Energy-saving operation control method for air conditioning system

Abstract

The invention provides an energy-saving operation control method for an air conditioning system, which has the characteristic of remarkably improving the load prediction precision and the system energy efficiency, and comprises the following steps of: performing double-path prediction of external environment load and internal personnel load in parallel and adopting a dynamic weighted fusion mechanism; the defects that multi-source disturbance factors are not fully considered and a prediction model is static and rigid in a traditional method are effectively overcome; further combining with a model predictive control framework, solving an optimal control sequence which takes energy consumption and comfort deviation as a joint objective function in a rolling manner under the condition of meeting physical constraints of the system, so as to realize dynamic optimization operation of the air conditioning system under complex and changeable working conditions; therefore, the total energy consumption of the system is greatly reduced while the comfort of the indoor environment is ensured.

Description

Technical Field

[0001] This invention relates to the field of air conditioning technology, and in particular to an energy-saving operation control method for air conditioning systems. Background Technology

[0002] With the development of the social economy and the improvement of people's living standards, the proportion of building energy consumption in the total social energy consumption continues to rise. Among them, heating, ventilation, and air conditioning (HVAC) systems, as the main energy-consuming equipment in buildings, have made energy-saving operation a crucial link in achieving the "dual carbon" goal. Traditional air conditioning systems mostly adopt control strategies based on fixed temperature setpoints or simple feedback, such as ON / OFF control and PID control, lacking the ability to actively respond to and predict load disturbances caused by dynamic changes in the external environment and indoor occupant activities. In actual operation, air conditioning systems often operate under partial load conditions. If the control strategy fails to adjust in time to follow load changes, it will lead to problems such as low system energy efficiency, large room temperature fluctuations, and poor comfort.

[0003] Existing air conditioning operation control methods fall into two categories. One is a control strategy based on feedforward compensation using environmental parameters (such as outdoor temperature and humidity, and solar radiation intensity). For example, this involves coarsely adjusting the start-up and shutdown times of the air conditioner or the water supply temperature by combining weather forecast data. However, this type of method often ignores the load impact caused by changes in the number and distribution of people indoors, resulting in insufficient control accuracy during periods of high or low population density. Another category is a strategy based on load prediction and control using single indicators such as historical energy consumption data or indoor CO2 concentration. For example, this involves predicting short-term loads through regression analysis or simple time series methods and adjusting equipment operation accordingly. However, because it does not fully consider the spatiotemporal coupling characteristics and nonlinear relationships between multi-source heterogeneous data, the prediction results are prone to deviation, especially in complex scenarios with high population mobility and sudden weather changes, where its adaptability is poor.

[0004] Furthermore, existing control models mostly employ static weights or empirical formulas to fuse multiple load components, failing to dynamically and adaptively adjust parameters based on date type (e.g., weekdays vs. holidays), time period (e.g., daytime vs. nighttime), or real-time operating conditions. This results in systematic errors between the composite load prediction curve and the actual value. At the control level, most strategies only optimize instantaneous energy consumption or comfort, lacking rolling optimization of system dynamic performance across multiple prediction time domains, making it difficult to achieve optimal global energy consumption while ensuring indoor thermal comfort. Simultaneously, due to the absence of an online model parameter correction mechanism, models are prone to drift after long-term operation, leading to a gradual degradation in prediction and control performance.

[0005] Therefore, existing control methods for air conditioning systems generally suffer from problems such as low load forecasting accuracy, insufficient consideration of multi-factor coupling, single optimization objective, and weak model adaptability. This results in the actual operating energy efficiency of the air conditioning system being far lower than the design value, causing energy waste and affecting indoor environmental quality. Against this backdrop, there is an urgent need in this field to develop an energy-saving control method for air conditioning systems that can integrate multi-source time-series data, achieve high-precision composite load forecasting, and dynamically generate optimal operating strategies based on a model predictive control framework. This method aims to effectively balance system energy consumption and indoor comfort, improving overall performance and economy. Summary of the Invention

[0006] The purpose of this invention is to provide an energy-saving operation control method for an air conditioning system to solve the problems existing in the prior art.

[0007] To achieve the above objectives, the present invention provides the following solution:

[0008] This invention provides an energy-saving operation control method for an air conditioning system, comprising the following steps:

[0009] S1. Data Acquisition and Preprocessing: Real-time acquisition of environmental time-series data and personnel time-series data; environmental time-series data includes at least outdoor temperature, outdoor humidity, and solar irradiance data within a preset historical time period; personnel time-series data includes at least carbon dioxide concentration in the building space, number of Wi-Fi device connections, or number of personnel monitored by infrared thermal imaging sensors within a preset historical time period; data cleaning, missing value imputation, and normalization are performed on the environmental time-series data and personnel time-series data respectively to obtain a standardized first dataset and second dataset.

[0010] S2, Dual-path composite load forecasting, executes external environmental load forecasting and internal personnel load forecasting in parallel; specifically including:

[0011] S2.1 Based on the first dataset, using the first preset prediction model, predict the external environmental load within the first preset time period in the future, and obtain the external environmental load prediction sequence. , where t is the future predicted time point;

[0012] S2.2 Based on the second dataset, using the second preset prediction model, predict the internal personnel load within the first preset time period in the future, and obtain the internal personnel load prediction sequence. ;

[0013] S2.3, Prediction sequence of external environmental loads and internal personnel load forecast sequence Dynamic weighted fusion is performed to generate a future composite load forecast curve. The calculation formula is as follows:

[0014] ;

[0015] in, This is the dynamic weighting coefficient for the external environmental load. The dynamic weighting coefficient for internal personnel workload. For dynamic bias terms, , and These are parameters that are adaptively adjusted based on the date type, time period, or historical operating conditions corresponding to time t.

[0016] S3, Optimal control strategy generation and execution; specifically including:

[0017] S3.1, Future composite load forecast curve As input to the model predictive control optimizer; the model predictive control optimizer takes the joint objective function J, which minimizes the total energy consumption of the air conditioning system and the deviation of indoor temperature comfort, within the second preset time period in the future, i.e., the control time domain P, as its optimization objective; the expression of the joint objective function J is:

[0018]

[0019] Where k is the discrete time step in the control time domain. To predict the system energy consumption at time k, To predict the indoor temperature at time k, The indoor comfort temperature setpoint is defined at time k, and α and β are preset weighting factors for balancing energy consumption and comfort.

[0020] S3.2 Under the condition of satisfying the physical constraints of each actuator of the air conditioning system, the joint objective function J is solved in a rolling manner to obtain a control sequence U*(t) = {u*(t), u*(t+1), ..., u*(t+P-1)} composed of a series of optimal control quantities in the control time domain P; the control quantities include at least one or more of the following: air conditioning compressor operating frequency, chilled water valve opening degree, or blower speed.

[0021] S3.3. The first control quantity u*(t) in the control sequence U*(t) is sent to the controller of the air conditioning system as the control command at the current moment and executed. Then, steps S1 to S3.3 are repeated in the next control cycle to achieve rolling optimization control.

[0022] Preferably, in step S2.1, the first preset prediction model is a long short-term memory network model, and the first dataset is used as the input of the long short-term memory network model. The learning and prediction of the correlation of the time series of external environmental loads are realized through the hidden layer state transmission of the model.

[0023] Preferably, in step S2.2, the second preset prediction model is an autoregressive integral moving average model, wherein the parameters in the autoregressive integral moving average model are the autoregressive order, the difference order, and the moving average order of the model, and the parameters are determined based on the analysis of the autocorrelation function and partial autocorrelation function of the second dataset.

[0024] Preferably, the environmental time series data also includes future short-term weather forecast data, which, together with the first dataset, serves as the input to the first preset prediction model to improve the accuracy of external environmental load prediction.

[0025] Preferably, in step S3.1, the system energy consumption Based on control quantity and predicted load Indoor temperature is calculated using a pre-set air conditioning system energy consumption model. Based on control quantity Forecasted load and the indoor temperature at the previous moment Calculated using a pre-set building thermodynamics model.

[0026] Preferably, in step S1, the normalization process uses the min-max normalization method to linearly transform the original data to the interval [0, 1], in order to eliminate the influence of different physical dimensions on model training.

[0027] The present invention also provides an energy-saving operation control device for an air conditioning system, comprising:

[0028] The data acquisition and preprocessing module is used to collect and acquire environmental time-series data and personnel time-series data in real time, and to perform data cleaning, missing value imputation and normalization on the data to obtain a standardized first dataset and a second dataset.

[0029] A dual-path composite load forecasting module, connected to the output of the data acquisition and preprocessing module, includes:

[0030] An external environmental load prediction unit is used to predict the external environmental load within a first preset time period based on the first dataset and using a first preset prediction model, thereby obtaining an external environmental load prediction sequence.

[0031] An internal personnel load prediction unit is used to predict the internal personnel load within the first preset time period based on the second dataset and using a second preset prediction model, so as to obtain an internal personnel load prediction sequence.

[0032] The composite load generation unit is used to dynamically weight and fuse the external environment load prediction sequence and the internal personnel load prediction sequence to generate a future composite load prediction curve.

[0033] An optimal control strategy generation and execution module, connected to the output of the dual-path composite load forecasting module, includes:

[0034] The model predictive control optimizer receives the future composite load prediction curve and uses the joint objective function of minimizing the total energy consumption of the air conditioning system and the indoor temperature comfort deviation within the future control time domain as the optimization objective. Under the condition of satisfying the physical constraints of the system, it solves the optimal control sequence by rolling.

[0035] The control command execution unit is used to extract the first control quantity in the optimal control sequence and send it as the control command at the current moment to the controller of the air conditioning system.

[0036] Preferably, the external environmental load prediction unit incorporates a long short-term memory network model, and the internal personnel load prediction unit incorporates an autoregressive integral moving average model.

[0037] Preferably, the expression for the joint objective function J set in the model predictive control optimizer is:

[0038]

[0039] Where k is the discrete time step in the control time domain. To predict the system energy consumption at time k, To predict the indoor temperature at time k, Let α and β be the indoor comfort temperature setpoint at time k, and let α and β be the preset weighting factors for balancing energy consumption and comfort.

[0040] Preferably, it also includes a model parameter adaptive update module, which is used to periodically collect actual operating data of the air conditioning system, compare the actual energy consumption and actual indoor temperature with the energy consumption and predicted indoor temperature of the prediction system, and, based on the error generated by the comparison, use gradient descent algorithm or particle swarm optimization algorithm to perform online correction and update of the model parameters in the first preset prediction model, the second preset prediction model and the model prediction control optimizer.

[0041] The present invention achieves the following beneficial technical effects compared to the prior art:

[0042] This invention provides an energy-saving operation control method for air conditioning systems, which significantly improves load prediction accuracy and system energy efficiency. By performing dual-path prediction of external environmental load and internal personnel load in parallel and adopting a dynamic weighted fusion mechanism, it effectively overcomes the shortcomings of traditional methods that do not fully consider multi-source disturbance factors and have static and rigid prediction models. Furthermore, by combining a model predictive control framework, it solves the optimal control sequence with energy consumption and comfort deviation as joint objective functions under the condition of satisfying the physical constraints of the system. This enables the dynamic optimization operation of the air conditioning system under complex and variable operating conditions, thereby significantly reducing the total energy consumption of the system while ensuring indoor environmental comfort. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 A flowchart of an energy-saving operation control method for an air conditioning system provided by the present invention. Detailed Implementation

[0045] The serial numbers assigned to components in this document, such as "first," "second," etc., are merely used to distinguish the described objects and have no sequential or technical meaning. The terms "connection" and "linkage" used in this application, unless otherwise specified, include both direct and indirect connections (linkages). In the description of this invention, it should be understood that the terms "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention.

[0046] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can mean that the first feature is in direct contact with the second feature, or that the first feature is in indirect contact with the second feature through an intermediate medium. Furthermore, "above," "over," and "on top" of the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature.

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] The purpose of this invention is to provide an energy-saving operation control method for an air conditioning system to solve the problems existing in the prior art.

[0049] Example 1:

[0050] To enable those skilled in the art to fully understand the energy-saving operation control method for an air conditioning system according to the present invention and to reproduce its technical effects, the technical solution is described in detail below with reference to the accompanying drawings and typical implementation scenarios. It should be noted that the embodiments described below are only for explaining the present invention and do not constitute any limitation on its scope of protection; without departing from the spirit of the present invention, those skilled in the art can make adaptive adjustments to specific parameters, hardware models, and communication protocols according to actual application conditions.

[0051] In this embodiment, the energy-saving operation control device for the air conditioning system is deployed in a building with twelve floors above ground and two floors underground, with a total building area of ​​approximately 3.6 × 10⁻⁶. 4 m 2 This is a modern office building. The building employs a variable air volume (VAV) central air conditioning system, with the cooling source consisting of two variable frequency centrifugal chillers and one magnetic levitation variable frequency chiller connected in parallel. Terminal units are equipped with variable air volume air handling units (AHUs) and district reheat coils. The system is interconnected with field sensors, actuators, and the energy management system (EMS) via the OPCUA unified data bus of the building automation system (BAS).

[0052] The data acquisition and preprocessing module collects real-time data from outdoor temperature and humidity sensors, solar total radiation meters, anemometers, and rain gauges via an RS-485 bus, with a sampling period of 60 seconds. Simultaneously, it uses the BACnet / IP protocol to read raw data from CO2 concentration monitors on each floor of the building, Wi-Fi AP connection counts, and the infrared thermal imaging array, with a sampling period of 30 seconds. To ensure data quality, the module integrates a sliding window-based Z-score outlier detection algorithm and a cubic spline interpolation missing value compensation algorithm. Then, a minimum-maximum normalization method is used to linearly map physical quantities of different dimensions to the [0,1] interval, forming a standardized first dataset (external environment) and a second dataset (internal personnel). All preprocessing is completed in the edge computing gateway and uploaded to the local server cluster via the MQTT protocol.

[0053] The dual-path composite load prediction module in the server cluster adopts a CPU-GPU heterogeneous computing architecture. The external environment load prediction unit deploys an offline-trained LSTM network with an input tensor dimension of (batch=64, time_steps=96, features=7), where the seven features correspond to normalized outdoor temperature, relative humidity, solar radiation intensity, wind speed, rainfall, and the highest / lowest temperature in the 24-hour weather forecast. The network consists of two hidden layers, each containing 128 LSTM units, and a fully connected output layer, using the Adam optimizer with an initial learning rate of 0.001. Batch normalization layers and Dropout layers are embedded to prevent overfitting. The training dataset is taken from historical records of the past year, divided into training, validation, and test sets in a 7:2:1 ratio. After training, the model's RMSE on the test set is 0.023 (normalized scale), corresponding to an actual cooling load error of approximately ±1.8%.

[0054] The internal personnel load prediction unit uses the ARIMA(2,1,2) model, whose parameters were determined through analysis of the autocorrelation function (ACF) and partial autocorrelation function (PACF) of the second dataset. To ensure that the differences between holidays and weekdays are fully represented, the model was split into two groups of data according to date type before training, and ARIMA models were built for each group. During the prediction phase, the model automatically switched according to the date type. The model output is a sequence of personnel heat load at 5-minute intervals for the next 4 hours (first preset duration).

[0055] The composite load generation unit calculates dynamic weight coefficients and bias terms in real time during each prediction cycle. Specifically, the system minimizes the mean square error between the predicted and actual composite load values ​​over the past 2 hours using an online gradient descent method, and updates the weights every 15 minutes to ensure the adaptability of the fusion strategy under scenarios such as weekday morning peaks, lunch break troughs, and nighttime overtime.

[0056] The model predictive control optimizer runs on an industrial PC, with its control time domain P set to 2 hours, prediction time domain N set to 4 hours, and sampling interval Δt = 5 minutes. The energy consumption term in the optimizer's joint objective function J is calculated using an air conditioning system energy consumption model. This model is constructed based on the chiller COP curve, water pump variable frequency efficiency curve, terminal VAV fan energy consumption curve, and cooling tower power consumption curve, and is fitted as an analytical function about the control variables using polynomial regression. The comfort deviation term is predicted by a first-order equivalent thermal parameter (ETP) building thermodynamic model, and the model parameters are corrected online using the building's thermal performance and a real-time identification algorithm.

[0057] The decision variables for the optimization problem include: the start-up and shutdown status of each chiller unit (Boolean value), operating frequency (continuous, range 20-100Hz), chilled water main valve opening (continuous, range 0-100%), and the air supply fan speed of each air conditioning unit (continuous, range 30-100%). Constraints include: minimum start-up and shutdown time of equipment (≥15min), chilled water supply and return temperature difference (≥5℃), upper limit of supply air temperature (≥12℃), upper limit of indoor CO2 concentration (≤1000ppm), and the coupling relationship between the terminal reheat valve opening and the supply air volume.

[0058] The optimized solver employs the IPOPT algorithm within the CasaADi framework, achieving an average single-run solution time of approximately 2.3 seconds on an Intel i7-12700 CPU, meeting the 5-minute rolling cycle requirement. After each solution run, only the first element of the control sequence U*(t) is sent to the BAS, enabling rolling optimization and feedback correction.

[0059] To prevent model drift, the system deploys an adaptive model parameter update module. This module automatically collects the previous day's actual operating data every 24 hours, including chiller power, chilled water flow rate, indoor temperature curves, and true cooling load, and calculates the 24-hour rolling average absolute percentage error (MAPE). If the MAPE > 8%, an online calibration procedure is triggered: the LSTM network uses mini-batch stochastic gradient descent (mini-batchSGD) to locally fine-tune the weights of the last two layers, with the learning rate decaying to 1 / 5 of the initial value; the ARIMA model re-estimates the coefficients using maximum likelihood estimation; and the building thermodynamics model in the model predictive control optimizer updates the heat capacity and thermal resistance parameters using recursive least squares (RLS). After calibration, the new version of the model is marked with a version number and hot-swapped to the production environment, without any downtime.

[0060] By deploying the aforementioned devices, precise control of energy-saving operation of the air conditioning system can be achieved. The control method is as follows: Figure 1 As shown, it includes the following steps:

[0061] S1. Data Acquisition and Preprocessing: Real-time acquisition of environmental time-series data and personnel time-series data; Environmental time-series data includes at least outdoor temperature, outdoor humidity, solar irradiance data within a preset historical time period, as well as short-term weather forecast data. The short-term weather forecast data and the first dataset are used as input to the first preset prediction model to improve the accuracy of external environmental load prediction; Personnel time-series data includes at least carbon dioxide concentration in building spaces, number of Wi-Fi device connections, or number of people monitored by infrared thermal imaging sensors within a preset historical time period; Data cleaning, missing value imputation, and normalization are performed on the environmental time-series data and personnel time-series data respectively to obtain a standardized first dataset and second dataset; The normalization process uses the min-max normalization method to linearly transform the original data to the interval [0, 1] to eliminate the influence of different physical dimensions on model training;

[0062] S2, Dual-path composite load forecasting, executes external environmental load forecasting and internal personnel load forecasting in parallel; specifically including:

[0063] S2.1 Based on the first dataset, using the first preset prediction model, predict the external environmental load within the first preset time period in the future, and obtain the external environmental load prediction sequence. , where t is the future prediction time point; the first preset prediction model is a long short-term memory network model, and the first dataset is used as the input of the long short-term memory network model. The learning and prediction of the correlation of the time series of external environmental load is realized through the hidden layer state transmission of the model.

[0064] S2.2 Based on the second dataset, using the second preset prediction model, predict the internal personnel load within the first preset time period in the future, and obtain the internal personnel load prediction sequence. The second preset prediction model is an autoregressive integral moving average model. The parameters in the autoregressive integral moving average model are the autoregressive order, the differencing order, and the moving average order of the model. The parameters are determined based on the analysis of the autocorrelation function and partial autocorrelation function of the second dataset.

[0065] S2.3, Prediction sequence of external environmental loads and internal personnel load forecast sequence Dynamic weighted fusion is performed to generate a future composite load forecast curve. The calculation formula is as follows:

[0066] ;

[0067] in, This is the dynamic weighting coefficient for the external environmental load. The dynamic weighting coefficient for internal personnel workload. For dynamic bias terms, , and These are parameters that are adaptively adjusted based on the date type, time period, or historical operating conditions corresponding to time t.

[0068] S3, Optimal control strategy generation and execution; specifically including:

[0069] S3.1, Future composite load forecast curve This is the input to the model predictive control optimizer; the model predictive control optimizer takes the joint objective function J, which minimizes the total energy consumption of the air conditioning system and the deviation of indoor temperature comfort, within the second preset time period in the future, i.e., the control time domain P, as its optimization objective; the expression of the joint objective function J is:

[0070] ;

[0071] Where k is the discrete time step in the control time domain. To predict the system energy consumption at time k, To predict the indoor temperature at time k, The indoor comfort temperature setpoint is defined at time k, where α and β are preset weighting factors for balancing energy consumption and comfort; system energy consumption. Based on control quantity and predicted load Indoor temperature is calculated using a pre-set air conditioning system energy consumption model. Based on control quantity Forecasted load and the indoor temperature at the previous moment Calculated using a pre-defined building thermodynamics model;

[0072] S3.2 Under the condition of satisfying the physical constraints of each actuator of the air conditioning system, the joint objective function J is solved in a rolling manner to obtain a control sequence U*(t) = {u*(t), u*(t+1), ..., u*(t+P-1)} composed of a series of optimal control quantities in the control time domain P; the control quantities include at least one or more of the following: air conditioning compressor operating frequency, chilled water valve opening degree, or blower speed.

[0073] S3.3. The first control quantity u*(t) in the control sequence U*(t) is sent to the controller of the air conditioning system as the control command at the current moment and executed. Then, steps S1 to S3.3 are repeated in the next control cycle to achieve rolling optimization control.

[0074] Taking a typical summer workday as an example, the system initiates the pre-cooling program at 06:00. As the weather forecast indicates a high of 38℃, the LSTM model predicts intense solar radiation between 09:00 and 11:00, causing a rapid increase in external environmental load. Meanwhile, the ARIMA model, based on Wi-Fi connection counts, predicts approximately 420 people will enter the building between 08:30 and 09:00, simultaneously increasing the internal personnel load. The composite load curve reaches its peak at approximately 1.45MW at 08:45. The model predictive control optimizer preemptively loads the two variable frequency centrifugal chillers to 85Hz at 07:30 and sets the chilled water supply temperature to 6.5℃. Simultaneously, it reduces the VAV box supply air temperature by floor zone to pre-store cooling. After 09:15, as solar radiation stabilized and some employees went out for meetings, the load decreased. The optimizer gradually reduced the chiller unit frequency to 60Hz and increased the chilled water supply temperature to 7.8℃. Under the premise of maintaining the room temperature at 23.5±0.5℃, the overall system efficiency ratio (COP_sys) reached 4.97 throughout the day, which is about 12.4% higher than the traditional PID control strategy, and the total power consumption was reduced by 1870kWh.

[0075] As can be seen from the above implementation methods, the present invention effectively solves the problems of insufficient prediction accuracy, response lag, and single optimization objective of traditional air conditioning control strategies in complex dynamic load environments through a closed-loop architecture of "dual-path prediction - dynamic weighting - model predictive control - online correction". It achieves coordinated optimization of energy consumption and comfort in real building scenarios. At the same time, its modular design makes it flexibly applicable to air conditioning projects of different scales and system forms, and has significant economic and social benefits.

[0076] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0077] It should be noted that the components mentioned in the above embodiments are all general standard parts or components known to those skilled in the art. Their structures and principles can be learned by those skilled in the art through technical manuals or conventional experimental methods.

[0078] This invention has illustrated its principles and implementation methods using specific examples. The descriptions of these embodiments are merely illustrative of the method and its core ideas; furthermore, those skilled in the art will recognize that modifications may be made to the specific implementation methods and application scope based on the principles of this invention. Therefore, the content of this specification should not be construed as limiting the invention.

Claims

1. An energy saving operation control method of an air conditioning system, characterized by, Includes the following steps: S1. Data Acquisition and Preprocessing: Real-time acquisition of environmental time-series data and personnel time-series data; Environmental time-series data shall include at least outdoor temperature, outdoor humidity and solar irradiance data within a preset historical time period; Personnel time-series data shall include at least carbon dioxide concentration in the building space, number of Wi-Fi devices connected or number of personnel monitored by infrared thermal imaging sensors within a preset historical time period. Data cleaning, missing value imputation, and normalization were performed on environmental time series data and personnel time series data respectively to obtain a standardized first dataset and second dataset; S2, Dual-path composite load forecasting, executes external environmental load forecasting and internal personnel load forecasting in parallel; specifically including: S2.1, based on the first data set, using a first preset prediction model, predicting the external environment load in the future first preset time length to obtain an external environment load prediction sequence where t is a future prediction time point; S2.2, based on the second data set, using a second preset prediction model, predicting the internal staff load in the future first preset time length to obtain an internal staff load prediction sequence ; S2.3, Prediction sequence of external environmental loads and internal personnel load forecast sequence Dynamic weighted fusion is performed to generate a future composite load forecast curve. The calculation formula is as follows: ;in, This is the dynamic weighting coefficient for the external environmental load. The dynamic weighting coefficient for internal personnel workload. For dynamic bias terms, , and These are parameters that are adaptively adjusted based on the date type, time period, or historical operating conditions corresponding to time t. S3, Optimal control strategy generation and execution; specifically including: S3.1, Future composite load forecast curve As input to the model predictive control optimizer; the model predictive control optimizer takes the joint objective function J, which minimizes the total energy consumption of the air conditioning system and the deviation of indoor temperature comfort, within the second preset time period in the future, i.e., the control time domain P, as its optimization objective; the expression of the joint objective function J is: ; Where k is the discrete time step in the control time domain. Predicting the system energy consumption at time k. To predict the indoor temperature at time k, The indoor comfort temperature setpoint is defined at time k, and α and β are preset weighting factors for balancing energy consumption and comfort. S3.2 Under the condition of satisfying the physical constraints of each actuator of the air conditioning system, the joint objective function J is solved in a rolling manner to obtain a control sequence U*(t) = {u*(t), u*(t+1), ..., u*(t+P-1)} composed of a series of optimal control quantities in the control time domain P; the control quantities include at least one or more of the following: air conditioning compressor operating frequency, chilled water valve opening degree, or blower speed. S3.

3. The first control quantity u*(t) in the control sequence U*(t) is sent to the controller of the air conditioning system as the control command at the current moment and executed. Then, steps S1 to S3.3 are repeated in the next control cycle to achieve rolling optimization control.

2. The energy-saving operation control method for an air conditioning system according to claim 1, characterized in that, In step S2.1, the first preset prediction model is a long short-term memory network model, and the first dataset is used as the input of the long short-term memory network model. The learning and prediction of the correlation of the time series of external environmental loads are realized through the hidden layer state transmission of the model.

3. The energy-saving operation control method for an air conditioning system according to claim 1, characterized in that, In step S2.2, the second preset prediction model is an autoregressive integral moving average model. The parameters in the autoregressive integral moving average model are the autoregressive order, the difference order, and the moving average order of the model. The parameters are determined based on the analysis of the autocorrelation function and partial autocorrelation function of the second dataset.

4. The energy-saving operation control method for an air conditioning system according to claim 1, characterized in that, Environmental time series data also includes future short-term weather forecast data. The future short-term weather forecast data and the first dataset are used together as input to the first preset prediction model to improve the accuracy of external environmental load prediction.

5. The energy-saving operation control method for an air conditioning system according to claim 1, characterized in that, In step S3.1, the system energy consumption Based on control quantity and predicted load Indoor temperature is calculated using a pre-set air conditioning system energy consumption model. Based on control quantity Forecasted load and the indoor temperature at the previous moment Calculated using a pre-set building thermodynamics model.

6. The energy-saving operation control method for an air conditioning system according to claim 1, characterized in that, In step S1, the normalization process uses the min-max normalization method to linearly transform the original data to the interval [0, 1], in order to eliminate the influence of different physical dimensions on model training.

7. An energy-saving operation control device for an air conditioning system, characterized in that, include: The data acquisition and preprocessing module is used to collect and acquire environmental time-series data and personnel time-series data in real time, and to perform data cleaning, missing value imputation and normalization processing on the data to obtain a standardized first dataset and a second dataset. A dual-path composite load forecasting module, connected to the output of the data acquisition and preprocessing module, includes: An external environmental load prediction unit is used to predict the external environmental load within a first preset time period based on the first dataset and using a first preset prediction model, thereby obtaining an external environmental load prediction sequence. An internal personnel load prediction unit is used to predict the internal personnel load within the first preset time period based on the second dataset and using a second preset prediction model, so as to obtain an internal personnel load prediction sequence. The composite load generation unit is used to dynamically weight and fuse the external environment load prediction sequence and the internal personnel load prediction sequence to generate a future composite load prediction curve. An optimal control strategy generation and execution module, connected to the output of the dual-path composite load forecasting module, includes: The model predictive control optimizer receives the future composite load forecast curve and uses the joint objective function of minimizing the total energy consumption of the air conditioning system and the indoor temperature comfort deviation within the future control time domain as the optimization objective. Under the condition of satisfying the physical constraints of the system, it solves the optimal control sequence by rolling. The control command execution unit is used to extract the first control quantity in the optimal control sequence and send it as the control command at the current moment to the controller of the air conditioning system.

8. The energy-saving operation control device for an air conditioning system according to claim 7, characterized in that, The external environmental load prediction unit incorporates a long short-term memory network model, while the internal personnel load prediction unit incorporates an autoregressive integral moving average model.

9. The energy-saving operation control device for an air conditioning system according to claim 7, characterized in that, The expression for the joint objective function J set in the model predictive control optimizer is: ; Where k is the discrete time step in the control time domain. To predict the system energy consumption at time k, To predict the indoor temperature at time k, Let α and β be the indoor comfort temperature setpoint at time k, and let α and β be the preset weighting factors for balancing energy consumption and comfort.

10. The energy-saving operation control device for an air conditioning system according to claim 7, characterized in that, It also includes a model parameter adaptive update module, which is used to periodically collect actual operating data of the air conditioning system, compare the actual energy consumption and actual indoor temperature with the energy consumption and predicted indoor temperature of the prediction system, and, based on the error generated by the comparison, use gradient descent algorithm or particle swarm optimization algorithm to perform online correction and update of the model parameters in the first preset prediction model, the second preset prediction model and the model prediction control optimizer.

Images