Reformed energy system optimization control method and system based on incremental learning

Abstract

An incremental learning-based optimization control method and system for modified energy systems includes: acquiring on-site data from the original energy system and processing it using multiple data processing methods to obtain an original energy system dataset; training a pre-selected data model using the normalized original energy system dataset, and selecting a system energy consumption and temperature model by comparing the fitting effects of different types of data models; for the modified new energy system, using an incremental learning algorithm to form a new energy system energy consumption and temperature model applicable to both old and new knowledge; and using the new energy system energy consumption and temperature model, according to different optimization objectives, combining appropriate optimization algorithms to solve for the optimal operating mode, operating state, and optimal system control parameters under different operating conditions to achieve energy saving. This invention utilizes the forward transfer of knowledge from the old energy system to quickly form a new energy system model, avoiding the problem of catastrophic forgetting.

Description

Technical Field

[0001] This invention relates to the field of energy-saving technology for energy systems, and specifically to a method and system for optimizing and controlling energy systems based on incremental learning. Background Technology

[0002] With continuous economic development and population growth, energy issues have become one of the most pressing challenges facing modern society. To improve the efficiency of energy systems and achieve energy conservation goals, some enterprises urgently need to upgrade their aging energy systems. However, rebuilding a new energy system requires a huge investment, leading many enterprises to prefer upgrading older systems to new, more efficient ones. Simultaneously, energy companies may also need to upgrade their energy systems to expand capacity in order to meet market demands and expand their business. However, a series of technical challenges have arisen during the energy system upgrade process, including:

[0003] Data models for older energy systems are built upon their specific structures and operational characteristics. As these systems upgrade to new, highly efficient ones, their structures, operating mechanisms, and thermodynamic laws undergo significant changes, rendering the older models inapplicable. Retraining models for these newer, more efficient systems is not only time-consuming and computationally expensive, but also carries the risk of catastrophic forgetting. The new model learns about the new, more efficient energy systems but gradually loses the thermodynamic knowledge embedded in the older models, knowledge that remains crucial for a comprehensive and accurate understanding and control of these systems.

[0004] Therefore, how to accurately fit the thermodynamic laws of new high-efficiency energy systems while effectively avoiding catastrophic amnesia and retaining relevant knowledge of old energy systems has become a key technical problem that urgently needs to be solved in the field of energy systems. Summary of the Invention

[0005] The purpose of this invention is to address the problems in the prior art by providing a method and system for optimizing and controlling energy systems based on incremental learning. This method utilizes the forward transfer of knowledge from the old energy system to quickly form a new energy system model, solving the problem of time-consuming remodeling, while retaining the knowledge of the old energy system to avoid catastrophic forgetting.

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

[0007] Firstly, a method for optimizing and controlling an energy system based on incremental learning is provided, including:

[0008] Data collected on-site from the original energy system was obtained, and the data was processed using multiple data processing methods to obtain the original energy system dataset.

[0009] The original energy system dataset is normalized, and a pre-selected data model is trained using the normalized original energy system dataset. By comparing the fitting effects of different types of data models, the model with the best fitting effect is selected as the system energy consumption and temperature model.

[0010] For the modified new energy system, an incremental learning algorithm is used based on the selected system energy consumption and temperature model to form a new energy system energy consumption and temperature model that can be applied to both old and new knowledge.

[0011] By utilizing a new energy system energy consumption and temperature model that is applicable to both old and new knowledge, and by matching appropriate optimization algorithms with different optimization objectives, the optimal operating mode, operating state, and optimal system control parameters under different working conditions can be solved to achieve energy conservation.

[0012] As a preferred approach, it also includes adjusting the hyperparameters of the incremental learning algorithm and selecting the best incremental learning algorithm from different types of incremental learning algorithms through effect comparison, so as to form a new energy system energy consumption and temperature model that can be applied to both old and new knowledge data.

[0013] As a preferred approach, the incremental learning algorithm includes any one or more combinations of methods based on regularized incremental learning algorithms, incremental learning algorithms based on dynamic adjustment strategies, memory playback, and complementary learning systems.

[0014] As a preferred approach, the regularized incremental learning algorithm strengthens constraints during model weight updates through regularization methods, thereby enabling the learning of new tasks while maintaining existing knowledge. The regularized incremental learning algorithm includes any one or more combinations of parametric regularization methods, distribution regularization methods, and regularization methods from a Bayesian perspective.

[0015] As a preferred approach, the incremental learning algorithm based on a dynamic adjustment strategy dynamically adjusts the network structure to adapt to the ever-changing environment. The incremental learning algorithm based on a dynamic adjustment strategy selectively trains the network and expands the network as needed to adapt to the learning of new tasks.

[0016] As a preferred approach, the memory replay and complementary learning system comprises two parts: the hippocampus and the neocortex. The hippocampus exhibits short-term adaptability and allows for the learning of new knowledge. The learned knowledge is then reintroduced into the neocortex over time to maintain long-term memory.

[0017] As a preferred option, the energy system includes any or a combination of refrigeration and cooling systems, thermal power generation systems, and data center cooling systems;

[0018] Data collected on-site from the primary energy system is acquired through sensors. The acquired data includes one or more thermodynamic parameters such as temperature, pressure, fluidity, and power.

[0019] Data processing methods include data denoising, data filtering, data repair, and data reduction; data denoising methods include any or a combination of moving average, smoothing filter, and outlier removal methods; data filtering methods include classifier-based or keyword-based data filtering methods; data reduction methods include any or a combination of principal component analysis, factor analysis, and singular value decomposition methods.

[0020] As a preferred approach, the data model is trained using any one or a combination of artificial neural networks (ANN), extreme gradient boosting (XGBoost), support vector regression (SVR), gated recurrent units (GRU), long short-term memory networks (LSTM), and lightweight gradient boosting machines (LightGBM).

[0021] When training a pre-selected data model using the normalized raw energy system dataset, the hyperparameters within the data model are continuously adjusted using the root mean square error (RMSE) and mean absolute percentage error (MAPE) indices. The system energy consumption and temperature models are selected by comparing the fitting effects of different types of data models.

[0022] The hyperparameters in the data model include the learning rate, number of weak learners, and maximum tree depth of XGBoost; the initial learning rate, batch size, and number of iterations of ANN; and the sample sampling ratio and column sampling ratio in LightGBM.

[0023] As a preferred option, the step of solving the optimal operating mode, operating state, and optimal system control parameters under different working conditions according to different optimization objectives and corresponding optimization algorithms to achieve energy saving includes any one or more combinations of Newton's method, gradient descent method, branch and bound algorithm, background segmentation algorithm, dynamic programming algorithm, genetic algorithm, particle swarm algorithm, and simulated annealing algorithm.

[0024] Secondly, an incremental learning-based energy system optimization control system is provided, comprising:

[0025] The dataset construction module is used to acquire field-collected data from the original energy system and process the field-collected data from the original energy system using multiple data processing methods to obtain the original energy system dataset.

[0026] The data model selection module is used to normalize the original energy system dataset and train a pre-selected data model using the normalized original energy system dataset. By comparing the fitting effects of different types of data models, the model with the best fitting effect is selected as the system energy consumption and temperature model.

[0027] The incremental learning module is used to generate a new energy system energy consumption and temperature model that can be applied to both old and new knowledge by using an incremental learning algorithm based on the selected system energy consumption and temperature model for the modified new energy system.

[0028] The model matching output module is used to utilize energy consumption and temperature models of new energy systems that are applicable to both old and new knowledge. Based on different optimization objectives, it combines corresponding optimization algorithms to solve for the optimal operating mode, operating state, and optimal system control parameters under different operating conditions, so as to achieve energy saving.

[0029] Compared with the prior art, the present invention has at least the following beneficial effects:

[0030] This invention employs an incremental learning algorithm that can learn new knowledge while retaining old knowledge, avoiding catastrophic forgetting. The method mimics the human concept of lifelong learning. For energy systems modified for various reasons, it selects the optimal machine learning model to establish an initial system energy consumption and temperature model. After the energy system is modified, the incremental learning algorithm updates the model, enabling the learning of new knowledge (the characteristics of the modified system) while retaining old knowledge (the characteristics of the original system). Utilizing a new energy system energy consumption and temperature model applicable to both old and new knowledge, and based on different optimization objectives, appropriate optimization algorithms are used to solve for the optimal operating mode, operating state, and optimal system control parameters under different operating conditions, achieving efficient operation of the energy system. This invention establishes an incremental model capable of continuous learning, ensuring the accuracy of thermodynamic fitting of the modified system while retaining the accuracy of knowledge from the old energy system, avoiding catastrophic forgetting. After obtaining an incremental learning model capable of accurately predicting both the thermodynamic knowledge of the new system and the knowledge of the old system, it can be combined with optimization algorithms such as genetic algorithms to quickly achieve intelligent control and energy-saving optimization of the energy system. This invention presents an efficient modeling method for modified energy systems based on incremental learning. This method can quickly model modified energy systems, thereby rapidly achieving optimized energy-saving control. It has a short training cycle and can significantly reduce system energy consumption. Attached Figure Description

[0031] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0032] Figure 1 This is a schematic diagram of the cold plate liquid-cooled data center refrigeration system structure before modification, as described in this embodiment of the invention.

[0033] Figure 2 This is a schematic diagram of the modified cold plate liquid-cooled data center refrigeration system according to an embodiment of the present invention;

[0034] Figure 3 This is a flowchart of the energy system optimization control method based on incremental learning according to an embodiment of the present invention;

[0035] Figure 4 This is a schematic diagram of the EWC regularized incremental learning algorithm according to an embodiment of the present invention.

[0036] In the attached diagram: 1-Cooling tower; 2-Primary side filter; 3-Plate heat exchanger; 4-No. 1 three-way valve; 5-Primary side water pump; 6-Primary side bypass valve; 7-Secondary side filter; 8-Server; 9-Secondary side water pump; 10-Chiller unit; 11-No. 2 three-way valve; 12-LCU back panel door; 13-Air-cooled side water pump; 14-Air-cooled side filter; 15-Air-cooled side bypass valve. Detailed Implementation

[0037] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, those skilled in the art can obtain other embodiments without creative effort.

[0038] This invention proposes a method for optimizing the control of an energy system based on incremental learning, comprising the following steps:

[0039] S1. Obtain on-site data from the original energy system and process the on-site data from the original energy system using multiple data processing methods to obtain the original energy system dataset;

[0040] S2. Normalize the original energy system dataset and use the normalized original energy system dataset to train a pre-selected data model. By comparing the fitting effects of different types of data models, select the one with the best fitting effect as the system energy consumption and temperature model.

[0041] S3. For the modified new energy system, an incremental learning algorithm is used based on the selected system energy consumption and temperature model to form a new energy system energy consumption and temperature model that can be applied to both old and new knowledge.

[0042] S4. Utilizing a new energy system energy consumption and temperature model that is applicable to both old and new knowledge, and based on different optimization objectives, using appropriate optimization algorithms, solve for the optimal operating mode, operating state, and optimal system control parameters under different working conditions to achieve energy conservation.

[0043] In one possible implementation, after step S3, the hyperparameters of the incremental learning algorithm are continuously adjusted using indicators such as accuracy and forgetting rate. The best incremental learning algorithm is selected from different types of incremental learning algorithms by comparing the effects, forming a new energy system energy consumption and temperature model that is applicable to both old and new knowledge data.

[0044] Furthermore, the incremental learning algorithm includes any one or more combinations of methods based on regularized incremental learning algorithms, incremental learning algorithms based on dynamic adjustment strategies, memory replay, and complementary learning systems.

[0045] Among them, the regularized incremental learning algorithm strengthens the constraints when updating model weights through regularization methods, thereby achieving the learning of new tasks while maintaining existing knowledge; the regularized incremental learning algorithm includes parametric regularization methods such as SI, R-EWC, and Rwalk; distribution regularization methods such as LwF, SLNID, and GL-GAN; and regularization methods from a Bayesian perspective such as PVI.

[0046] Incremental learning algorithms based on dynamic adjustment strategies adapt to constantly changing environments by dynamically adjusting the network structure. These algorithms selectively train the network and expand it as needed to accommodate new tasks. Examples of such algorithms include CWR, DEN, PN, and ACLF.

[0047] The memory replay and complementary learning system comprises two parts: the hippocampus and the neocortex. The hippocampus exhibits short-term adaptability and allows for rapid learning of new knowledge, which is then reintroduced into the neocortex over time to maintain long-term memory. Algorithms for the memory replay and complementary learning system include episodic memory methods such as EMR, LGM, and MER, and dual-memory learning methods such as BIIL and GDM.

[0048] The incremental learning algorithm in this embodiment of the invention can also be called a continuous learning algorithm or a lifelong learning algorithm.

[0049] In one possible implementation, the energy system described in step S1 includes any or a combination of refrigeration systems, thermal power generation systems, and data center cooling systems.

[0050] Data collected on-site from the primary energy system is acquired through sensors. The acquired data includes one or more thermodynamic parameters such as temperature, pressure, fluidity, and power.

[0051] Data processing methods include data denoising, data filtering, data repair, and data reduction; data denoising methods include any or a combination of moving average, smoothing filter, and outlier removal methods; data filtering methods include classifier-based or keyword-based data filtering methods; data reduction methods include any or a combination of principal component analysis, factor analysis, and singular value decomposition methods.

[0052] In one possible implementation, the training of the data model in step S2 employs any or a combination of artificial neural networks (ANN), extreme gradient boosting (XGBoost), support vector regression (SVR), gated recurrent units (GRU), long short-term memory networks (LSTM), and lightweight gradient boosting machines (LightGBM).

[0053] When training a pre-selected data model using the normalized raw energy system dataset, the hyperparameters within the data model are continuously adjusted using the root mean square error (RMSE) and mean absolute percentage error (MAPE) indices. The system energy consumption and temperature models are selected by comparing the fitting effects of different types of data models.

[0054] Hyperparameters in the data model include the learning rate, number of weak learners, and maximum tree depth of XGBoost; the initial learning rate, batch size, and number of iterations of ANN; and the sample sampling ratio and column sampling ratio of LightGBM.

[0055] In one possible implementation, the transformation of the original old energy system includes, but is not limited to, adding energy equipment such as cooling towers, chillers, and water pumps due to factory expansion; and adding valves to enable interaction between multiple subsystems within the old energy system due to policy requirements, thereby improving the energy efficiency of the old energy system.

[0056] In one possible implementation, step S4, which involves using appropriate optimization algorithms to solve for the optimal operating mode, operating state, and optimal system control parameters under different operating conditions to achieve energy saving, includes, but is not limited to, mathematical methods such as Newton's method and gradient descent; precise algorithms such as branch and bound algorithms, background segmentation algorithms, and dynamic programming; and metaheuristic algorithms such as genetic algorithms, particle swarm optimization algorithms, and simulated annealing algorithms.

[0057] The following content uses the retrofitting of an actual cold plate liquid-cooled data center refrigeration system as an example to illustrate the application of the incremental learning-based energy system optimization control method of this invention.

[0058] like Figure 1 and Figure 2 As shown, the cold plate liquid-cooled data center refrigeration system of this invention includes an external refrigeration equipment cooling tower 1, which utilizes the direct / indirect contact between the refrigerant and air to exchange heat and generate steam. The steam evaporates and carries away heat, achieving heat dissipation through evaporation, convection, and radiation to dissipate waste heat generated in industrial processes or refrigeration and air conditioning. The primary side filter 2, secondary side filter 7, and air-cooled side filter 14 filter impurities in the pipes. The plate heat exchanger 3 exchanges heat between the primary and secondary fluids of the liquid-cooled section, cooling the secondary hot fluid with the cold fluid generated by the primary side refrigeration equipment, thereby cooling the server. The primary side water pump 5, secondary side water pump 9, and air-cooled side water pump 13 provide energy for fluid flow. The primary side bypass valve 6 and air-cooled side bypass valve 15... The server 8 has various memory chips, including high-power-density chips that are directly cooled by liquid through cold plate contact and low-power-density chips that are cooled by air through a fan inside the air-liquid heat exchanger 12. The external refrigeration equipment, the chiller unit 10, cools the refrigerant in the air-cooled section through a vapor compression refrigeration cycle. The LCU back panel door (air-liquid heat exchanger) 12 has a fan inside, which carries the cooling fluid in the pipes into the computer room to cool the servers. The No. 1 three-way valve 4 and the No. 2 three-way valve 11 are components added to the upgraded refrigeration system. In the old refrigeration system, the chiller unit 10 only cooled the air-cooled section, and the cooling tower 1 only cooled the liquid-cooled section. The two three-way valves connect the air-cooled and liquid-cooled sections, forming three modes: low-temperature mode, normal mode, and high-temperature mode. The normal mode is as follows: Figure 1 As shown in the original refrigeration system structure diagram, 10 chiller units provide cooling for the air-cooled section, cooling the servers in the server room through the fans in the air-liquid heat exchanger 12. 1 cooling tower provides cooling for the liquid-cooled section, cooling the chips in the server through the cold plates inside the server. In low-temperature mode, the chiller units are turned off, and the cooling tower cools the air-cooled and liquid-cooled sections through three-way valves 4 and 11. In high-temperature mode, the cooling capacity of cooling tower 1 cannot meet the demand of the liquid-cooled side, and the excess cooling capacity generated by the chiller units is supplemented to the liquid-cooled section through three-way valves 4 and 11.

[0059] like Figure 3 As shown, the energy system optimization control method based on incremental learning in this embodiment of the invention includes the following steps:

[0060] 1) Figure 1 and Figure 2 A typical cold plate liquid-cooled data center cooling system is equipped with a variety of sensors to acquire data. Figure 1Data was collected on-site from the original energy system before the upgrade. Sensor data included thermodynamic parameters such as temperature, pressure, flow rate, power, and frequency. After obtaining the data from the pre-upgrade refrigeration system, multiple data processing methods were applied, including data denoising methods such as moving average, smoothing filter, and outlier removal; data filtering methods based on classifiers and keywords; and data reduction methods such as principal component analysis, factor analysis, and singular value decomposition, to form the original system dataset.

[0061] 2) Normalize the obtained raw dataset using the following formula:

[0062]

[0063] Where x0 represents normalized data; x represents original data; x min x is the minimum value in the original data. max This represents the maximum value of the original data.

[0064] The hyperparameters within data models such as BPNN, LSTM, and GRU are continuously adjusted using the RMSE accuracy metric. The system energy consumption and temperature model with the best fit is selected by comparing different types of data models. Input parameters include ambient temperature and humidity, cooling capacity of the air-cooled section, cooling capacity of the liquid-cooled section, fan and pump speeds, and chiller load. Output parameters include system power, PUE, and chip temperature. The RMSE calculation method is shown in the following formula:

[0065]

[0066] Where RMSE is the root mean square error, which can be used to evaluate the model accuracy; N is the amount of test set data; and Yi is the actual value of system energy consumption and temperature. These are the predicted values ​​from the neural network model.

[0067] 3) To improve the PUE of the data center cooling system, the cooling system from Figure 1 The original old cold plate liquid-cooled data center cooling system shown has been transformed into... Figure 2 The new cold-plate liquid-cooled data center refrigeration system shown has three modes: low-temperature mode, conventional mode, and high-temperature mode. At this point, the system energy consumption and temperature data models established in step 2) are no longer applicable to the low-temperature and high-temperature modes. Re-collecting data for both modes and retraining the models is time-consuming and labor-intensive, and the three models are detrimental to subsequent optimization and control, significantly increasing optimization time. The old energy system belongs to a conventional operating mode within the new energy system; therefore, the data from the old energy system still has value and represents old knowledge within incremental learning. The data from the newly added low-temperature and high-temperature modes in the upgraded new energy system represents new knowledge within incremental learning.

[0068] Using the model developed in step 2) applicable to the original, outdated cold-plate liquid-cooled data center cooling system, and combined with an incremental learning algorithm (taking regularized EWC as an example), a model can accurately predict the thermodynamic behavior of the original, outdated system (conventional mode) as well as the thermodynamic behavior of the low-temperature mode and conventional mode in the new system. The objective function of the EWC model includes a penalty term for the model parameters between the old and new tasks, effectively mitigating the forgetting of previously learned knowledge relevant to the current task. A schematic diagram of the EWC effect is shown below. Figure 4 As shown. The area inside the circle indicates that the hyperparameters at this point have good prediction accuracy for the corresponding task, and the intersection of the two circles indicates that the hyperparameters at this point perform well for all tasks. The parameters of the conventional pattern data model learned in step 2) are θ. A If the newly collected low-temperature and high-temperature model data are modeled independently, knowledge of the conventional model will be forgotten (as shown by the dense dashed lines in the figure). If the EWC method is used to constrain the model hyperparameters, applying strong constraints to hyperparameters important for conventional model prediction and weak constraints to hyperparameters unimportant for conventional model prediction, hyperparameters that can accurately predict both conventional and new high-temperature / low-temperature models can be found (as shown by the sparse dashed lines in the figure). If the hyperparameters are constrained with the same strength, and this trade-off constraint is too strong, the performance for both tasks (conventional model and high-temperature / low-temperature model) may not reach the optimal level (as shown by the solid lines in the figure). The objective function of EWC is shown in the following equation:

[0069]

[0070] Where L(θ) is the EWC loss function, θ represents the hyperparameter, and L... B (θ) is the loss function for the high-temperature / low-temperature mode task, λ represents the trade-off between the old and new tasks, i represents the subscript index of the parameter, and F represents the Fisher matrix, which can determine the importance of hyperparameters to the old task (normal mode).

[0071] 4) Using indicators such as accuracy and forgetting rate, the hyperparameters in the EWC incremental learning algorithm are continuously adjusted and the best parameter values ​​are selected. This retains the old knowledge of the original energy system (conventional mode) and can accurately fit the data of the new energy system (low temperature mode / high temperature mode) to learn new knowledge, forming a new system energy consumption and temperature model based on the incremental learning algorithm that can accurately fit both old and new knowledge data.

[0072] 5) Utilizing a new system energy consumption and temperature model based on incremental learning algorithms, and in accordance with the optimization objective of minimizing energy consumption in the new energy system, combined with novel and efficient optimization algorithms such as genetic algorithms, the optimal operating mode / operating state and optimal system control parameters under different working conditions are calculated to achieve maximum energy saving.

[0073] Another embodiment of the present invention proposes a modified energy system optimization control system based on incremental learning, comprising:

[0074] The dataset construction module is used to acquire field-collected data from the original energy system and process the field-collected data from the original energy system using multiple data processing methods to obtain the original energy system dataset.

[0075] The data model selection module is used to normalize the original energy system dataset and train a pre-selected data model using the normalized original energy system dataset. By comparing the fitting effects of different types of data models, the model with the best fitting effect is selected as the system energy consumption and temperature model.

[0076] The incremental learning module is used to generate a new energy system energy consumption and temperature model that can be applied to both old and new knowledge by using an incremental learning algorithm based on the selected system energy consumption and temperature model for the modified new energy system.

[0077] The model matching output module is used to utilize energy consumption and temperature models of new energy systems that are applicable to both old and new knowledge. Based on different optimization objectives, it combines corresponding optimization algorithms to solve for the optimal operating mode, operating state, and optimal system control parameters under different operating conditions, so as to achieve energy saving.

[0078] Another embodiment of the present invention provides an electronic device comprising:

[0079] A memory for storing at least one instruction; and a processor for executing the instructions stored in the memory to implement the incremental learning-based energy system optimization control method.

[0080] Another embodiment of the present invention provides a computer-readable storage medium storing at least one instruction, which is executed by a processor in an electronic device to implement the incremental learning-based energy system optimization control method.

[0081] For example, the instructions stored in the memory can be divided into one or more modules / units. These modules / units are stored in a computer-readable storage medium and executed by the processor to complete the incremental learning-based energy system optimization control method of the present invention. The one or more modules / units can be a series of computer-readable instruction segments capable of performing specific functions, which describe the execution process of the computer program on the server.

[0082] The electronic device may be a smartphone, laptop, PDA, or cloud server, among other computing devices. It may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the electronic device may also include more or fewer components, or combinations of certain components, or different components; for example, it may also include input / output devices, network access devices, buses, etc.

[0083] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0084] The memory can be an internal storage unit of the server, such as a hard drive or RAM. Alternatively, it can be an external storage device, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card. Furthermore, the memory can include both internal and external storage units. The memory is used to store computer-readable instructions and other programs and data required by the server. It can also be used to temporarily store data that has been output or will be output.

[0085] It should be noted that the information interaction and execution process between the above-mentioned module units are based on the same concept as the method embodiment. For details on their specific functions and technical effects, please refer to the method embodiment section. They will not be repeated here.

[0086] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0087] If the integrated unit is implemented as 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, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying the computer program code to a photographing device / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0088] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0089] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for optimizing the control of an energy system based on incremental learning, characterized in that, include: Data collected on-site from the original energy system was obtained, and the data was processed using multiple data processing methods to obtain the original energy system dataset. The original energy system dataset is normalized, and a pre-selected data model is trained using the normalized original energy system dataset. By comparing the fitting effects of different types of data models, the model with the best fitting effect is selected as the system energy consumption and temperature model. For the modified new energy system, an incremental learning algorithm is used based on the selected system energy consumption and temperature model to form a new energy system energy consumption and temperature model that can be applied to both old and new knowledge. By utilizing a new energy system energy consumption and temperature model that is applicable to both old and new knowledge, and by matching appropriate optimization algorithms with different optimization objectives, the optimal operating mode, operating state, and optimal system control parameters under different working conditions can be solved to achieve energy conservation.

2. The method for optimizing and controlling an energy system based on incremental learning according to claim 1, characterized in that, It also includes adjusting the hyperparameters of the incremental learning algorithm and selecting the best incremental learning algorithm from different types of incremental learning algorithms through effect comparison, so as to form a new energy system energy consumption and temperature model that can be applied to both old and new knowledge data.

3. The method for optimizing control of an energy system based on incremental learning according to claim 2, characterized in that, Incremental learning algorithms include any one or more combinations of methods based on regularized incremental learning algorithms, incremental learning algorithms based on dynamic adjustment strategies, memory playback, and complementary learning systems.

4. The energy system optimization control method based on incremental learning according to claim 3, characterized in that, Regularized incremental learning algorithms strengthen constraints when updating model weights through regularization methods, thereby enabling the learning of new tasks while maintaining existing knowledge. The regularized incremental learning algorithm includes any one or more combinations of parametric regularization methods, distribution regularization methods, and regularization methods from a Bayesian perspective.

5. The energy system optimization control method based on incremental learning according to claim 3, characterized in that, Incremental learning algorithms based on dynamic adjustment strategies adapt to constantly changing environments by dynamically adjusting the network structure. These algorithms selectively train the network and expand it as needed to adapt to learning new tasks.

6. The method for optimizing control of an energy system based on incremental learning according to claim 3, characterized in that, The memory replay and complementary learning system consists of two parts: the hippocampus and the neocortex. The hippocampus exhibits short-term adaptability and allows the learning of new knowledge. The learned knowledge is then put back into the neocortex over time to maintain long-term memory.

7. The method for optimizing control of an energy system based on incremental learning according to claim 1, characterized in that, Energy systems include any one or more combinations of refrigeration and cooling systems, thermal power generation systems, and data center cooling systems; Data collected on-site from the primary energy system is acquired through sensors. The acquired data includes one or more thermodynamic parameters such as temperature, pressure, fluidity, and power. Data processing methods include data denoising, data filtering, data repair, and data reduction; data denoising methods include any or a combination of moving average, smoothing filter, and outlier removal methods; data filtering methods include classifier-based or keyword-based data filtering methods; data reduction methods include any or a combination of principal component analysis, factor analysis, and singular value decomposition methods.

8. The method for optimizing control of an energy system based on incremental learning according to claim 1, characterized in that, The training of the data model employs any one or a combination of artificial neural networks (ANN), extreme gradient boosting (XGBoost), support vector regression (SVR), gated recurrent units (GRU), long short-term memory networks (LSTM), and lightweight gradient boosting machines (LightGBM). When training a pre-selected data model using the normalized raw energy system dataset, the hyperparameters within the data model are continuously adjusted using the root mean square error (RMSE) and mean absolute percentage error (MAPE) indices. The system energy consumption and temperature models are selected by comparing the fitting effects of different types of data models. The hyperparameters in the data model include the learning rate, number of weak learners, and maximum tree depth of XGBoost; the initial learning rate, batch size, and number of iterations of ANN; and the sample sampling ratio and column sampling ratio in LightGBM.

9. The method for optimizing control of an energy system based on incremental learning according to claim 1, characterized in that, In the step of solving the optimal operating mode, operating state, and optimal system control parameters under different working conditions according to different optimization objectives and corresponding optimization algorithms to achieve energy saving, the optimization algorithms include any one or more combinations of Newton's method, gradient descent method, branch and bound algorithm, background segmentation algorithm, dynamic programming algorithm, genetic algorithm, particle swarm algorithm, and simulated annealing algorithm.

10. A modified energy system optimization control system based on incremental learning, characterized in that, include: The dataset construction module is used to acquire field-collected data from the original energy system and process the field-collected data from the original energy system using multiple data processing methods to obtain the original energy system dataset. The data model selection module is used to normalize the original energy system dataset and train a pre-selected data model using the normalized original energy system dataset. By comparing the fitting effects of different types of data models, the model with the best fitting effect is selected as the system energy consumption and temperature model. The incremental learning module is used to generate a new energy system energy consumption and temperature model that can be applied to both old and new knowledge by using an incremental learning algorithm based on the selected system energy consumption and temperature model for the modified new energy system. The model matching output module is used to utilize energy consumption and temperature models of new energy systems that are applicable to both old and new knowledge. Based on different optimization objectives, it combines corresponding optimization algorithms to solve for the optimal operating mode, operating state, and optimal system control parameters under different operating conditions, so as to achieve energy saving.

Images