational data of the central air conditioning system, including outdoor meteorological parameters, indoor environmental parameters, equipment operating status parameters, and historical load data; A load forecasting model for a central air conditioning system is constructed, and the building's heating and cooling load for the next moment is predicted based on the load forecasting model. Based on the predicted building heating and cooling loads, and combined with the preset load classification rules, the loads are divided into several load categories; For each load category, with the goal of minimizing total system energy consumption or maximizing system energy efficiency ratio, and under the constraints of thermal comfort and equipment safety, the operating frequency and number of hot and cold water pumps are optimized to obtain the optimal operating parameters corresponding to that load category. When the system is running, it acquires the system operation data at the current moment, performs load prediction and classification, calls the optimized operation parameters of the corresponding load category, and implements variable frequency control on the hot and cold water pumps to achieve energy-saving and optimized operation of the central air conditioning system.

[0008] This invention is based on a closed-loop control principle of prediction, classification, optimization, and execution. It constructs a BP neural network load prediction model by collecting multi-source operational data to accurately predict the heating and cooling loads at the next moment. Furthermore, it divides complex operating conditions into typical load ranges using load classification rules, and optimizes the optimal operating frequency and number of pumps to start / stop for each category offline or online, forming a condition-parameter mapping strategy. During system operation, it matches the load category in real time and calls the corresponding optimized parameters, implementing independent frequency conversion control of the pumps through a frequency converter to achieve refined "on-demand water supply." This method significantly reduces the energy consumption of the pumps and the overall system while ensuring indoor thermal comfort and equipment safety, improving operational efficiency. It also avoids the problems of relying on mechanistic models and heavy computational burden in traditional control methods, demonstrating excellent energy-saving effects and engineering applicability.

[0009] As a preferred approach, the construction of the central air conditioning system load prediction model specifically includes: The collected historical operational data is preprocessed, including correlation analysis, anomaly identification and interpolation, data smoothing and noise reduction, and normalization. A load forecasting model is constructed using a data-driven black-box model, wherein the black-box model is a BP neural network model; The preprocessed data is divided into training and testing sets to train and validate the BP neural network model until the model's prediction accuracy meets the preset threshold.

[0010] As a preferred approach, the correlation analysis employs the Pearson correlation analysis method to screen input variables that are significantly correlated with building heating and cooling loads. The abnormal data identification uses box plot detection, and the interpolation processing uses linear interpolation, with the following formula: ;in, For the first One missing data point. and These are two normal data points before and after the missing value. , , These are the corresponding data sampling times; The data smoothing and noise reduction uses a moving average smoothing algorithm; The normalization formula is: ;in, The values ​​are normalized. The original value, and These are the minimum and maximum values ​​of the variable in the training set.

[0011] As a preferred embodiment, the input layer parameters of the BP neural network model include: the current outdoor dry-bulb temperature, outdoor relative humidity, hourly time, day of the week, and the current load value; the output layer parameter is the predicted load value for the next time step; the hidden layer is a single-layer structure with a certain number of neurons. Calculate using the following formula: ;in, The number of neurons in the input layer. The number of neurons in the output layer. It is a constant, and its range is [value range missing]. The activation function used is the Sigmoid function, and the training function uses the Levenberg-Marquardt algorithm.

[0012] As a preferred approach, the load classification rules include interval classification or classification based on decision tree regression; If the interval classification method is used, then based on the load factor: ; The load is divided into several intervals, among which This is the current load value. This represents the historical maximum load value. If a classification method based on decision tree regression is adopted, the outdoor temperature, outdoor relative humidity, hourly time, weekday time, and the previous load are used as input variables, and the current load is used as the output variable to construct a decision tree regression model. The load is divided into several categories by splitting variables and splitting points.

[0013] As a preferred embodiment, the classification method based on decision tree regression further includes a pruning step: defining a subtree loss function: ;in, To predict the error based on the training data, This represents the number of leaf nodes in the subtree. This is a complexity parameter; Each internal node is calculated recursively. Pruning gain: ;in, For nodes As a loss for a single node, For subtree with root The loss, For subtree The number of leaf nodes, select The smallest subtree is pruned to generate a subtree sequence, and the optimal subtree is selected through cross-validation.

[0014] As a preferred embodiment, the constraints include: Indoor thermal environment constraints: summer: ; winter: ; in, Indoor temperature, Indoor relative humidity; Cold and heat source unit water supply temperature constraints: ;in, For water supply temperature, and The minimum and maximum permissible water supply temperatures; Hot and cold water pump frequency constraints: ;in, For the water pump operating frequency, The lowest permissible frequency, in Hz; Cooling water supply temperature constraints: ;in, For cooling water supply temperature, and The minimum and maximum allowable cooling water supply temperatures.

[0015] As a preferred approach, the optimization process is implemented using an optimization algorithm: When the number of optimization parameters is no more than two, a grid search algorithm or a genetic algorithm is used. When the number of optimization parameters is greater than 3, particle swarm optimization, sparrow search algorithm or joint optimization algorithm shall be adopted. The objective function is to minimize the total energy consumption of the system. : ;in, For host power consumption, For water pump energy consumption, These are end-point energy consumption figures, all in kW. Or maximize the system's energy efficiency ratio: ;in, Provides cooling or heating for the system, measured in kW.

[0016] As a preferred embodiment, the variable frequency control of the hot and cold water pump is implemented based on a frequency converter, and the controller determines the target frequency from the optimized operating parameters. Send control commands to the frequency converter, and the frequency converter adjusts the motor speed, thereby changing the water pump flow rate and realizing "quantity regulation"; When the main unit's water supply temperature is constant, the pumps can be independently controlled by adjusting the pump frequency and the number of pumps. When the main unit's water supply temperature is adjustable, the water supply temperature and water pump frequency are optimized in a coordinated manner to achieve coupled control of "quality regulation" and "quantity regulation".

[0017] An energy-saving optimization system for a central air conditioning system based on an independent pump control strategy, comprising: The data acquisition module is used to collect the operation data of the central air conditioning system, including outdoor meteorological parameters, indoor environmental parameters, equipment operating status parameters, and historical load data. The first construction module is used to construct a load prediction model for the central air conditioning system and to predict the building's heating and cooling load for the next moment based on the load prediction model. The second construction module is used to divide the load into several load categories based on the predicted building heating and cooling loads and in combination with preset load classification rules. The first optimization module is used to optimize the operating frequency and number of hot and cold water pumps for each load category, with the goal of minimizing the total system energy consumption or maximizing the system energy efficiency ratio, while meeting the constraints of thermal comfort and equipment safety, and obtain the optimized operating parameters corresponding to that load category. The execution control module is used to acquire the system operation data in real time when the system is running, perform load prediction and classification, call the optimized operation parameters of the corresponding load category, and implement variable frequency control of the hot and cold water pumps to achieve energy-saving and optimized operation of the central air conditioning system.

[0018] The present invention has at least the following beneficial effects: This invention collects multi-source data, including outdoor meteorological parameters, indoor environmental parameters, and equipment operating status, to construct a load prediction model based on a BP neural network. This model accurately predicts the building's heating and cooling load for the next moment, improving the system's responsiveness to dynamic load changes. Combining load classification rules, the operating conditions are divided into several load categories, and energy consumption optimization is performed in advance for each category to obtain the optimal operating frequency and number combination of the corresponding hot and cold water pumps. This avoids the energy waste problems of traditional fixed-frequency or simple variable-frequency control, significantly improving the system's operating energy efficiency.

[0019] Furthermore, this invention aims to minimize total system energy consumption or maximize energy efficiency ratio. Under multiple constraints, including indoor thermal comfort (e.g., 24-27℃ in summer, 18-22℃ in winter), main unit water supply temperature, pump frequency lower limit, and cooling water temperature, it employs intelligent optimization algorithms such as grid search, genetic algorithms, or particle swarm optimization to optimize parameters, ensuring the safety and feasibility of energy-saving control. Independent pump control is implemented through frequency converters for both hot and cold water pumps, achieving precise quantity regulation based on the target frequency. When the water supply temperature is adjustable, it can be expanded to "quality-quantity coupled" control, enhancing the system's adaptability. The overall solution requires no precise mechanistic model, is applicable to various building air conditioning systems, and demonstrates excellent energy-saving effects and promising engineering application prospects. Attached Figure Description

[0020] To reveal the technical details of the embodiments of the present invention, the accompanying drawings involved in the embodiments will be briefly described below. It should be emphasized that these drawings only present several embodiments of the present invention and should not be considered as defining the scope of the invention. For those skilled in the art, other related drawings can still be derived based on these drawings without inventive effort.

[0021] Figure 1 This is a schematic diagram of a method for optimizing energy conservation in a central air conditioning system based on an independent pump control strategy. Figure 2 A schematic diagram of the process of building a load forecasting model. Detailed Implementation

[0022] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings, but the scope of protection of the present invention is not limited to the following description.

[0023] In the following description, embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. However, it should be understood that the present disclosure is not limited to the specific forms shown herein. Rather, it should be understood to encompass various variations, equivalents, and / or alternatives to the embodiments of the present disclosure. In illustrating the drawings, the same reference numerals will be used to denote similar components.

[0024] It should be clarified that while the following description provides detailed specific information to aid in a comprehensive understanding of the exemplary embodiments, those skilled in the art will recognize that the exemplary embodiments can be implemented even without these specific details. For example, the system may be illustrated using block diagrams to avoid excessive detail that could obscure the clarity of the example. In other cases, to maintain the clarity of the example, unnecessary details of well-known processes, structures, and techniques may be omitted.

[0025] See Figure 1An energy-saving optimization method for central air conditioning systems based on an independent pump control strategy, applied to a central air conditioning system controller, includes the following steps: Collect operational data of the central air conditioning system, including outdoor meteorological parameters, indoor environmental parameters, equipment operating status parameters, and historical load data; A load forecasting model for a central air conditioning system is constructed, and the building's heating and cooling load for the next moment is predicted based on the load forecasting model. Based on the predicted building heating and cooling loads, and combined with the preset load classification rules, the loads are divided into several load categories; For each load category, with the goal of minimizing total system energy consumption or maximizing system energy efficiency ratio, and under the constraints of thermal comfort and equipment safety, the operating frequency and number of hot and cold water pumps are optimized to obtain the optimal operating parameters corresponding to that load category. When the system is running, it acquires the system operation data at the current moment, performs load prediction and classification, calls the optimized operation parameters of the corresponding load category, and implements variable frequency control on the hot and cold water pumps to achieve energy-saving and optimized operation of the central air conditioning system.

[0026] The central air conditioning system controller collects operational data such as outdoor meteorological parameters, indoor environmental parameters, equipment operating status, and historical loads. Based on a BP neural network, a load prediction model is constructed to predict the building's heating and cooling load at the next moment, achieving forward-looking control. Subsequently, combined with preset load classification rules (such as by load rate range or decision tree regression method), the predicted load is divided into several typical load categories, simplifying complex operating conditions into manageable operating modes. For each load category, under the constraints of indoor thermal comfort, main unit water supply temperature, and minimum pump frequency, the operating frequency and number of hot and cold water pumps are optimized offline or online with the goal of minimizing total system energy consumption or maximizing energy efficiency ratio, generating corresponding optimized operating parameters and forming a strategy mapping. During system operation, current data is acquired in real time, load prediction and classification are performed, and the optimal parameters for the corresponding category are automatically called. The pumps are independently controlled by frequency converters to dynamically adjust the water flow, achieving "on-demand water supply" and avoiding energy waste in traditional control. Thus, while ensuring comfort and safety, the system's operating energy efficiency is significantly improved.

[0027] In a preferred embodiment, the construction of the central air conditioning system load prediction model specifically includes: The collected historical operational data is preprocessed, including correlation analysis, anomaly identification and interpolation, data smoothing and noise reduction, and normalization. A load forecasting model is constructed using a data-driven black-box model, wherein the black-box model is a BP neural network model; The preprocessed data is divided into training and testing sets to train and validate the BP neural network model until the model's prediction accuracy meets the preset threshold.

[0028] In a preferred embodiment, the correlation analysis employs the Pearson correlation analysis method to screen input variables that are significantly correlated with building heating and cooling loads; The abnormal data identification uses box plot detection, and the interpolation processing uses linear interpolation, with the following formula: ;in, For the first One missing data point. and These are two normal data points before and after the missing value. , , These are the corresponding data sampling times; The data smoothing and noise reduction uses a moving average smoothing algorithm; The normalization formula is: ;in, The values ​​are normalized. The original value, and These are the minimum and maximum values ​​of the variable in the training set.

[0029] In a preferred embodiment, the input layer parameters of the BP neural network model include: the current outdoor dry-bulb temperature, outdoor relative humidity, hourly time, day of the week, and the current load value; the output layer parameter is the predicted load value for the next time step; the hidden layer is a single-layer structure with a certain number of neurons. Calculate using the following formula: ; in, The number of neurons in the input layer. The number of neurons in the output layer. It is a constant, and its range is [value range missing]. The activation function used is the Sigmoid function, and the training function uses the Levenberg-Marquardt algorithm.

[0030] This embodiment uses the central air conditioning system of a commercial complex as an example to illustrate the process of constructing a load forecasting model based on Pearson correlation analysis, box plot anomaly detection, linear interpolation, moving average smoothing, normalization, and BP neural network (see [link to documentation]). Figure 2 ).

[0031] 1. Data Collection

[0032] Historical operational data for six consecutive months (including typical summer and winter operating periods) was collected from the Building Automation System (BAS), with a sampling frequency of once every 15 minutes, totaling approximately 29,000 records. The raw data includes: Outdoor meteorological parameters: dry bulb temperature (°C), relative humidity (%); Time characteristics: Hourly time (0-23), Weekday time (1-7, marked as weekday / weekend); System operating parameters: Total cooling load of the chiller unit (kW); Other auxiliary parameters: solar radiation, average indoor temperature, etc. (used for screening after correlation analysis).

[0033] 2. Data Preprocessing

[0034] (1) Pearson correlation analysis The Pearson correlation coefficients between each variable and the current cooling load were calculated, and the results are as follows: variable Correlation coefficient Outdoor dry bulb temperature 0.81 outdoor relative humidity 0.43 Hours 0.68 Weekday 0.32 Current load value 0.95 With a significant correlation threshold of 0.3, outdoor dry-bulb temperature, outdoor relative humidity, hourly time, day of the week, and current load value were selected as input variables for the BP neural network.

[0035] (2) Abnormal data identification and interpolation processing Outlier detection using box plots: Calculate the interquartile range (IQR) for each variable. Data points with values ​​less than Q1-1.5IQR or greater than Q3+1.5IQR were identified as anomalous. Approximately 680 anomalous or missing data points were identified.

[0036] For intervals with no more than 3 consecutive missing points, linear interpolation is used for repair. The interpolation formula is: ;in: For the first One missing data point; These are two normal data points before and after the missing point; The corresponding data sampling time (unit: minutes).

[0037] If a temperature sequence is missing at 10:15, and the normal values ​​before and after it are 26.2℃ at 10:00 and 26.8℃ at 10:30, then the interpolation value is:

[0038] (3) Data smoothing and noise reduction The cooling load sequence was smoothed using a 5-point moving average method, with the following formula: ;in, Indicates at a point in time Smoothed heating and cooling load values ​​at the location (unit: kW); Indicates the first [number]th [item] in the original load sequence The original heating and cooling load values ​​(unit: kW) at each sampling time; This is the time index of the current sampling point; The variable to be summed has a range of values. arrive That is, taking the current point as the center, taking two data points before and after it, for a total of 5 adjacent samples.

[0039] It effectively suppresses measurement noise and short-term fluctuations, improving the stability of model training.

[0040] (4) Data normalization All input variables are min-max normalized based on the extreme values ​​of the training set, using the following formula: ;in The normalized value, The original value, , These are the minimum and maximum values ​​of the variable in the training set. After normalization, all data are mapped to the interval [0,1].

[0041] 3. Construction and Training of BP Neural Network Model

[0042] Network architecture design: Number of neurons in the input layer (Corresponding to 5 input variables); Number of neurons in the output layer (Predicting the load at the next moment); The hidden layer is a single layer, and the number of neurons is calculated using an empirical formula. ; Pick ,have to Rounded up to 6, the hidden layer is set to 6 neurons.

[0043] Activation function: The hidden layer uses the Sigmoid function. ; Training function: The Levenberg-Marquardt algorithm (trainlm) is adopted, which has the advantages of fast convergence and high accuracy; Data partitioning: The preprocessed data was divided into a training set (approximately 23,200 records) and a test set (approximately 5,800 records) in an 8:2 ratio. Model training objective: Mean squared error. The maximum number of training rounds is 1000.

[0044] 4. Model Validation and Results

[0045] After training, the model's prediction performance on the test set is as follows: Mean relative error (MRE): 5.8% Root mean square error (RMSE): 8.2kW Coefficient of determination (R²): 0.93 The predicted values ​​and measured values ​​show a high degree of consistency in trend, meeting the preset accuracy requirements (MRE<8%). The model training is completed and saved for subsequent load prediction and pump group optimization control.

[0046] In a preferred embodiment, the load classification rule includes interval classification or a classification method based on decision tree regression; If the interval classification method is used, then based on the load factor: ; The load is divided into several intervals, among which This is the current load value. This represents the historical maximum load value. If a classification method based on decision tree regression is adopted, the outdoor temperature, outdoor relative humidity, hourly time, weekday time, and the previous load are used as input variables, and the current load is used as the output variable to construct a decision tree regression model. The load is divided into several categories by splitting variables and splitting points.

[0047] In a preferred embodiment, the classification method based on decision tree regression further includes a pruning step: defining a subtree loss function. ;in, To predict the error based on the training data, This represents the number of leaf nodes in the subtree. This is a complexity parameter; Each internal node is calculated recursively. Pruning gain: ;in, For nodes As a loss for a single node, For subtree with root The loss, For subtree The number of leaf nodes, select The smallest subtree is pruned to generate a subtree sequence, and the optimal subtree is selected through cross-validation.

[0048] This embodiment is based on the central air conditioning system of a comprehensive office building with a building area of ​​25,000㎡. The sampling period is 15 minutes, and 6 consecutive months of operation data (approximately 24,000 records) are collected to verify two load classification methods: interval classification and decision tree regression-based classification.

[0049] 1. Interval-based Classification (1) Calculate the load factor. Select the historical maximum cooling load. kW (taken from the peak value of the hottest day in summer). Measured cooling load at each moment. Calculate the load factor:

[0050] (2) Divide the load range. Based on operating experience and energy efficiency characteristics, the load rate is divided into three typical operating conditions: Low load area:

[0051] Medium load area:

[0052] High-load area:

[0053] (3) Examples of classification results Record the cooling load at 10:00 on a certain day kW, then ; It was classified as a low-load category.

[0054] This method is simple and intuitive, suitable for buildings with relatively stable load change patterns, and has a clear classification logic, making it easy to deploy in engineering projects.

[0055] 2. Decision Tree Regression-based Classification (1) Model building The following variables are used as input features: Outdoor dry-bulb temperature (°C) Outdoor relative humidity (%) Hourly time (0~23) Weekdays (Monday to Sunday) Load value at the previous moment (kW) Output variable: Current cooling load (kW) The CART (Classification and Regression Tree) algorithm is used to construct a regression tree model, with the goal of automatically learning data splitting rules to divide the load into several typical categories.

[0056] (2) Model training and segmentation process Training was performed using 80% of the data (approximately 19,000 records), with the following settings: Maximum depth: 5 Minimum number of leaf node samples: 100 Split Criterion: Minimize Mean Squared Error (MSE) After training, the model generates multiple segmentation rules. For example: If the outdoor temperature is <28°C, enter the left subtree. If the short time period is ≥18, it is classified as a "low-load zone". If the short time interval is ∈ [10, 15] and the temperature is ≥ 32°C, it is classified as a "high-load zone". (3) Classification and naming The final decision tree divides the input space into four leaf nodes, corresponding to four load patterns. Category A: High load (>2500kW), mostly occurring between 10:00-16:00 on weekdays, with high temperature and humidity; Category B: Medium load (1200-2500kW), during off-peak hours on weekdays; Category C: Low load (<1200kW), at night or on weekends; Category D: Very low load (<600kW), during holidays or early morning startup phases.

[0057] (4) Runtime classification process The system acquires current environmental and time parameters in real time, inputs them into the trained decision tree model, automatically matches them to a certain leaf node, obtains the corresponding load category label, and uses it to call the preset pump group optimization parameters.

[0058] In a preferred embodiment, the constraints include: Indoor thermal environment constraints: summer: ; winter: ; in, Indoor temperature, Indoor relative humidity; Cold and heat source unit water supply temperature constraints: ;in, For water supply temperature, and The minimum and maximum allowable water supply temperatures for the equipment; Hot and cold water pump frequency constraints: ;in, For the water pump operating frequency, The lowest permissible frequency, in Hz; This is the minimum permissible frequency to ensure the system's minimum circulating flow rate and prevent cavitation and motor overheating. This constraint prevents the pump from operating at excessively low frequencies, which could lead to insufficient flow, increased vibration, or poor heat dissipation.

[0059] Cooling water supply temperature constraints: ;in, For cooling water supply temperature, and The minimum and maximum allowable cooling water supply temperatures.

[0060] In the energy-saving optimization method for the central air conditioning system in this embodiment, setting constraints is a key step to ensure the safety, reliability, and feasibility of the optimized control. The system categorizes operating conditions into typical classes such as high, medium, and low through load prediction and classification, and optimizes the pump frequency and number for each class to achieve optimal energy consumption. However, without reasonable physical and environmental boundary constraints, the optimization process may generate control strategies that, while low in energy consumption, are not feasible, leading to equipment damage or deterioration of comfort.

[0061] For example, under high-load conditions in summer (such as a load rate exceeding 90%), if the optimization algorithm only aims to minimize energy consumption and ignores constraints, it may tend to reduce the pump operating frequency or increase the supply water temperature to reduce the number of chiller units in operation. However, too low a pump frequency will lead to insufficient water flow, causing the chiller evaporator to freeze; too high a supply water temperature will result in insufficient heat exchange at the terminal, making it difficult to achieve the desired indoor temperature. In addition, if the indoor temperature exceeds the summer comfort range (24~27℃), it will directly affect the thermal comfort of people and violate the basic function of the air conditioning system.

[0062] Therefore, this invention sets multi-dimensional constraints: indoor thermal environment constraints ensure comfort; main unit water supply temperature constraints ensure efficient equipment operation; water pump frequency constraints prevent cavitation or insufficient flow caused by low-frequency operation; and cooling water temperature constraints prevent shutdown caused by excessive condensation pressure. These constraints serve as boundary conditions for the optimization problem, ensuring that the obtained water pump operating parameters are both energy-efficient and safe under different load categories.

[0063] By embedding constraints into the optimization model, this invention achieves a synergy between energy saving and safe, stable operation. Especially within the step-by-step optimization framework based on load classification, different categories can be configured with differentiated constraint thresholds, enhancing control flexibility. Ultimately, when the system invokes the corresponding category's optimization parameters during real-time operation, it ensures that the frequency converter control commands remain within acceptable ranges for both the equipment and the environment, truly achieving sustainable, energy-efficient, and optimized operation.

[0064] In a preferred embodiment, to improve the accuracy of the total system energy consumption prediction and thus enhance the reliability of the optimization results, this embodiment refines the energy consumption model of the hot and cold water pumps. Traditional methods often use linear or fixed efficiency models to estimate pump energy consumption, neglecting the nonlinear coupling characteristics of the motor-pump-pipeline system under variable frequency operation. Therefore, we propose a comprehensive pump energy consumption model based on actual operating characteristics, comprehensively considering the pump efficiency curve, motor efficiency changes, and dynamic characteristics of pipeline impedance.

[0065] The formula for the comprehensive energy consumption model of the water pump is as follows: ;in, This indicates the pump shaft power, measured in kW. This represents the density of water, with a value of approximately kg / m³; Represents gravitational acceleration, with values ​​ranging from 1 to 10. m / s²; This indicates the pump head, expressed in meters (m). This indicates the water pump flow rate, measured in m³ / h. This represents pump efficiency, a dimensionless quantity, obtained from the pump performance curve, and is expressed as flow rate. The function; Motor efficiency is a dimensionless quantity, usually expressed as frequency. The empirical function can be expressed as: ,in The coefficients are the fitting coefficients for motor efficiency.

[0066] The model is determined through real-time acquisition or table lookup. and Combined with head and traffic Relationship (usually) , By constructing a dynamic energy consumption model that varies with frequency (using the pipeline impedance coefficient), the accuracy of pump energy consumption calculation is significantly improved, providing a more accurate basis for subsequent optimization with the goal of minimizing the total system energy consumption.

[0067] In a preferred embodiment, to further quantify the energy-saving potential of independent pump control and to evaluate the rationality of system hydraulic distribution, the present invention introduces a hydraulic balance index. Energy-saving potential index This is used to guide the dynamic adjustment of optimization strategies. The hydraulic balance degree is defined. The normalized value of the ratio of the most unfavorable loop flow to the average loop flow: ;in, This represents the hydraulic balance, with a value range of (0,1). The closer it is to 1, the more balanced the hydraulic distribution of the system. This represents the branch flow with the minimum flow rate among all terminal branches, expressed in units of... ; This represents the average flow rate of all terminal branches, in units of... .

[0068] Based on hydraulic balance, an energy-saving potential index is defined. as follows: ;in, The energy-saving potential index reflects the energy-saving potential that can be tapped under current operating conditions by optimizing pump control. This indicates the current operating frequency of the water pump, in Hz. When the energy-saving potential index threshold (e.g., 0.3) is reached, it indicates that the system has significant hydraulic imbalance and the pump operating frequency is too high, indicating significant energy-saving potential. In this case, the system will prioritize using a high-precision optimization algorithm (such as particle swarm optimization) to optimize parameters. Otherwise, a lightweight algorithm will be used to maintain operation. This mechanism enables adaptive selection of optimization strategies, improving the intelligence level and energy-saving response capability of the control system.

[0069] In a preferred embodiment, to prevent cavitation or overheating caused by internal flow instability during low-frequency variable frequency operation of the water pump, this invention introduces a safety boundary criterion based on flow stability. The criterion defines the water pump's frequency... During operation, the Reynolds number at the impeller inlet is: ;in For inlet flow rate, This is the flow velocity-frequency proportionality coefficient. The hydraulic diameter, Let Re be the kinematic viscosity of the fluid. At the critical Reynolds number (taken as 3000), the flow tends towards laminar flow, which easily leads to instability. Therefore, the minimum safe operating frequency can be obtained: If the target frequency is lower than The system will automatically increase the frequency or the number of operating units to ensure flow stability.

[0070] In a preferred embodiment, to compensate for the efficiency decrease caused by long-term operation of the water pump, this embodiment introduces a dynamic efficiency decay compensation factor. The current actual efficiency is defined as: ;in , Cumulative operating time (years) The initial efficiency of the water pump. This represents the maximum attenuation ratio (e.g., 0.1). The decay rate constant (take) The system is updated periodically based on measured data. and This enables adaptive correction of the energy consumption model, ensuring long-term energy-saving effects.

[0071] In a preferred embodiment, the optimization process is implemented using an optimization algorithm: When the number of optimization parameters is no more than two, a grid search algorithm or a genetic algorithm is used. When the number of optimization parameters is greater than 3, particle swarm optimization, sparrow search algorithm or joint optimization algorithm shall be adopted. The objective function is to minimize the total energy consumption of the system. : ;in, For host power consumption, For water pump energy consumption, These are end-point energy consumption figures, all in kW. Or maximize the system's energy efficiency ratio: ;in, Provides cooling or heating for the system, measured in kW.

[0072] In the energy-saving optimization method of this invention, an appropriate intelligent optimization algorithm is adopted for different numbers of optimization parameters to ensure efficient solution of the optimal control strategy under various constraints. When there are few parameters to be optimized (such as only pump frequency and number of pumps), the system complexity is low, and a grid search algorithm or genetic algorithm can be used for global traversal or evolutionary optimization. For example, under a certain medium load category, if it is only necessary to determine the operating frequency (30~50 Hz) and start / stop combination (0 / 1) of two pumps, all feasible solutions can be exhausted by grid search with a step size of 1 Hz, and the total energy consumption of the system can be quickly located by combining the energy consumption model. The smallest combination is simple to calculate and yields accurate results.

[0073] However, when the number of optimization variables increases (e.g., optimizing more than three parameters such as the frequency of multiple water pumps, the main unit's water supply temperature, and the speed of the terminal fan), the search space grows exponentially, making traditional methods computationally inefficient. In this case, the system automatically switches to swarm intelligence optimization algorithms such as Particle Swarm Optimization (PSO) or Sparrow Search Algorithm (SSA). These algorithms achieve efficient global search by simulating swarm behavior, avoiding getting trapped in local optima. For example, under high-load conditions, the system needs to collaboratively optimize five variables: the frequency of three water pumps, the setpoint of the main unit's outlet water temperature, and the speed of the cooling tower fan. Using PSO, it can converge to a near-optimal solution within hundreds of iterations, significantly reducing computation time and making it suitable for offline training or online rolling optimization in practical engineering applications.

[0074] The objective function is flexibly set according to the control objective: if energy saving is the core objective, then minimize the total energy consumption of the system. If a greater focus is on energy efficiency, then maximize the system's energy efficiency ratio:

[0075] During peak electricity pricing periods, prioritize selecting "Minimize". The primary objective is to reduce operating costs; however, during the equipment evaluation period, the objective of "maximizing EERs" is adopted to improve overall energy efficiency. Through adaptive matching of multiple objectives and algorithms, this invention achieves efficient and accurate optimization under different load categories, providing reliable parameter support for subsequent independent frequency conversion control of pumps.

[0076] In a preferred embodiment, the method further includes: real-time monitoring of whether indoor temperature and humidity constraints are met; if the constraints are violated, the method triggers a re-optimization calculation process, updates the load prediction model and energy consumption simulation model using the latest collected operating data, and re-executes the optimization to generate new optimized operating parameters.

[0077] In the actual operation of a central air conditioning system, although the optimized control strategy is generated based on accurate load forecasting and strict constraints, external disturbances (such as sudden weather changes, dense crowds, and equipment failures) can still cause key parameters to exceed limits. Therefore, this invention introduces a real-time monitoring and feedback re-optimization mechanism to ensure that the system always operates within a safe, comfortable, and efficient range.

[0078] For example, one afternoon, the system predicted a load of 2800 kW, classifying it as a high-load condition, and invoked preset optimization parameters: starting three hot and cold water pumps, setting the operating frequency to 42 Hz, and setting the main unit's water supply temperature to 7.0℃. After the control command was issued, the system continuously monitored the terminal environment through indoor temperature and humidity sensors. One hour later, monitoring data showed that the indoor temperature in some areas had risen to 27.8℃, exceeding the summer comfort limit of 27℃, while the relative humidity reached 72%, slightly exceeding the 70% limit. At this point, the system determined that the thermal comfort constraint had been violated and immediately triggered a re-optimization process.

[0079] The system automatically performs the following operations: First, it incrementally updates the BP neural network load prediction model using the latest collected outdoor temperature, humidity, and actual load data to correct subsequent load trends. Second, it reconstructs the energy consumption simulation model based on the current equipment operating status. Finally, based on the updated model, it re-executes optimization calculations to find a new combination of pump frequency and number of pumps while satisfying all constraints. After re-optimization, the system decides to increase the pump frequency to 46 Hz and add one more pump to increase water flow and reduce the supply air temperature difference, thereby improving the terminal cooling effect.

[0080] Through this feedback mechanism, the system has upgraded from open-loop optimization to closed-loop adaptive optimization. It not only promptly corrects constraint violations but also improves the accuracy of subsequent control through model updates, effectively addressing practical engineering challenges such as sudden load fluctuations or model mismatches, and ensuring the continuous achievement of the dual goals of energy conservation and comfort.

[0081] In a preferred embodiment, the variable frequency control of the hot and cold water pump is implemented based on a frequency converter, and the controller determines the target frequency from the optimized operating parameters. Send control commands to the frequency converter, and the frequency converter adjusts the motor speed, thereby changing the water pump flow rate and realizing "quantity regulation"; When the main unit's water supply temperature is constant, the pumps can be independently controlled by adjusting the pump frequency and the number of pumps. When the main unit's water supply temperature is adjustable, the water supply temperature and water pump frequency are optimized in a coordinated manner to achieve coupled control of "quality regulation" and "quantity regulation".

[0082] In the actual operation of central air conditioning systems, variable frequency control of hot and cold water pumps is a key step in achieving energy-saving optimization. This invention achieves precise regulation of pump flow through the collaboration of the controller and the frequency converter. For example, during a medium-load period in an office building (such as 10:00 AM), if the system is determined to be in a "medium-load condition" based on load prediction and classification, the corresponding optimization parameter, the target frequency, will be invoked. Two pumps are in operation. The controller sends the target frequency to the pump's inverter in analog (e.g., 4~20mA) or communication protocol (e.g., Modbus) format. Upon receiving the command, the inverter adjusts the power supply frequency of the drive motor, reducing the motor speed to 76% of the rated speed (since frequency is proportional to speed), thereby reducing the pump flow rate and achieving "flow regulation." At this time, the main unit's water supply temperature remains constant at 7.0℃. The system matches the load demand by independently adjusting the pump's operating parameters, avoiding the energy waste caused by the "valve throttling" of traditional fixed-frequency pumps.

[0083] When the system possesses more advanced control capabilities, it can further achieve "quality-quantity coupled control." For example, during low-load periods at night (such as 22:00), the predicted load is only 30% of the maximum value, and the system classifies it as a "low-load category." At this time, the main unit's water supply temperature is no longer fixed. The controller, through collaborative optimization, determines to increase the water supply temperature from 7.0℃ to 9.0℃, while simultaneously reducing the pump frequency to 32Hz, and reducing the number of operating pumps to one. Increasing the water supply temperature (i.e., "quality regulation") reduces the main unit's cooling intensity and improves its energy efficiency ratio; while reducing the pump frequency ("quantity regulation") reduces distribution energy consumption. The synergistic effect of these two measures significantly reduces the overall energy consumption of the system.

[0084] The "quality-quantity coupled control" mode fully leverages the advantages of independent pump control: the water pump no longer passively follows the main unit, but actively adjusts its flow rate according to the optimization strategy, while simultaneously linking with the main unit's temperature setting. The controller monitors operational feedback in real time, and if it detects room temperature deviation or insufficient flow, it can trigger a re-optimization mechanism to dynamically adjust the target frequency and supply water temperature. Through the precise execution of the frequency converter, this invention achieves an upgrade from "extensive regulation" to "precise on-demand supply," significantly improving the operational energy efficiency and adaptive capabilities of the central air conditioning system.

[0085] An energy-saving optimization system for a central air conditioning system based on an independent pump control strategy, comprising: The data acquisition module is used to collect the operation data of the central air conditioning system, including outdoor meteorological parameters, indoor environmental parameters, equipment operating status parameters, and historical load data. The first construction module is used to construct a load prediction model for the central air conditioning system and to predict the building's heating and cooling load for the next moment based on the load prediction model. The second construction module is used to divide the load into several load categories based on the predicted building heating and cooling loads and in combination with preset load classification rules. The first optimization module is used to optimize the operating frequency and number of hot and cold water pumps for each load category, with the goal of minimizing the total system energy consumption or maximizing the system energy efficiency ratio, while meeting the constraints of thermal comfort and equipment safety, and obtain the optimized operating parameters corresponding to that load category. The execution control module is used to acquire the system operation data in real time when the system is running, perform load prediction and classification, call the optimized operation parameters of the corresponding load category, and implement variable frequency control of the hot and cold water pumps to achieve energy-saving and optimized operation of the central air conditioning system.

[0086] A large commercial center has implemented refined energy-saving management of its hot and cold water pumps. The system consists of a data acquisition module, a first construction module, a second construction module, a first optimization module, and an execution control module, working collaboratively to form a closed-loop control system encompassing perception, prediction, decision-making, and execution. In actual operation, the data acquisition module acquires real-time data on outdoor temperature and humidity, indoor temperature and humidity, chiller status, pump operating frequency, and historical load through a sensor network, and uploads this data to the central controller. The first construction module establishes a load prediction model based on a BP neural network, training and validating it using historical data from the past 7 days to predict the building's cooling load for the next moment, with a prediction accuracy (MRE) controlled within 6%. For example, at 9:00 AM on a certain day, the system predicts that the load at 10:00 AM will be 2100 kW. Subsequently, the second construction module, based on preset load classification rules (such as division by load rate range), classifies 2100 kW (accounting for 65.6% of the maximum load of 3200 kW) as a "medium load category." The first optimization module is then activated. Under constraints such as an indoor temperature of 24-27℃, a water pump frequency of no less than 30 Hz, and a main unit water supply temperature of 6.5-7.5℃, the system aims to minimize total energy consumption. Using a particle swarm optimization algorithm, it optimizes the operating frequency and start-stop combinations of the two parallel water pumps, arriving at the optimal strategy: running both pumps at a frequency of 36 Hz. Finally, at 10:00, the execution control module invokes this optimized parameter, sending a 4-20mA control signal to the pump motor via the frequency converter to adjust its speed and achieve precise flow matching. The system continuously monitors the operating status. If the indoor temperature approaches the 27℃ upper limit, a re-optimization process is triggered, dynamically adjusting the frequency to 38 Hz to ensure comfort. Through this system, water pump distribution energy consumption is reduced by approximately 25%, and the overall air conditioning system energy efficiency is significantly improved, achieving safe, stable, and efficient energy-saving operation.

[0087] In a preferred embodiment, the first building module further includes: The preprocessing submodule is used to perform correlation analysis, anomaly identification and interpolation, data smoothing and noise reduction, and normalization on the collected historical operational data. The model training submodule is used to use the BP neural network model as a black box model, divide the preprocessed data into training set and test set, and perform model training and validation. The model validation submodule is used to evaluate the model's prediction accuracy. When the mean relative error (MRE) or root mean square error (RMSE) meets the preset threshold, the model is saved.

[0088] The first construction module collaboratively builds a high-precision load forecasting model through three functional sub-modules. The preprocessing sub-module first systematically processes the collected historical operational data to improve data quality. Pearson correlation analysis is used to screen input variables significantly related to building heating and cooling loads, eliminating redundant features and reducing model complexity. Box plots are used to identify and remove outlier data points to prevent erroneous data from interfering with model training. Linear interpolation is used to repair missing or outlier values, ensuring data continuity. Moving averages are used to smooth and reduce noise in the load sequence, suppressing measurement noise and short-term fluctuations. Finally, the data is normalized to eliminate dimensional differences and accelerate subsequent model convergence.

[0089] The model training submodule employs a backpropagation (BP) neural network as a data-driven black-box prediction model, fully leveraging its powerful nonlinear fitting capabilities to capture the complex relationship between meteorological parameters, temporal characteristics, and building load. The preprocessed data is divided into training and testing sets, and the network weights are continuously adjusted using the backpropagation algorithm to ensure the model output closely approximates the actual load value, achieving accurate load prediction for the next time step.

[0090] The model validation submodule is responsible for evaluating the predictive performance of the trained model, typically using mean relative error (MRE) or root mean square error (RMSE) as evaluation metrics. When the error metric meets a preset threshold, it indicates that the model has sufficient predictive accuracy and can be saved and used in practice; otherwise, it will be fed back to the training phase to adjust the network structure or retrain. This mechanism ensures the reliability of load prediction results, providing an accurate basis for subsequent load classification and optimal control.

[0091] In a preferred embodiment, the second building module uses interval classification or decision tree regression-based classification to classify the load; If the interval classification method is used, then based on the load factor:

[0092] Divide the load into zones; If a classification method based on decision tree regression is adopted, the outdoor temperature, outdoor relative humidity, hourly time, weekday time, and load at the previous moment are used as input variables to construct a decision tree regression model, and the load is divided into several categories through splitting rules.

[0093] The core principle of the second building module is to divide the continuously changing building heating and cooling loads into several typical operating condition categories, providing a clear basis for subsequent classification and optimization of operating modes. This module supports two classification methods, each suitable for different engineering scenarios.

[0094] When using the interval classification method, its principle is based on the relative quantification of load factor. This is achieved by calculating the current load. Compared to the historical maximum load ratio The load is divided into three typical ranges: low, medium, and high (e.g., For low load, Medium load, (For high loads). This method is simple in principle, computationally efficient, and suitable for buildings with stable load characteristics and strong operational regularity, facilitating rapid project deployment. When using a classification method based on decision tree regression, the principle is to automatically learn the load change pattern under multi-factor coupling using a data-driven approach. Using outdoor temperature, relative humidity, time characteristics (hours, days of the week), and the previous load as input variables, and the current load as the output, a regression tree model is constructed. The optimal splitting variable and splitting point (e.g., outdoor temperature) are selected recursively. And hours This method divides the input space into multiple sub-regions, with each leaf node corresponding to a load pattern. It captures nonlinear relationships and complex influencing factors, enabling more refined and adaptive load condition identification, and is suitable for complex and ever-changing building operating environments. Both methods discretize continuous loads into a finite number of categories, effectively reducing the complexity of optimization control and enabling "category-based" energy-saving strategy management, laying the foundation for subsequent independent optimization for different load categories.

[0095] Alternatively, the second module can be used to classify continuously changing building heating and cooling loads into several typical operating condition categories, providing a basis for subsequent classification and optimization of operating modes. This module supports two classification methods. The first is the interval classification method: calculating the current load... Compared to the historical maximum load ratio ,according to The numerical values ​​are used to classify working conditions. For example, when... It was a low-load operating condition. It was operating under medium load conditions. The first method is a high-load condition. The threshold can be adjusted according to the building type and this data. The method is simple and efficient, suitable for scenarios with strong load regularity. The second method is a decision tree-based classification method: using outdoor temperature, relative humidity, time characteristics (hour, weekday), and the previous load as input variables, and a predefined load condition category (such as low, medium, high) as the output label, a classification decision tree model is constructed. The model learns from historical data and automatically forms a classification decision tree model, such as "if the outdoor temperature...". and Hour The discrimination rule of "high load condition" divides the input space into multiple regions, with each leaf node corresponding to a load mode. This method can capture multi-factor coupling and nonlinear relationships, and is suitable for complex and ever-changing operating environments. Both methods discretize continuous loads into a finite number of categories, reducing the complexity of optimization control and realizing energy-saving management based on "classification", providing a foundation for subsequent independent optimization for different load categories.

[0096] In a preferred embodiment, the first optimization module employs a grid search algorithm, genetic algorithm, particle swarm optimization algorithm, or sparrow search algorithm to minimize the total system energy consumption while satisfying constraints such as indoor temperature and humidity, main unit water supply temperature, water pump frequency, and cooling water temperature. ; Or maximize the system's energy efficiency ratio: With the goal of optimizing the operating parameters of hot and cold water pumps, the following steps are taken.

[0097] The first optimization module, based on load classification results, seeks the optimal combination of pump operating parameters for each typical operating condition, while satisfying system safety and comfort constraints, thereby achieving the control objective of energy saving or optimal energy efficiency. This module employs multiple intelligent optimization algorithms, adaptively selecting the appropriate algorithm based on the complexity of the optimization variables. When the number of optimization parameters is small (e.g., only pump frequency and number), a grid search algorithm or genetic algorithm is used to accurately find the optimal value in a low-dimensional space through global traversal or evolutionary search. When multiple variables are involved in collaborative optimization (e.g., multiple pump frequencies, water supply temperature, start-stop status, etc.), a faster-converging and more efficient swarm intelligence algorithm, such as Particle Swarm Optimization (PSO) or Sparrow Search algorithm, is used. It efficiently approximates the global optimum in high-dimensional space. The optimization process aims to minimize the total energy consumption of the system. (in For host power consumption, Energy consumption for water pump distribution (Energy consumption of end-of-line fans) or maximizing the system's energy efficiency ratio (EERs) ( Using cooling / heating as the objective function, the energy-saving effect is ensured to be quantifiable. All optimization processes are conducted under strict constraints, including indoor temperature and humidity ranges (e.g., 24-27°C in summer), main unit water supply temperature limits, lower limits for water pump frequency (to prevent cavitation), and safe cooling water temperature ranges, to avoid equipment failure or environmental exceedances due to excessive energy saving. Through this optimization mechanism, the system generates a set of optimal operating parameters (e.g., target frequency, number of operating units) for each load category, forming a working condition and strategy mapping table. This provides a basis for subsequent variable frequency control, achieving an energy-saving operation mode of on-demand energy supply and precise regulation.

[0098] In a preferred embodiment, a closed-loop optimization system encompassing classification, optimization, invocation, control, and feedback is constructed to achieve refined energy-saving operation of the central air conditioning chilled and hot water pumps. The system operates on a 15-minute control cycle. It first collects data such as outdoor dry-bulb temperature, relative humidity, indoor temperature and humidity, equipment operating status, and historical load. This data is then input into a pre-processed and validated BP neural network load prediction model to predict the building's cooling and heating load for the next moment. Subsequently, based on the load factor (in The load is divided into three typical operating conditions (the highest historical load): low load ( ), medium load ( ), high load The classification results are used to match preset optimization strategies. Before the system went live, offline optimization calculations were performed for each load condition to ensure that optimal parameters could be quickly invoked during online operation. The optimization process aims to minimize the system's total energy consumption. (Sum of energy consumption of main unit, water pump, and fan) or maximize system energy efficiency ratio (EERs) The objective was to conduct the experiment while meeting several constraints, including indoor temperature and humidity (24-27°C in summer). 、 40%~70%RH), main unit water supply temperature (6.5~7.5°C), water pump frequency ( Cooling water temperature (28~35°C), etc. Different optimization algorithms are used for different load categories: Low load conditions: fewer optimization parameters (usually only the frequency of one water pump), a grid search algorithm is used to traverse the range of 30~40Hz in 1 Hz step size to find the combination with the lowest energy consumption; Medium load condition: This involves the combination of frequency and number of 2 water pumps, and a genetic algorithm is used for evolutionary optimization. High-load conditions: There are many variables (more than 3 water pumps, frequency, and main unit temperature working together), and the particle swarm algorithm is used to efficiently search for the global optimum.

[0099] After optimization is completed, the system generates and stores the optimal operating parameter tables for each category. For example: Low load category: Run 1 water pump, target frequency 32 Hz, water supply temperature set to 8.0°C (slightly higher is allowed to reduce the operating intensity of the main unit); Medium load category: 2 water pumps in operation, target frequency 38Hz, water supply temperature 7.0°C; High load category: 3 water pumps in operation, target frequency 46Hz, supply water temperature 6.5°C (to ensure heat exchange capacity).

[0100] During online operation, the controller directly calls the corresponding parameters based on the current load category and sends the target frequency command to the frequency converter of the designated water pump via the communication interface. The frequency converter adjusts the motor power supply frequency, changing the pump speed and flow rate to achieve quantity regulation. The water pumps operate independently, no longer relying on the start / stop logic of the main unit, truly achieving independent pump control. For example, when the system determines that it has entered a medium load condition, it automatically starts two water pumps and sets the frequency to 38Hz, avoiding the energy waste caused by valve throttling in traditional fixed-frequency pumps.

[0101] To ensure long-term operational stability, the system is equipped with a feedback correction module. It monitors indoor temperature and humidity in real time. If the temperature exceeds 27°C or the humidity is above 70%, it determines that the current control strategy fails to meet comfort requirements, possibly due to a sudden load increase or model bias. In this case, the system immediately triggers a re-optimization process: updating the load prediction model with the latest data and re-performing the optimization calculation to generate a new combination of pump frequency and number, which is then put into use in the next control cycle. This mechanism enables the system to be adaptive, maintaining optimal energy efficiency and comfort even under environmental changes, equipment aging, or holiday / holiday conditions.

[0102] In a preferred embodiment, a feedback correction module is further included, which is used to monitor in real time whether the indoor temperature and humidity exceed a preset threshold. If they do, a feedback signal is sent to the first construction module and the first optimization module to trigger model updates and parameter re-optimization, so as to ensure that the system operates in an optimal and safe state.

[0103] The feedback correction module is a key component of this system to achieve closed-loop control and adaptive optimization. Its core principle is to dynamically correct the prediction model and optimization strategy by monitoring the operating status in real time, so as to cope with the risk of control failure caused by sudden environmental changes, load disturbances or model deviations.

[0104] The feedback correction module continuously collects indoor temperature and humidity sensor data to determine whether it exceeds the preset comfort range (e.g., 24~27℃, 40%~70% in summer). Once it detects an indoor temperature exceeding 27℃ or humidity exceeding 70%, it determines that the current control strategy fails to meet the thermal environment requirements, possibly caused by load forecasting deviations, equipment performance degradation, or external disturbances (such as crowds or sudden weather changes). At this point, the module immediately generates a feedback signal and sends it to the first construction module and the first optimization module.

[0105] Upon receiving the feedback signal, the first construction module incrementally updates or retrains the load prediction model using the latest collected operational data, improving the model's adaptability to the current operating conditions. Simultaneously, the first optimization module re-executes optimization calculations based on the updated prediction results and the current system state, generating new combinations of pump operating frequencies and numbers while satisfying various constraints. Through this monitoring, feedback, and correction mechanism, the system achieves an upgrade from open-loop optimization to closed-loop adaptive control, effectively ensuring the dual stability of energy-saving performance and indoor environmental quality.

[0106] While preferred embodiments of the present invention have been described, those skilled in the art, upon grasping the basic innovative concepts, may propose further adjustments and improvements to these embodiments. Therefore, the appended claims are intended to cover both the preferred embodiments and all adjustments and improvements that fall within the scope of this invention. The foregoing is merely a preferred example of the invention and does not constitute a limitation thereof. It should be emphasized that any modifications, equivalent substitutions, or optimizations made within the spirit and principles of this invention should be considered within the protection scope of this invention.

Claims

1. An energy-saving optimization method for a central air conditioning system based on an independent pump control strategy, characterized in that, Includes the following steps: Collect operational data of the central air conditioning system, including outdoor meteorological parameters, indoor environmental parameters, equipment operating status parameters, and historical load data; A load forecasting model for a central air conditioning system is constructed, and the building's heating and cooling load for the next moment is predicted based on the load forecasting model. Based on the predicted building heating and cooling loads, and combined with the preset load classification rules, the loads are divided into several load categories; For each load category, with the goal of minimizing total system energy consumption or maximizing system energy efficiency ratio, and under the constraints of thermal comfort and equipment safety, the operating frequency and number of hot and cold water pumps are optimized to obtain the optimal operating parameters corresponding to that load category. When the system is running, it acquires the system operation data at the current moment, performs load prediction and classification, calls the optimized operation parameters of the corresponding load category, and implements variable frequency control on the hot and cold water pumps to achieve energy-saving and optimized operation of the central air conditioning system.

2. The energy-saving optimization method for a central air conditioning system based on an independent pump control strategy according to claim 1, characterized in that, The construction of the central air conditioning system load prediction model specifically includes: The collected historical operational data is preprocessed, including correlation analysis, anomaly identification and interpolation, data smoothing and noise reduction, and normalization. A load forecasting model is constructed using a data-driven black-box model, wherein the black-box model is a BP neural network model; The preprocessed data is divided into training and testing sets to train and validate the BP neural network model until the model's prediction accuracy meets the preset threshold.

3. The energy-saving optimization method for a central air conditioning system based on an independent pump control strategy according to claim 2, characterized in that, The correlation analysis employed Pearson correlation analysis to screen input variables that were significantly correlated with building heating and cooling loads. The abnormal data identification uses box plot detection, and the interpolation processing uses linear interpolation, with the following formula: ;in, For the first One missing data point. and These are two normal data points before and after the missing value. , , These are the corresponding data sampling times; The data smoothing and noise reduction uses a moving average smoothing algorithm. The normalization formula is: ;in, The values ​​are normalized. The original value, and These are the minimum and maximum values ​​of the variable in the training set.

4. The energy-saving optimization method for a central air conditioning system based on an independent pump control strategy according to claim 1, characterized in that, The input layer parameters of the BP neural network model include: the current outdoor dry-bulb temperature, outdoor relative humidity, hour, day of the week, and the current load value; the output layer parameter is the predicted load value for the next time step; the hidden layer is a single-layer structure with a certain number of neurons. Calculate using the following formula: ;in, The number of neurons in the input layer. The number of neurons in the output layer. It is a constant, and its range is [value range missing]. The activation function used is the Sigmoid function, and the training function uses the Levenberg-Marquardt algorithm.

5. The energy-saving optimization method for a central air conditioning system based on an independent pump control strategy according to claim 1, characterized in that, The load classification rules include interval classification or classification based on decision tree regression; If the interval classification method is used, then based on the load factor: ； The load is divided into several intervals, among which This is the current load value. This represents the historical maximum load value. If a classification method based on decision tree regression is adopted, the outdoor temperature, outdoor relative humidity, hourly time, weekday time, and the previous load are used as input variables, and the current load is used as the output variable to construct a decision tree regression model. The load is divided into several categories by splitting variables and splitting points.

6. The energy-saving optimization method for a central air conditioning system based on an independent pump control strategy according to claim 5, characterized in that, The classification method based on decision tree regression also includes a pruning step: defining the subtree loss function: ;in, To predict the error based on the training data, This represents the number of leaf nodes in the subtree. This is a complexity parameter; Each internal node is calculated recursively. Pruning gain: ;in, For nodes As a loss for a single node, For subtree with root The loss, For subtree The number of leaf nodes, select The smallest subtree is pruned to generate a subtree sequence, and the optimal subtree is selected through cross-validation.

7. The energy-saving optimization method for a central air conditioning system based on an independent pump control strategy according to claim 1, characterized in that, The constraints include: Indoor thermal environment constraints: summer: ; winter: ; in, Indoor temperature, Indoor relative humidity; Cold and heat source unit water supply temperature constraints: ;in, For water supply temperature, and The minimum and maximum permissible water supply temperatures; Hot and cold water pump frequency constraints: ;in, For the water pump operating frequency, The lowest permissible frequency, in Hz; Cooling water supply temperature constraints: ;in, For cooling water supply temperature, and The minimum and maximum allowable cooling water supply temperatures.

8. The energy-saving optimization method for a central air conditioning system based on an independent pump control strategy according to claim 1, characterized in that, The optimization process is implemented using an optimization algorithm: When the number of optimization parameters is no more than two, a grid search algorithm or a genetic algorithm is used. When the number of optimization parameters is greater than 3, particle swarm optimization, sparrow search algorithm or joint optimization algorithm shall be adopted. The objective function is to minimize the total energy consumption of the system. : ;in, For host power consumption, For water pump energy consumption, All figures represent end-point energy consumption, in kW. Or maximize the system's energy efficiency ratio: ;in, Provides cooling or heating for the system, measured in kW.

9. The energy-saving optimization method for a central air conditioning system based on an independent pump control strategy according to claim 1, characterized in that, The variable frequency control of the hot and cold water pump is implemented based on a frequency converter, and the controller determines the target frequency from the optimized operating parameters. Send control commands to the frequency converter, and the frequency converter adjusts the motor speed, thereby changing the water pump flow rate and realizing "quantity regulation"; When the main unit's water supply temperature is constant, the pumps can be independently controlled by adjusting the pump frequency and the number of pumps. When the main unit's water supply temperature is adjustable, the water supply temperature and water pump frequency are optimized in a coordinated manner to achieve coupled control of "quality regulation" and "quantity regulation".

10. An energy-saving optimization system for a central air conditioning system based on an independent pump control strategy, characterized in that, include: The data acquisition module is used to collect the operation data of the central air conditioning system, including outdoor meteorological parameters, indoor environmental parameters, equipment operating status parameters, and historical load data. The first construction module is used to construct a load prediction model for the central air conditioning system and to predict the building's heating and cooling load for the next moment based on the load prediction model. The second construction module is used to divide the load into several load categories based on the predicted building heating and cooling loads and in combination with preset load classification rules. The first optimization module is used to optimize the operating frequency and number of hot and cold water pumps for each load category, with the goal of minimizing the total system energy consumption or maximizing the system energy efficiency ratio, while meeting the constraints of thermal comfort and equipment safety, and obtain the optimized operating parameters corresponding to that load category. The execution control module is used to acquire the system operation data in real time when the system is running, perform load prediction and classification, call the optimized operation parameters of the corresponding load category, and implement variable frequency control of the hot and cold water pumps to achieve energy-saving and optimized operation of the central air conditioning system.

Images