A method for predicting the energy efficiency ratio of chiller units

Abstract

This invention belongs to the field of energy-saving optimization technology for HVAC systems, and discloses a method for predicting the energy efficiency ratio (EER) of chillers, thereby improving the accuracy of EER prediction. The method includes: Step 1, acquiring chiller sensor data and calculating the variables required to train a convolutional neural network model for predicting the EER based on the sensor data; Step 2, constructing a convolutional neural network model for predicting the EER of a single-compressor chiller based on the variables calculated in Step 1; Step 3, substituting the real-time sensor data collected from the operating chiller into the convolutional neural network model constructed in Step 2 to predict the EER of the chiller.

Description

Technical Field

[0001] This invention belongs to the field of energy-saving optimization technology for HVAC systems, and specifically relates to a method for predicting the energy efficiency ratio of a chiller unit. Background Technology

[0002] HVAC systems account for over 50% of total energy consumption during building operation. In data centers, chillers, as core energy-consuming components, consume over 40% of the HVAC system's energy, and their COP (Coefficient of Performance) directly determines the potential for energy optimization in data center buildings. Accurately predicting the COP of chillers is a crucial prerequisite for achieving system energy-saving control. Sensors, as the core component for acquiring chiller operating data, directly impact the accuracy and practicality of the prediction model due to the quality of their data and the rationality of variable selection. Current mainstream technologies for predicting the COP of data center chillers have significant limitations, specifically as follows: 1. Physics-based models: These are built based on thermodynamics and heat and mass transfer equations, requiring complex physical parameters (such as geometric dimensions and refrigerant properties). The modeling process is lengthy and difficult to calibrate. Furthermore, they do not consider the impact of sensor data deviations on model parameter calibration, thus limiting their practicality.

[0003] 2. Data-driven models: These models rely entirely on data and are computationally efficient, but lack physical constraints and have weak generalization ability. In recent years, machine learning methods (such as artificial neural networks, long short-term memory neural networks, and generative adversarial networks) have been introduced to capture nonlinear relationships, but they still face problems such as insufficient generalization performance and poor interpretability. First, these models heavily depend on large amounts of high-quality data; insufficient or low-quality data will affect their performance, limiting generalization ability and making them prone to overfitting and poor adaptability. Second, the models have poor interpretability and lack transparent intermediate process mechanisms.

[0004] 3. Traditional prediction models: These models rely on sensor-collected data for modeling. While computationally efficient, they have drawbacks: First, existing models suffer from variable redundancy, which can introduce sensor noise, while missing variables lead to incomplete information. At the same time, they lack error correction mechanisms for core variables, and sensor biases are directly transmitted to the prediction results, resulting in decreased accuracy.

[0005] Second, the model has poor adaptability to the features of sensor data: the temperature, flow rate and other data collected by the sensors of the chiller unit have significant temporal coupling and local correlation, but traditional multinomial regression, BP neural network and other models are difficult to capture such complex features and can only learn simple mapping relationships; while single convolutional neural network models cannot match the physical characteristics of sensor data, resulting in insufficient feature extraction.

[0006] Third, the generalization ability and adaptability of the models to engineering scenarios are weak: existing neural network-based models mostly use raw sensor data for modeling directly without combining physical mechanisms such as energy conservation for constraints; at the same time, they do not consider non-standardized sensor data in engineering sites and coupling errors when multiple sensors work together, resulting in insufficient practicality of the models and difficulty in dealing with complex scenarios. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide a method for predicting the energy efficiency ratio of chillers, thereby improving the accuracy of energy efficiency ratio prediction for chillers.

[0008] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: A method for predicting the energy efficiency ratio of a chiller unit includes: Step 1: Obtain sensor data from the chiller unit, and calculate the variables required to train the convolutional neural network model for predicting the energy efficiency ratio of the chiller unit based on the sensor data. Step 2: Based on the variables calculated in Step 1, construct a convolutional neural network model for predicting the energy efficiency ratio of a single-compressor chiller unit; Step 3: Substitute the real-time sensor data of the operating chiller unit into the convolutional neural network model constructed in Step 2 to predict the energy efficiency ratio of the chiller unit.

[0009] As a preferred example, in step 1, the chiller unit sensor data includes: chilled water supply temperature of the chiller unit. T chw,out Unit: °C; Chilled water return temperature of chiller unit T chw,in Unit: °C; chilled water flow rate of chiller unit M chw Unit: m 3 / h; Cooling water supply temperature of chiller unit T cw,out Unit: °C; Cooling water return temperature of chiller unit T cw,in Unit: °C; Cooling water flow rate of chiller unit M cw Unit: m 3 / h; Evaporation temperature of the chiller unit T ev Unit: °C; Condensation temperature of chiller unit T cn Unit: °C; Chiller unit operating power P , Unit: kW.

[0010] As a preferred example, in step 1, the variables required to calculate the convolutional neural network model for predicting the energy efficiency ratio of the chiller unit include: Using equation (1), calculate the variable load-thermodynamic difficulty product A of the convolutional neural network model for training the chiller unit's energy efficiency ratio prediction; using equation (2), calculate the variable flow-temperature coupling ratio B of the convolutional neural network model for training the chiller unit's energy efficiency ratio prediction; using equation (3), calculate the variable unit load power consumption offset C of the convolutional neural network model for training the chiller unit's energy efficiency ratio prediction; using equation (4), calculate the variable terminal difference-available temperature difference ratio D of the convolutional neural network model for training the chiller unit's energy efficiency ratio prediction; using equation (5), calculate the variable thermal balance residual E of the convolutional neural network model for training the chiller unit's energy efficiency ratio prediction; using equation (6), calculate the variable chiller unit operating energy efficiency ratio (COP) of the convolutional neural network model for training the chiller unit's energy efficiency ratio prediction: Equation (1) Equation (2) Equation (3) Equation (4) Equation (5) Equation (6) In the formula, c Specific heat capacity of water, unit: J / (kg·℃); Q 0 The rated cooling capacity on the nameplate of the chiller unit is in kW. P 0 Rated power consumption of the chiller unit, unit: kW.

[0011] As a preferred example, in step 2, the load-thermodynamic difficulty product A, flow-temperature difference coupling ratio B, unit load power consumption offset C, end difference-available temperature difference ratio D, and thermal balance residual E obtained in step 1 are used as physical neurons of the hidden layer of the convolutional neural network and embedded into the hidden layer of the convolutional neural network; the calculation and propagation between the physical neurons follow equations (1) to (6).

[0012] As a preferred example, step 3 includes: Step 301: Collect the chilled water supply temperature of the operating chiller unit. T chw,out Chilled water return temperature of chiller unit T chw,in chilled water flow rate of chiller unit M chw Cooling water return temperature of chiller unit T cw,in Cooling water flow rate of chiller unit M cw Evaporation temperature of chiller unitT ev Condensation temperature of chiller unit T cn and chiller unit operating power P ; Step 302: Calculate the energy efficiency ratio of the chiller unit based on the data collected in step 301 and the convolutional neural network model constructed in step 302.

[0013] As a preferred example, step 302 includes: Step 3021: Obtain the cooling capacity of the chiller unit according to formula (12): Equation (12) In the formula, Q chw This indicates the cooling capacity of the chiller unit, in kW. Density of water, unit: kg / m³; Step 3022: Obtain the initial boundary conditions for the condensing heat load according to equation (13); Equation (13) In the formula, This represents the condensation heat load under the initial boundary conditions, in kW. Step 3023: Obtain the initial cooling water supply temperature of the chiller unit according to formula (14); Equation (14) In the formula, This indicates the initial cooling water supply temperature of the chiller unit, in °C. This represents the initial condensing heat load calculated by the convolutional neural network model, in kW. Step 3024: Set the initial cooling water supply temperature and chilled water supply temperature of the chiller unit. T chw,out Chilled water return temperature of chiller unit T chw,in chilled water flow rate of chiller unit M chw Cooling water return temperature of chiller unit T cw,in Cooling water flow rate of chiller unit M cw Evaporation temperature of chiller unit T ev Condensation temperature of chiller unit T cn Chiller unit operating power P As input parameters, these parameters are used in the convolutional neural network model constructed in step 302 to obtain the chiller unit's operating energy efficiency ratio. COPmodel ; Step 3025: Obtain the condensation heat load calculated by the convolutional neural network model according to equation (15); Equation (15) In the formula, This represents the condensation heat load calculated by the convolutional neural network model, in kW. COP model This represents the chiller unit's operating energy efficiency ratio calculated by a convolutional neural network model; it has no unit. Step 3026: Based on the condensing heat load obtained in step 3025, determine whether the energy balance coefficient obtained according to formula (16) is less than or equal to the threshold. If yes, the latest chiller unit operating energy efficiency ratio obtained in step 3024 is used as the predicted value of the chiller unit operating energy efficiency ratio. If no, iterate until the energy balance coefficient obtained according to formula (16) is less than or equal to the threshold. Equation (16) In the formula, This represents the energy balance coefficient and has no unit. This represents the condensing heat load obtained in step 3025, in kW. The initial value is obtained from equation (13).

[0014] As a preferred example, step 3026 involves iteration, including: Return to step 3023 and use the latest result obtained in step 3025. Replace step 3023 (14) And calculate the new chiller unit cooling water supply temperature according to formula (14); In step 3024, the new chiller cooling water supply temperature replaces the initial chiller cooling water supply temperature and is used as the input parameter of the convolutional neural network model to obtain the new chiller operating energy efficiency ratio. When performing step 3025, the chiller unit operating energy efficiency ratio in equation (15) is... COP model The condensing heat load was calculated by replacing the energy efficiency ratio of the new chiller unit with a new convolutional neural network model.

[0015] As a preferred example, the threshold is 3%.

[0016] Compared with existing technologies, the chiller unit energy efficiency ratio prediction method of the present invention can improve the prediction accuracy of chiller unit energy efficiency ratio. The method includes: Step 1, acquiring chiller unit sensor data and calculating the variables required to train the convolutional neural network model for chiller unit energy efficiency ratio prediction based on the chiller unit sensor data; Step 2, constructing a convolutional neural network model for single-compressor chiller unit energy efficiency ratio prediction based on the variables calculated in Step 1; Step 3, substituting the chiller unit sensor data collected in real time from the operating chiller unit into the convolutional neural network model constructed in Step 2 to predict the chiller unit energy efficiency ratio. This method combines the advantages of both physical mechanism and data-driven approaches, effectively compensating for the shortcomings of existing physics-based modeling and data-driven modeling, while also considering measurement errors caused by sensor aging, environmental corrosion, and drift deviation, thus constructing a chiller unit energy efficiency ratio prediction method with strong universality. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method according to an embodiment of the present invention; Figure 2 Line graphs showing the predicted energy efficiency ratio and actual energy efficiency ratio of the chiller unit in this embodiment of the invention; Figure 3 This is a scatter plot showing the predicted energy efficiency ratio and the actual energy efficiency ratio of the chiller unit in an embodiment of the present invention. Figure 4 Line graphs showing the predicted energy efficiency ratio and the actual energy efficiency ratio for existing convolutional neural network models; Figure 5 Fitted scatter plots of the predicted energy efficiency ratio and the actual energy efficiency ratio for existing convolutional neural network models. Detailed Implementation

[0018] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings.

[0019] like Figure 1 As shown, an embodiment of the present invention provides a method for predicting the energy efficiency ratio of a chiller unit, comprising: Step 1: Obtain sensor data from the chiller unit, and calculate the variables required to train the convolutional neural network model for predicting the energy efficiency ratio of the chiller unit based on the sensor data. Step 2: Based on the variables calculated in Step 1, construct a convolutional neural network model for predicting the energy efficiency ratio of a single-compressor chiller unit; Step 3: Substitute the real-time sensor data of the operating chiller unit into the convolutional neural network model constructed in Step 2 to predict the energy efficiency ratio of the chiller unit.

[0020] The method described above, based on sensor variables and convolutional neural networks, is a method for predicting the energy efficiency ratio (EER) of chillers. It identifies, extracts, and filters sensor data, calculates the variables required for training the model, constructs a convolutional neural network model adapted to the characteristics of chiller sensor data, and then builds an iterative calculation framework for predicting the EER based on the law of conservation of energy. This iteratively updates the model's input variables, ultimately achieving accurate prediction of the chiller's EER. This method includes sensor data filtering, core variable filtering, model structure adaptation, and physical mechanism constraints. It effectively solves the problems of imbalanced input variable filtering in existing models and the impact of sensor data bias on prediction accuracy. It also addresses the shortcomings of traditional models in extracting the temporal-spatial coupling features of sensor data. Furthermore, it solves the problems of weak generalization ability and poor adaptability to sensor data in engineering scenarios in existing models, significantly improving the accuracy, stability, and engineering practicality of chiller EER prediction.

[0021] As a preferred example, in step 1, the chiller unit sensor data includes: chilled water supply temperature of the chiller unit. T chw,out Unit: °C; Chilled water return temperature of chiller unit T chw,in Unit: °C; chilled water flow rate of chiller unit M chw Unit: m 3 / h; Cooling water supply temperature of chiller unit T cw,out Unit: °C; Cooling water return temperature of chiller unit T cw,in Unit: °C; Cooling water flow rate of chiller unit M cw Unit: m 3 / h; Evaporation temperature of the chiller unit T ev Unit: °C; Condensation temperature of chiller unit T cn Unit: °C; Chiller unit operating power P The unit is kW. These parameters characterize the chiller's cooling capacity, power consumption, load status, heat transfer intensity, and cycle thermodynamic state. They are necessary inputs for calculating and predicting the chiller's energy efficiency ratio, while ensuring that redundant data is avoided. For example, three quantities on the chilled water side (return water temperature, supply water temperature, and flow rate) are used to calculate the cooling capacity, while the cooling side can be used for energy balance verification and reflect heat dissipation capacity. Other physical quantities, such as ambient temperature and humidity, are unrelated to the chiller's thermodynamic performance; introducing them would lead to data redundancy, increase computational load, and even cause overfitting. This preferred example does not use compressor pressure data, mainly because compressor pressure data is easily affected by pipelines and is less stable than temperature.

[0022] As a preferred example, in step 1, the variables required to calculate the convolutional neural network model for predicting the energy efficiency ratio of the chiller unit include: Using equation (1), calculate the variable load-thermodynamic difficulty product A of the convolutional neural network model for training the chiller unit's energy efficiency ratio prediction; using equation (2), calculate the variable flow-temperature coupling ratio B of the convolutional neural network model for training the chiller unit's energy efficiency ratio prediction; using equation (3), calculate the variable unit load power consumption offset C of the convolutional neural network model for training the chiller unit's energy efficiency ratio prediction; using equation (4), calculate the variable terminal difference-available temperature difference ratio D of the convolutional neural network model for training the chiller unit's energy efficiency ratio prediction; using equation (5), calculate the variable thermal balance residual E of the convolutional neural network model for training the chiller unit's energy efficiency ratio prediction; using equation (6), calculate the variable chiller unit operating energy efficiency ratio (COP) of the convolutional neural network model for training the chiller unit's energy efficiency ratio prediction: Equation (1) Equation (2) Equation (3) Equation (4) Equation (5) Equation (6) In the formula, c Specific heat capacity of water, unit: J / (kg·℃); Q 0 The rated cooling capacity on the nameplate of the chiller unit is in kW. P 0 Rated power consumption of the chiller unit, unit: kW.

[0023] The variables used to construct the convolutional neural network model based on the chiller unit sensor data selected in the above preferred examples include load-thermodynamic difficulty product A, flow-temperature difference coupling ratio B, unit load power consumption offset C, terminal difference-available temperature difference ratio D, and thermal balance residual E.

[0024] The load-thermodynamic difficulty product A is obtained by coupling the actual cooling load of the chiller unit with the temperature ratio characterizing the thermodynamic irreversibility of the condenser-evaporator side. The cooling capacity of the chiller unit in the first term of Equation (1) is determined by the chilled water flow rate, the inlet and outlet temperature difference, and the specific heat capacity, reflecting the actual heat transfer load on the evaporator side; the temperature difference in the second term of Equation (1) reflects the thermodynamic difficulty between the condensing temperature and the evaporating temperature. The larger the temperature difference, the lower the theoretical energy efficiency ratio of the Carnot cycle, and the more significant the irreversible heat transfer loss of the system. The load-thermodynamic difficulty product A is obtained by multiplying the two, which reflects the influence of the load on the unit's operating state. At the same time, it incorporates the constraint of thermodynamic irreversibility, which can more comprehensively characterize the operating difficulty and load characteristics of the unit under different operating conditions. In the energy efficiency ratio prediction, the variable of the load-thermodynamic difficulty product A can distinguish different operating conditions, avoid the defect that a single load index cannot reflect thermodynamic constraints, thereby improving the model's ability to capture the energy efficiency ratio change law under different operating conditions, avoiding overfitting of the model in physically infeasible regions, and improving the prediction accuracy.

[0025] The flow-temperature coupling ratio B is obtained by coupling the temperature difference ratio between the chilled water side and the cooling water side, and the flow rate ratio between the chilled water and the cooling water side, and characterizes the degree of heat flux density matching between the chilled water and the cooling water side. In heat transfer, the evaporator and condenser of a chiller unit transfer heat through chilled water and cooling water, respectively. According to the law of conservation of energy, when heat loss is ignored, the heat flow on both sides should satisfy an approximate balance. The temperature difference ratio in the first term of equation (2) reflects the temperature difference between the fluids on both sides, and the flow rate ratio in the second term of equation (2) reflects the flow distribution ratio. The product of the two can characterize the matching of heat flow transfer. In energy efficiency ratio prediction, the flow-temperature coupling ratio B variable can reflect the influence of operations such as flow rate regulation and water temperature control on heat flow matching, reflect the heat transfer status of the evaporator and condenser, and thus improve the accuracy of the model's prediction of energy efficiency ratio under changes in flow rate and temperature.

[0026] The power consumption offset C per unit load is obtained by coupling the relative power consumption offset with the unit cooling load. In engineering thermodynamics, the ratio of power consumption P to cooling capacity Q0 of a chiller unit determines the COP, while the ratio of the first term in equation (3) reflects the degree of deviation of the current power consumption from the baseline operating condition P0. This deviation is due to irreversible losses such as changes in heat transfer efficiency, refrigerant side resistance loss, and compressor efficiency decay. Multiplying the power consumption offset by the unit cooling load yields the additional power consumption loss per unit load, which is related to the influence of irreversibility on the energy efficiency ratio in thermodynamics. In energy efficiency ratio prediction, the power consumption offset C per unit load can reflect the degree of energy efficiency ratio degradation when the unit deviates from the design operating condition, reflect normal and abnormal operating conditions, and improve the model's predictive ability and accuracy under abnormal changes in energy efficiency ratio.

[0027] The terminal temperature difference-usable temperature difference ratio D is composed of the ratio of the inlet and outlet terminal temperature differences between cooling water and chilled water. In heat transfer, the usable heat transfer temperature difference of the evaporator is T. cw,in -Tchw,in (i.e., the temperature difference between the cooling water return and the chilled water return), the actual heat transfer end difference is T. cw,out -T chw,out (That is, the temperature difference between cooling water supply and chilled water supply). The ratio of the two reflects the sufficiency of heat transfer within the heat exchanger: when the ratio approaches 1, it indicates high heat exchanger efficiency; if the ratio deviates from 1, it indicates a decrease in heat transfer efficiency due to thickening of the heat transfer boundary layer, scaling, uneven flow distribution, etc. In energy efficiency ratio prediction, the terminal difference-available temperature difference ratio D can quantify the heat transfer efficiency level of the evaporator and condenser, reflecting the impact of factors such as heat exchanger fouling and operating condition fluctuations on the utilization of the heat transfer temperature difference, and improving the accuracy of energy efficiency ratio prediction under long-term operating conditions.

[0028] The thermal balance residual E is obtained by normalizing the heat flow deviation between the chilled water side and the cooling water side. Utilizing the principles of heat transfer and energy conservation, under ideal conditions, the heat absorbed by the evaporator of a chiller unit (heat released on the chilled water side) should equal the heat released by the condenser (heat absorbed on the cooling water side). The difference between the two is the thermal balance residual. This residual arises from factors such as sensor measurement errors, pipe heat loss, and refrigerant heat storage, and its value reflects the reliability of the system operating data and the integrity of the heat transfer process. The normalized E eliminates the influence of load, enabling comparison of thermal balance deviations under different operating conditions. In energy efficiency ratio prediction, the thermal balance residual E characterizes data quality and system integrity, helping the model identify abnormal operating conditions with unbalanced heat flow, filtering invalid data caused by measurement errors or system leaks, and reflecting the impact of the integrity of the heat transfer process on the energy efficiency ratio, thereby improving the model's predictive stability and accuracy under complex operating conditions.

[0029] The Coefficient of Performance (COP) of a chiller unit is the ratio of cooling capacity to input power, representing the unit's energy efficiency. In engineering thermodynamics, the COP is the efficiency ratio of the actual refrigeration cycle to the Carnot cycle, influenced by various irreversible factors such as the heat transfer efficiency of the evaporator and condenser, the isentropic efficiency of the compressor, and throttling losses. These factors are indirectly reflected through variables A, B, C, D, and E. In this method, COP serves as the target variable for prediction, forming a feature-target mapping relationship with variables A, B, C, D, and E. Variables A, B, C, D, and E extract core physical features affecting the COP from multiple perspectives, providing interpretable physical constraints for the model to learn the COP's variation patterns. This avoids the overfitting and insufficient generalization problems of purely data-driven models, thus significantly improving the accuracy and reliability of COP prediction.

[0030] Preferably, in step 2, the load-thermodynamic difficulty product A, flow-temperature difference coupling ratio B, unit load power consumption offset C, terminal difference-available temperature difference ratio D, and thermal balance residual E obtained in step 1 are used as physical neurons of the hidden layer of the convolutional neural network and embedded into the hidden layer of the convolutional neural network; the calculation and propagation between the physical neurons follow equations (1) to (6). During operation, the chilled water supply temperature, chilled water return temperature, chilled water flow rate, cooling water return temperature, cooling water flow rate, evaporation temperature, condensation temperature, operating power, and the initial cooling water supply temperature of the chiller obtained by equation (14) are input to the convolutional neural network. The convolutional neural network outputs the predicted energy efficiency ratio (COP).

[0031] The computation and propagation between physical neurons strictly follow equations (1) to (6) to ensure that the variable correlation conforms to the physical mechanism of chiller unit operation. Simultaneously, the convolutional neural network structure is specifically optimized to address the temporal coupling and local correlation of chiller unit sensor data: Convolutional Neural Network Input Layer Design: A sequential input layer is used to receive the initial input data. A normalization method is set to standardize the input data and improve the stability of model training. Core Feature Extraction Layer: A feature extraction network consisting of two residual blocks is constructed. Each residual block contains two levels of one-dimensional convolutional layers. The first level of convolutional layer has an inflation factor (exponentially increasing by 2 with the residual block number), and the kernel size is set to 3. "causal" padding is used to avoid data leakage. At the same time, overfitting is suppressed through a layer normalization layer, a ReLU activation layer, and a spatial dropout layer (dropout factor set to 0.05). Residual Connection Design: After the first residual block achieves input dimension matching through a 1×1 convolutional layer, it is fused with the output of the second level convolutional layer. The second residual block directly fuses the input with the output of the second level convolutional layer to enhance feature propagation and alleviate the gradient vanishing problem.

[0032] Convolutional Neural Network Output Layer Design: The extracted high-dimensional features are mapped to a single output dimension through a fully connected layer, and a regression layer is used to complete the prediction of the energy efficiency ratio of the chiller unit.

[0033] This type of convolutional neural network model is adapted to the characteristics of sensor data, which helps to enhance feature extraction capabilities. Addressing the local correlations and temporal coupling of sensor data, the convolutional neural network structure can deeply mine the temporal-spatial coupling patterns inherent in variables, significantly improving the accuracy of COP prediction.

[0034] Preferably, step 3 includes: Step 301: Collect the chilled water supply temperature of the operating chiller unit. T chw,out Chilled water return temperature of chiller unitT chw,in chilled water flow rate of chiller unit M chw Cooling water return temperature of chiller unit T cw,in Cooling water flow rate of chiller unit M cw Evaporation temperature of chiller unit T ev Condensation temperature of chiller unit T cn and chiller unit operating power P ; Step 302: Calculate the energy efficiency ratio of the chiller unit based on the data collected in step 301 and the convolutional neural network model constructed in step 302.

[0035] In the above steps, the chilled water supply temperature, chilled water return temperature, chilled water flow rate, cooling water return temperature, cooling water flow rate, evaporation temperature, condensation temperature, and operating power of the operating chiller are collected from the chiller unit. These data are then substituted into the convolutional neural network model constructed in step 302 to calculate the predicted value of the chiller unit's energy efficiency ratio.

[0036] Preferably, step 302 includes: Step 3021: Obtain the cooling capacity of the chiller unit according to formula (12): Equation (12) In the formula, Q chw This indicates the cooling capacity of the chiller unit, in kW. Density of water, unit: kg / m³; Step 3022: Obtain the initial boundary conditions for the condensing heat load according to equation (13); Equation (13) In the formula, This represents the condensation heat load under the initial boundary conditions, in kW. Step 3023: Obtain the initial cooling water supply temperature of the chiller unit according to formula (14); Equation (14) In the formula, This indicates the initial cooling water supply temperature of the chiller unit, in °C. This represents the initial condensing heat load calculated by the convolutional neural network model, in kW. Step 3024: Set the initial cooling water supply temperature and chilled water supply temperature of the chiller unit. Tchw,out Chilled water return temperature of chiller unit T chw,in chilled water flow rate of chiller unit M chw Cooling water return temperature of chiller unit T cw,in Cooling water flow rate of chiller unit M cw Evaporation temperature of chiller unit T ev Condensation temperature of chiller unit T cn Chiller unit operating power P As input parameters, these parameters are used in the convolutional neural network model constructed in step 302 to obtain the chiller unit's operating energy efficiency ratio. COP model ; Step 3025: Obtain the condensation heat load calculated by the convolutional neural network model according to equation (15); Equation (15) In the formula, This represents the condensation heat load calculated by the convolutional neural network model, in kW. COP model This represents the chiller unit's operating energy efficiency ratio calculated by a convolutional neural network model; it has no unit. Step 3026: Based on the condensing heat load obtained in step 3025, determine whether the energy balance coefficient obtained according to equation (16) is less than or equal to the threshold. If yes, use the latest chiller unit operating energy efficiency ratio obtained in step 3024 as the predicted value of the chiller unit operating energy efficiency ratio. If no, iterate until the energy balance coefficient obtained according to equation (16) is less than or equal to the threshold. Preferably, the threshold is 3%.

[0037] Equation (16) In the formula, This represents the energy balance coefficient and has no unit. This indicates the condensation heat load obtained in step 3025. This represents the condensing heat load obtained in step 3025, in kW. The initial value is obtained from equation (13). This is the difference in condensing heat load calculated between two consecutive iterations in step 3025. (First calculation) hour, The initial value is obtained from equation (13).

[0038] Preferably, in step 3026, the iteration includes: Return to step 3023 and use the latest result obtained in step 3025. Replace step 3023 (14) And calculate the new chiller unit cooling water supply temperature according to formula (14); In step 3024, the new chiller cooling water supply temperature replaces the initial chiller cooling water supply temperature and is used as the input parameter of the convolutional neural network model to obtain the new chiller operating energy efficiency ratio. When performing step 3025, the chiller unit operating energy efficiency ratio in equation (15) is... COP model The condensing heat load was calculated by replacing the energy efficiency ratio of the new chiller unit with a new convolutional neural network model.

[0039] In the above preferred embodiment, the chiller supply water temperature is selected as the iterative parameter. Other sensor parameters are based on those obtained from the initial operation of the chiller and remain unchanged throughout the iteration process. The cooling water supply temperature determines the condensing pressure and temperature of the condenser, and is a thermodynamic parameter affecting the chiller's compression ratio, compressor power consumption, and cycle efficiency ratio. In engineering thermodynamics, for every 1°C increase in condensing temperature, the unit's COP decreases by approximately 2% to 4%, and its sensitivity to efficiency ratio is higher than that of chilled water side parameters or flow rate parameters. Furthermore, the cooling water supply temperature is easier to measure than parameters such as chilled water side temperature and flow rate.

[0040] The method of this invention filters sensor data and calculates the variables required to train a chiller unit model; it constructs a convolutional neural network model for predicting the energy efficiency ratio (EER) of a single-compressor chiller unit; simultaneously, based on the law of conservation of energy, it uses the convolutional neural network model to construct an iterative framework for predicting the EER of the chiller unit, thus achieving EER prediction. This method considers the differences in physical process mechanisms among different types of chiller units, while avoiding increased training costs and sensor noise introduction caused by variable redundancy. It effectively solves the problems of limited engineering application, poor generalization ability and interpretability, and insufficient versatility of current chiller unit models.

[0041] The method of this invention addresses multiple requirements, including sensor data quality optimization, feature depth extraction, and physical mechanism constraints. It combines sensor data processing, convolutional neural network modeling, and an energy conservation iterative framework to construct a chiller unit energy efficiency ratio prediction system. This effectively solves the problems of imbalanced input variable selection in existing models, the impact of sensor data bias on prediction accuracy, and the shortcomings of traditional models in extracting temporal-spatial coupling features from sensor data. It also addresses the poor adaptability of existing models to sensor data in engineering scenarios. Specifically, the method of this invention has the following significant advantages: First, the synergy between sensor data optimization and variable selection enhances the reliability of the prediction basis. By identifying, extracting, and selecting core sensor data, we avoid increased training costs and sensor noise caused by variable redundancy, while also compensating for information loss due to insufficient variables. At the same time, we reduce the propagation impact of biases such as sensor aging and drift, providing data support for the prediction of chiller unit energy efficiency ratio.

[0042] Second, by integrating physical mechanism constraints with engineering scenario adaptation, the practical value of the model is enhanced. An iterative prediction framework is built based on the law of conservation of energy, and the model's generalization ability under different working conditions is improved through physical constraints; at the same time, the non-standardized data from sensors and multi-sensor coupling errors are taken into account, improving the model's adaptability to complex operating scenarios in engineering sites and making it more practical.

[0043] The following is an example.

[0044] This embodiment focuses on a single-unit chiller with a capacity of 400 refrigeration tons (approximately 1407 kW).

[0045] The working principle of a single-unit chiller is as follows: It adopts a single-stage vapor compression refrigeration cycle, in which the refrigerant circulates in the direction of "compressor-condenser-throttling device-evaporator-compressor". The low-temperature, low-pressure gaseous refrigerant is compressed into a high-temperature, high-pressure gaseous state by the compressor and then enters the condenser, where it releases heat to the cooling water and condenses into a high-pressure liquid state. The liquid refrigerant is throttled by the throttling device and becomes a low-temperature, low-pressure gas-liquid mixture. It then enters the evaporator to absorb heat from the chilled water and vaporizes, returning to a low-temperature, low-pressure gaseous state and returning to the compressor, completing one cycle. This achieves the purpose of transferring the heat of the chilled water to the cooling water and discharging it.

[0046] Detailed information about the chiller unit in this embodiment is shown in Table 1.

[0047] Table 1 Overview of Chiller Unit Information

[0048] The method of this invention is used to predict the energy efficiency ratio of the single-head chiller unit, including: Step 1: Obtain sensor data from the chiller unit, and calculate the variables required to train the convolutional neural network model for predicting the energy efficiency ratio of the chiller unit based on the sensor data.

[0049] A single-unit chiller with a capacity of 400 refrigeration tons (approximately 1407 kW) was used to collect and measure the following variables through experiments: chilled water supply temperature of the chiller unit. T chw,out Chilled water return temperature of chiller unit T chw,in chilled water flow rate of chiller unit M chwCooling water supply temperature of chiller unit T cw,out Cooling water return temperature of chiller unit T cw,in Cooling water flow rate of chiller unit M cw Evaporation temperature of chiller unit T ev Condensation temperature of chiller unit T cn and chiller unit operating power P Using the aforementioned equations (1)-(6), the variables required for training the chiller unit model are calculated as follows: A. Load-thermodynamic difficulty product, B. Flow-temperature difference coupling ratio, C. Unit load power consumption offset, D. Terminal difference-available temperature difference ratio, E. Thermal balance residual, and COP of chiller unit operation.

[0050] Step 2: Based on the variables calculated in Step 1, construct a convolutional neural network model for predicting the energy efficiency ratio of a single-compressor chiller unit.

[0051] The chilled water supply temperature of the chiller unit measured in step 1 T chw,out Chilled water return temperature of chiller unit T chw,in chilled water flow rate of chiller unit M chw Cooling water supply temperature of chiller unit T cw,out Cooling water return temperature of chiller unit T cw,in Cooling water flow rate of chiller unit M cw Evaporation temperature of chiller unit T ev condensing temperature of chiller unit T cn and chiller unit operating power P As initial input data, the load-thermodynamic difficulty product A, flow-temperature difference coupling ratio B, unit load power consumption offset C, terminal difference-available temperature difference ratio D, and thermal balance residual E calculated in step 1 are used as physical neurons of the hidden layer of the neural network and embedded into the hidden layer of the convolutional neural network. The calculation and propagation between these physical neurons strictly follow equations (1) to (6) to ensure that the variable correlation conforms to the physical mechanism of chiller unit operation.

[0052] Meanwhile, the convolutional neural network structure was specifically optimized to address the temporal coupling and local correlation of the chiller unit sensor data: Convolutional Neural Network Input Layer Design: A sequential input layer is used to receive the initial input data. A normalization method is set to standardize the input data and improve the stability of model training. Core Feature Extraction Layer: A feature extraction network consisting of two residual blocks is constructed. Each residual block contains two levels of one-dimensional convolutional layers. The first level of convolutional layer has an inflation factor (exponentially increasing by 2 with the residual block number), and the kernel size is set to 3. "causal" padding is used to avoid data leakage. At the same time, overfitting is suppressed through layer normalization layers, ReLU activation layers, and spatial dropout layers (dropout factor set to 0.05). Residual Connection Design: After the first residual block achieves input dimension matching through a 1×1 convolutional layer, it is fused with the output of the second level convolutional layer. The second residual block directly fuses the input with the output of the second level convolutional layer to strengthen feature propagation and alleviate the gradient vanishing problem. Convolutional Neural Network Output Layer Design: The extracted high-dimensional features are mapped to a single output dimension through a fully connected layer, and a regression layer is used to complete the prediction of the energy efficiency ratio of the chiller unit.

[0053] The model uses the chiller unit's operating energy efficiency ratio (COP) as the output target. During training, the root mean square error (RMSE) shown in equation (7), the mean absolute error (MAE) shown in equation (8), the mean deviation error (MBE) shown in equation (9), and the coefficient of determination (R²) shown in equation (10) are introduced as performance evaluation indicators. The goal is to minimize the deviation between the predicted and actual values. Iterative training is conducted using the Adam optimizer (initial learning rate set to 0.002), with a batch size of 32 and a maximum training epoch of 1000, until the model converges. Finally, an improved convolutional neural network model for predicting the COP of a single-compressor chiller unit, adapted to the characteristics of chiller unit operating data, is constructed to achieve in-depth mining of the temporal-spatial coupling characteristics of sensor data.

[0054] Equation (7) Equation (8) Equation (9) Equation (10) Equation (11) In the formula, The COP value of the chiller unit for the kth sample is the true value. The predicted COP value for the chiller unit of the kth sample; This represents the total number of samples in the corresponding dataset; This is the average of the true values ​​of all samples.

[0055] Step 3: Substitute the real-time sensor data of the operating chiller unit into the convolutional neural network model constructed in Step 2 to predict the energy efficiency ratio of the chiller unit.

[0056] A single-head chiller unit with a capacity of 400 refrigeration tons (approximately 1407 kW) is given an initial boundary condition for the condensing heat load according to equation (13), and the initial chiller unit cooling water supply temperature is calculated using equation (14). This yields the initial input parameters for the convolutional neural network model. After model calculation, the initial chiller unit operating energy efficiency ratio is obtained. The condensing heat load calculated by the model is obtained based on equation (15). The energy balance coefficient obtained from equation (16) is used to determine whether the energy balance coefficient is less than or equal to a threshold of 3%. If it is greater than the threshold of 3%, iteration is performed until the energy balance coefficient is less than or equal to the threshold of 3%. During the iteration process, the latest obtained... In the substitution (14) And according to Equation (14), calculate the new chiller cooling water supply temperature; replace the initial chiller cooling water supply temperature with the new chiller cooling water supply temperature as the input parameter of the convolutional neural network model to obtain the new chiller operating energy efficiency ratio; replace the chiller operating energy efficiency ratio calculated by the convolutional neural network model in Equation (15) with the new chiller operating energy efficiency ratio to obtain the new condensing heat load calculated by the convolutional neural network model; obtain the new energy balance coefficient according to Equation (16) and make a judgment again; repeat the above process until the new energy balance coefficient is less than or equal to the threshold 3%.

[0057] Comparative Example: An existing convolutional neural network model is used. In this model, the load-thermodynamic difficulty product (A), flow-temperature difference coupling ratio (B), unit load power consumption offset (C), terminal difference-available temperature difference ratio (D), and thermal balance residual (E) are not embedded as physical neurons in the hidden layers of the convolutional neural network. Only the collected chilled water supply temperature of the chiller unit is used. T chw,out Chilled water return temperature of chiller unit T chw,in chilled water flow rate of chiller unit M chw Cooling water supply temperature of chiller unit T cw,out Cooling water return temperature of chiller unit T cw,in Cooling water flow rate of chiller unit M cw Chiller unit operating power P As input data, COP is output from the convolutional neural network.

[0058] Table 2 provides an overview of convolutional neural network models as a comparative example.

[0059] For a 400-ton (approximately 1407 kW) single-unit chiller under stable operating conditions, 4366 sets of model input data were collected and input into the above-mentioned embodiments and comparative examples, resulting in 4366 COP data points. Figure 2 and Figure 3 The lines are drawn from data obtained using an embodiment of the method of the present invention. Figure 4 and Figure 5 The lines are drawn using data obtained from a comparative scale using an existing convolutional neural network model. Figure 2 In the figure, the horizontal axis represents the collected prediction samples, and the vertical axis represents the chiller unit's operating energy efficiency ratio (COP). The red line represents the actual value of the chiller unit's operating energy efficiency ratio (COP), and the blue line represents the predicted value of the chiller unit's operating energy efficiency ratio (COP) obtained by the embodiment of the method of the present invention. Figure 3 In the figure, the horizontal axis represents the actual COP value of a single-unit chiller, and the vertical axis represents the predicted COP value of the chiller obtained by the embodiment of the method of the present invention. Figure 3 In the diagram, the blue dot represents the intersection of two lines: a straight line perpendicular to the horizontal axis drawn from the actual COP value of the chiller unit and a straight line perpendicular to the vertical axis drawn from the predicted COP value obtained from the model according to this invention. The red dashed line represents the ideal fit line, which means that the predicted COP value and the actual COP value of the chiller unit are equal. The ideal fit line is formed by drawing a straight line perpendicular to the horizontal axis from the actual COP value and a straight line perpendicular to the vertical axis from the predicted COP value obtained from the model according to this invention. The intersection of these two lines is a straight line with a slope of 1.

[0060] Figure 4 In the figure, the horizontal axis represents the collected prediction samples, and the vertical axis represents the chiller unit's operating energy efficiency ratio (COP). The red line represents the actual value of the chiller unit's operating energy efficiency ratio (COP), and the blue line represents the predicted value of the chiller unit's operating energy efficiency ratio (COP) obtained using the comparative method. Figure 5 In the figure, the horizontal axis represents the actual COP value of a single-unit chiller, and the vertical axis represents the predicted COP value of the chiller obtained using the comparative method. Figure 5In the diagram, the blue dot represents the intersection of two lines: a line perpendicular to the horizontal axis drawn using the actual COP value of the chiller unit and a line perpendicular to the vertical axis drawn using the predicted COP value obtained from the comparative model; the blue dot is the intersection of these two lines. The red dashed line represents the ideal fitting line, which means that the predicted COP value and the actual COP value of the chiller unit are equal. The ideal fitting line is formed by drawing a line perpendicular to the horizontal axis using the actual COP value and a line perpendicular to the vertical axis using the predicted COP value obtained from the comparative model; the intersection of these two lines is a line with a slope of 1. The actual COP value of the chiller unit is calculated using measured data according to equation (6).

[0061] from Figure 2 and Figure 3 It can be seen that, for single-unit chillers, compared with the actual COP values ​​obtained by the method of this invention and existing neural network models, the predicted values ​​of the COP obtained by the method of this invention are closer to the actual values. This indicates that the method of this invention has the highest prediction accuracy and can better handle abnormal data. Compared with existing neural network models, the method of this invention has a wider range of applications and higher practical value.

[0062] This invention is not limited to the above embodiments. Based on the technical solutions disclosed in this invention, those skilled in the art can make some substitutions and modifications to some of the technical features without creative effort, and all such substitutions and modifications are within the protection scope of this invention.

Claims

1. A method for predicting the energy efficiency ratio of a chiller unit, characterized in that, The method includes: Step 1: Obtain sensor data from the chiller unit, and calculate the variables required to train the convolutional neural network model for predicting the energy efficiency ratio of the chiller unit based on the sensor data. Step 2: Based on the variables calculated in Step 1, construct a convolutional neural network model for predicting the energy efficiency ratio of a single-compressor chiller unit; Step 3: Substitute the real-time sensor data of the operating chiller unit into the convolutional neural network model constructed in Step 2 to predict the energy efficiency ratio of the chiller unit. In step 1, the chiller unit sensor data includes: chilled water supply temperature of the chiller unit. T chw,out Unit: °C; Chilled water return temperature of chiller unit T chw,in Unit: °C; chilled water flow rate of chiller unit M chw Unit: m 3 / h; Cooling water supply temperature of chiller unit T cw,out Unit: °C; Cooling water return temperature of chiller unit T cw,in Unit: °C; Cooling water flow rate of chiller unit M cw Unit: m 3 / h; Evaporation temperature of the chiller unit T ev Unit: °C; Condensation temperature of chiller unit T cn Unit: °C; Chiller unit operating power P Unit: kW; In step 1, the variables required to calculate and train the convolutional neural network model for predicting the energy efficiency ratio of the chiller unit include: Using equation (1), calculate the variable load-thermodynamic difficulty product A of the convolutional neural network model for training the chiller unit's energy efficiency ratio prediction; using equation (2), calculate the variable flow-temperature coupling ratio B of the convolutional neural network model for training the chiller unit's energy efficiency ratio prediction; using equation (3), calculate the variable unit load power consumption offset C of the convolutional neural network model for training the chiller unit's energy efficiency ratio prediction; using equation (4), calculate the variable terminal difference-available temperature difference ratio D of the convolutional neural network model for training the chiller unit's energy efficiency ratio prediction; using equation (5), calculate the variable thermal balance residual E of the convolutional neural network model for training the chiller unit's energy efficiency ratio prediction; using equation (6), calculate the variable chiller unit operating energy efficiency ratio (COP) of the convolutional neural network model for training the chiller unit's energy efficiency ratio prediction: Equation (1) Equation (2) Equation (3) Equation (4) Equation (5) Equation (6) In the formula, c Specific heat capacity of water, unit: J / (kg·℃); Q 0 The rated cooling capacity on the nameplate of the chiller unit is in kW. P 0 Rated power consumption of the chiller unit, unit: kW.

2. The method for predicting the energy efficiency ratio of a chiller unit according to claim 1, characterized in that, In step 2, the load-thermodynamic difficulty product A, flow-temperature difference coupling ratio B, unit load power consumption offset C, end difference-available temperature difference ratio D, and thermal balance residual E obtained in step 1 are used as physical neurons of the hidden layer of the convolutional neural network and embedded into the hidden layer of the convolutional neural network; the calculation and propagation between the physical neurons follow equations (1) to (6).

3. The method for predicting the energy efficiency ratio of a chiller unit according to claim 2, characterized in that, Step 3 includes: Step 301: Collect the chilled water supply temperature of the operating chiller unit. T chw,out Chilled water return temperature of chiller unit T chw,in chilled water flow rate of chiller unit M chw Cooling water return temperature of chiller unit T cw,in Cooling water flow rate of chiller unit M cw Evaporation temperature of chiller unit T ev Condensation temperature of chiller unit T cn and chiller unit operating power P ; Step 302: Calculate the energy efficiency ratio of the chiller unit based on the data collected in step 301 and the convolutional neural network model constructed in step 302.

4. The method for predicting the energy efficiency ratio of a chiller unit according to claim 3, characterized in that, Step 302 includes: Step 3021: Obtain the cooling capacity of the chiller unit according to formula (12): Equation (12) In the formula, Q chw This indicates the cooling capacity of the chiller unit, in kW. Density of water, unit: kg / m³; Step 3022: Obtain the initial boundary conditions for the condensing heat load according to equation (13): Equation (13) In the formula, This represents the condensation heat load under the initial boundary conditions, in kW. Step 3023: Obtain the initial cooling water supply temperature of the chiller unit according to formula (14); Equation (14) In the formula, This indicates the initial cooling water supply temperature of the chiller unit, in °C. This represents the initial condensing heat load calculated by the convolutional neural network model, in kW. Step 3024: Set the initial cooling water supply temperature and chilled water supply temperature of the chiller unit. T chw,out Chilled water return temperature of chiller unit T chw,in chilled water flow rate of chiller unit M chw Cooling water return temperature of chiller unit T cw,in Cooling water flow rate of chiller unit M cw Evaporation temperature of chiller unit T ev Condensation temperature of chiller unit T cn Chiller unit operating power P As input parameters, these parameters are used in the convolutional neural network model constructed in step 302 to obtain the chiller unit's operating energy efficiency ratio. COP model ; Step 3025: Obtain the condensation heat load calculated by the convolutional neural network model according to equation (15): Equation (15) In the formula, This represents the condensation heat load calculated by the convolutional neural network model, in kW. COP model This represents the chiller unit's operating energy efficiency ratio calculated by a convolutional neural network model; it has no unit. Step 3026: Based on the condensing heat load obtained in step 3025, determine whether the energy balance coefficient obtained according to formula (16) is less than or equal to the threshold. If yes, the latest chiller unit operating energy efficiency ratio obtained in step 3024 is used as the predicted value of the chiller unit operating energy efficiency ratio. If no, iterate until the energy balance coefficient obtained according to formula (16) is less than or equal to the threshold. Equation (16) In the formula, This represents the energy balance coefficient and has no unit. This represents the condensing heat load obtained in step 3025, in kW. The initial value is obtained from equation (13).

5. The method for predicting the energy efficiency ratio of a chiller unit according to claim 4, characterized in that, In step 3026, iteration is performed, including: Return to step 3023 and use the latest result obtained in step 3025. Replace step 3023 (14) And calculate the new chiller unit cooling water supply temperature according to formula (14); In step 3024, the new chiller cooling water supply temperature replaces the initial chiller cooling water supply temperature and is used as the input parameter of the convolutional neural network model to obtain the new chiller operating energy efficiency ratio. When performing step 3025, the chiller unit operating energy efficiency ratio in equation (15) is... COP model The condensing heat load was calculated by replacing the energy efficiency ratio of the new chiller unit with a new convolutional neural network model.

6. The method for predicting the energy efficiency ratio of a chiller unit according to claim 4, characterized in that, The threshold is 3%.

Images