An air conditioning water system control method and system based on predictive optimal control

Abstract

The application discloses an air conditioner water system control method and system based on predictive optimal control and belongs to the technical field of energy-saving control. The method comprises the following steps: collecting operation data and meteorological data, and performing short-time cold load prediction; based on the operation data, an equipment energy consumption model is established; equipment operation constraints are obtained; a global optimization model is constructed through the equipment energy consumption model, taking the minimization of system total energy consumption as an optimization target and taking each equipment operation constraint as a constraint condition; an optimal control strategy is obtained by solving the global optimization model; a safety boundary check is performed; and if the safety boundary check is satisfied, a control instruction is issued to control each equipment. The application can predict short-time cold load, establish a global optimization model of the minimum system total energy consumption and solve an optimal control strategy, realizes the minimization of system total energy consumption, ensures the timeliness of the control strategy, and guarantees the safety of system operation by issuing a control instruction generated by the control strategy satisfying the safety evaluation condition.

Description

Technical Field

[0001] This invention relates to the field of energy-saving control technology, and more specifically to a control method and system for air conditioning water systems based on predictive optimization control. Background Technology

[0002] With the continuous improvement of energy structure transformation and energy efficiency management requirements, the optimized operation of integrated energy systems is receiving increasing attention. As a key energy-consuming component, the operating efficiency of the air conditioning water system directly affects the energy utilization efficiency and operational economy of the entire system.

[0003] In existing integrated energy systems, the control strategies for air conditioning water systems are mostly focused on meeting immediate cooling load demands and maintaining safe equipment operation. Their optimization objectives are often limited to local equipment efficiency or static operating conditions. However, the system's cooling load and equipment operating points are constantly changing dynamically. Traditional control strategies based on fixed rules or simple feedback struggle to achieve optimal system-level energy efficiency across the entire operating range. On one hand, load prediction relies heavily on field sensor data, but due to measurement accuracy issues (e.g., ±0.2℃ error in temperature measurement, significant errors in flow measurement, and data transmission processing lag), it's difficult to accurately reflect the system's true dynamic demands, leading to control commands being either premature or delayed. On the other hand, most strategies fail to fully consider the strong coupling characteristics between chillers, pumps, cooling towers, and other equipment, as well as efficiency differences under partial loads. Their control commands are often based on local optima rather than the overall system energy efficiency ratio (i.e., the ratio of total cooling capacity to total power consumption). This results in the system operating in a suboptimal state for extended periods under complex real-world conditions, causing unnecessary energy consumption. Furthermore, traditional methods lack a mechanism for proactively assessing system safety boundaries before generating control commands. In pursuing energy efficiency, they may overlook equipment constraints and hydraulic safety, posing operational risks.

[0004] Therefore, there is an urgent need for a control method for air conditioning water systems that can accurately predict demand, optimize the system's overall energy efficiency ratio as the core goal, and take into account operational safety. Summary of the Invention

[0005] In view of the above problems, the present invention is proposed to provide a predictive optimization control-based air conditioning water system control method and system that overcomes or at least partially solves the above problems.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] In a first aspect, embodiments of the present invention provide a control method for an air conditioning water system based on predictive optimization control, comprising: Collect operational data from multiple devices in the data center's air conditioning water system and outdoor meteorological data; Based on multiple operational data and meteorological data, an LSTM neural network is used to perform short-term prediction of cooling load; Based on the operational data, establish an energy consumption model corresponding to each of the devices; A global optimization model is constructed with minimizing the total system energy consumption as the optimization objective and the operating constraints of each device as the constraints. The optimal control strategy for controlling the air conditioning water system is obtained by solving the global optimization model using a global optimization algorithm. A safety boundary check is performed on the optimal control strategy and the predicted cooling load to determine whether the control commands corresponding to the optimal control strategy meet the safe operating conditions. If the conditions are met, the control command is issued to optimize the operation of each device in the air conditioning water system.

[0008] In one embodiment, the step of using an LSTM neural network to perform short-term cold load forecasting based on multiple operational data and meteorological data includes: Based on the dual-source fusion data calibration strategy, multiple operating data are calibrated to obtain the calibrated cold load; Based on the time data, the temperature data and humidity data in the meteorological data, and the calibrated cooling load measurement value, a feature vector within a preset time window is constructed. The feature vector is input into the LSTM neural network to determine the predicted cold load within a preset time period, and the predicted cold load for generating the control strategy in the next control cycle is predicted in a rolling manner according to the preset time period.

[0009] In one embodiment, the step of calibrating multiple operating data sets based on a dual-source fusion data calibration strategy to obtain the calibrated cold load includes: Based on the chilled water supply and return water temperature difference and flow rate measurements from the chilled water pump, the first cooling load, determined by direct measurement, is:

[0010] in, As the first cooling load, The specific heat capacity of water, The density of water, This is the measured value of chilled water flow rate. The temperature difference between the chilled water supply and return water; Based on the real-time power value and real-time energy efficiency ratio of the cooling water unit, the second cooling load is determined by power back-calculation. The second cooling load is:

[0011] in, As the second cooling load, This refers to the real-time power value of the cooling water unit. Real-time energy efficiency ratio; Using the error propagation formula, the first uncertainty corresponding to the first cooling load and the second uncertainty corresponding to the second cooling load are determined respectively; Based on the error covariance, the first uncertainty and the second uncertainty are weighted and fused to obtain a first weight corresponding to the first cooling load and a second weight corresponding to the second cooling load; The calibrated cooling load is calculated based on the first weight, the first cooling load, the second weight, and the second cooling load. The calibrated cooling load is:

[0012] in, For the calibrated cooling load, As the first weight, As the first cooling load, As the second weight, This is the second cooling load.

[0013] In one embodiment, determining the first uncertainty corresponding to the first cooling load and the second uncertainty corresponding to the second cooling load using the error propagation formula includes: Based on the first cooling load and the first error propagation formula, the first uncertainty is determined. The first error propagation formula is as follows:

[0014] in, The first uncertainty is... The standard deviation of flow measurement. For the standard deviation of temperature difference measurement, The specific heat capacity of water, The density of water, To address the temperature difference between the chilled water supply and return water, This is the measured flow rate of the chilled water. The second uncertainty is determined based on the second cooling load and the second error propagation formula. The second error propagation formula is as follows:

[0015] in, The second uncertainty, The standard deviation of power measurement. The standard deviation of the energy efficiency ratio estimate. This refers to the real-time power value of the cooling water unit. This refers to the real-time energy efficiency ratio.

[0016] In one embodiment, after calculating the calibrated cooling load based on the first weight, the first cooling load, the second weight, and the second cooling load, the method further includes: Compare whether the current load rate under the current operating conditions is less than the load rate threshold; If the current load rate is less than the load rate threshold, the first weight is corrected according to the low load correction coefficient, and the second weight is updated according to the corrected first weight. The calibrated cooling load is updated based on the corrected first weight and the updated second weight; Check if the sensor is malfunctioning; If the sensor malfunctions, the first weight is reset to zero, and the second weight is set to a preset value. The calibrated cooling load is updated according to the preset value; The relative deviation is determined based on the calibrated cooling load, the first cooling load, the second cooling load, and the relative deviation calculation formula, and it is determined whether the relative deviation is less than the relative deviation threshold. If the relative deviation is less than the relative deviation threshold, the first weight and the second weight are adjusted according to historical reference data, and the calibrated cooling load is updated.

[0017] In one embodiment, establishing a device energy consumption model corresponding to each of the devices based on the operational data includes: The step of establishing an energy consumption model for each device based on the operational data includes: Based on a multinomial regression model, the chilled water supply temperature, cooling water supply temperature, and cooling capacity of the chiller unit, an energy consumption model for the chiller unit is constructed. The energy consumption model for the chiller unit is as follows:

[0018] in, For chiller unit power, For chilled water supply temperature, For cooling water supply temperature, For cooling capacity, , , , , , , , , and All are model coefficients; Based on the similarity law of water pumps, the power of the chilled water pump, and the water flow rate of the chilled water pump, an energy consumption model for the chilled water pump is constructed. The energy consumption model for the chilled water pump is as follows:

[0019] in, For chilled water pump power, This refers to the rated power of the chilled water pump. This refers to the chilled water flow rate. This refers to the rated flow rate of the chilled water pump. Based on the similarity law of water pumps, the power of the cooling water pump, and the water flow rate of the cooling water pump, an energy consumption model for the cooling water pump is constructed. The energy consumption model for the cooling water pump is as follows:

[0020] in, For cooling water pump power, This refers to the rated power of the cooling water pump. For cooling water flow rate, This is the rated flow rate of the cooling water pump; A cooling tower energy consumption model is constructed based on the cooling tower power and the number of operating cooling towers. The cooling tower energy consumption model is as follows:

[0021] in, For cooling tower power, The rated power of the cooling tower For the number of runs, This represents the total number of cooling towers.

[0022] In one embodiment, the step of constructing a global optimization model with the goal of minimizing the total system energy consumption and the operational constraints of each device as the constraint condition includes: The energy consumption of each device is superimposed to construct the global optimization model;

[0023] in, The total power of the air conditioning water system. Let i be the power of the i-th chiller unit. Let i be the power of the i-th chilled water pump. Let the power of the kth cooling water pump be... The power of the lth cooling tower; The constraints include cooling balance constraints, operating parameter constraints, flow balance constraints, and temperature constraints, wherein: The cooling balance constraint is: ,in For the predicted cooling load; The operating parameter constraints are as follows: ,in, For the operating parameters of device n, Let n be the allowable range of operating parameters for device n; The flow balance constraint is: , This refers to the chilled water flow rate. The hydraulic balance condition corresponding to the chilled water flow rate; , For cooling water flow rate, The hydraulic balance condition corresponding to the cooling water flow rate; The temperature constraint is: , For chilled water supply temperature, The permissible supply temperature for chilled water; , For cooling water supply temperature, This refers to the permissible supply temperature of the cooling water.

[0024] In one embodiment, the step of solving the global optimization model using a global optimization algorithm to obtain the optimal control strategy for controlling the air conditioning water system includes: Based on the set operating data of each device, N particles containing multiple sets of set operating data are randomly generated, and the position and velocity of the particles are initialized; where N is an integer greater than 1; Based on the global optimization model and the positions of each particle, the current fitness value for evaluating the total system energy consumption under the current set operating data is determined. The historical fitness value of each particle is traversed to obtain the historical best fitness value. The current fitness value of each particle is compared with the historical best fitness value to determine the individual optimal position of each particle. Iterate through the updated fitness values ​​of particles at each position to determine the global optimal position; The position and velocity of each particle are updated according to the update formula, and the optimal control strategy is output when the convergence condition is met.

[0025] In one embodiment, the update formula includes a velocity update formula and a position update formula, wherein: The speed update formula is:

[0026] in, Let i be the updated particle velocity. For inertial weights, Let be the particle velocity of particle i at time t. and As a learning factor, and A random number within the interval [0, 1] Let be the position of particle i at time t. For the optimal position of an individual, The globally optimal position; The location update formula is:

[0027] in, Let i be the updated particle position. Let be the position of particle i at time t. Let i be the updated particle velocity.

[0028] Secondly, embodiments of the present invention provide a control system for an air conditioning water system based on predictive optimization control, employing the following technical solution including a cloud platform and a control terminal. The cloud platform includes a prediction module, and the control terminal includes a data acquisition module, an optimization calculation module, a safety assessment module, and an instruction issuance module, wherein: The data acquisition module is used to collect operational data from multiple devices in the data center's air conditioning water system and outdoor meteorological data. The prediction module is used to perform short-term prediction of cooling load based on multiple operational data and meteorological data, using an LSTM neural network. The optimization calculation module is used to establish an energy consumption model for each device based on the operating data; construct a global optimization model with minimizing the total energy consumption of the system as the optimization objective and the operating constraints of each device as the constraint condition; and obtain the optimal control strategy for controlling the air conditioning water system by calling the global optimization algorithm to solve the global optimization model. The safety assessment module is used to perform safety boundary checks on the optimal control strategy and the predicted cooling load to determine whether the control command corresponding to the optimal control strategy meets the safe operating conditions; if the safe operating conditions are met, the command issuance module is triggered. The instruction issuing module is used to issue the control instructions to achieve optimized control and operation of each device in the air conditioning water system.

[0029] The beneficial effects of the above-described technical solutions provided in the embodiments of the present invention include at least the following: The system collects operational and meteorological data from multiple devices in the air conditioning water system. Short-term predictions of the cooling load for the next moment are then made based on this data, enabling timely issuance of control commands. An energy consumption model is established based on the operational data, and a global optimization model is built from two dimensions: minimizing total system energy consumption and minimizing the operational constraints of each device. This ensures that the output of the optimization model minimizes energy consumption while meeting the normal operating requirements of the devices. A global optimization algorithm is then used to solve the global optimization model to obtain the optimal control strategy that considers multiple dimensions, thus preventing increased energy consumption due to the air conditioning water system operating in a suboptimal state. Finally, a safety boundary check is performed on the optimal control strategy, and corresponding commands are issued only when the optimal control strategy meets safety requirements. This ensures the safety of the air conditioning water system when operating according to the optimal control strategy and effectively reduces operational risks. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0031] Figure 1 This is a schematic diagram of the air conditioning water system control method based on predictive optimization control provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the short-term prediction process for cooling load provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the data calibration process based on the dual-source fusion data calibration strategy provided in an embodiment of the present invention. Detailed Implementation

[0032] 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.

[0033] This invention discloses a control method for an air conditioning water system based on predictive optimization control, such as... Figure 1 As shown, the method includes steps S101 to S107, wherein: Step S101: Collect operating data of multiple devices in the data center air conditioning water system and outdoor meteorological data.

[0034] Specifically, the equipment includes at least a chiller unit, a chilled water pump, a cooling water pump, and a cooling tower.

[0035] Operating data includes the power of the chiller unit, chilled water supply temperature, cooling water supply temperature, and cooling capacity; the power, flow rate, and frequency of the chilled water pumps and cooling water pumps; the power, fan frequency, and number of operating cooling towers; and meteorological data including outdoor temperature and humidity data.

[0036] The operational data and meteorological data are respectively the equipment sensors, temperature sensors and humidity sensors corresponding to each device. For example, they can be collected at a collection frequency of once every 3 minutes and transmitted, stored and processed through PLC and communication network. This embodiment of the invention does not limit the placement of the sensors, and the number of sensors can be multiple.

[0037] Step S102: Based on multiple operational and meteorological data, use an LSTM neural network to perform short-term prediction of cooling load.

[0038] Among them, such as Figure 2 The diagram shown illustrates the process of cooling load prediction provided in this embodiment of the invention. The first and second cooling loads are calculated based on temperature difference flow measurement and power back-calculation values, respectively. These are then weighted and fused using error covariance to obtain the calibrated cooling load. A feature vector within a preset time window is constructed based on outdoor temperature, humidity, time, and historical cooling load sequences. This feature vector is then input into an LSTM neural network. For example, the LSTM neural network predicts the cooling load within 15 minutes using the input feature vector, updating the input feature vector every 15 minutes, re-predicting, and re-outputting the prediction result, thereby achieving rolling prediction.

[0039] In this embodiment of the invention, the LSTM prediction model training adopts the Adam adaptive learning rate optimization algorithm. The initial learning rate is set to 0.001 during the training process. The specific training process includes: (1) dividing the training data into batches of 32, with each batch containing 32 samples; (2) feeding a batch of input sequences into the LSTM prediction model and obtaining the prediction output through forward propagation; (3) calculating the loss value between the prediction output and the true value of each training data using the mean squared error loss function; (4) calculating the gradient of the loss function with respect to the parameters of each layer using the backpropagation algorithm; (5) updating the model parameters according to the gradient using the Adam optimization algorithm; (6) repeating steps (2) to (5) until all training samples are traversed to complete one training round. Training stops when the LSTM prediction model meets the convergence condition.

[0040] The LSTM prediction model uses mean squared error (MSE) as the loss function, calculated as follows:

[0041] in, For output dimensions, This represents the actual load value. To predict load values, this loss function penalizes large errors more severely, which helps the model converge quickly.

[0042] Among them, the convergence condition is: (1) Maximum training rounds constraint: training stops when the number of training rounds reaches the preset upper limit (such as 100 rounds); (2) Optimal model selection strategy: After each training round, calculate the loss value of the model on the validation set. When the validation loss value of the current round is lower than the historical best validation loss value, save the current model parameters as the optimal model. (3) The training termination condition is to reach the maximum number of training rounds and finally output the model parameters corresponding to the minimum validation loss.

[0043] Understandably, the Adam adaptive learning rate optimization algorithm can leverage the advantages of first-order and second-order momentum when training LSTM prediction models, adaptively adjusting the learning rate of each parameter to accelerate model convergence.

[0044] The specific process for short-term cooling load forecasting includes: obtaining the calibrated cooling load by performing data calibration based on a dual-source fusion data calibration strategy and multiple operational data. Specifically, this includes: calculating the first cooling load based on temperature difference flow rate measurement, i.e., calculating the first cooling load based on the temperature difference and flow rate measurement of the chilled water supply and return water from the chilled water pump. The first cooling load is:

[0045] in, As the first cooling load, The specific heat capacity of water, =4.186kJ / ( ), The density of water, =1000 / , This is a measurement of the flow rate of chilled water, in units of... s, The temperature difference between chilled water supply and return water, in units of , , This refers to the chilled water return temperature. The chilled water supply temperature is determined by measuring the temperature difference and flow rate of the chilled water supply and return water. This allows for direct knowledge of the terminal load demand and provides a fast response time.

[0046] Next, the second cooling load is calculated based on power feedback. That is, the cooling capacity is inferred from the real-time power value and real-time energy efficiency ratio of the cooling water unit. Determining the second cooling load using a method independent of temperature and flow sensors can reduce the influence of sensors. The second cooling load is:

[0047] in, As the second cooling load, This is the real-time power value of the cooling water unit, in kW. This refers to the real-time energy efficiency ratio.

[0048] The real-time energy efficiency ratio is estimated based on the performance curve of the chiller unit, and the estimation formula is as follows:

[0049] in, Rated energy efficiency ratio This is a performance correction function. For chilled water supply temperature, For cooling water supply temperature, This refers to the load factor.

[0050] The performance correction function is obtained by fitting the manufacturer's performance data or historical operating data of the chiller unit. The performance correction function is as follows:

[0051] in, , , , , , , All coefficients are polynomials. It is understandable that by inferring the second cooling load from the power, this can be used as a backup cooling load measurement method in case of sensor failure.

[0052] Calculating the first uncertainty based on the first cooling load measured directly and the second uncertainty based on the second cooling load calculated from the power back-calculation, it can be understood that the uncertainty can accurately measure the reliability of the measurement results corresponding to the two methods.

[0053] The first uncertainty is calculated based on the error propagation formula, including:

[0054] in, The standard deviation of flow measurement. The value is determined based on the accuracy class of the flow sensor, and ranges from 0.5% to 2% of the rated flow. For the standard deviation of temperature difference measurement, The value is determined based on the accuracy class of the temperature sensor, and the range is 0.1℃ to 0.3℃. The specific heat capacity of water, The density of water, To address the temperature difference between the chilled water supply and return water, This is the measured flow rate of chilled water.

[0055] The second uncertainty of the second cooling load based on power back-calculation is:

[0056] in, The second uncertainty, The standard deviation of power measurement. The value is determined based on the accuracy class of the power sensor, and ranges from 0.5% to 1% of the rated power. The standard deviation of the energy efficiency ratio estimate. The value was determined based on the fitting accuracy of the chiller unit's performance curve, and ranges from 0.2% to 0.5%. This represents the real-time power value of the cooling water unit. This refers to the real-time energy efficiency ratio.

[0057] Then, the first uncertainty and the second uncertainty are weighted and fused according to the error covariance to obtain the first weight and the second weight corresponding to the above two methods. In this embodiment of the invention, the fused weight is inversely proportional to the uncertainty; the larger the uncertainty, the smaller the corresponding weight. Therefore, the weighted fused weight has theoretical optimality. The first weight based on the directly measured first cooling load and the second weight based on the power back-calculation method for the second cooling load are respectively: (1) (2) in, As the first weight, It is the second weight, and .

[0058] After recalculating the fused cooling load, it can be understood that when the two measurement methods are independent, the fusion weight is inversely proportional to their respective variances. Therefore, the variance after fusion is minimized, and the calibrated cooling load is:

[0059] in, For the calibrated cooling load, As the first weight, As the first cooling load, As the second weight, This is the second cooling load.

[0060] Understandably, by adopting a dual-source fusion data calibration strategy and by weighting and fusing the first cooling load measured by temperature difference flow rate and the second cooling load based on power back-calculation, the accumulation of sensor errors can be effectively eliminated and the accuracy of the calibrated cooling load can be improved.

[0061] Furthermore, the accuracy of the calibrated cooling load can be verified by calculating the fused uncertainty. The fused uncertainty is:

[0062] in, The uncertainty after fusion is understood to be less than the uncertainty of any single method. This indicates that the accuracy of the combined cooling load is higher than that of any single method.

[0063] To further improve the accuracy of the fused cooling load, this embodiment of the invention can also adaptively adjust the weights based on three dimensions: operating conditions, sensor malfunction, and consistency deviation. Specifically, this includes: comparing whether the current load rate is less than a load rate threshold; when it is less than the load rate threshold, it indicates that the system is in a low-load condition. At this time, the relative error of the first cooling load obtained based on the direct measurement method increases, so a low-load correction coefficient is introduced to adjust the first weight. The adjustment formula is as follows:

[0064] in, The first weight after correction. For correction factor, .

[0065] Then, update the second weight based on the corrected first weight. The updated second weight is:

[0066] in, This is the updated second weight.

[0067] The calibration cooling load corresponding to the low-load condition is determined based on the updated second weight, and the calibration cooling load corresponding to the low-load condition is then used for subsequent data processing calculations.

[0068] The first relative deviation is determined by comparing the cooling load under low load conditions with the cooling load before adaptive weight adjustment, based on the relative deviation calculation formula. If the first relative deviation is less than the relative deviation threshold, then the weights and cooling load under low load conditions are used. The relative deviation calculation formula is as follows: Alternatively, determine if the sensor is malfunctioning. If the sensor is malfunctioning, reset the first weight corresponding to the direct measurement method to zero, and set the second weight based on the power back-calculation method to 1, then calculate the calibrated cooling load. Specifically, if the rate of change of the data detected by the sensor exceeds a rate of change threshold or the data detected by the sensor exceeds a preset value range, the sensor is determined to be malfunctioning. The sensor can be a temperature sensor or a flow sensor; this embodiment of the invention does not limit the rate of change threshold.

[0069] Alternatively, the relative deviations arising from the direct measurement method and the power back-calculation method can be calculated using the following formula:

[0070] in, This represents the relative deviation.

[0071] The relative deviation is compared with a relative deviation threshold. When the relative deviation is greater than the relative deviation threshold, historical cooling load data corresponding to each method is obtained. The accuracy of each method is determined based on the historical cooling load data, and the first weight and the second weight are adjusted according to the accuracy of each method. In this embodiment of the invention, the relative deviation threshold can be 20%.

[0072] Based on time data, temperature data, humidity data from meteorological data, and calibrated cooling load measurements, a feature vector within a preset time window is constructed, specifically including: In this embodiment of the invention, the preset time window is 1 hour. The feature vector includes the outdoor temperature sequence, humidity sequence, time sequence, and cooling load sequence of the past hour. It is understood that, due to the thermal inertia of the building itself, the cooling load of multiple previous times will affect the cooling load of the current time. Therefore, it is necessary to use the sequence of load influencing factors over a past period as the basis for prediction.

[0073] The feature vector is then input into an LSTM neural network to predict the cooling load for the next 15 minutes. In this embodiment, the LSTM neural network employs three gating mechanisms—a forget gate, an input gate, and an output gate—to address the vanishing gradient problem during long-sequence training, achieving long-term memory of temporal information. The forget gate determines the degree to which historical information from the previous cell state is retained or discarded; the input gate filters the current input information and generates candidate states to update the current cell state; and the output gate controls the output degree based on the updated cell state, generating the hidden state for the current time step. These three gating mechanisms work together, enabling the model to adaptively retain important load features while forgetting irrelevant noise. This embodiment solves the vanishing gradient problem by setting three gate functions, ensuring long-term information transmission. Furthermore, the LSTM neural network possesses long short-term memory capabilities, making it suitable for building cooling load prediction scenarios.

[0074] The expression for the forget gate is:

[0075] in, The output of the forget gate at time t. It is the sigmoid activation function. Here is the weight matrix for the forget gate. This is the hidden state from the previous moment. For the current input, This is the bias vector for the forget gate; The expression for the input gate is:

[0076] in, The input gate outputs at time t. It is the sigmoid activation function. Here is the weight matrix of the input gate. This is the hidden state from the previous moment. For the current input, The bias vector for the input gate; The expression for the output gate is:

[0077] in, The output of the output gate at time t. It is the sigmoid activation function. Here is the weight matrix of the output gate. This is the hidden state from the previous moment. For the current input, This is the bias vector for the output gate.

[0078] Furthermore, the LSTM neural network model prediction input is updated every 15 minutes to achieve rolling prediction and optimization of the cold load. The prediction step size is set to 15 minutes, and the input features are updated every 15 minutes to re-predict the cold load for the next 15 minutes, so that the prediction results can guide the generation of the control strategy in the next control optimization cycle.

[0079] Step S103: Based on the operating data, establish the energy consumption model corresponding to each device.

[0080] Specifically, the equipment energy consumption models include chiller unit energy consumption models, chilled water pump energy consumption models, cooling water pump energy consumption models, and cooling tower energy consumption models.

[0081] The chiller unit energy consumption model can be constructed based on the chilled water supply temperature, cooling water supply temperature, and cooling capacity of the chiller unit, using a multinomial regression model. The chiller unit energy consumption model is as follows:

[0082] in, For chiller unit power, For chilled water supply temperature, For cooling water supply temperature, For cooling capacity, , , , , , , , , and All of these are model coefficients obtained by fitting historical operating data; The energy consumption model of a chilled water pump can be constructed using the power-to-flow ratio and based on the pump similarity law. The chilled water pump energy consumption model is as follows:

[0083] in, For chilled water pump power, This refers to the rated power of the chilled water pump. This refers to the chilled water flow rate. This refers to the rated flow rate of the chilled water pump. The cooling water pump energy consumption model can be constructed using the ratio of the cooling water pump's power to its flow rate, based on the pump similarity law. The cooling water pump energy consumption model is as follows:

[0084] in, For cooling water pump power, This refers to the rated power of the cooling water pump. For cooling water flow rate, This is the rated flow rate of the cooling water pump; Cooling tower energy consumption models can be constructed using cooling tower power and the number of operating towers. The cooling tower energy consumption model is as follows:

[0085] in, For cooling tower power, The rated power of the cooling tower For the number of runs, This represents the total number of cooling towers.

[0086] Step S104: Obtain the operating constraints of each device, and construct a global optimization model with the goal of minimizing the total energy consumption of the system and the operating constraints of each device as the conditions.

[0087] Specifically, the device operation constraints of each device can be obtained from the data information repository.

[0088] The global optimization model is:

[0089] in, The total power of the air conditioning water system. Let i be the power of the i-th chiller unit. Let i be the power of the i-th chilled water pump. Let the power of the kth cooling water pump be... The power of the lth cooling tower.

[0090] The constraints of the global optimization model include cooling balance constraints, operating parameter constraints, flow balance constraints, and temperature constraints, among which: The cooling balance constraint is: This means that the total cooling capacity of the air conditioning water system meets the cooling load demand of the terminals. To predict cooling load; The operating parameters are constrained as follows: ,in, For the operating parameters of device n, This refers to the normal operating parameter range of equipment n, that is, the operating parameters of each piece of equipment are within the normal operating parameter range. For example, the load rate of a chiller unit should be between 30% and 100%. The flow balance constraint is that the chilled water flow rate and the cooling water flow rate must meet the hydraulic balance requirements of the air conditioning water system, wherein, , This refers to the chilled water flow rate. The hydraulic balance condition corresponding to the chilled water flow rate; , For cooling water flow rate, The hydraulic balance condition corresponding to the cooling water flow rate; Temperature constraints are: , For chilled water supply temperature, This refers to the permissible supply temperature range for chilled water. , For cooling water supply temperature, This refers to the permissible supply temperature range of the cooling water; where, ; .

[0091] This application employs a modeling method that combines physical mechanisms with data-driven approaches, resulting in model parameters with clear physical meanings.

[0092] Step S105: Solve the global optimization model using a global optimization algorithm to obtain the optimal control strategy for controlling the air conditioning water system.

[0093] Specifically, Particle Swarm Optimization (PSO) is a global optimization algorithm based on swarm intelligence. It boasts advantages such as fast convergence, few parameters, and ease of implementation. Therefore, PSO is used to solve the global optimization model. The specific process of solving the global optimization model using PSO includes: initializing the particle swarm. Multiple particles are randomly generated, where each particle represents a set of control variables. Each set of control variables contains the set operating data for each device, such as the chiller outlet water temperature setpoint, the water pump set frequency, and the cooling tower fan speed set frequency; and the position and velocity of each particle are randomly initialized.

[0094] Calculate the fitness value. In this embodiment of the invention, the global optimization model is the objective function in the particle swarm optimization algorithm. Therefore, the position of each particle is substituted into the objective function to obtain the current fitness value of each particle, where the current fitness value is the total energy consumption of the air conditioning water system. It can be understood that the fitness value is used to evaluate the quality of the particle solution. In this embodiment of the invention, the lower the current fitness value, the lower the total energy consumption of the system, indicating that the set of operating parameters is better.

[0095] Update the individual and global optimal values ​​for each particle. For each particle, iterate through all its historical fitness values ​​to obtain the historical best fitness value. If the particle's current fitness value is better than the historical best fitness value, then the current fitness value is determined as the particle's optimal fitness value; otherwise, the historical best fitness value is still determined as the particle's optimal fitness value. Iterate through all particles and, based on the optimal fitness values ​​of each particle, determine the optimal particle from all particles. Compare the current fitness value of the optimal particle with the historical global best fitness value to update the global optimal position.

[0096] Then, the particle velocity and particle position are updated using velocity update formulas and position update formulas. The velocity update formula is as follows:

[0097] in, Let i be the updated particle velocity. For inertial weights, Let be the particle velocity of particle i at time t. and As a learning factor, and A random number within the interval [0, 1] Let be the position of particle i at time t. Let be the optimal position of particle i. This is the globally optimal position.

[0098] The position update formula is:

[0099] in, Let i be the updated particle position. Let be the position of particle i at time t. Let i be the updated particle velocity.

[0100] The system determines whether the maximum number of iterations or fitness value has been reached. If convergence has occurred, the global optimal solution is output as the optimal control strategy. Otherwise, iterative optimization continues until the maximum number of iterations or fitness value converges. The optimal control strategy includes at least: the optimal outlet water temperature setpoint for the chiller unit, the optimal number of operating cooling units, the optimal operating frequency of the chilled water pumps and cooling water pumps, the optimal fan operating frequency of the cooling tower, and the optimal number of operating cooling towers. Global optimization control can effectively reduce the system's total energy consumption.

[0101] Step S106: Perform a safety boundary check on the optimal control strategy and the predicted cooling load to determine whether the control commands corresponding to the optimal control strategy meet the safe operating conditions.

[0102] Specifically, safety boundary checks include temperature boundary checks, flow boundary checks, equipment start-up and shutdown checks, and load matching checks.

[0103] Temperature boundary checks include: chilled water supply temperature not lower than the minimum supply temperature; cooling water temperature not higher than the maximum supply temperature.

[0104] Flow boundary checks include ensuring that both chilled water flow and cooling water flow are within the hydraulic balance conditions corresponding to each piece of equipment, in order to avoid pump overload or cavitation.

[0105] Equipment start-up and shutdown checks include ensuring that the equipment start-up and shutdown sequence conforms to preset safety logic. For example, the start-up and shutdown interval for a chiller unit should be no less than five minutes. The preset safety logic may be the same or different for different equipment.

[0106] Load matching checks include ensuring that the deviation between the cooling capacity of the air conditioning water system and the predicted cooling load is within a preset deviation range to avoid insufficient or excessive cooling. In this embodiment of the invention, the preset deviation range is... .

[0107] If the safe operating conditions are met, proceed to step S107; if the safe operating conditions are not met, execute the abnormal rollback strategy, record the abnormal type, abnormal time, and abnormal parameters; and switch to the preset safe operating mode corresponding to the abnormal parameters according to the abnormal type and abnormal parameters, while using the control parameters corresponding to the preset safe operating mode for control.

[0108] Anomaly alarm information is sent to the equipment on the maintenance personnel's side. After confirmation from the maintenance personnel, the normal optimized control mode is restored. Through safety assessments and anomaly rollback mechanisms, the safe and stable operation of the air conditioning water system can be ensured.

[0109] Step S107: If satisfied, issue a control command to optimize the operation of each device in the air conditioning water system.

[0110] Specifically, when the total system load rate is in the range of 70% to 80% and the LSTM predicts a load change rate greater than 2% / 15min, the global optimization algorithm determines that the load is rising sharply, outputs a command to add chillers in advance, and controls the start-up of the standby chiller unit with the shortest cumulative running time based on the principle of balanced wear. When the total system load rate is less than 40% and the load change rate is less than -1% / 15min, a command to reduce the number of chillers is output, controlling the chiller unit with the longest cumulative running time to stop operating. When the total system load rate is in the range of 40% to 50%, the current number of operating units is maintained, thus forming a lag loop for adding and reducing chillers to avoid frequent start-ups and shutdowns of chiller units at critical load points.

[0111] The chilled water pump employs a dual closed-loop variable frequency control system, with a minimum operating frequency of 30Hz and a maximum of 50Hz, and a frequency adjustment step of 0.5Hz. In the differential pressure protection closed loop, the target differential pressure is set to 0.08MPa. When the actual supply and return water differential pressure is lower than 0.08MPa for 180 seconds, the global optimization algorithm will calculate the frequency increase and execute it, with a maximum single frequency increase not exceeding 1.5Hz.

[0112] Node-RED is a process-based visualization programming tool with advantages such as high development efficiency, easy extensibility, and support for multiple communication protocols. Therefore, control commands can be issued to various devices in the field through control terminals built using the Node-RED framework. Furthermore, Node-RED's seamless integration with various devices can improve the level of automation.

[0113] It is understood that the air conditioning water system control method based on predictive optimization control of the present invention is applied to a control terminal built on the Node-RED framework. The control terminal includes a data acquisition module, an optimization calculation module, a safety assessment module, an instruction issuance module, and an exception handling module.

[0114] The data acquisition module communicates with end devices via Modbus and BACnet protocols to achieve real-time acquisition of operational data from each device. The optimization calculation module uses a particle swarm optimization algorithm to perform global optimization and generate the optimal control strategy. The safety assessment module performs safety boundary checks on the control commands generated based on the optimal control strategy to ensure the safety of the air conditioning water system under the current control commands. The command issuance module issues the safety-assessed control commands to the controllers of each device to control their operation. The anomaly handling module switches to a safe operation mode and sends alarm information when an anomaly is detected.

[0115] The control terminal performs optimization calculations every preset time interval, such as 15 minutes, and generates control commands, which are then sent to the controllers of each device to achieve optimized system operation.

[0116] Based on the above embodiments, operational data and meteorological data of multiple devices in the air conditioning water system are collected, and short-term prediction of the cooling load for the next moment is made using the operational data and meteorological data, so that control commands can be issued in a timely manner based on the predicted cooling load; then, an energy consumption model is established based on the operational data, and a global optimization model is established from two dimensions: minimizing the total energy consumption of the system and the operational constraints of each device, so that the output of the optimization model can minimize energy consumption while meeting the normal operation of the equipment.

[0117] The global optimization model is then solved using a global optimization algorithm to obtain the optimal control strategy that takes into account multiple dimensions, thereby avoiding increased energy consumption due to the air conditioning water system being in a non-optimal operating state. The optimal control strategy is then subjected to a safety boundary check, and the corresponding command is issued when the optimal control strategy meets the safety requirements, thereby ensuring the safety of the air conditioning water system when operating according to the optimal control strategy and effectively reducing operational risks.

[0118] In one specific embodiment of the present invention, the operating parameters of the chiller unit, the chilled water pump, the cooling water pump, the cooling tower, and the outdoor meteorological parameters in the data center air conditioning water system are collected in real time, and the collected operating data is summarized into a dataset.

[0119] A dual-source fusion data calibration strategy and operating parameters are used to calibrate the cooling load to determine the accuracy of the cooling load data. At the same time, the accuracy of the cooling load is further improved by adjusting the weights according to the actual operating conditions. The accurate cooling load data is then input into an LSTM neural network for short-term cooling load prediction, and a rolling optimization strategy is implemented.

[0120] Based on the collected operating parameters of each device, an energy consumption model for each device is established to form a total system energy consumption model. Using a particle swarm optimization algorithm, the global optimization model, constructed with minimizing total system energy consumption as the optimization objective and device operating constraints as the constraints, is solved to obtain the optimal control strategy. Control commands are generated according to the optimal control strategy, and safety boundary checks are performed to ensure that each device can operate safely and stably under the optimal control strategy. After the control commands pass the safety boundary check, they are issued to each device, and prediction and control are executed every 15 minutes to achieve optimized operation of the air conditioning water system.

[0121] The above embodiments introduce a predictive optimization control-based air conditioning water system control method from the perspective of the method flow. The following embodiments introduce a predictive optimization control-based air conditioning water system control system from the perspective of virtual modules or simulation units. For details, please refer to the following embodiments.

[0122] This invention provides a predictive optimization control-based air conditioning water system control system, which may include a cloud platform and a control terminal. The cloud platform includes a prediction module, and the control terminal includes a data acquisition module, an optimization calculation module, a safety assessment module, and a command issuance module, wherein: The data acquisition module is used to collect operational data from multiple devices in the data center's air conditioning water system and outdoor meteorological data. The prediction module is used to perform short-term prediction of cooling load based on multiple operational data and the meteorological data, using an LSTM neural network. The optimization calculation module is used to establish an energy consumption model for each device based on the operating data; to construct a global optimization model with minimizing the total energy consumption of the system as the optimization objective and the operating constraints of each device as the constraint conditions; and to obtain the optimal control strategy for controlling the air conditioning water system by calling the global optimization algorithm to solve the global optimization model. The safety assessment module is used to perform safety boundary checks on the optimal control strategy and the predicted cooling load to determine whether the control commands corresponding to the optimal control strategy meet the safe operating conditions; if the safe operating conditions are met, the command issuance module is triggered. The instruction issuing module is used to issue control instructions to optimize the operation of various devices in the air conditioning water system.

[0123] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0124] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A control method for an air conditioning water system based on predictive optimization control, characterized in that, include: Collect operational data from multiple devices in the data center's air conditioning water system and outdoor meteorological data; Based on multiple operational data and meteorological data, an LSTM neural network is used to perform short-term prediction of cooling load; Based on the operational data, establish an energy consumption model corresponding to each of the devices; A global optimization model is constructed with minimizing the total system energy consumption as the optimization objective and the operating constraints of each device as the constraints. The optimal control strategy for controlling the air conditioning water system is obtained by solving the global optimization model using a global optimization algorithm. A safety boundary check is performed on the optimal control strategy and the predicted cooling load to determine whether the control commands corresponding to the optimal control strategy meet the safe operating conditions. If the conditions are met, the control command is issued to optimize the operation of each device in the air conditioning water system.

2. The air conditioning water system control method based on predictive optimization control as described in claim 1, characterized in that, The method of using an LSTM neural network to perform short-term cold load forecasting based on multiple operational data and meteorological data includes: Based on the dual-source fusion data calibration strategy, multiple operating data are calibrated to obtain the calibrated cold load; Based on the time data, the temperature data and humidity data in the meteorological data, and the calibrated cooling load measurement value, a feature vector within a preset time window is constructed. The feature vector is input into the LSTM neural network to determine the predicted cold load within a preset time period, and the predicted cold load for generating the control strategy in the next control cycle is predicted in a rolling manner according to the preset time period.

3. The air conditioning water system control method based on predictive optimization control as described in claim 2, characterized in that, The method of calibrating multiple operational data based on a dual-source fusion data calibration strategy to obtain the calibrated cold load includes: Based on the chilled water supply and return water temperature difference and flow rate measurements from the chilled water pump, the first cooling load, determined by direct measurement, is: in, As the first cooling load, The specific heat capacity of water, The density of water, This is the measured value of chilled water flow rate. The temperature difference between the chilled water supply and return water; Based on the real-time power value and real-time energy efficiency ratio of the cooling water unit, the second cooling load is determined by power back-calculation. The second cooling load is: in, As the second cooling load, This refers to the real-time power value of the cooling water unit. Real-time energy efficiency ratio; Using the error propagation formula, the first uncertainty corresponding to the first cooling load and the second uncertainty corresponding to the second cooling load are determined respectively; Based on the error covariance, the first uncertainty and the second uncertainty are weighted and fused to obtain a first weight corresponding to the first cooling load and a second weight corresponding to the second cooling load; The calibrated cooling load is calculated based on the first weight, the first cooling load, the second weight, and the second cooling load. The calibrated cooling load is: in, For the calibrated cooling load, As the first weight, As the first cooling load, As the second weight, This is the second cooling load.

4. The air conditioning water system control method based on predictive optimization control as described in claim 3, characterized in that, The step of determining the first uncertainty corresponding to the first cooling load and the second uncertainty corresponding to the second cooling load using the error propagation formula includes: Based on the first cooling load and the first error propagation formula, the first uncertainty is determined. The first error propagation formula is as follows: in, The first uncertainty is... The standard deviation of flow measurement. For the standard deviation of temperature difference measurement, The specific heat capacity of water, The density of water, To address the temperature difference between the chilled water supply and return water, This is the measured flow rate of the chilled water. The second uncertainty is determined based on the second cooling load and the second error propagation formula. The second error propagation formula is as follows: in, The second uncertainty, The standard deviation of power measurement. The standard deviation of the energy efficiency ratio estimate. This refers to the real-time power value of the cooling water unit. This refers to the real-time energy efficiency ratio.

5. The air conditioning water system control method based on predictive optimization control as described in claim 3, characterized in that, After calculating the calibrated cooling load based on the first weight, the first cooling load, the second weight, and the second cooling load, the method further includes: Compare whether the current load rate under the current operating conditions is less than the load rate threshold; If the current load rate is less than the load rate threshold, the first weight is corrected according to the low load correction coefficient, and the second weight is updated according to the corrected first weight. The calibrated cooling load is updated based on the corrected first weight and the updated second weight; Check if the sensor is malfunctioning; If the sensor malfunctions, the first weight is reset to zero, and the second weight is set to a preset value. The calibrated cooling load is updated according to the preset value; The relative deviation is determined based on the calibrated cooling load, the first cooling load, the second cooling load, and the relative deviation calculation formula, and it is determined whether the relative deviation is less than the relative deviation threshold. If the relative deviation is less than the relative deviation threshold, the first weight and the second weight are adjusted according to historical reference data, and the calibrated cooling load is updated.

6. The air conditioning water system control method based on predictive optimization control as described in claim 1, characterized in that, The step of establishing an energy consumption model for each device based on the operational data includes: Based on a multinomial regression model, the chilled water supply temperature, cooling water supply temperature, and cooling capacity of the chiller unit, an energy consumption model for the chiller unit is constructed. The energy consumption model for the chiller unit is as follows: in, For chiller unit power, For chilled water supply temperature, For cooling water supply temperature, For cooling capacity, , , , , , , , , and All are model coefficients; Based on the similarity law of water pumps, the power of the chilled water pump, and the water flow rate of the chilled water pump, an energy consumption model for the chilled water pump is constructed. The energy consumption model for the chilled water pump is as follows: in, For chilled water pump power, This refers to the rated power of the chilled water pump. This refers to the chilled water flow rate. This refers to the rated flow rate of the chilled water pump. Based on the similarity law of water pumps, the power of the cooling water pump, and the water flow rate of the cooling water pump, an energy consumption model for the cooling water pump is constructed. The energy consumption model for the cooling water pump is as follows: in, For cooling water pump power, This refers to the rated power of the cooling water pump. For cooling water flow rate, This is the rated flow rate of the cooling water pump; A cooling tower energy consumption model is constructed based on the cooling tower power and the number of operating cooling towers. The cooling tower energy consumption model is as follows: in, For cooling tower power, The rated power of the cooling tower For the number of runs, This represents the total number of cooling towers.

7. The air conditioning water system control method based on predictive optimization control as described in claim 2, characterized in that, The global optimization model, which aims to minimize the total energy consumption of the system and uses the operational constraints of each device as constraints, includes: The energy consumption of each device is superimposed to construct the global optimization model; in, The total power of the air conditioning water system. Let i be the power of the i-th chiller unit. Let i be the power of the i-th chilled water pump. Let the power of the kth cooling water pump be... The power of the lth cooling tower; The constraints include cooling balance constraints, operating parameter constraints, flow balance constraints, and temperature constraints, wherein: The cooling balance constraint is: ,in For the predicted cooling load; The operating parameter constraints are as follows: ,in, For the operating parameters of device n, Let n be the allowable range of operating parameters for device n; The flow balance constraint is: , This refers to the chilled water flow rate. The hydraulic balance condition corresponding to the chilled water flow rate; , For cooling water flow rate, The hydraulic balance condition corresponding to the cooling water flow rate; The temperature constraint is: , For chilled water supply temperature, The permissible supply temperature for chilled water; , For cooling water supply temperature, This refers to the permissible supply temperature of the cooling water.

8. The air conditioning water system control method based on predictive optimization control as described in claim 1, characterized in that, The step of solving the global optimization model using a global optimization algorithm to obtain the optimal control strategy for controlling the air conditioning water system includes: Based on the set operating data of each device, N particles containing multiple sets of set operating data are randomly generated, and the position and velocity of the particles are initialized; where N is an integer greater than 1; Based on the global optimization model and the positions of each particle, the current fitness value for evaluating the total system energy consumption under the current set operating data is determined. The historical fitness value of each particle is traversed to obtain the historical best fitness value. The current fitness value of each particle is compared with the historical best fitness value to determine the individual optimal position of each particle. Iterate through the updated fitness values ​​of particles at each position to determine the global optimal position; The position and velocity of each particle are updated according to the update formula, and the optimal control strategy is output when the convergence condition is met.

9. The air conditioning water system control method based on predictive optimization control as described in claim 8, characterized in that, The update formula includes a velocity update formula and a position update formula, wherein: The speed update formula is: in, Let i be the updated particle velocity. For inertial weights, Let be the particle velocity of particle i at time t. and As a learning factor, and A random number within the interval [0, 1] Let be the position of particle i at time t. For the individual's optimal position, The globally optimal position; The location update formula is: in, Let i be the updated particle position. Let be the position of particle i at time t. Let i be the updated particle velocity.

10. A control system for an air conditioning water system based on predictive optimization control, characterized in that, The system includes a cloud platform and a control terminal. The cloud platform includes a prediction module, and the control terminal includes a data acquisition module, an optimization calculation module, a security assessment module, and a command issuance module. The data acquisition module is used to collect operational data from multiple devices in the data center's air conditioning water system and outdoor meteorological data. The prediction module is used to perform short-term prediction of cooling load based on multiple operational data and meteorological data, using an LSTM neural network. The optimization calculation module is used to establish an energy consumption model for each device based on the operating data; construct a global optimization model with minimizing the total energy consumption of the system as the optimization objective and the operating constraints of each device as the constraint condition; and obtain the optimal control strategy for controlling the air conditioning water system by calling the global optimization algorithm to solve the global optimization model. The safety assessment module is used to perform safety boundary checks on the optimal control strategy and the predicted cooling load to determine whether the control command corresponding to the optimal control strategy meets the safe operating conditions; if the safe operating conditions are met, the command issuance module is triggered. The instruction issuing module is used to issue the control instructions to achieve optimized control and operation of each device in the air conditioning water system.

Images