Cold station operation control method, device, equipment and medium

Abstract

The application discloses a cold station operation control method and device, equipment and medium, and relates to the technical fields of energy management, artificial intelligence and the like. The cold station operation control method comprises the following steps: acquiring state variables, wherein the state variables comprise cold station operation state data and environment state data; when making a decision based on the state variables, determining a matched probability distribution mode based on the data type of the control variables, and obtaining a control variable probability distribution matched with the probability distribution mode; performing data sampling on the control variable probability distribution to obtain a control action corresponding to the control variables; and controlling the cold station operation based on the control action.

Description

Technical Field

[0001] This invention relates to the fields of energy management, artificial intelligence and other technologies, and in particular to a method, device, equipment and medium for controlling the operation of a chiller plant. Background Technology

[0002] As a crucial component of building energy consumption, chiller plants account for a significant proportion of total building energy consumption. Statistics show that buildings consume approximately 40% of global energy demand, with central air conditioning accounting for 40%–50% of total building energy consumption. Chiller plant energy consumption constitutes a major portion of air conditioning energy consumption, accounting for over 30% of total building electricity consumption. Therefore, optimizing chiller plant operation and control to improve energy efficiency is of great significance for reducing building energy consumption and operating costs. Optimized chiller plant control not only contributes to energy conservation and emission reduction but also enhances building comfort and operational efficiency, promoting sustainable development.

[0003] The proposed cold storage plant operation control methods are highly complex, resulting in low computational efficiency; they are also not very adaptable to inputs and are difficult to cope with changing operating conditions; in addition, they lack online adaptability and cannot adjust and optimize strategies online to deal with emergencies. Summary of the Invention

[0004] The embodiments of the present invention aim to at least partially solve one of the technical problems in the related art. Therefore, one object of the present invention is to provide a cooling plant operation control method, apparatus, equipment, and medium, which achieves efficient energy utilization and improves the intelligence level of the cooling plant.

[0005] This invention provides a method for controlling the operation of a chilled plant. The method includes: acquiring state variables, wherein the state variables include chilled plant operation state data and environmental state data; when making decisions based on the state variables, determining a matching probability distribution method based on the data type of the control variables, and obtaining a control variable probability distribution that matches the probability distribution method; sampling the control variable probability distribution to obtain the control action corresponding to the control variable; and controlling the operation of the chilled plant based on the control action.

[0006] For example, the data type of the control variable includes at least one of continuous, discrete, and on / off types. Determining the matching probability distribution method based on the data type of the control variable includes: when the data type of the control variable is continuous, determining the probability distribution method as a normal probability distribution method; when the data type of the control variable is discrete, determining the probability distribution method as a categorical probability distribution method; and when the data type of the control variable is on / off, determining the probability distribution method as a Bernoulli probability distribution method.

[0007] For example, sampling the probability distribution of control variables to obtain the control actions corresponding to the control variables includes: using preset safety rule data as sampling constraints to sample the probability distribution of control variables to obtain the control actions corresponding to the control variables.

[0008] For example, the chiller plant operation status data and environmental status data include current data, historical data and future data. The current data includes at least one of cooling load, equipment parameters, system parameters and weather. The historical data includes at least one of cooling load and weather. The future data includes at least one of cooling load and weather.

[0009] For example, the chiller plant operation control method is applied to a decision control model, which includes an input layer, an intermediate layer, and an output layer. The input layer is used to input state variables; the intermediate layer is used to make decisions based on the state variables; and the output layer is used to determine the matching probability distribution method based on the data type of the control variables, obtain the probability distribution of the control variables that matches the probability distribution method, sample the data of the probability distribution of the control variables, obtain the control actions corresponding to the control variables, and output the control actions.

[0010] For example, the method further includes: acquiring training samples, wherein the training samples include historical state variables; training a decision control model based on the training samples; wherein training the decision control model based on the training samples includes: making a decision based on the historical state variables to obtain a target control action; obtaining a loss function value based on the probability and reward of the target control action according to the gradient loss function; obtaining parameter gradient information based on the loss function value; and updating the model parameters of the decision control model based on the parameter gradient information.

[0011] For example, the reward is accumulated and processed to obtain the accumulated discounted reward, and the gradient loss function represents the sum of the product of the accumulated discounted reward and the logarithm of the probability.

[0012] For example, training a decision control model based on training samples further includes: determining the probability of the target control action corresponding to the historical state variable based on the policy function; and determining the reward of the target control action corresponding to the historical state variable based on the reward function.

[0013] Another embodiment of the present invention provides a chiller plant operation control device, comprising: an acquisition module for acquiring state variables, wherein the state variables include chiller plant operation state data and environmental state data; a first acquisition module for determining a matching probability distribution method based on the data type of the control variable when making decisions based on the state variables, and obtaining a control variable probability distribution matching the probability distribution method; a second acquisition module for sampling the control variable probability distribution data to obtain the control action corresponding to the control variable; and a control module for controlling the operation of the chiller plant based on the control action.

[0014] The present invention provides an electronic device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the method of any of the above embodiments.

[0015] The present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method of any of the above embodiments.

[0016] In the above embodiments, the chiller plant operation control method includes: acquiring state variables, wherein the state variables include chiller plant operation state data and environmental state data; when making decisions based on state variables, determining a matching probability distribution method based on the data type of control variables, and obtaining a control variable probability distribution matching the probability distribution method; sampling the control variable probability distribution to obtain the control action corresponding to the control variable; and controlling the chiller plant operation based on the control action. This method, by acquiring state variables including chiller plant operation state data and environmental state data, can comprehensively grasp the real-time operation status of the chiller plant and external environmental conditions, helping to make decisions that are more in line with reality. Determining the matching probability distribution method based on the data type of control variables fully considers the characteristics of the control variables, making the decision-making process more scientific and reasonable, improving the accuracy and reliability of the decisions. Obtaining the control action corresponding to the control variable by sampling the control variable probability distribution makes the control action more flexible and adaptable, better able to cope with various complex and changing situations during chiller plant operation, improving the operating efficiency and stability of the chiller plant, and achieving efficient energy utilization.

[0017] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0018] Figure 1 A flowchart of a chiller plant operation control method provided for an embodiment of the present invention;

[0019] Figure 2This is a schematic diagram of the decision control model structure provided in an embodiment of the present invention;

[0020] Figure 3 This is a flowchart illustrating the reasoning and decision-making process of the decision control model provided in this embodiment of the invention.

[0021] Figure 4 This is a flowchart illustrating the training logic of the decision control model provided in an embodiment of the present invention.

[0022] Figure 5 A comparison chart of the effects of the decision control model provided in the embodiments of the present invention;

[0023] Figure 6 A block diagram of a chiller plant operation control device provided for another embodiment of the present invention;

[0024] Figure 7 A block diagram of an electronic device provided for another embodiment of the present invention. Detailed Implementation

[0025] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0026] As a crucial component of building energy consumption, chiller plants account for a significant proportion of total building energy consumption. Statistics show that buildings consume approximately 40% of global energy demand, with central air conditioning accounting for 40%–50% of total building energy consumption. Chiller plant energy consumption constitutes a major portion of air conditioning energy consumption, accounting for over 30% of total building electricity consumption. Therefore, optimizing chiller plant operation and control to improve energy efficiency is of great significance for reducing building energy consumption and operating costs. Optimized chiller plant control not only contributes to energy conservation and emission reduction but also enhances building comfort and operational efficiency, promoting sustainable development.

[0027] In the field of chiller plant optimization control, the proposed technologies demonstrate the diversity and cutting-edge nature of technological development in this area. One related technology proposes an energy-saving control method based on multi-agent systems in chiller plants. This method constructs building agents and air conditioning unit agents to perform global optimization control of the chiller plant, comprehensively considering building load, environmental parameters, and air conditioning unit operating status to minimize overall energy consumption. Another related technology demonstrates an optimization control method based on event-driven genetic algorithms. This method periodically acquires historical data from central air conditioning chiller plant equipment, constructs a power model, and performs real-time optimization calculations based on event trigger points. This effectively avoids the problems of long global optimization calculation time and numerous parameters, thus improving optimization efficiency.

[0028] With the development of artificial intelligence technology, another related technology has proposed a Transformer-based energy-saving control method for data centers. This method constructs a PUE (Power Usage Effectiveness) prediction model to comprehensively analyze data center chiller stations, terminal equipment, and environmental indicators, achieving online optimization of operating parameters and significantly improving energy-saving performance. Another related technology proposes a solution based on adaptive algorithms. This solution collects environmental and chiller station equipment data, uses multiple machine learning models for cooling load prediction, and selects the optimal prediction model through an adaptive algorithm, achieving precise energy-saving control of chiller units.

[0029] In terms of model fusion and self-learning, another related technology proposes an optimized control method for central air conditioning refrigeration stations based on a self-learning fusion model. This method combines a mechanistic model and a data model, utilizing a self-learning architecture to continuously optimize the model's accuracy and adaptability, thereby achieving overall energy efficiency optimization of the refrigeration station system. Another related technology provides a high-efficiency refrigeration station control system. This system monitors refrigeration station parameters in real time through sensors and optimizes control strategies and parameter configurations using an iterative optimization module. Simultaneously, the system also includes temperature control devices, data storage modules, fault detection and protection modules, etc., realizing intelligent control and efficient operation of the refrigeration station.

[0030] Regarding the comprehensive optimization of air conditioning systems, another related technology proposes an optimized control method for the energy consumption of air conditioning air and water systems. This method collects sensor data, uses a long short-term memory neural network for load prediction, and simultaneously performs coordinated control of the air and water systems, achieving comprehensive system optimization and high-efficiency energy saving. Another related technology demonstrates an energy-saving control method for cooling water systems based on overall energy efficiency optimization. This method constructs a historical database, monitors the refrigeration station environment and system operating parameters in real time, and utilizes historical data matching and parameter optimization steps to achieve optimal overall energy efficiency of the cooling water system. Simultaneously, the system includes an evaluation module to quantitatively assess the energy-saving effect.

[0031] Despite significant progress in the optimization and control of cooling plants, some shortcomings remain. For example, some models are highly complex, potentially leading to low computational efficiency; some systems lack adaptability to inputs, making it difficult to cope with varying operating conditions; furthermore, some systems lack online adaptability, failing to adjust optimization strategies online to address unforeseen circumstances. Therefore, future optimization and control technologies for cooling plants require further exploration and innovation in areas such as reducing model complexity, improving input adaptability, and enhancing online adaptability.

[0032] In view of this, the present application proposes a chiller plant operation control method. This method deeply integrates the latest advances in deep learning and deep reinforcement learning with profound expertise in the field of HVAC. It is specifically designed to address a series of complex challenges encountered in chiller plant energy efficiency optimization. Based on deep neural networks and stochastic policy gradients, it significantly enhances the intelligence level of chiller plant operation and provides strong technical support for achieving energy-efficient utilization and reducing operating costs.

[0033] Figure 1 A flowchart of a chiller plant operation control method provided for an embodiment of the present invention.

[0034] like Figure 1 As shown, the chiller plant operation control method 100 includes steps S110 to S140.

[0035] Step S110: Obtain state variables, including cold station operating status data and environmental status data.

[0036] For example, the operating status data of the chiller plant includes the chiller load, equipment parameters (start and stop status, frequency, load rate, cooling efficiency, etc. of the chiller, chilled pump and cooling pump, cooling tower) and system parameters (temperature, pressure, flow rate, etc. of the cooling side and the chiller side), and environmental status data includes weather data (temperature, humidity, air enthalpy, rainfall, etc.).

[0037] Step S120: When making decisions based on state variables, determine the matching probability distribution method based on the data type of the control variables, and obtain the control variable probability distribution that matches the probability distribution method.

[0038] For example, the data types of control variables include continuous control variables, discrete control variables, and on / off control variables, and the probability distribution methods include normal probability distribution, categorical probability distribution, and Bernoulli probability distribution. Decision-making based on state variables refers to inputting the state variables into a decision control model, which may include models based on deep neural networks.

[0039] Step S130: Sample the probability distribution of the control variables to obtain the control actions corresponding to the control variables.

[0040] For example, the probability distribution of the control variable (normal probability distribution, categorical probability distribution, Bernoulli probability distribution, etc.) has a corresponding probability for each possible value. Data sampling can be carried out by randomly selecting some specific values ​​as control actions according to the rules of this probability distribution. Control actions include, for example, the setpoints of freezing and cooling temperatures, the number of operating chillers, water pumps, and cooling towers, the operating frequency of water pumps and cooling towers, and the start and stop of equipment.

[0041] Step S140: Control the operation of the chiller plant based on the control action.

[0042] For example, based on the obtained control actions (set values ​​of freezing and cooling temperatures, number of operating chillers, water pumps, and cooling towers, operating frequency of water pumps and cooling towers, and start / stop of equipment), corresponding control is performed on the chiller plant equipment, thereby controlling the operation of the chiller plant system.

[0043] According to the embodiments of this application, by acquiring state variables including chiller plant operating status data and environmental status data, it is possible to comprehensively grasp the real-time operating status of the chiller plant and external environmental conditions, which helps to make decisions that are more in line with the actual situation. By determining the matching probability distribution method according to the data type of the control variables, the characteristics of the control variables are fully considered, making the decision-making process more scientific and reasonable, and improving the accuracy and reliability of the decision. By sampling the probability distribution of the control variables to obtain the corresponding control actions, the control actions are more flexible and adaptable, which can better cope with various complex and changing situations during the operation of the chiller plant, improve the operating efficiency and stability of the chiller plant, and realize the efficient use of energy.

[0044] The data type of the control variable includes at least one of continuous, discrete, and on / off types. The probability distribution method for matching is determined based on the data type of the control variable, including: when the data type of the control variable is continuous, determining the probability distribution method as a normal probability distribution method; when the data type of the control variable is discrete, determining the probability distribution method as a categorical probability distribution method; and when the data type of the control variable is on / off, determining the probability distribution method as a Bernoulli probability distribution method.

[0045] Specifically, for continuous control variables, such as temperature settings, a normal distribution can be used to describe them. The normal distribution is a continuous probability distribution that is symmetrical about the mean. The peak of the curve corresponds to the mean, and the curve gradually declines as it moves away from the mean. This distribution is completely determined by two parameters: the mean (μ), representing the average level of the data and corresponding to the expected value of the temperature setting; and the variance (σ). 2 Standard deviation (σ) or standard deviation (σ): indicates the degree of dispersion of the data, that is, the range of fluctuation of the temperature setpoint around the mean.

[0046] For discrete control variables, such as the operating status or working mode of equipment, a category distribution can be used to describe them. Category distributions are suitable for representing random variables with a finite number of possible values, each with a corresponding probability. In a chiller plant control system, the number of equipment units and their operating modes can be considered as a finite number of categories, each corresponding to a certain probability distribution.

[0047] For binary switching variables, such as the start / stop state of equipment, the Bernoulli distribution can be used to describe them. The Bernoulli distribution is a discrete distribution, also known as the "0-1 distribution" or "two-point distribution," which has only two possible values: 0 and 1, corresponding to the off and on states of the equipment, respectively. The parameter of the Bernoulli distribution is the probability p of success, i.e., the probability of the equipment starting. In a chiller plant control system, this probability can be adjusted according to the equipment's operating strategy and requirements.

[0048] The probability distribution methods used in this application are illustrated using normal probability distribution, categorical probability distribution, and Bernoulli probability distribution as examples. This is not intended to limit the application. The probability distribution description of the control variables will not be limited to the above variables, as long as it can clearly describe the probability of the control action to be taken.

[0049] The above embodiments can adapt to various control variables, covering common data types such as continuous, discrete, and switch types, and meet the data analysis needs in various practical scenarios, with strong versatility and applicability.

[0050] Data sampling of the probability distribution of control variables to obtain the corresponding control actions includes: using preset safety rule data as sampling constraints to sample the probability distribution of control variables and obtain the corresponding control actions.

[0051] Specifically, based on the obtained probability distribution of the control variables, random variable sampling can be performed to obtain the output of the optimized control variables. For example, sampling a normal distribution with a mean of 5.6℃ and a standard deviation of 0.2℃ can yield an optimized chilled water setpoint of 5.7℃, although other temperatures are possible, but it will most likely be close to 5.6℃. Sampling the number of chilled water pumps in operation will most likely result in the optimized state requiring 3 chilled water pumps to be in operation. Regarding the action of cooling pump No. 1, the result will most likely be that cooling pump No. 1 is turned on, but there is also about a 20% probability that it will not be turned on. An example output is that, under optimized state, the output should be: the cooling water outlet temperature should be set to 5.6℃, 3 chilled water pumps should be turned on, and cooling pump No. 1 should be turned on.

[0052] The output control actions are filtered. For example, if the predefined safety rules stipulate that four refrigeration pumps must be turned on when the outdoor temperature is 30°C or above, then the above output does not meet the predetermined safety rules. Therefore, the above output distribution needs to be randomly sampled again until it meets the safety rules.

[0053] In the above embodiments, the safety rule data reflects the reasonable boundaries and constraints of the system operation. Using this as a constraint for sampling ensures that the sampled control actions are based on safety and reasonable premises. This helps to make more scientific and reasonable decisions on the basis of meeting safety requirements and avoids unreasonable control actions caused by blind sampling.

[0054] The operating status data and environmental status data of the chiller plant include current data, historical data and future data. Current data includes at least one of the following: cooling load, equipment parameters, system parameters and weather conditions. Historical data includes at least one of the following: cooling load and weather conditions. Future data includes at least one of the following: cooling load and weather conditions.

[0055] For example, chiller plant operating status data includes at least one of the following: cooling load, equipment parameters, and system parameters. Environmental status data includes weather conditions. The chiller plant's cooling load includes the current value, historical value, and future value of the current cooling load. The current value is the chiller plant load measured in real time, typically expressed as cooling capacity per unit time (e.g., tons of cooling per hour or kilowatts), which can be directly measured or calculated using sensors such as flow meters and thermometers. Short-term historical values ​​are load data over a past period (e.g., minutes or hours), used to analyze load trends and potentially predict short-term load fluctuations. Future forecast values ​​are based on historical data and weather forecasts, using predictive models (e.g., time series analysis, machine learning algorithms) to derive future load predictions, which are crucial for energy dispatch and optimized operating strategies.

[0056] Equipment parameters: These include the operating parameters of the main equipment, including the chiller's start / stop status (indicating whether the chiller is running); frequency (the operating frequency of the chiller compressor, affecting cooling capacity and energy consumption); load rate (the ratio of the chiller's actual load to its rated load); and power (the electrical energy consumed by the chiller). The chilled water pump and cooling pump's start / stop status (indicating whether the pump is running); frequency (the pump speed, affecting water flow and energy consumption); and the cooling tower's start / stop status (indicating whether the cooling tower is running); fan frequency (the speed of the cooling tower fan, affecting heat dissipation and energy consumption); and cooling efficiency (the cooling tower's ability to cool hot water to the set temperature).

[0057] System parameters include: cooling-side temperature (cooling water temperature, affecting the chiller's cooling effect); pressure (cooling system pressure, which must be maintained within a safe range); and flow rate (cooling water flow rate, ensuring sufficient cooling capacity). Refrigeration-side temperature (chilled water temperature, determining the cooling system's output temperature); pressure (refrigeration system pressure, also needing to be maintained within a safe range); and flow rate (chilled water flow rate, affecting the overall efficiency of the cooling system).

[0058] The weather data includes key status parameters such as the current, historical, and future outdoor temperatures of the cooling station. Current values ​​are real-time measurements of outdoor temperature, humidity, and enthalpy, which directly affect the cooling station's cooling demand. Weather forecasts are based on future weather predictions from meteorological stations, including temperature, humidity, and rainfall, used to influence and adjust the cooling station's operating strategies to cope with upcoming weather changes. Historical values ​​represent weather data over a past period (e.g., the past few hours), used to input weather patterns on the cooling station's load and improve the accuracy of optimization decisions.

[0059] The state variables (cooling plant operating status data and environmental status data) in this application are described above and are not intended to limit the application. They are not limited to the above types. Any variable that can provide information for optimizing control decisions can be used as the system's input variables. The above variables can be read from sensors, building automation systems, cooling plant group control systems, etc.

[0060] The above embodiments provide a rich information foundation for the decision-making of cooling plants by integrating current, historical and future data. They take into account the current operating status of the equipment, refer to the equipment life cycle and failure status in historical data, and combine future load demand and development trends to make more scientific and reasonable decisions.

[0061] The chiller plant operation control method is applied to a decision control model, which includes an input layer, an intermediate layer, and an output layer. The input layer is used to input state variables; the intermediate layer is used to make decisions based on the state variables; and the output layer is used to determine the matching probability distribution based on the data type of the control variables, obtain the probability distribution of the control variables that matches the probability distribution, sample the data of the probability distribution of the control variables, obtain the control actions corresponding to the control variables, and output the control actions.

[0062] For example, the decision control model includes a neural network module that can take the state variables of the cooling station as input and output optimized control actions.

[0063] The input layer of the neural network is input after preprocessing the state variables of the above-mentioned cold station. Typical preprocessing methods include normalization of continuous variables and encoding of discrete variables, such as applying one-hot encoding or label encoding.

[0064] The intermediate layers of a neural network can use common neural network structures. A typical implementation can be defined using several fully connected layers.

[0065] The output layer of a neural network uses a series of probability distribution functions to describe the probability of taking a certain action. Different probability distributions are used for continuous control variables, discrete control variables, and on / off control variables. For continuous control variables, a normal probability distribution is used; for discrete control variables, a category distribution is used; and for on / off control variables, a Bernoulli probability distribution is used. Random sampling is performed based on the probability distribution parameters to obtain the output of the control variable (the control action).

[0066] The outputs of the control variables in the decision control model include: Refrigeration temperature setpoint: determines the output temperature of the refrigeration system, affecting cooling performance and user-side temperature control. The decision control model needs to dynamically adjust the refrigeration temperature setpoint based on factors such as load demand, outdoor weather conditions, and system efficiency to achieve optimal cooling performance and energy balance; Cooling temperature setpoint: affects the chiller's cooling efficiency and energy consumption. The model needs to optimize the cooling temperature setpoint based on the chiller's performance characteristics, the cooling tower's heat dissipation capacity, and the overall system's heat balance requirements; Number of chillers operating: based on the size and trend of the cooling load, the model needs to determine how many chillers to start to meet the cooling demand, while considering the chiller's energy efficiency ratio and operating costs; Number of water pumps operating: the water pumps are responsible for circulating chilled water and... Cooling water needs to meet the required flow rate and the system's hydraulic balance requirements; Number of operating cooling towers: Cooling towers are used for heat dissipation, and the model determines how many cooling towers to start to achieve the best heat dissipation effect; Water pump operating frequency: By adjusting the water pump speed (i.e., operating frequency), the water flow rate can be controlled, thereby affecting the efficiency and energy consumption of the cooling system; Cooling tower fan operating frequency: The speed of the cooling tower fan directly affects the heat dissipation effect and energy consumption, and the decision control model needs to adjust the operating frequency of the cooling tower fan to achieve the best heat dissipation effect and energy consumption balance; Equipment start-up and shutdown: The start-up and shutdown of key equipment such as a specific chiller or water pump.

[0067] This application uses the above-mentioned output control variables as an example for illustration, and is not intended to limit the application in any specific way. The output variables of the decision control model are not limited to the above-mentioned output variables. All variables that can affect the state of the cold station and can be output can be defined as output control actions.

[0068] Figure 2 This is a schematic diagram of the decision control model structure provided in an embodiment of the present invention.

[0069] like Figure 2 As shown, the input layer of the decision control model takes into account state variables such as outdoor temperature, outdoor humidity, and chiller load. After inference and decision-making by the intermediate layer, the output layer outputs the chilled water temperature setpoint distribution μ, the standard deviation of the chilled water temperature setpoint distribution σ, the probability distribution of starting pump 1 p1, and the probability distribution of starting n pumps q. nwait.

[0070] Figure 3 The following is a flowchart of the reasoning and decision-making process of the decision control model provided in the embodiments of the present invention.

[0071] like Figure 3 As shown, the decision control model reasoning decision process includes steps S310 to S350.

[0072] Step S310: Organize the operating status variables of the cooling plant and input them into the decision neural network.

[0073] Step S320: The neural network outputs a series of probability distributions for control variables.

[0074] Step S330: Sample the probability distribution to obtain a series of control variables.

[0075] Step S340: Determine whether the combination of control variables meets the constraint rules. If it does, proceed to step S350; otherwise, repeat step S330 until the constraint rules are met.

[0076] Step S350: Output the control variables. This inference cycle ends, and we begin the next cycle.

[0077] Specifically, during neural network control inference, the input state variables are mapped to the probability distribution of the output. The output is calculated using the neural network's input. For example, given an outdoor temperature of 30℃ and outdoor humidity of 60%, after preprocessing such as normalization and combining other state variables, the output is used in the decision-making neural network to obtain the parameters of the control variable distribution. For instance, the chilled water outlet temperature distribution under optimized conditions has a mean of 5.6℃ and a standard deviation of 0.2℃, which is a random variable conforming to a normal distribution. Under optimized conditions, the probability of operating 3 out of 4 chilled water pumps is 70%, and the probabilities of operating 1, 2, and 4 pumps are 5%, 15%, and 10%, respectively, which are random variables conforming to a categorical distribution. Under optimized conditions, the probability of operating cooling pump 1 is 20%, and the probability of not operating cooling pump 1 is 80%, which are random variables conforming to a Bernoulli distribution. By sampling based on the above probability distribution, the output of the optimized control variable can be obtained. The control action of the output can be filtered. For example, if the predefined safety rule stipulates that four refrigeration pumps must be turned on when the outdoor temperature is 30°C or above, the above output does not meet the predetermined safety rule. Therefore, the above output distribution needs to be randomly sampled again until it meets the safety rule.

[0078] The method further includes: acquiring training samples, wherein the training samples include historical state variables; training a decision control model based on the training samples; wherein training the decision control model based on the training samples includes: making decisions based on historical state variables to obtain target control actions; obtaining loss function values ​​based on the probability and reward of the target control actions according to the gradient loss function; obtaining parameter gradient information based on the loss function values; and updating the model parameters of the decision control model based on the parameter gradient information.

[0079] For example, the training samples include historical state variables, which are quantities that change over time, i.e., recorded as s. t s t This is a set of outdoor weather, chiller station and equipment status at a given time t. The target control actions include temperature setting, number of units activated, set frequency, and starting / stopping a specific piece of equipment; it is a set of quantities that change over time. The control action recorded at time t is denoted as a. t .

[0080] During neural network training, the gradient loss function is defined as shown in equation (1):

[0081] loss(t) = -G t ·logπ θ (a t |s t (1)

[0082] In formula (1), π θ (a t |s t ) is in state s t Next, take action a t The gradient loss function incentivizes the policy network to assign higher probabilities to state-action pairs that yield high rewards, thereby reducing the loss value. G t As a reward.

[0083] The training of the decision control model based on training samples also includes: determining the probability of the target control action corresponding to the historical state variable based on the policy function; and determining the reward of the target control action corresponding to the historical state variable based on the reward function.

[0084] For example, the policy function is the probability distribution of the output control action after probability sampling and rule filtering through a neural network, denoted as π. θ θ represents the set of model parameters, which includes not only the weight parameters of each layer of the neural network, but also the distribution parameters of the random variables in the output layer, such as μ and σ for a normal distribution, and the probabilities p of each option in class distributions and Bernoulli distributions, etc. θ (a t |s t) is in state s t Next, take action a t The probability of.

[0085] Payoff function: At a certain time t, in state s t Next, use the strategy function π θ The resulting profit is denoted as G. t (i.e., reward), G t The definition of a function can include, but is not limited to, the following forms: the Coefficient of Performance (COP) of the system at this moment and for a future period; the negative of the total power of the system at this moment and for a future period; and a benefit function that considers comfort and energy saving. The function should conform to the overall indicators of energy consumption, energy efficiency, and comfort; the larger the benefit value, the better.

[0086] The reward is accumulated and processed to obtain the cumulative discounted reward. The gradient loss function represents the sum of the product of the cumulative discounted reward and the logarithm of the probability.

[0087] Cumulative discount rewards (usually represented as G) t (2) is the weighted sum of all future rewards starting from the current time step t, and its mathematical form is shown in formula (2):

[0088]

[0089] r t+k The immediate reward obtained at time step t+k is γ, which is a discount factor (0≤γ≤1) used to adjust the importance of future rewards. The smaller γ is, the more attention is paid to short-term rewards; the larger γ is, the more attention is paid to long-term rewards. The cumulative discounted reward Gt reflects the reward from the current state s. t Starting point, the total potential future gains.

[0090] In practical applications, the gradient loss function is usually the sum of the cumulative discounted reward and the logarithm of the action probability over all time steps T. For example, the gradient loss function for T time steps in a trajectory (episode) can be expressed as formula (3):

[0091]

[0092] The goal of the gradient loss function in formula (3) is to maximize the cumulative discounted reward of the entire trajectory.

[0093] Figure 4 The flowchart illustrates the training logic of the decision control model provided in this embodiment of the invention.

[0094] like Figure 4 As shown, the training logic flow of the decision control model includes steps S410 to S440.

[0095] Step S410, Status Input: Input the operating status of the chiller system and environment.

[0096] Step S420: Optimize the decision model, based on a deep neural network model.

[0097] Step S430: Control output, output the control actions of the chiller plant system.

[0098] Step S440, Model Training and Update: Based on historical data and according to the definition of the optimization objective, the model parameters are updated using the stochastic gradient descent algorithm.

[0099] Using the above training logic, the policy function π is... θ The training process includes the following steps:

[0100] (a) Initialize the policy network: Initialize the parameters of the policy network using random weights.

[0101] (b) Data acquisition and interaction: In each training iteration, select an action based on the current policy function, execute the action, and observe the system's feedback (state and reward).

[0102] (c) Store the trajectory: Store the state, actions and reward information of each round to form a trajectory.

[0103] (d) Calculate the total reward (cumulative discount reward): Calculate the total reward for each state-action pair using the Monte Carlo method.

[0104] (e) Calculate loss and gradient: Calculate the loss for each state-action pair according to the loss function, and use the backpropagation algorithm to calculate the gradient of the loss with respect to the policy network parameters.

[0105] (f) Update the policy network: Use an optimizer (such as Adam) to update the parameters of the policy network based on the gradient.

[0106] (g) Repeated iteration: Repeat the above steps until the policy network converges or the preset number of training rounds is reached.

[0107] Through the above training process, the optimized control policy function π can be obtained. θ (a t |s t ).

[0108] The decision control model proposed in this application is simple and efficient, unlike traditional complex mechanism-based models. It only requires abstracting key state variables and control action variables based on the actual operating characteristics of the chiller plant, constructing a concise yet practical model framework. This characteristic reduces the complexity of model construction and improves the model's versatility and practicality. It adapts to various control variable inputs, exhibiting extremely high flexibility and capable of handling multiple types of control variables. Whether it's continuously changing control setpoints, such as fine-tuning of temperature, flow rate, and pressure, or discrete control decisions, such as start-up and shutdown control of chillers, chiller plants, and chiller towers, and dynamic optimization of the number of operating units, it can respond accurately and achieve optimized control. Online update capability: To cope with the dynamic changes in the chiller plant's operating environment and the challenges of different operating conditions, a stochastic policy gradient algorithm is incorporated, enabling the system to utilize historical data accumulated during operation to optimize and update the deep neural network model online in real time. This self-learning and self-adaptive capability ensures that the model always remains in optimal condition, effectively improving the continuity and accuracy of chiller plant energy efficiency optimization.

[0109] For ease of understanding, this application provides a specific embodiment that uses simplified input and output, which is not intended to limit the scope of this application.

[0110] The input to the decision control model is the environmental state of the chiller plant, specifically the wet-bulb temperature (Twb) and cooling load (CL). These two state parameters are acquired in real time by sensors and used as input to the neural network.

[0111] The output of the decision control model consists of the control actions of the chiller plant, namely the chiller setpoint temperature (Tchws_set) and the cooling tower operating frequency (t_f). These two control parameters are directly applied to the chiller plant's control system to regulate its operating status.

[0112] A simple fully connected neural network is used, consisting of two layers: Input layer: receives OBS_SIZE (set to 2, corresponding to Twb and CL) input nodes and outputs 128 nodes. Intermediate layer: a three-layer fully connected layer with 128 nodes. Output layer: receives 128 input nodes and outputs ACT_SIZE*2 (set to 4, corresponding to the mean and standard deviation of the two actions, i.e., the mean and standard deviation of Tchws_set and t_f) nodes. The parameters of the output layer represent the mean (μ) and standard deviation (σ) of each action, thus forming a normal distribution for action sampling. Transfer function: (activation function) is the ReLU function, used to increase the network's non-linearity.

[0113] The revenue function is calculated based on the chiller's operating efficiency (such as COP (Coefficient of Performance)) and energy consumption. Higher efficiency and lower energy consumption result in higher revenue. The loss function is the policy gradient loss, which is the sum of the products of the negative cumulative discount reward and the log probability of the action. The policy parameters are optimized by maximizing the cumulative discount reward.

[0114] The training process is as follows: The deep learning framework used is PyTorch; the optimizer chosen is the Adam optimizer, with a learning rate of 0.001. In each training epoch, the model selects an action based on the current state, executes the action, and obtains a new state and reward. The reward is accumulated and processed as a cumulative discounted reward. Then, the loss is calculated, backpropagation is performed, and the network parameters are updated. The training process continues until a preset stopping condition (such as the number of training epochs or a convergence criterion) is reached.

[0115] Through training based on policy gradients, the policy model can learn to adjust control parameters (Tchws_set and t_f) in real time according to the historical operating data of the chiller plant, thereby optimizing the chiller plant's operating efficiency and reducing energy consumption. In practical applications, this model can make decisions based on real-time environmental conditions, achieving intelligent control of the chiller plant.

[0116] Figure 5 A comparison chart of the effects of the decision control model provided in the embodiments of the present invention.

[0117] like Figure 5 As shown, the system can automatically output the operating frequency of the cooling tower based on the outdoor wet-bulb temperature, thereby improving the system's COP. The yellow curve represents the optimized effect, while the green curve represents the baseline effect. During the test period, energy consumption can be reduced by 3%-10% compared to the baseline condition.

[0118] Through continuous training and optimization, the model's decision-making ability will gradually improve, and the operating efficiency and energy-saving effect of the cooling plant will be significantly improved. Simultaneously, data recording and analysis can further verify the model's optimization effectiveness and practical application value.

[0119] This application not only provides a novel and efficient solution for energy efficiency optimization of chiller plants, but also significantly enhances the intelligence level of chiller plant operation through its model simplicity, broad adaptability of control variable inputs, and online update capabilities. This provides strong technical support for achieving energy-efficient utilization, reducing operating costs, and promoting sustainable development goals. The model has a simple and clear structure, does not rely on complex mechanistic models, and can be updated online in real time based on historical operating data, continuously adapting to the actual operating conditions of the chiller plant. Through this optimization process, the decision control model can achieve real-time regulation and optimization of the chiller plant, effectively improving its operating efficiency and achieving significant energy-saving and consumption-reducing effects.

[0120] Figure 6 A block diagram of a chiller plant operation control device provided for another embodiment of the present invention.

[0121] This invention provides a chiller plant operation control device 600. Please refer to [link / reference]. Figure 6 The cooling plant operation control device 600 includes: an acquisition module 610, a first acquisition module 620, a second acquisition module 630, and a control module 640.

[0122] For example, the acquisition module 610 is used to acquire state variables, wherein the state variables include cold station operating status data and environmental status data.

[0123] For example, the first obtaining module 620 is used to determine the matching probability distribution method based on the data type of the control variable when making decisions based on the state variable, and to obtain the probability distribution of the control variable that matches the probability distribution method.

[0124] For example, the second obtaining module 630 is used to sample the probability distribution of the control variable to obtain the control action corresponding to the control variable.

[0125] For example, control module 640 is used to control the operation of the chiller plant based on control actions.

[0126] For example, the data type of the control variable includes at least one of continuous, discrete, and switch types. The first obtaining module 620 is further configured to determine the probability distribution mode as a normal probability distribution mode when the data type of the control variable is continuous; determine the probability distribution mode as a categorical probability distribution mode when the data type of the control variable is discrete; and determine the probability distribution mode as a Bernoulli probability distribution mode when the data type of the control variable is switch type.

[0127] For example, the second obtaining module 630 is further configured to use preset safety rule data as sampling constraints to sample the probability distribution of the control variable and obtain the control action corresponding to the control variable.

[0128] For example, the chiller plant operation status data and environmental status data include current data, historical data and future data. The current data includes at least one of cooling load, equipment parameters, system parameters and weather. The historical data includes at least one of cooling load and weather. The future data includes at least one of cooling load and weather.

[0129] For example, the chiller plant operation control device is applied to a decision control model, which includes an input layer, an intermediate layer, and an output layer. The input layer is used to input state variables; the intermediate layer is used to make decisions based on the state variables; and the output layer is used to determine the matching probability distribution method based on the data type of the control variables, obtain the probability distribution of the control variables that matches the probability distribution method, sample the data of the probability distribution of the control variables, obtain the control actions corresponding to the control variables, and output the control actions.

[0130] For example, the chiller plant operation control device further includes: acquiring training samples, wherein the training samples include historical state variables; training a decision control model based on the training samples; wherein training the decision control model based on the training samples includes: making a decision based on the historical state variables to obtain a target control action; obtaining a loss function value based on the probability and reward of the target control action according to the gradient loss function; obtaining parameter gradient information based on the loss function value; and updating the model parameters of the decision control model based on the parameter gradient information.

[0131] For example, the reward is accumulated and processed to obtain the accumulated discounted reward, and the gradient loss function represents the sum of the product of the accumulated discounted reward and the logarithm of the probability.

[0132] For example, training a decision control model based on training samples further includes: determining the probability of the target control action corresponding to the historical state variable based on the policy function; and determining the reward of the target control action corresponding to the historical state variable based on the reward function.

[0133] It is understood that for a detailed description of the cooling plant power prediction device 500, please refer to the description of the cooling plant power prediction method above, and it will not be repeated here.

[0134] Figure 7 A block diagram of an electronic device provided for another embodiment of the present invention.

[0135] This application provides an electronic device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0136] like Figure 7 As shown, for ease of understanding, an embodiment of this application illustrates a specific electronic device 700.

[0137] Electronic device 700 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic device 700 may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0138] like Figure 7 As shown, device 700 includes a computing unit 701, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 702 or a computer program loaded from storage unit 708 into random access memory (RAM) 703. RAM 703 may also store various programs and data required for the operation of electronic device 700. The computing unit 701, ROM 702, and RAM 703 are interconnected via bus 704. Input / output (I / O) interface 705 is also connected to bus 704.

[0139] Multiple components in electronic device 700 are connected to I / O interface 705. These components include: input unit 706, such as a keyboard or mouse; output unit 707, such as various types of displays or speakers; storage unit 708, such as a disk or optical disk; and communication unit 709, such as a network interface card (NIC), modem, or wireless transceiver. Communication unit 709 allows electronic device 700 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0140] The computing unit 701 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 701 performs the various methods described above. For example, in some embodiments, any one or more of the various methods described above can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as storage unit 708. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 700 via ROM 702 and / or communication unit 709. When the computer program is loaded into RAM 703 and executed by the computing unit 701, one or more steps of any one or more of the various methods described above can be performed. Alternatively, in other embodiments, the computing unit 701 can be configured to perform any one or more of the various methods described above by any other suitable means (e.g., by means of firmware).

[0141] The present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method of any of the above embodiments.

[0142] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this invention, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0143] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0144] In the description of this invention, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this invention, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0145] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this invention and simplifying the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0146] Furthermore, the terms "first," "second," etc., used in the embodiments of this invention are for descriptive purposes only and should not be construed as indicating or implying relative importance, or implicitly specifying the number of technical features indicated in this embodiment. Therefore, features defined with terms such as "first" and "second" in the embodiments of this invention can explicitly or implicitly indicate that the embodiment includes at least one of those features. In the description of this invention, the word "multiple" means at least two or more, such as two, three, four, etc., unless otherwise explicitly specified in the embodiments.

[0147] In this invention, unless otherwise explicitly specified or limited in the embodiments, the terms "installation," "connection," "joining," and "fixing" appearing in the embodiments should be interpreted broadly. For example, a connection can be a fixed connection, a detachable connection, or an integral part; it can also be a mechanical connection, an electrical connection, etc. Of course, it can also be a direct connection, or an indirect connection through an intermediate medium, or it can be the internal communication of two components, or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific implementation.

[0148] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can mean that the first feature is in direct contact with the second feature, or that the first feature is in indirect contact with the second feature through an intermediate medium. Furthermore, "above," "over," and "on top" of the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature.

[0149] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A cold station operation control method characterized by, The method comprises: acquiring state variables, wherein the state variables comprise cold station operation state data and environment state data; when making a decision based on the state variables, determining a matching probability distribution mode based on the data type of the control variable, and obtaining a control variable probability distribution matched with the probability distribution mode; sampling data of the control variable probability distribution to obtain a control action corresponding to the control variable; controlling the cold station operation based on the control action; The method is applied to a decision control model comprising an input layer, an intermediate layer and an output layer, wherein: the input layer is configured to input the state variables; the intermediate layer is configured to make a decision based on the state variables; the output layer is configured to determine a matching probability distribution mode based on the data type of the control variable, obtain a control variable probability distribution matched with the probability distribution mode, sample data of the control variable probability distribution to obtain a control action corresponding to the control variable, and output the control action; The method further comprises: acquiring training samples, wherein the training samples comprise historical state variables; training the decision control model based on the training samples; wherein the training of the decision control model based on the training samples comprises: making a decision based on the historical state variables to obtain a target control action; obtaining a loss function value based on the probability of the target control action and the reward of the target control action according to a gradient loss function; obtaining parameter gradient information based on the loss function value; updating the model parameters of the decision control model based on the parameter gradient information.

2. The method of claim 1, wherein, The data type of the control variable comprises at least one of continuous type, discrete type and on-off type, and the determination of the matching probability distribution mode based on the data type of the control variable comprises: when the data type of the control variable is continuous, determining that the probability distribution mode is a normal probability distribution mode; when the data type of the control variable is discrete, determining that the probability distribution mode is a category probability distribution mode; when the data type of the control variable is on-off, determining that the probability distribution mode is a Bernoulli probability distribution mode.

3. The method of claim 1, wherein, The sampling of the control variable probability distribution to obtain a control action corresponding to the control variable comprises: sampling data of the control variable probability distribution to obtain a control action corresponding to the control variable with a preset safety rule data as a sampling constraint.

4. The method of claim 1, wherein, The cold station operation state data and the environment state data comprise current data, historical data and future data, the current data comprises at least one of refrigeration load, equipment parameter, system parameter and weather, the historical data comprises at least one of refrigeration load and weather, and the future data comprises at least one of refrigeration load and weather.

5. The method of claim 1, wherein, The reward is obtained after accumulation and processing to obtain a cumulative discounted reward, and the gradient loss function represents the sum of the product of the cumulative discounted reward and the logarithm of the probability.

6. The method according to claim 1 or 5, characterized in that, The training of the decision control model based on the training samples further comprises: determining the probability of the target control action corresponding to the historical state variables based on a policy function; Based on the reward function, a reward of the target control action corresponding to the historical state variable is determined.

7. A cold station operation control device characterized by comprising: The device implements the method of any one of claims 1-6, and the device comprises: An acquisition module is configured to acquire state variables, wherein the state variables comprise cold station operation state data and environment state data; A first obtaining module is configured to, when making a decision based on the state variables, determine a matched probability distribution mode based on a data type of a control variable, and obtain a control variable probability distribution matched with the probability distribution mode; A second obtaining module is configured to sample data of the control variable probability distribution to obtain a control action corresponding to the control variable; A control module is configured to control cold station operation based on the control action. 8.An electronic device comprising a memory and a processor, the memory storing a computer program, wherein, The processor executes the computer program to implement the steps of the method of any one of claims 1-6.

9. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the method of any one of claims 1-6.

Images