Water chilling unit load distribution method and device

Through the combination of singular spectrum analysis and reinforcement learning model, the problems of high complexity and low flexibility of the load distribution method of chiller units are solved, and more efficient load distribution and energy efficiency improvement are achieved.

CN119022413AActive Publication Date: 2024-11-26GUANGDONG UNIV OF TECH

Patent Information

Application Number
CN202411286150.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-13
Publication Date
2024-11-26
Estimated Expiration
2044-09-13

AI Technical Summary

Technical Problem

The traditional chiller load distribution method is based on metaheuristic algorithm, which leads to an exponential increase in overall complexity, resulting in low load distribution flexibility and lag problems.

Method used

The total cooling load sequence is decomposed and reconstructed by using a singular spectrum analysis algorithm. Combined with the preset load prediction model and reinforcement learning model, the load rate of the chiller unit is output, and the flexibility and energy efficiency of load distribution are improved.

Benefits of technology

Through the combination of singular spectrum analysis and reinforcement learning model, the flexibility and energy efficiency of chiller load distribution are improved, energy consumption is reduced, and the problems of high complexity and hysteresis of traditional methods are solved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119022413A_ABST
    Figure CN119022413A_ABST
Patent Text Reader

Abstract

The invention discloses a water chilling unit load distribution method and device, and relates to the technical field of water chilling units, and the method comprises the steps: obtaining the target working condition data of a plurality of water chilling units at a plurality of operation moments related to the current operation moment; corresponding cooling loads are calculated according to all the target working condition data, and a total cooling load sequence is constructed according to all the operation moments on the basis of all the cooling loads; decomposing and reconstructing the total cooling load sequence by adopting a singular spectrum analysis algorithm, and determining a reconstructed total cooling load sequence and a noise component; performing load prediction according to the reconstructed total cooling load sequence through a preset total cooling load prediction model, and outputting a predicted total cooling load in combination with the noise component; and outputting the load rate of each water chilling unit through a preset load distribution control reinforcement learning model based on the target working condition data and the cooling load at the current operation moment and the predicted total cooling load. In the whole load distribution process of the water chilling unit, the load distribution flexibility of the water chilling unit can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of water chillers, and in particular to a method and device for distributing loads of water chillers. Background Art

[0002] According to data from the United Nations Environment Programme, buildings account for about 40% of the world's total energy consumption, of which heating, cooling and ventilation systems (HVAC) account for about 50% of the total energy consumption of buildings. As the main cooling equipment in the HVAC system, the energy consumption of chillers usually exceeds half of the total energy consumption of the HVAC system. Therefore, optimizing the load distribution of chillers is of great significance for energy saving.

[0003] Traditional chiller load allocation methods are mainly based on meta-heuristic algorithms. In order to ensure that the results do not fall into local optimality, additional mechanisms are usually needed to ensure the effectiveness of global search, which leads to an exponential increase in overall complexity. When environmental conditions change, there is an obvious lag problem from the issuance of control instructions to the arrival of chilled water at the terminal equipment and the realization of temperature regulation, resulting in low flexibility in load allocation of chillers. Summary of the invention

[0004] The present invention provides a method and device for load distribution of a chiller, which solves the technical problem that the traditional load distribution method of a chiller is mainly based on a meta-heuristic algorithm, which usually requires the design of an additional mechanism to ensure the effectiveness of the global search, thereby causing the overall complexity to grow exponentially and resulting in low flexibility in load distribution of the chiller.

[0005] A first aspect of the present invention provides a chiller load distribution method, comprising:

[0006] Respectively obtain target operating condition data of multiple chillers at multiple operating times associated with the current operating time;

[0007] Calculating the corresponding cooling load according to each of the target operating condition data, and constructing a total cooling load sequence based on each of the cooling loads according to each of the operating moments;

[0008] Decomposing and reconstructing the total cooling load sequence by using a singular spectrum analysis algorithm to determine a reconstructed total cooling load sequence and a noise component;

[0009] Performing load forecasting according to the reconstructed total cooling load sequence by using a preset total cooling load forecasting model, and outputting a forecasted total cooling load in combination with the noise component;

[0010] Based on the target operating condition data and cooling load at the current operating moment and the predicted total cooling load, the load rate of each of the chillers is output through a preset load distribution control reinforcement learning model.

[0011] Optionally, before the step of respectively acquiring target operating condition data of multiple chillers at multiple operating times associated with the current operating time, the step includes:

[0012] Collecting initial operating condition data of multiple chillers at multiple operating times within a preset time period at the current operating time;

[0013] Data elimination and data completion are performed on each of the initial operating condition data to obtain corresponding target operating condition data.

[0014] Optionally, the step of calculating the corresponding cooling load according to each target operating condition data, and constructing a total cooling load sequence based on each cooling load at each operating time includes:

[0015] Extracting a target chilled water supply temperature, a target chilled water return temperature and a target chilled water flow rate from each of the target operating condition data respectively;

[0016] Performing a difference calculation on each of the target chilled water return temperatures and the corresponding target chilled water supply temperatures to obtain a plurality of corresponding chilled water temperature differences;

[0017] Performing multiplication operation on each of the chilled water temperature differences and the associated target chilled water flow rate and water specific heat, outputting the cooling load of each of the chillers at each of the operating moments;

[0018] According to each of the operating times, the cooling loads of each of the chillers are summed to determine the corresponding total cooling load;

[0019] A total cooling load sequence is constructed based on each of the total cooling loads according to the corresponding operating times.

[0020] Optionally, the step of using a singular spectrum analysis algorithm to decompose and reconstruct the total cooling load sequence to determine the reconstructed total cooling load sequence and noise components includes:

[0021] Projecting the total cooling load sequence into a multi-dimensional lag vector, and using the multi-dimensional lag vector to form a trajectory matrix;

[0022] Performing singular value decomposition based on the trajectory matrix to obtain multiple singular components;

[0023] Calculating a component contribution value of each of the singular components according to the singular value of each of the singular components;

[0024] Select singular components whose component contribution values ​​are greater than or equal to a preset contribution threshold and reconstruct them into a new trajectory matrix, perform diagonal averaging on the new trajectory matrix, and output a reconstructed total cooling load sequence;

[0025] The singular component whose component contribution value is less than a preset contribution threshold is selected as the noise component.

[0026] Optionally, the step of outputting the load rate of each of the chillers through a preset load distribution control reinforcement learning model based on the target operating condition data at the current operating moment, the cooling load and the predicted total cooling load comprises:

[0027] Extracting a target chilled water return temperature, a target cooling water return temperature and a target cooling water inlet temperature from the target operating condition data at the current operating time;

[0028] Using the target chilled water return temperature and the target cooling water inlet temperature associated with the current operating time and the predicted total cooling load as state space parameters;

[0029] The state space parameters are input into a preset load distribution control reinforcement learning model for iterative calculation, and a reward value is calculated based on the target chilled water return temperature and the target cooling water return temperature corresponding to the current operating moment and the predicted total cooling load through a preset reward function. With the goal of maximizing the reward value, the load rate of each chiller is output.

[0030] Optionally, the reward function includes:

[0031]

[0032] in,

[0033] ;

[0034]

[0035] In the formula, is the reward value, is the average performance coefficient, For The exponential function with base , For the Chillers, is the load rate, For the The penalty value when a chiller exceeds the specified load rate range. is the first balance coefficient, is the second balance coefficient, is the third balance coefficient, is the total cooling load, is the actual power consumption of a single chiller, is the cooling load, is the performance coefficient, is the energy efficiency coefficient, is the chilled water return temperature, is the cooling water return temperature.

[0036] A second aspect of the present invention provides a chiller load distribution device, comprising:

[0037] The operating condition acquisition module is used to respectively obtain target operating condition data of multiple chillers at multiple operating times associated with the current operating time;

[0038] A sequence construction module, used to calculate the corresponding cooling load according to each target operating condition data, and to construct a total cooling load sequence according to each operating time based on each cooling load;

[0039] A sequence reconstruction module, used to decompose and reconstruct the total cooling load sequence using a singular spectrum analysis algorithm to determine a reconstructed total cooling load sequence and a noise component;

[0040] A load prediction module, configured to perform load prediction according to the reconstructed total cooling load sequence by using a preset total cooling load prediction model, and output a predicted total cooling load in combination with the noise component;

[0041] The load distribution module is used to output the load rate of each of the chillers through a preset load distribution control reinforcement learning model based on the target operating condition data and cooling load at the current operating time and the predicted total cooling load.

[0042] A third aspect of the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the chiller load distribution method as described in any one of the above items.

[0043] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed, implements the chiller load distribution method as described in any one of the above items.

[0044] A fifth aspect of the present invention provides a computer program product, comprising a computer program / instruction, which, when executed by a processor, implements the chiller load distribution method as described in any one of the above items.

[0045] It can be seen from the above technical solutions that the present invention has the following advantages:

[0046] The above technical scheme of the present invention provides a method for distributing loads of chillers, including: respectively obtaining target operating condition data of multiple operating moments associated with multiple chillers at the current operating moment; calculating the corresponding cooling load according to each target operating condition data, and constructing a total cooling load sequence according to each operating moment based on each cooling load; using a singular spectrum analysis algorithm to decompose and reconstruct the total cooling load sequence, and determine the reconstructed total cooling load sequence and noise components; using a preset total cooling load prediction model to perform load prediction based on the reconstructed total cooling load sequence, and outputting the predicted total cooling load in combination with the noise component; based on the target operating condition data at the current operating moment, the cooling load and the predicted total cooling load, outputting the load rate of each chiller through a preset load distribution control reinforcement learning model. In the load distribution process of the entire chiller, the fluctuation characteristics of the total cooling load are analyzed based on the singular spectrum analysis algorithm. The total cooling load prediction model can better predict future load changes, and this change is fed back to the load distribution control reinforcement learning model to use the chiller load rate distribution as the control action to achieve optimal control of multiple chillers, which can improve the load distribution flexibility of the chiller, which is beneficial to improving the energy efficiency of the chiller system and reducing energy consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0048] Figure 1 A flow chart of the steps of a chiller load distribution method provided by an embodiment of the present invention;

[0049] Figure 2 A structural block diagram of a chiller load distribution device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] An embodiment of the present invention provides a method and device for distributing load to a chiller, which is used to solve the technical problem that the traditional method for distributing load to a chiller is mainly based on a metaheuristic algorithm, which usually requires the design of an additional mechanism to ensure the effectiveness of the global search, thereby causing the overall complexity to grow exponentially and resulting in low flexibility in load distribution of the chiller.

[0051] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] See also Figure 1 , Figure 1 A flow chart of the steps of a chiller load distribution method provided in an embodiment of the present invention.

[0053] The present invention provides a method for distributing load of a chiller, comprising:

[0054] Step 101: respectively obtain target operating condition data of multiple chillers at multiple operating times associated with the current operating time.

[0055] Before step 101, the following steps are included:

[0056] Collecting initial operating condition data of multiple chillers at multiple operating times within a preset time period at the current operating time;

[0057] The initial working condition data are processed by data elimination and data completion to obtain the corresponding target working condition data.

[0058] It should be noted that the initial operating condition data refers to data that can reflect the operating conditions of the chiller, including but not limited to the chilled water supply temperature, chilled water return temperature, chilled water flow rate, cooling water return temperature and cooling water inlet temperature of the chiller, while the target operating condition data refers to the data after data preprocessing such as data elimination and data completion on the initial operating condition data, and correspondingly includes the target chilled water supply temperature, target chilled water return temperature, target chilled water flow rate, target cooling water return temperature and target cooling water inlet temperature.

[0059] When multiple units need to be load coordinated and distributed, the initial operating data of each unit at the current operating time and multiple operating times within a preset time before the current operating time are obtained to establish an operating data set, and the data set is preprocessed to ensure the accuracy of the data. Data preprocessing includes data removal and data completion. Among them, data removal refers to the removal of logical error data and outlier data, and the removal of logical error data refers to invalid data that does not conform to physical laws or business logic, such as: the situation where the chilled water supply temperature is higher than the chilled water return temperature due to sensor failure, data recording error or system configuration error, and outlier data is also called abnormal value, which refers to the value of the operating data set that deviates significantly from other data points. The reason for this phenomenon may be due to sensor noise. Abnormal situations such as noise, network delay or data transmission error occur, such as: the water temperature has an extreme sudden increase or decrease in a short period of time, which is much higher or lower than the historical data of the same period, and there is no reasonable explanation or background factors; Among them, the data completion processing can include linear interpolation completion and spline interpolation completion. In the specific implementation, when there is missing data of some operating times of a chiller in the operating condition data set, and the data of the data type to which the missing data belongs is in an operating stage with relatively stable data changes, the linear interpolation method can be used for completion. This method is simple and effective and is suitable for situations where data missing is not serious and changes are gentle. When there is a nonlinear trend in a certain type of data of a chiller in the operating condition data set, the spline interpolation method can be used for completion. This method can better fit the nonlinear data changes.

[0060] Step 102: Calculate the corresponding cooling load according to each target operating condition data, and construct a total cooling load sequence based on each cooling load at each operating time.

[0061] It should be noted that the total cooling load sequence refers to a sequence formed by arranging multiple total cooling loads in chronological order.

[0062] Step 102 includes the following sub-steps:

[0063] S11. Extracting a target chilled water supply temperature, a target chilled water return temperature and a target chilled water flow rate from each target operating condition data respectively.

[0064] S12. Perform a difference calculation on each target chilled water return temperature and the corresponding target chilled water supply temperature to obtain a plurality of corresponding chilled water temperature differences.

[0065] S13, performing multiplication operation on each chilled water temperature difference value and the associated target chilled water flow rate and water specific heat, and outputting the cooling load of each chiller at each operating time.

[0066] It should be noted that the calculation process of cooling load specifically includes:

[0067]

[0068] In the formula, For the Chillers, is the cooling load, is the specific heat of water, is the chilled water flow rate, is the chilled water return temperature, The chilled water supply temperature.

[0069] S14. According to each operating time, the cooling load of each chiller is summed up to determine the corresponding total cooling load.

[0070] It should be noted that the total cooling load The calculation process specifically includes:

[0071]

[0072] S15. Construct a total cooling load sequence based on each total cooling load according to the corresponding operating time.

[0073] It should be noted that, according to the target operating data of each unit at each operating moment, according to the above-mentioned calculation process of calculating the cooling load, after determining the cooling load of each unit at each operating moment, taking each operating moment as a unit, the cooling load of each unit corresponding to each operating moment is calculated according to the above-mentioned calculation process of calculating the total cooling load, to obtain the total cooling load corresponding to each operating moment, and sort the total cooling loads according to their corresponding operating moments to construct a total cooling load sequence.

[0074] Step 103: Decompose and reconstruct the total cooling load sequence using a singular spectrum analysis algorithm to determine a reconstructed total cooling load sequence and a noise component.

[0075] It should be noted that the singular spectrum analysis (SSA) algorithm is a time series analysis and forecasting technique, which is usually regarded as a method to identify and extract oscillatory components from the original sequence, thereby decomposing the original sequence into the sum of interpretable components, including trend, periodic, quasi-periodic components or noise information.

[0076] Step 103 includes the following sub-steps:

[0077] S21. Project the total cooling load sequence into a multi-dimensional lag vector, and use the multi-dimensional lag vector to form a trajectory matrix.

[0078] S22. Perform singular value decomposition based on the trajectory matrix to obtain multiple singular components.

[0079] It should be noted that the singular value decomposition of the trajectory matrix can be understood as obtaining multiple singular components composed of the product of left singular vectors, right singular vectors and singular values, and different singular components can be used to characterize noise components, trend components and possible periodic components.

[0080] S23. Calculate the component contribution value of each singular component according to the singular value of each singular component.

[0081] It should be noted that the contribution of each singular component of the total cooling load sequence can be measured by the following formula:

[0082]

[0083] In the formula, is the component contribution value, For the A strange quantity, is the number of singular components, are singular values, For the The singular values ​​of the singular components.

[0084] S24, selecting singular components whose component contribution values ​​are greater than or equal to a preset contribution threshold and reconstructing them into a new trajectory matrix, performing diagonal averaging on the new trajectory matrix, and outputting a reconstructed total cooling load sequence.

[0085] S25. Select a singular component whose component contribution value is less than a preset contribution threshold as a noise component.

[0086] It should be noted that the singular components are sorted from large to small according to their component contribution values, and the singular components whose component contribution values ​​are less than the preset contribution threshold are taken as noise components. The singular components whose component contribution values ​​are greater than or equal to the preset contribution threshold are selected to characterize the periodic data and trend data to reconstruct a new trajectory matrix. Diagonal averaging is performed based on the new trajectory matrix to restore it to a one-dimensional time series component to obtain the reconstructed total cooling load sequence, which can eliminate the influence of noise and is beneficial to improving the prediction accuracy.

[0087] Step 104: perform load prediction based on the reconstructed total cooling load sequence using a preset total cooling load prediction model, and output the predicted total cooling load in combination with the noise component.

[0088] It should be noted that, by training and testing the existing time series prediction model, such as the Transformer model, to obtain the preset total cooling load prediction model, in specific implementation, according to the similar process of the above steps 101 to 103, firstly, a plurality of historical reconstructed total cooling load sequences corresponding to the current operating moments are obtained, and the maximum and minimum values ​​of each historical reconstructed total cooling load sequence are standardized so that the standardized value is between [0, 1]. The standardization process is as follows:

[0089]

[0090] In the formula, is the original data, is the original data after standardization, is the minimum value of the sequence of the original data, is the maximum value of the sequence of the original data;

[0091] The standardized historical reconstructed total cooling load series are divided into training sets and test sets. The training set is used to train the time series prediction model, and the test set is used to verify the accuracy of the time series prediction model, thereby obtaining the total cooling load prediction model. The evaluation indicators of prediction accuracy can be root mean square error (RMSE), mean absolute error (MAE) and mean absolute percentage error (MAPE), including:

[0092]

[0093] In the formula, For the Running time, is the number of running times, is the true value, is the predicted value;

[0094] The reconstructed total cooling load sequence corresponding to the current operating time is input into the total cooling load prediction model for load prediction. After the total cooling load to be adjusted is output, the noise component data decomposed by the singular spectrum analysis algorithm is added back to the total cooling load to be adjusted, thereby obtaining the predicted total cooling load to achieve data authenticity.

[0095] Step 105: Based on the target operating condition data at the current operating time, the cooling load and the predicted total cooling load, the load rate of each chiller is output through a preset load distribution control reinforcement learning model.

[0096] Step 105 includes the following sub-steps:

[0097] S31, extracting a target chilled water return temperature, a target cooling water return temperature, and a target cooling water inlet temperature from the target operating condition data at the current operating time;

[0098] S32, using the target chilled water return temperature and the target cooling water inlet temperature associated with the current operating time and the predicted total cooling load as state space parameters;

[0099] S33. Input the state space parameters into the preset load distribution control reinforcement learning model for iterative calculation, and calculate the reward value based on the target chilled water return temperature and the target cooling water return temperature corresponding to the current operating time and the predicted total cooling load through the preset reward function. With the goal of maximizing the reward value, output the load rate of each chiller.

[0100] It should be noted that in this embodiment, the optimal chiller load allocation problem is transformed into a Markov (MDP) decision process based on deep reinforcement learning. The MDP model of a chiller system composed of multiple chillers can be expressed as a quaternion, where is the state parameter related to the chiller system, is the control variable of the chiller system, For chillers in state Make the following move Rewards received, is the state transition probability;

[0101] In the load distribution control reinforcement learning model, the design of the state space needs to include a complete description of the problem so that the agent can make the best decision based on the state change. At the same time, the dimension of the state space should be simplified as much as possible, but the key information of the problem should not be lost. For the chiller system, the cooling water inlet temperature can reflect the changes in outdoor meteorological parameters, and the chilled water return temperature can reflect the changes in indoor load to a certain extent. The cooling load can greatly affect the operating efficiency of the chiller and is not controlled by the system. It is one of the most important state parameters of the cold source system. Therefore, the state space parameters are specifically expressed as follows:

[0102]

[0103] In the formula, To predict the total cooling load, is the chilled water return temperature, is the cooling water inlet temperature;

[0104] At the same time, in the load distribution control reinforcement learning model, the design of the action space needs to take into account the feasibility of the problem and the operation, and the action space should be ensured to have sufficient accuracy so that the intelligent agent can find the optimal solution in it. The discretization accuracy of the action should take into account the exploration cost. To achieve higher accuracy, a larger action space is required, and the intelligent agent also needs longer time to converge. Therefore, the action space includes: in each control time step, the intelligent agent uses the load rate PLR ​​of the chiller as the control action, and the lower limit of the load rate is set to 30%, because when the chiller runs at a low load rate, it will cause unstable operation of the equipment and even cause equipment failure. The upper limit of the load rate is 100%, and the discrete step length is set according to the task requirements;

[0105] In addition, the reward function of the load distribution control reinforcement learning model is designed to include:

[0106]

[0107] in,

[0108] ;

[0109] Where; is the reward value; is the average coefficient of performance, which represents the average COP of multiple chillers; For An exponential function with a base value of . For the Chillers; is the load rate; For the The penalty value for a chiller when it exceeds the specified load rate range. The higher the degree of excess, the greater the penalty. If it is within the specified range, a positive reward will be given. is the first balance coefficient, is the second balance coefficient, is the third balance coefficient, which is mainly used to balance the reward obtained by the agent between the operating efficiency of the chiller and the constraints;

[0110] In the specific implementation, the intelligent agent calculates the reward function based on different state space values ​​through iterative calculations with the environment, takes the highest reward value obtained as the optimal control point in this state and outputs the corresponding control action, thereby issuing the optimal control command to each chiller, including: creating two neural networks, one for estimating the Q value and the other as the target network; initializing the experience replay pool, setting the replay pool capacity, and defining the number of samples drawn from the pool during each training; obtaining the initial state of the system, and setting the chilled water return temperature , Cooling water inlet temperature and total cooling load as input states; through Strategy selection action ; Execute actions and observe the results, and select the action Applied to the environment, the system responds and moves to a new state , collect the immediate reward for this action and the new state of the system ; Current experience Store in the experience replay pool; regularly randomly extract a batch of samples from the experience replay pool for training; update the weights of the Q network through the back propagation algorithm (such as gradient descent) to minimize the loss function; regularly copy the weights of the Q network to the target Q network; send the control action as the optimal control instruction to the chiller;

[0111] Based on the above steps, in the actual iterative calculation process of the load distribution control reinforcement learning model, after determining the predicted cooling load allocated to each chiller based on the predicted total cooling load, each predicted cooling load is used to further determine the load rate of each chiller. In specific implementation, the load rate can be calculated by the ratio of the cooling load to the rated cooling load. In this process:

[0112] The performance coefficient of each chiller is calculated by using the target chilled water return temperature and the associated target cooling water return temperature and the predicted cooling load at the current operating time. The performance coefficient of each chiller can be calculated by establishing an energy efficiency model of the chiller. The energy efficiency model of the chiller is established with the cooling load, chilled water return temperature and cooling water return temperature as input data and the performance coefficient as output data. The functional relationship between the chilled water return temperature, cooling water return temperature, cooling load and performance coefficient is determined by training the energy efficiency model, and the energy efficiency coefficient of the functional relationship is calculated, so as to determine the energy efficiency model as follows:

[0113]

[0114] In the formula, For the Chillers; is the performance coefficient; is the energy efficiency coefficient; is the cooling load; is the chilled water return temperature; is the cooling water return temperature;

[0115] Next, the predicted cooling loads are calculated by ratio with the corresponding performance coefficient to obtain the actual power consumption of each chiller, as shown in the following process:

[0116]

[0117] In the formula, For the Chillers; is the actual power consumption of a single chiller; is the cooling load; is the coefficient of performance;

[0118] Subsequently, the average performance coefficient is determined by performing a ratio operation between the predicted total cooling load and the sum of the actual power consumption, and then the reward value is calculated based on the average performance coefficient and the load rate through the above-mentioned preset reward function, and the load rate of each chiller is output as a control action with the goal of maximizing the reward value; wherein, the calculation process of the average performance coefficient can be referred to as follows:

[0119]

[0120] In the formula, is the total cooling load; For the Chillers; is the actual power consumption of a single chiller; is the average performance coefficient.

[0121] In this embodiment, during the load distribution process of the entire chiller, the fluctuation characteristics of the total cooling load are analyzed based on the singular spectrum analysis algorithm. The total cooling load prediction model can better predict future load changes, and this change is fed back to the load distribution control reinforcement learning model to use the chiller load rate distribution as the control action to achieve optimal control of multiple chillers, which can improve the load distribution flexibility of the chiller, which is beneficial to improving the energy efficiency of the chiller system and reducing energy consumption.

[0122] See also Figure 2 , Figure 2 A structural block diagram of a chiller load distribution device provided in an embodiment of the present invention.

[0123] The present invention provides a chiller load distribution device, comprising:

[0124] The operating condition acquisition module 201 is used to respectively obtain target operating condition data of multiple chillers at multiple operating times associated with the current operating time;

[0125] A sequence building module 202 is used to calculate the corresponding cooling load according to each target operating condition data, and to build a total cooling load sequence according to each operating time based on each cooling load;

[0126] A sequence reconstruction module 203 is used to decompose and reconstruct the total cooling load sequence using a singular spectrum analysis algorithm to determine a reconstructed total cooling load sequence and a noise component;

[0127] A load prediction module 204 is used to perform load prediction according to the reconstructed total cooling load sequence by using a preset total cooling load prediction model, and output the predicted total cooling load in combination with the noise component;

[0128] The load distribution module 205 is used to output the load rate of each chiller through a preset load distribution control reinforcement learning model based on the target operating condition data at the current operating time, the cooling load and the predicted total cooling load.

[0129] Furthermore, the working condition collection module 201 is also used for:

[0130] Collecting initial operating condition data of multiple chillers at multiple operating times within a preset time period at the current operating time;

[0131] The initial working condition data are processed by data elimination and data completion to obtain the corresponding target working condition data.

[0132] Furthermore, the sequence construction module 202 is specifically used for:

[0133] Extracting target chilled water supply temperature, target chilled water return temperature and target chilled water flow rate from each target operating condition data respectively;

[0134] Performing a difference calculation on each target chilled water return temperature and the corresponding target chilled water supply temperature to obtain a plurality of corresponding chilled water temperature differences;

[0135] The cooling load of each chiller at each operating time is output by multiplying the temperature difference of each chilled water with the associated target chilled water flow rate and water specific heat;

[0136] According to each operating time, the cooling load of each chiller is summed up to determine the corresponding total cooling load;

[0137] A total cooling load sequence is constructed based on each total cooling load according to the corresponding operating time.

[0138] Furthermore, the sequence reconstruction module 203 is specifically used for:

[0139] The total cooling load sequence is projected into a multi-dimensional lag vector, and the multi-dimensional lag vector is used to form a trajectory matrix;

[0140] Perform singular value decomposition based on the trajectory matrix to obtain multiple singular components;

[0141] Calculate the component contribution value of each singular component according to the singular value of each singular component;

[0142] Select singular components whose component contribution values ​​are greater than or equal to a preset contribution threshold and reconstruct them into a new trajectory matrix, perform diagonal averaging on the new trajectory matrix, and output the reconstructed total cooling load sequence;

[0143] The singular component whose component contribution value is less than a preset contribution threshold is selected as the noise component.

[0144] Furthermore, the load distribution module 205 is specifically used for:

[0145] Extracting the target chilled water return temperature, the target cooling water return temperature and the target cooling water inlet temperature from the target operating condition data at the current operating time;

[0146] The target chilled water return temperature, target cooling water inlet temperature and predicted total cooling load associated with the current operating time are used as state space parameters;

[0147] The state space parameters are input into the preset load distribution control reinforcement learning model for iterative calculation, and the reward value is calculated based on the target chilled water return temperature and the target cooling water return temperature corresponding to the current operating time and the predicted total cooling load through the preset reward function. With the goal of maximizing the reward value, the load rate of each chiller is output.

[0148] Furthermore, the reward function includes:

[0149]

[0150] in, ;

[0151] ;

[0152] In the formula, is the reward value, is the average performance coefficient, For The exponential function with base , For the Chillers, is the load rate, For the The penalty value when a chiller exceeds the specified load rate range. is the first balance coefficient, is the second balance coefficient, is the third balance coefficient, is the total cooling load, is the actual power consumption of a single chiller, is the cooling load, is the performance coefficient, is the energy efficiency coefficient, is the chilled water return temperature, is the cooling water return temperature.

[0153] An embodiment of the present invention further provides a computer device, including a memory and a processor, wherein a computer program is stored in the memory; when the computer program is executed by the processor, the processor executes the steps of the chiller load distribution method of any of the above embodiments.

[0154] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program / instruction is stored. When the computer program / instruction is executed by a processor, the steps of the chiller load distribution method as in any of the above embodiments are implemented.

[0155] An embodiment of the present invention further provides a computer program product, including a computer program / instruction, which implements the steps of the chiller load distribution method as described in any of the above embodiments when the computer program / instruction is executed by a processor.

[0156] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and modules can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0157] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0158] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0159] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0160] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0161] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for distributing load of a chiller, characterized in that: include: Respectively obtain target operating condition data of multiple chillers at multiple operating times associated with the current operating time; Calculating the corresponding cooling load according to each of the target operating condition data, and constructing a total cooling load sequence based on each of the cooling loads according to each of the operating moments; Decomposing and reconstructing the total cooling load sequence by using a singular spectrum analysis algorithm to determine a reconstructed total cooling load sequence and a noise component; Performing load forecasting according to the reconstructed total cooling load sequence by using a preset total cooling load forecasting model, and outputting a forecasted total cooling load in combination with the noise component; Based on the target operating condition data and cooling load at the current operating moment and the predicted total cooling load, the load rate of each of the chillers is output through a preset load distribution control reinforcement learning model.

2. The chiller load distribution method according to claim 1, characterized in that: Before respectively acquiring target operating condition data of multiple chillers at multiple operating times associated with the current operating time, the method includes: Collecting initial operating condition data of multiple chillers at multiple operating times within a preset time period at the current operating time; Data elimination and data completion are performed on each of the initial operating condition data to obtain corresponding target operating condition data.

3. The chiller load distribution method according to claim 1, characterized in that: The step of calculating the corresponding cooling load according to each of the target operating condition data, and constructing a total cooling load sequence based on each of the cooling loads according to each of the operating moments, comprises: Extracting a target chilled water supply temperature, a target chilled water return temperature and a target chilled water flow rate from each of the target operating condition data respectively; Performing a difference calculation on each of the target chilled water return temperatures and the corresponding target chilled water supply temperatures to obtain a plurality of corresponding chilled water temperature differences; Performing multiplication operation on each of the chilled water temperature differences and the associated target chilled water flow rate and water specific heat, outputting the cooling load of each of the chillers at each of the operating moments; According to each of the operating times, the cooling loads of each of the chillers are summed to determine the corresponding total cooling load; A total cooling load sequence is constructed based on each of the total cooling loads according to the corresponding operating times.

4. The chiller load distribution method according to claim 1, characterized in that: The step of using a singular spectrum analysis algorithm to decompose and reconstruct the total cooling load sequence to determine the reconstructed total cooling load sequence and noise components includes: Projecting the total cooling load sequence into a multi-dimensional lag vector, and using the multi-dimensional lag vector to form a trajectory matrix; Performing singular value decomposition based on the trajectory matrix to obtain multiple singular components; Calculating a component contribution value of each of the singular components according to the singular value of each of the singular components; Select singular components whose component contribution values ​​are greater than or equal to a preset contribution threshold and reconstruct them into a new trajectory matrix, perform diagonal averaging on the new trajectory matrix, and output a reconstructed total cooling load sequence; The singular component whose component contribution value is less than a preset contribution threshold is selected as the noise component.

5. The chiller load distribution method according to claim 1, characterized in that: The step of outputting the load rate of each of the chillers through a preset load distribution control reinforcement learning model based on the target operating condition data at the current operating time, the cooling load and the predicted total cooling load comprises: Extracting a target chilled water return temperature, a target cooling water return temperature and a target cooling water inlet temperature from the target operating condition data at the current operating time; Using the target chilled water return temperature and the target cooling water inlet temperature associated with the current operating time and the predicted total cooling load as state space parameters; The state space parameters are input into a preset load distribution control reinforcement learning model for iterative calculation, and a reward value is calculated based on the target chilled water return temperature and the target cooling water return temperature corresponding to the current operating moment and the predicted total cooling load through a preset reward function. With the goal of maximizing the reward value, the load rate of each chiller is output.

6. The chiller load distribution method according to claim 5, characterized in that: The reward function includes: in, ; In the formula, is the reward value, is the average performance coefficient, For The exponential function with base , For the Chillers, is the load rate, For the The penalty value when a chiller exceeds the specified load rate range. is the first balance coefficient, is the second balance coefficient, is the third balance coefficient, is the total cooling load, is the actual power consumption of a single chiller, is the cooling load, is the performance coefficient, is the energy efficiency coefficient, is the chilled water return temperature, is the cooling water return temperature.

7. A chiller load distribution device, characterized in that: include: The operating condition acquisition module is used to respectively obtain target operating condition data of multiple chillers at multiple operating times associated with the current operating time; A sequence construction module, used to calculate the corresponding cooling load according to each target operating condition data, and to construct a total cooling load sequence according to each operating time based on each cooling load; A sequence reconstruction module, used to decompose and reconstruct the total cooling load sequence using a singular spectrum analysis algorithm to determine a reconstructed total cooling load sequence and a noise component; A load prediction module, configured to perform load prediction according to the reconstructed total cooling load sequence by using a preset total cooling load prediction model, and output a predicted total cooling load in combination with the noise component; The load distribution module is used to output the load rate of each of the chillers through a preset load distribution control reinforcement learning model based on the target operating condition data and cooling load at the current operating time and the predicted total cooling load.

8. A computer device, characterized in that: It comprises a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the steps of the chiller load distribution method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by the processor, the steps of the chiller load distribution method according to any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by the processor, the steps of the chiller load distribution method according to any one of claims 1 to 6 are implemented.

Citation Information

Patent Citations

  • Intelligent park air conditioner load regulation and control method and system based on deep reinforcement learning

    CN114322208A

  • Water chilling unit group control optimization method based on load prediction

    CN116398994A

  • Control method of building cold source system

    CN117366799A

  • Energy-saving optimization method and system for industrial circulating water system

    CN118469103A

  • Air conditioning system and control method therefor

    WO2024156177A1

Cited By

  • Water chilling unit multi-unit load allocation system and method

    CN121557582A