A method and system for energy saving control of electromechanical devices

Abstract

The application discloses an electromechanical equipment energy-saving control method and system. The electromechanical equipment is started, and the electromechanical equipment is controlled based on an expert strategy in a first time period. For a second time after the first time period, a first control parameter of the expert strategy in the second time and a second control parameter of the second time predicted based on an AI model are acquired. The first control parameter and the second control parameter are respectively input into an energy consumption evaluation model. First simulation energy consumption corresponding to the first control parameter and second simulation energy consumption corresponding to the second control parameter are determined based on the energy consumption evaluation model. According to the first simulation energy consumption and the second simulation energy consumption, a virtual energy-saving rate corresponding to the AI model is determined. If the virtual energy-saving rate is greater than a preset energy-saving rate threshold, it is indicated that the energy-saving effect of controlling the electromechanical equipment based on the control parameter predicted based on the AI model is better than that of the expert strategy. The electromechanical equipment is controlled by switching from the expert strategy to the AI model, so that the energy-saving effect of the electromechanical equipment is better.

Description

Technical Field

[0001] This application relates to the field of industrial energy conservation and automatic control technology, and in particular to an energy-saving control method and system for electromechanical equipment. Background Technology

[0002] In the field of industrial energy conservation, especially for energy-saving control schemes for complex electromechanical equipment such as HVAC (e.g., chiller rooms, heat pump systems) and air compressor stations, related technologies generally rely on expert strategies for energy-saving control of electromechanical equipment. This approach depends on fixed rules and thresholds preset by engineers based on their experience (e.g., unit start-up and shutdown control based on return water temperature). The problem with this approach is that the rules are static and rigid, unable to perceive and adapt to complex operating conditions such as equipment performance degradation, dynamic load fluctuations, and seasonal changes, resulting in poor energy-saving effects. Summary of the Invention

[0003] This application provides an energy-saving control method and system for electromechanical equipment to solve the problem of poor energy-saving effect of electromechanical equipment in related technologies.

[0004] In a first aspect, this application provides an energy-saving control method for electromechanical equipment, the method comprising:

[0005] When the electromechanical equipment is started, it is controlled based on expert strategies during the first time period.

[0006] For a second time period following the first time period, obtain the first control parameters of the expert strategy for the second time period and the second control parameters of the second time period predicted by the AI ​​model; input the first control parameters and the second control parameters into the energy consumption assessment model respectively, and determine the first simulated energy consumption corresponding to the first control parameters and the second simulated energy consumption corresponding to the second control parameters based on the energy consumption assessment model; determine the virtual energy saving rate corresponding to the AI ​​model based on the first simulated energy consumption and the second simulated energy consumption; if the virtual energy saving rate is greater than a preset energy saving rate threshold, switch from the expert strategy to the AI ​​model to control the electromechanical equipment.

[0007] The above technical solution has the following advantages or beneficial effects:

[0008] This application addresses the problem that expert-based energy-saving control of electromechanical equipment suffers from static and rigid rules, failing to perceive and adapt to complex operating conditions such as equipment performance degradation, dynamic load fluctuations, and seasonal changes, resulting in poor energy-saving performance. It proposes a control scheme for electromechanical equipment combining expert strategies and AI models. Specifically, the technical solution proposed in this application, which switches from expert strategies to AI models, optimizes energy-saving control of electromechanical equipment. Specifically, during equipment startup, the equipment is first controlled based on expert strategies within a first time period, mitigating the cold-start risk of AI models and achieving better energy-saving control in the initial startup phase. Then, for a second time period following the first, the first control parameters of the expert strategy and the second control parameters predicted by the AI ​​model are obtained. These first and second control parameters are input into an energy consumption assessment model, which determines the first simulated energy consumption corresponding to the first control parameters and the second simulated energy consumption corresponding to the second control parameters. Based on the first and second simulated energy consumptions, the virtual energy-saving rate corresponding to the AI ​​model is determined. If the virtual energy saving rate is greater than the preset energy saving rate threshold, it means that the energy saving effect of controlling the electromechanical equipment based on the control parameters predicted by the AI ​​model is better than that of the expert strategy. At this time, the expert strategy is switched to the AI ​​model to control the electromechanical equipment, that is, the control of the electromechanical equipment is based on the control parameters predicted by the AI ​​model, thereby making the energy saving effect of the electromechanical equipment better.

[0009] Secondly, this application provides an energy-saving control system for electromechanical equipment, the system comprising: a platform service layer, a core decision engine layer, and an equipment and control layer;

[0010] The platform service layer is used to maintain the AI ​​model and the energy consumption assessment model;

[0011] The core decision engine layer is used for starting up electromechanical equipment. During a first time period, it controls the electromechanical equipment based on an expert strategy. For a second time period following the first time period, it acquires the first control parameters of the expert strategy for the second time period and the second control parameters predicted by the AI ​​model for the second time period. It then inputs the first and second control parameters into an energy consumption assessment model, and determines the first simulated energy consumption corresponding to the first control parameters and the second simulated energy consumption corresponding to the second control parameters based on the energy consumption assessment model. Based on the first and second simulated energy consumption, it determines the virtual energy saving rate corresponding to the AI ​​model. If the virtual energy saving rate is greater than a preset energy saving rate threshold, it switches from the expert strategy to the AI ​​model to control the electromechanical equipment.

[0012] The equipment and control layer is used to maintain the operating status and collect operating data of the electromechanical equipment.

[0013] Thirdly, this application provides an energy-saving control device for electromechanical equipment, the device comprising:

[0014] The first control module is used to start the electromechanical equipment and control the electromechanical equipment based on expert strategies during a first time period.

[0015] The second control module is configured to, for a second time following the first time period, acquire first control parameters of the expert strategy for the second time period and second control parameters of the second time period predicted by the AI ​​model; input the first control parameters and the second control parameters into an energy consumption assessment model, and determine the first simulated energy consumption corresponding to the first control parameters and the second simulated energy consumption corresponding to the second control parameters based on the energy consumption assessment model; determine the virtual energy saving rate corresponding to the AI ​​model based on the first simulated energy consumption and the second simulated energy consumption; if the virtual energy saving rate is greater than a preset energy saving rate threshold, switch from the expert strategy to the AI ​​model to control the electromechanical equipment.

[0016] Fourthly, this application provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus;

[0017] Memory, used to store computer programs;

[0018] A processor, used to execute a program stored in memory, implements the method described.

[0019] Fifthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described herein.

[0020] In a sixth aspect, this application provides a computer program product comprising an executable program that is executed by a processor to implement the method described. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This application provides a schematic diagram of the energy-saving control process for electromechanical equipment.

[0023] Figure 2A schematic diagram illustrating the process of predicting the second control parameter at the second time based on an AI model, as provided in this application;

[0024] Figure 3 A schematic diagram illustrating the training process of the AI ​​model provided in this application;

[0025] Figure 4 A schematic diagram illustrating the training process of the energy consumption assessment model provided in this application;

[0026] Figure 5 A schematic diagram illustrating the first process of switching from expert strategy to AI model for controlling electromechanical equipment, as provided in this application;

[0027] Figure 6 A schematic diagram illustrating the second process of switching from expert strategy to AI model for controlling electromechanical equipment, provided in this application;

[0028] Figure 7 A schematic diagram of the first type of energy-saving control system for electromechanical equipment provided in this application;

[0029] Figure 8 A schematic diagram of the second type of energy-saving control system for electromechanical equipment provided in this application;

[0030] Figure 9 A detailed architecture diagram of the energy-saving control system for electromechanical equipment provided in this application;

[0031] Figure 10 A schematic diagram of the structure of the energy-saving control device for electromechanical equipment provided in this application;

[0032] Figure 11 A schematic diagram of the electronic device structure provided in this application. Detailed Implementation

[0033] To make the objectives and implementation methods of this application clearer, the exemplary implementation methods of this application will be clearly and completely described below with reference to the accompanying drawings of the exemplary embodiments of this application. Obviously, the exemplary embodiments described are only some embodiments of this application, and not all embodiments.

[0034] It should be noted that the brief descriptions of terms in this application are only for the convenience of understanding the embodiments described below, and are not intended to limit the embodiments of this application. Unless otherwise stated, these terms should be understood in their ordinary and common meaning.

[0035] The terms "first," "second," "third," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar or related objects or entities, and do not necessarily imply a specific order or sequence, unless otherwise specified. It should be understood that such terms are interchangeable where appropriate.

[0036] The terms “comprising” and “having”, and any variations thereof, are intended to cover but not exclude inclusion, for example, a product or device that includes a range of components is not necessarily limited to all of the components that are clearly listed, but may include other components that are not clearly listed or that are inherent to such product or device.

[0037] The term "module" refers to any known or subsequently developed hardware, software, firmware, artificial intelligence, fuzzy logic, or combination of hardware and / or software code that is capable of performing the functions associated with that element.

[0038] Finally, it should be noted that the above 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 or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0039] For ease of explanation, the above description has been provided in conjunction with specific embodiments. However, the above exemplary discussion is not intended to be exhaustive or to limit the embodiments to the specific forms disclosed above. Various modifications and variations can be obtained based on the above teachings. The selection and description of the above embodiments are for the purpose of better explaining the principles and practical applications, thereby enabling those skilled in the art to better utilize the described embodiments and various different variations of embodiments suitable for specific use considerations.

[0040] In the field of industrial energy conservation, especially for complex electromechanical systems such as HVAC (e.g., chiller rooms, heat pump systems) and air compressor stations, the current mainstream energy-saving control technologies mainly fall into two categories, both of which have significant drawbacks:

[0041] ① Control based on expert systems:

[0042] This approach relies on fixed rules and thresholds preset by engineers based on their experience (such as unit start-up and shutdown control based on return water temperature). Its advantage lies in the stability, reliability, and transparency of the control logic; however, its inherent disadvantage is that the rules are static and rigid, unable to perceive and adapt to complex operating conditions such as equipment performance degradation, dynamic load fluctuations, and seasonal changes, resulting in a clear ceiling to its energy-saving potential and making it difficult to achieve deep energy savings.

[0043] ② Optimization control based on pure AI models:

[0044] This approach utilizes big data and machine learning models for system modeling and global optimization, theoretically possessing greater energy-saving potential. However, in practical industrial applications, this method has revealed three fatal flaws:

[0045] Cold start challenge: In the early stages of project deployment, due to a lack of sufficient and high-quality historical data, the AI ​​model cannot be put into effective work immediately, causing the system to fall into a long debugging period and miss the valuable initial energy-saving window.

[0046] Data quality trap: If the system's original historical operating data originated from an energy-inefficient operating mode, the AI ​​model trained on it will "learn" and solidify this inefficient mode, falling into a vicious cycle of "garbage in, garbage out", which fundamentally limits its optimization potential.

[0047] Safety and trust risks: The "black box" decision-making process of AI lacks transparency and poses a risk of outputting instructions that do not conform to physical safety constraints, which reduces its acceptance in industrial settings.

[0048] In addition, existing AI energy-saving solutions have two major bottlenecks in their implementation models:

[0049] The pitfalls of customized development: This approach heavily relies on "workshop-style" project customization. Each new project requires algorithm experts to handle the entire process, from data cleaning and feature engineering to model selection, training, debugging, and code-level deployment. This model results in long development cycles, high labor costs, and difficulties in technology consolidation and standardization, constituting a key bottleneck hindering the large-scale promotion and replication of the technology.

[0050] Model stagnation and performance degradation: Even if a project is successfully deployed, the lack of an efficient model maintenance process means that the system generally lacks the ability to update the model in the long term and at low cost. Once deployed, the model tends to become stagnant and cannot be adjusted to keep up with changes in the system itself. As a result, its energy-saving effect gradually diminishes as equipment ages and operating conditions change, making it impossible to achieve sustainable energy-saving benefits.

[0051] In summary, the industry currently lacks a systematic and complete solution that not only needs to overcome the technical challenges of "AI cold start and data traps" but also needs to completely reverse the high-cost model of "customized project development" from an engineering perspective and ensure the long-term sustainable maintenance of energy-saving effects.

[0052] The core of this application lies in a "dynamic transfer of control mechanism based on quantitative assessment and gradual deployment." Upon system startup, an expert system takes the lead, immediately achieving a basic energy saving of 8%-15% and cultivating a high-quality dataset for the AI ​​model. Subsequently, the AI ​​is trained in "shadow mode" and passes the Bootstrap-based Virtual Energy Saving Rate (VES) confidence interval test (first-level test). After passing, it enters a cross-running phase, undergoing entity verification through similar day pairing and paired sample t-tests (second-level test). The entire process is scheduled by a dynamic decision-maker and ensured by a security verification module, ultimately achieving a safe and smooth transition to full AI control. After going live, the system monitors performance through a self-learning module, implementing concept drift detection and automatic rollback, forming a closed loop of continuous optimization.

[0053] Figure 1 The schematic diagram of the energy-saving control process for electromechanical equipment provided in this application includes the following steps:

[0054] S101: The electromechanical equipment is started, and the electromechanical equipment is controlled based on expert strategies during the first time period;

[0055] S102: For the second time following the first time period, obtain the first control parameters of the expert strategy for the second time and the second control parameters of the second time predicted by the AI ​​model; input the first control parameters and the second control parameters into the energy consumption assessment model respectively, and determine the first simulated energy consumption corresponding to the first control parameters and the second simulated energy consumption corresponding to the second control parameters based on the energy consumption assessment model; determine the virtual energy saving rate corresponding to the AI ​​model based on the first simulated energy consumption and the second simulated energy consumption; if the virtual energy saving rate is greater than the preset energy saving rate threshold, switch from the expert strategy to the AI ​​model to control the electromechanical equipment.

[0056] The energy-saving control method for electromechanical equipment provided in this application is applied to electronic equipment, which may be personal computer PC, computer, smart terminal, server and other equipment.

[0057] When electromechanical equipment starts up, it is controlled based on expert strategies during the initial time period, such as 10 days or 15 days. Controlling the equipment based on expert strategies means controlling it using expert strategy control parameters. These parameters are control parameters defined by experts for controlling the equipment. The expert strategy control parameters stored in the electronic equipment can be updated according to user commands. Electromechanical equipment generally refers to a system that includes chillers, water pumps, cooling towers, heat pumps, air compressors, sensors, meters, etc. Furthermore, the expert strategy control parameters stored in the electronic equipment can be a single set or multiple sets; for example, different weather conditions or environmental conditions correspond to different expert strategy control parameters. During the initial time period, the equipment is controlled based on the expert strategies corresponding to the weather and environmental conditions at each time point. Each time point is, for example, each day.

[0058] For the second time period following the first time period, such as the day after the first time period, the first control parameters of the expert strategy for the second time period are obtained. The second control parameters for the second time period are then predicted based on the AI ​​model. Optionally, historical load data, historical weather data, historical building population density data, historical time series data, historical operating data of electromechanical equipment, and environmental temperature and humidity data within the historical time period are input into the AI ​​model, and the AI ​​model outputs the predicted second control parameters for the second time period.

[0059] The electronic device is equipped with a trained energy consumption assessment model. This model extracts features from input control parameters and outputs predicted energy consumption data (or simulated energy consumption) corresponding to those parameters. First and second control parameters are input into the energy consumption assessment model, and the model determines the first simulated energy consumption corresponding to the first control parameter and the second simulated energy consumption corresponding to the second control parameter. Then, based on the first and second simulated energy consumptions, the virtual energy saving rate corresponding to the AI ​​model is determined. Optionally, the difference between the first and second simulated energy consumptions is calculated, and the ratio of this difference to the first simulated energy consumption is used as the virtual energy saving rate corresponding to the AI ​​model. If the virtual energy saving rate is greater than a preset energy saving rate threshold, the expert strategy switches to the AI ​​model to control the electromechanical equipment. Controlling the electromechanical equipment based on the AI ​​model means controlling the equipment based on the control parameters predicted by the AI ​​model.

[0060] This application addresses the problem that expert-based energy-saving control of electromechanical equipment suffers from static and rigid rules, failing to perceive and adapt to complex operating conditions such as equipment performance degradation, dynamic load fluctuations, and seasonal changes, resulting in poor energy-saving performance. It proposes a control scheme for electromechanical equipment combining expert strategies and AI models. Specifically, the technical solution proposed in this application, which switches from expert strategies to AI models, optimizes energy-saving control of electromechanical equipment. Specifically, during equipment startup, the equipment is first controlled based on expert strategies within a first time period, mitigating the cold-start risk of AI models and achieving better energy-saving control in the initial startup phase. Then, for a second time period following the first, the first control parameters of the expert strategy and the second control parameters predicted by the AI ​​model are obtained. These first and second control parameters are input into an energy consumption assessment model, which determines the first simulated energy consumption corresponding to the first control parameters and the second simulated energy consumption corresponding to the second control parameters. Based on the first and second simulated energy consumptions, the virtual energy-saving rate corresponding to the AI ​​model is determined. If the virtual energy saving rate is greater than the preset energy saving rate threshold, it means that the energy saving effect of controlling the electromechanical equipment based on the control parameters predicted by the AI ​​model is better than that of the expert strategy. At this time, the expert strategy is switched to the AI ​​model to control the electromechanical equipment, that is, the control of the electromechanical equipment is based on the control parameters predicted by the AI ​​model, thereby making the energy saving effect of the electromechanical equipment better.

[0061] Figure 2 The process diagram for predicting the second control parameter at the second time based on the AI ​​model provided in this application includes the following steps:

[0062] S201: Input the historical load data, historical weather data, historical building population density data and historical time series prior to the second time into the load prediction sub-model in the AI ​​model, and predict the load demand for the second time based on the load prediction sub-model;

[0063] S202: Input the load demand, the real-time operating data of the electromechanical equipment, and the ambient temperature and humidity data into the energy consumption prediction sub-model in the AI ​​model, and predict the second control parameters when the energy consumption of the electromechanical equipment is at its lowest at the second time based on the energy consumption prediction sub-model.

[0064] The AI ​​model includes a load forecasting sub-model and an energy consumption forecasting sub-model. The load forecasting sub-model is used to output the load demand for future times based on input historical load data, historical weather data, historical building occupancy density data, and historical time series. In this application, historical load data, historical weather data, historical building occupancy density data, and historical time series prior to a second time point are input into the load forecasting sub-model of the AI ​​model, and the load demand for the second time point is predicted based on the load forecasting sub-model. The energy consumption forecasting sub-model is used to output control parameters for the future time point when the energy consumption of the electromechanical equipment is at its lowest, based on input load demand, real-time operating data of electromechanical equipment, and environmental temperature and humidity data. In this application, the load demand for the second time point predicted based on the load forecasting sub-model, the real-time operating data of the electromechanical equipment, and the environmental temperature and humidity data are input into the energy consumption forecasting sub-model of the AI ​​model, and the energy consumption forecasting sub-model predicts the second control parameters for the second time point when the energy consumption of the electromechanical equipment is at its lowest.

[0065] Figure 3 The training process diagram of the AI ​​model provided in this application includes the following steps:

[0066] S301: Using the operating data of the electromechanical equipment within the first time period as the first training set, inputting the sample historical load data, sample historical weather data, sample historical building personnel density data, and sample historical time series from the first training set into the load prediction sub-model to be trained in the AI ​​model, and predicting the sample load demand for the fourth time period based on the load prediction sub-model; determining the first loss value based on the actual load value and the sample load demand for the fourth time period; and training the load prediction sub-model based on the first loss value.

[0067] S302: Input the sample load demand, the sample operation data of the electromechanical equipment, and the sample environment temperature and humidity data into the energy consumption prediction sub-model to be trained in the AI ​​model, and predict the sample energy consumption value at the fourth time based on the energy consumption prediction sub-model; determine the second loss value based on the actual energy consumption value at the fourth time and the sample energy consumption value; train the energy consumption prediction sub-model based on the second loss value.

[0068] It should be noted that the electromechanical equipment includes various devices such as chillers, water pumps, cooling towers, heat pump units, and air compressors. Different AI models are used to obtain the control parameters of different devices. The AI ​​models for different devices all include load prediction sub-models and energy consumption prediction sub-models.

[0069] For different equipment in the electromechanical equipment, when training the load prediction sub-model in the AI ​​model corresponding to the equipment, a first loss value is determined based on the actual load value at the fourth time and the sample load demand; during the iterative training process, when the first loss value meets the requirements, or when the number of iterations reaches the threshold, the load prediction sub-model training is determined to be complete.

[0070] For different types of equipment in electromechanical systems, there are two scenarios when training the energy consumption prediction sub-model in the corresponding AI model for that equipment. Scenario 1: The equipment lacks a clear physical mechanism, meaning there is no physical relationship between control parameters and energy consumption. In this case, the sample load demand, sample operating data of the electromechanical equipment, and sample environmental temperature and humidity data are input into the energy consumption prediction sub-model to be trained. Based on the energy consumption prediction sub-model, the sample energy consumption value at the fourth time point is predicted. A second loss value is determined based on the actual energy consumption value at the fourth time point and the sample energy consumption value. During iterative training, when the second loss value meets the requirements, or the number of iterations reaches a threshold, the energy consumption prediction sub-model training is considered complete. Scenario 2: The equipment has a clear physical mechanism, meaning there is a physical relationship between control parameters and energy consumption, such as a chiller unit having a clear physical mechanism. At this point, the physical relationship between the equipment's control parameters and energy consumption is used as the foundational sub-model for the energy consumption prediction sub-model. Then, the least squares method is used to determine the fitting coefficients for each term in the physical relationship. Subsequently, sample load demand, sample operating data of electromechanical equipment, and sample environmental temperature and humidity data are input into the energy consumption prediction sub-model to be trained. Based on the energy consumption prediction sub-model, the sample energy consumption value at the fourth time point is predicted. Based on the actual energy consumption value at the fourth time point and the sample energy consumption value, the second loss value is determined. During iterative training, when the second loss value meets the requirements, or the number of iterations reaches a threshold, the energy consumption prediction sub-model training is considered complete. In scenario two, using the physical relationship between the equipment's control parameters and energy consumption as the foundational sub-model improves the training efficiency and accuracy of the energy consumption prediction sub-model. Furthermore, scenario two trains the residual between the energy consumption obtained from the foundational sub-model and a more accurate energy consumption prediction. Compared to the scheme that only uses the physical relationship between the equipment's control parameters and energy consumption for energy consumption prediction, this further improves the accuracy of energy consumption prediction.

[0071] Figure 4 A schematic diagram illustrating the training process of the energy consumption assessment model provided in this application, which includes the following steps:

[0072] S401: Input the sample control parameters from the second training set into the energy consumption assessment model to be trained, and predict the sample simulated energy consumption corresponding to the sample control parameters based on the energy consumption assessment model;

[0073] S402: Determine a third loss value based on the simulated energy consumption of the sample and the label energy consumption corresponding to the sample control parameters; train the energy consumption assessment model based on the third loss value; wherein the sample data stored in the second training set is different from that in the first training set used to train the AI ​​model, and the model structure of the energy consumption assessment model is different from that of the AI ​​model.

[0074] In this application, a third loss value is determined based on the simulated energy consumption of the samples and the labeled energy consumption corresponding to the sample control parameters. During iterative training, the energy consumption assessment model is considered complete when the third loss value meets the requirements or the number of iterations reaches a threshold. The sample data stored in the second training set of this application differs from that in the first training set used to train the AI ​​model, and the model structure of the energy consumption assessment model differs from that of the AI ​​model. This is to avoid the energy consumption assessment model being overly reliant on the accuracy of the control parameters predicted by the AI ​​model, which could lead to poor accuracy and objectivity in energy consumption assessment and consequently affect the accuracy of the triggering conditions for switching from expert strategy to AI model control.

[0075] Figure 5 The first schematic diagram provided in this application illustrates the process of switching from expert strategy to AI model to control electromechanical equipment. If the virtual energy saving rate is greater than a preset energy saving rate threshold, the process of switching from expert strategy to AI model to control the electromechanical equipment includes the following steps:

[0076] S501: Within a preset evaluation period, determine the virtual energy saving rate sequence corresponding to the AI ​​model, determine the confidence interval of the virtual energy saving rate based on the sample statistical algorithm, and determine the virtual energy saving rate corresponding to the lower limit of the confidence interval based on the virtual energy saving rate sequence.

[0077] S502: If the virtual energy saving rate corresponding to the lower limit of the preset evaluation period built-in information interval for a first consecutive number of cycles is greater than the preset energy saving rate threshold, the expert strategy is switched to the AI ​​model to control the electromechanical equipment.

[0078] For example, a virtual energy-saving rate is calculated every hour, with a preset evaluation period of 24 hours. This yields 24 virtual energy-saving rates within the preset evaluation period. These 24 virtual energy-saving rates are then sorted in ascending order to obtain the virtual energy-saving rate sequence corresponding to the AI ​​model. A confidence interval for the virtual energy-saving rate is determined based on a sample statistical algorithm; for example, this confidence interval is a 95% confidence interval (2.5% to 97.5%). The lower limit of the confidence interval is then determined based on the virtual energy-saving rate sequence. The virtual energy-saving rate corresponding to the lower limit of the 95% confidence interval is the 2.5% virtual energy-saving rate from left to right among the 24 virtual energy-saving rates. It should be noted that if the 2.5% position is a decimal, it can be rounded up or down to obtain the corresponding virtual energy-saving rate.

[0079] If the virtual energy-saving rate corresponding to the lower limit of the built-in confidence interval within a first consecutive preset evaluation period is greater than the preset energy-saving rate threshold, the expert strategy will switch to the AI ​​model to control the electromechanical equipment. The first number is, for example, 3 or 4, and the preset energy-saving rate threshold is, for example, 2% or 3%.

[0080] In this application, if the virtual energy saving rate exceeds a preset energy saving rate threshold, before the expert strategy switches to the AI ​​model to control the electromechanical equipment, the method further includes:

[0081] During the second time period, the electromechanical equipment is controlled by a combination of expert strategies and AI models; wherein, the combination of expert strategies and AI models is used alternately to control the electromechanical equipment at preset time intervals.

[0082] The second time period is, for example, 30 days, 35 days, etc. The preset time interval is, for example, 1 day. That is to say, within the second time period, the expert strategy is used to control the electromechanical equipment on one day, the AI ​​model is used to control the electromechanical equipment on the next day, the expert strategy is used to control the electromechanical equipment on the day after that, and so on, alternating between the expert strategy and the AI ​​model to control the electromechanical equipment.

[0083] Figure 6 This application provides a second schematic diagram of the process for switching from expert strategy to AI model control of electromechanical equipment. After the electromechanical equipment is controlled using a cross-control method of expert strategy and AI model in the second time period, and before switching from expert strategy to AI model control of the electromechanical equipment, the method further includes the following steps:

[0084] S601: For at least one of the AI ​​model control times, determine the expert strategy control time that matches the ambient temperature and humidity data of the time; determine the actual energy consumption of the electromechanical equipment based on the actual energy consumption of the electromechanical equipment during the AI ​​model control time and the actual energy consumption of the electromechanical equipment during the expert strategy control time;

[0085] S602: Determine the control time of the at least one AI model, the average value of the actual energy saving rate corresponding to the AI ​​model, and obtain the average energy saving rate;

[0086] S603: If the average energy saving rate is greater than the preset energy saving rate threshold, the process of switching from the expert strategy to the AI ​​model to control the electromechanical equipment is carried out.

[0087] In the second time period, a combination of expert strategy and AI model control is employed to manage the electromechanical equipment, resulting in multiple AI model control times (e.g., 15 days). For the control time of at least one AI model, the control time of the expert strategy, matched with the ambient temperature and humidity data, is determined. This matching satisfies two conditions: Condition 1, the difference in ambient temperature and humidity data is within a preset temperature and humidity error ratio range; Condition 2, the time interval between the two matches is less than a preset duration. The preset temperature and humidity error ratio range is, for example, 0 to 5%; the preset duration is, for example, 3 days, 5 days, etc.

[0088] Based on the actual energy consumption of the time-controlled electromechanical equipment controlled by the matched AI model and the actual energy consumption of the time-controlled electromechanical equipment controlled by the expert strategy, the actual energy-saving rate corresponding to the AI ​​model is determined. Optionally, the difference between the actual energy consumption of the time-controlled electromechanical equipment controlled by the expert strategy and the actual energy consumption of the time-controlled electromechanical equipment controlled by the AI ​​model is calculated, and the ratio of this difference to the actual energy consumption of the time-controlled electromechanical equipment controlled by the expert strategy is determined as the actual energy-saving rate corresponding to the AI ​​model.

[0089] Then, the control time of at least one AI model is determined, and the average value of the actual energy saving rate corresponding to the AI ​​model is obtained to obtain the average energy saving rate. If the average energy saving rate is greater than a preset energy saving rate threshold, the subsequent process of switching from the expert strategy to the AI ​​model to control the electromechanical equipment is performed. The preset energy saving rate threshold is, for example, 1%, 2%, etc.

[0090] In this application, after switching from the expert strategy to the AI ​​model to control the electromechanical equipment, the method further includes:

[0091] Based on the load demand predicted by the load forecasting sub-model in the AI ​​model at the third time and the actual load value at the third time, the mean absolute percentage error (MAPE) of the load forecast is determined; wherein, the third time includes multiple load forecasting times.

[0092] If the second consecutive number of MAPE values ​​are greater than the preset error value, the AI ​​model will switch to the expert strategy to control the electromechanical equipment.

[0093] Optionally, for multiple load forecast times included in the third time period, the load forecasting sub-model in the AI ​​model predicts the load demand at each forecast time. Based on the load demand and the actual load value at that forecast time, the absolute percentage error value for that load forecast time is determined. Specifically, the difference between the actual load value and the predicted load demand at that forecast time is calculated, and the ratio of this difference to the actual load value is determined as the absolute percentage error value for that load forecast time. Then, the average of the absolute percentage error values ​​for each of the multiple load forecast times is determined as the Mean Absolute Percentage Error (MAPE). A second quantity, such as 5 or 10, is used. Optionally, the average of a second consecutive number of MAPE values ​​can also be determined. If this average value exceeds a preset error value, the AI ​​model switches to an expert strategy to control the electromechanical equipment. The preset error value is, for example, 10%.

[0094] It should be noted that the AI ​​model undergoes periodic optimization training according to a preset cycle, such as 10 days or 15 days. After switching from the AI ​​model to expert strategy to control the electromechanical equipment, the AI ​​model needs to be optimized and trained again. Once the requirements are met, the system can then switch back from expert strategy to the AI ​​model to control the electromechanical equipment.

[0095] Figure 7 A schematic diagram of a first type of energy-saving control system for electromechanical equipment provided in this application, the system comprising: a platform service layer 11, a core decision engine layer 12, and an equipment and control layer 13;

[0096] The platform service layer 11 is used to maintain the AI ​​model and the energy consumption assessment model;

[0097] The core decision engine layer 12 is used for starting up electromechanical equipment. During a first time period, it controls the electromechanical equipment based on an expert strategy. For a second time period following the first time period, it acquires the first control parameters of the expert strategy for the second time period and the second control parameters predicted by the AI ​​model for the second time period. It inputs the first control parameters and the second control parameters into an energy consumption assessment model, and determines the first simulated energy consumption corresponding to the first control parameters and the second simulated energy consumption corresponding to the second control parameters based on the energy consumption assessment model. It determines the virtual energy saving rate corresponding to the AI ​​model based on the first simulated energy consumption and the second simulated energy consumption. If the virtual energy saving rate is greater than a preset energy saving rate threshold, it switches from the expert strategy to the AI ​​model to control the electromechanical equipment.

[0098] The device and control layer 13 is used to maintain the operating status and collect operating data of the electromechanical equipment.

[0099] Figure 8 A schematic diagram of a second type of energy-saving control system for electromechanical equipment provided in this application, wherein the system further includes: an application interaction layer 14;

[0100] The application interaction layer 14 is used to provide a visual configuration interface, which provides graphical tools for drag-and-drop configuration of electromechanical equipment topology, data points, control strategies, basic information of model algorithms, as well as display of system operation data, energy-saving effect statistics, model performance indicators and alarm information.

[0101] The energy-saving control method for electromechanical equipment provided in this application belongs to the field of industrial energy conservation and automatic control technology. Specifically, it relates to an energy-saving optimization control method and system for electromechanical equipment based on the integration of artificial intelligence and expert system, which is particularly suitable for electromechanical equipment systems with variable load requirements such as chiller rooms, heat pumps, and air compressor stations.

[0102] This application proposes a dynamic control transfer mechanism based on quantitative assessment and gradual deployment. First, a virtual energy-saving rate is calculated in parallel in the background, and significance is determined using statistical confidence intervals. After passing the assessment, a testing phase is conducted to verify the cross-execution of dual control strategies. Finally, with the guarantee of an independent security verification module, a safe, automatic, and gradual transition of control from the expert system to the AI ​​system is achieved. This method systematically solves the three major challenges of AI: "cold start," "data quality traps," and "security trust." This application constructs a platform system that achieves fully automated closed-loop management. Through standardized configuration templates and automated model pipelines, this system integrates data collection, processing, training, deployment, monitoring, updating, and rollback into an organic whole. This design upgrades AI energy-saving technology from a highly expert-dependent "manual workshop-style customization" to a scalable, replicable "standardized product," and ensures the long-term sustainability of energy-saving effects through online monitoring and a self-learning closed loop.

[0103] This application achieves the immediacy, depth, and sustainability of energy-saving benefits through a "safe and controllable dynamic transfer of responsibilities," completely solving the challenges of implementing AI in industrial scenarios. Immediate Startup and Security Assurance: In the initial stage of operation, the system is led by an expert system. With the protection of a built-in security boundary library, it immediately achieves 8%-15% basic energy savings, perfectly avoiding the cold start risks and security trust crises of pure AI solutions. Deep Optimization and Smooth Transition: Once the AI ​​model has matured through evaluation, the system automatically transfers control through a dynamic decision-maker. All AI commands undergo mandatory review by an independent security verification module before being issued, ensuring that global optimization is performed within an absolutely safe range. This achieves an additional 5%-15% energy saving improvement on top of the expert system, with a smooth process that is imperceptible to the user. Continuous Evolution and Risk Mitigation: The platform's self-learning closed loop can maintain optimal energy-saving status in the long term. More importantly, once the system detects any anomalies or performance degradation, the automatic rollback mechanism immediately returns control to the expert system, providing a final risk mitigation for the entire optimization process and ensuring the absolute reliability of the system throughout its entire lifecycle. This application realizes the productization and large-scale replication of the technology, greatly reducing the application threshold: through the platform-based standard design, the AI ​​energy-saving technology is transformed from "manual workshop-style customization" that highly depends on algorithm experts to "standardized product deployment" that can be configured by engineers. It can be adapted to various types of equipment, which greatly reduces the deployment cycle and manpower cost of new projects and solves the core bottleneck of large-scale promotion.

[0104] Figure 9This is a detailed architecture diagram of the energy-saving control system for electromechanical equipment provided in this application. The energy-saving optimization control platform provided in this application, as the physical carrier of the dynamic responsibility transfer method, is designed to achieve an end-to-end automated closed loop from data perception to intelligent decision-making and then to safe execution. The platform adopts a layered design, with each layer's modules working collaboratively to specifically support the subsequent core methodological processes. The overall operation of the system follows the following... Figure 9 The architecture shown.

[0105] 1. Application interaction layer (platform interface).

[0106] This layer serves as the window for user interaction with the system, showcasing the platform's configurability, operability, and observability.

[0107] Visual configuration interface: Provides graphical tools for drag-and-drop configuration of basic information such as device topology, data points, and control strategies, enabling standardized and rapid replication of project implementation.

[0108] Algorithm and Model Management: Allows users to select or upload algorithms, manually trigger model training, and configure the goals and constraints of the optimization algorithm, giving the platform a high degree of flexibility and scalability.

[0109] Control mode panel: Provides a manual mode selection switch (expert / AI) and an emergency intervention entry, ensuring that users have ultimate control over the system at all times, enhancing the system's credibility and security.

[0110] Panoramic monitoring dashboard: Real-time display of system operation data, energy-saving effect statistics, model performance indicators and alarm information, making the entire system's operating status completely transparent and realizing comprehensive monitoring of the optimization process and effect.

[0111] 2. Platform service layer.

[0112] This layer is the core capability center supporting the system's intelligence and automation, and is divided into two major sections: data management and modeling training.

[0113] ① Data Management Center:

[0114] Data access and interfaces: Supports multiple industry standard protocols such as Modbus, BACnet, and MQTT, solving the problem of unified access for multi-source heterogeneous devices.

[0115] The data preprocessing and standardization module performs cleaning, anomaly identification and imputation, and normalization on the raw data, ensuring the data quality for subsequent model training.

[0116] Feature engineering and feature library management: Automatically build and manage feature libraries based on feature mapping relationship tables, providing a ready-to-use and standardized data foundation for model training and self-learning in the core methods.

[0117] ② Modeling and Training Center:

[0118] Automated training pipeline: Extracts data from the feature repository, calls algorithms and formulas from open algorithm libraries or mechanism formula libraries, automatically completes model training, evaluation, and selection, and publishes the winning model to the model repository. This process automates model iteration.

[0119] Open Algorithm Library: It has a rich set of built-in machine learning, deep learning and other algorithms, providing basic algorithmic support for various prediction and optimization problems.

[0120] Mechanism Formula Library: Built-in gray box mechanism models of various electromechanical equipment such as compressors, pumps, and cooling towers. Users can select appropriate formulas and fit parameters through data, balancing model accuracy and physical interpretability.

[0121] Model Repository: Versioned storage, metadata management, and performance tracking of all trained models, and automatically marks the best-performing "champion model" for one-click deployment, realizing lifecycle management of models.

[0122] 3. Core decision engine layer.

[0123] This layer is the system's "intelligent brain," responsible for executing the core "dynamic transfer of control mechanism based on confidence assessment" of this application.

[0124] ①Expert system engine:

[0125] Policy rule base: Stores a large number of verified energy-saving control policies for different equipment types (such as host start-up and shutdown, water pump frequency adjustment) to ensure the stability and efficiency of the system's basic operation.

[0126] Boundary constraint library: Defines the upper and lower limits of safe operating parameters for all devices and systems, and is the cornerstone of safe system operation.

[0127] ②AI Optimization Engine: With the support of the deployed model, it performs load forecasting and equipment energy efficiency forecasting, and performs global optimization calculations based on these to find the operating point with the lowest total system energy consumption.

[0128] ③ Dynamic Decision Maker: This is the core control logic unit of this application. Based on the confidence assessment results of the AI ​​model, it automatically decides whether to use the instructions of the expert system or the instructions of the AI ​​optimization engine in the current control cycle, thereby realizing the dynamic, smooth, and safe transfer of control.

[0129] ④ Security Verification Module: As the final security line, it performs mandatory verification on all control commands to be issued (regardless of their source) to ensure that they absolutely comply with the security rules in the boundary constraint library and prevent the output of any unsafe commands.

[0130] ⑤ Self-learning: This module drives the continuous evolution of the system. It continuously monitors the deviation between model predictions and actual results. When performance degradation is detected, it automatically triggers the retraining process of the platform service layer or executes model rollback, forming a complete self-optimization closed loop.

[0131] 4. Equipment and control layer.

[0132] This layer contains all the controlled physical devices (chillers, water pumps, cooling towers, heat pumps, air compressors, etc.) and sensing elements (sensors, meters). It is both the final executor of control commands and the source of real-time operating data required by the entire system.

[0133] In addition, the system may include a data persistence layer.

[0134] ① Time-series database: efficiently stores and queries all timestamped device operation data, supporting real-time monitoring and historical data analysis.

[0135] ② Relational database: Stores structured data such as system configuration information, model metadata, and user information.

[0136] The core approach involved in this application is the dynamic transfer of control based on quantitative assessment and gradual deployment.

[0137] ① System initialization and expert system leadership.

[0138] System initialization: After the system starts, the dynamic decision-maker of the core decision engine layer sets the control mode to "expert system dominant".

[0139] Expert Control and Data Labeling: The expert system engine generates control commands based on the policy rule base and boundary constraint base, and issues them for execution through the device and control layers. Simultaneously, the platform service layer's data management center collects operational data, labels it as "expert control period data," and stores it in the data persistence layer. This stage involves automated data processing and cleaning, and differentiated processing of data points of varying importance based on preset rules. Furthermore, the system automatically constructs a feature repository based on the feature mapping relationship tables required by each model, providing a standardized data foundation for subsequent model training and self-learning. The core objective of this process is to utilize the expert system to build a high-quality initial dataset with a clear energy-saving orientation for the AI ​​model.

[0140] ② AI parallel training and model metric calculation.

[0141] After the system starts up and runs stably for a specific period (e.g., 14 days), the automated training pipeline will be automatically triggered, and the system will enter "shadow mode." In this mode, the platform executes the following model training and evaluation processes in parallel:

[0142] Model Training: The pipeline extracts expert-controlled data from the data persistence layer to train the following two types of core AI models:

[0143] Demand forecasting models (such as cooling load forecasting): The input features of the model are determined based on a preset feature mapping table, including historical load, outdoor temperature and humidity, holiday markers, and timestamps, etc. The output is the predicted cooling load value for a specific future period (such as one hour). The system automatically calls algorithms from the open algorithm library for training and selects the champion model based on preset model evaluation indicators (such as root mean square error RMSE).

[0144] Equipment energy efficiency model: A hybrid training strategy of gray box and black box models is adopted. For equipment with a clear physical mechanism (such as chiller units), the basic model is built first using the international standard formulas (such as COP polynomials) defined in the mechanism formula library, and then the black box model is used to fit its residuals; for equipment with an unclear mechanism (such as cooling towers), the black box model is used directly for training.

[0145] Model Updates and Management: Trained models are automatically published to the model repository and marked. Model retraining cycles can be automatically triggered based on preset frequencies (e.g., once every 7 days for load forecasting models) or data increments, ensuring continuous iterative optimization of the model.

[0146] ③ Transfer decision based on statistical confidence level.

[0147] The transfer decision in this application does not rely on indicators at a single point in time, but is based on a statistical hypothesis testing framework. This mechanism includes the following steps:

[0148] A. Model preparation and pre-screening.

[0149] The system first ensures that the AI ​​models used for evaluation have reached a basic performance threshold, as a prerequisite for initiating quantitative evaluation. For example:

[0150] The load forecasting model has a mean absolute percentage error (MAE) of less than 15% on the validation set, and the equipment energy efficiency model has a goodness of fit greater than 0.85 on the validation set. This pre-screening step aims to eliminate obviously unqualified models and avoid unnecessary computational overhead. The mean absolute percentage error (MAE) refers to the average of multiple absolute percentage error values.

[0151] B. Calculation and independence assurance of virtual energy saving rate.

[0152] After passing the preliminary screening, the system enters the formal quantitative evaluation cycle. Its core is calculating the virtual energy saving rate, which is defined as:

[0153] VES = (E_expert_sim - E_AI_sim) / E_expert_sim × 100%;

[0154] in:

[0155] E_expert_sim represents the simulated energy consumption based on the historical control sequences of the expert system.

[0156] E_AI_sim represents the simulated energy consumption under the exact same historical operating condition sequence, using an AI control strategy.

[0157] To ensure the fairness and unbiasedness of the evaluation, the calculation of E_AI_sim must be performed by an evaluator independent of the AI ​​optimization engine model training process. This independent evaluator should use differentiated model architectures and training datasets to eliminate evaluation bias caused by model 'self-proof' at the system level.

[0158] C. Decision-making based on statistical inference.

[0159] To base decisions on a solid mathematical and statistical foundation, the system employs the following process:

[0160] Constructing a sample sequence: Within a sliding evaluation window (e.g., the past 24 hours), the system calculates a set of virtual energy-saving rate values ​​{VE}. VE , ..., VE }

[0161] Perform statistical tests: The dynamic decision maker uses the Bootstrap resampling method to perform statistical inference on the above samples and calculates the 95% confidence interval of VES.

[0162] The system determines that the energy-saving advantage of AI is statistically significant and generates a "recommend transfer" instruction only when the lower limit of the confidence interval is greater than the preset significance threshold y (e.g., 2%) for multiple consecutive evaluation periods (e.g., 3 windows).

[0163] At this point, the system has completed its first level of verification, namely virtual verification based on historical data.

[0164] ④ Progressive handover and entity verification.

[0165] After passing virtual verification, the system enters the "gradual handover" phase, which verifies the energy-saving effect and stability of AI control in actual operation through a combination of daily cross-operation and similar day pairing.

[0166] A. Cross-operation and data acquisition.

[0167] The system performs daily cross-operation according to a preset cycle (e.g., 14–30 days), with the rule being "1 day of expert control (E), 1 day of AI control (A)", and this cycle continues.

[0168] The system collects key data such as total energy consumption, average daily outdoor temperature and humidity, and operating mode daily, and excludes days with malfunctions and extreme weather.

[0169] B. Pairing based on similar dates.

[0170] For each AI-controlled day, match similar days within the expert-controlled days, using the following matching criteria:

[0171] The average daily outdoor temperature deviation is ≤ 5%;

[0172] Daily average relative humidity deviation ≤ 5%;

[0173] Each AI day prioritizes matching the most recent expert day to form a valid pairing sample.

[0174] C. Statistics and judgment of energy-saving effect.

[0175] For each paired sample (one-to-one), calculate the actual energy saving: d = Ereal(E) - Ereal(A).

[0176] Calculate the average energy saving rate and perform a paired-samples t-test based on all paired samples.

[0177] Transfer judgment conditions (must be met simultaneously):

[0178] Number of paired samples n≥n min (Recommendation ≥ 14);

[0179] Average energy saving rate r≥r min (e.g., ≥ 1%); (at this time, it is in cross-operation) before switching to AI control; Energy saving rate = (Ereal(E) - Ereal(A)) / Ereal(E) × 100%.

[0180] Ereal(E) refers to the actual energy consumption of time-based electromechanical equipment controlled by expert strategies, while Ereal(A) refers to the actual energy consumption of time-based electromechanical equipment controlled by AI models.

[0181] D. Safety monitoring and decision-making.

[0182] AI control commands are still reviewed by a security verification module before being issued to ensure they do not exceed the boundaries.

[0183] If the verification passes and there are no operational anomalies, the system switches to full AI control mode; otherwise, the verification is extended or the system reverts to expert control.

[0184] ⑤ Full AI dominance and closed-loop evolution.

[0185] A. Full handover and system switchover.

[0186] After passing physical verification, the dynamic decision-maker officially switches the system control mode to "AI system-led". At this point, the AI ​​optimization engine takes full control of the system, generating all control commands based on real-time data and optimization models, and executing them through the device and control layers.

[0187] B. Continuous monitoring and performance tracking.

[0188] Once the system enters full AI operation, the self-learning module initiates a continuous performance monitoring mechanism to track the prediction accuracy of key AI models, such as calculating the mean absolute percentage error (MAPE) of the cooling load prediction model. The system supports setting independent performance thresholds for different models, typically stricter than those in the initial screening stage (e.g., requiring MAPE < 10% during full operation). All operating statuses and performance indicators are visible in real time on the panoramic monitoring dashboard, providing operators with transparent insights.

[0189] C. Concept drift detection and automatic rollback.

[0190] When the self-learning module detects a continuous deterioration in model performance (e.g., MAPE exceeding the threshold for three consecutive days), it determines that concept drift has occurred and immediately sends an alarm and rollback request to the dynamic decision-maker. Based on this, the dynamic decision-maker completes a control switch within minutes, rolling back the system to the "expert system-led" mode to ensure uninterrupted control. Simultaneously, the system automatically triggers the automated training pipeline for model retraining. Once the new model's performance recovers and passes virtual verification (Phase 3), the system can skip entity verification (Phase 4) and directly revert to AI-led mode, or execute a shortened version of entity verification according to policy settings to accelerate the system's return to its optimal state.

[0191] This stage establishes a complete "monitoring-decision-rollback-learning" closed loop, giving the system the adaptive ability to cope with changes in operating conditions and maintain efficient and stable operation in the long run.

[0192] The energy-saving control process for electromechanical equipment provided in this application is illustrated below with a specific example.

[0193] This platform is deployed in a central air conditioning system comprising 2 chillers, 3 chilled water pumps, 3 cooling pumps, and 4 cooling towers. Engineers configure the system topology and data points by dragging and dropping through the application's interactive layer's visual topology building module.

[0194] After platform initialization, the expert system engine in the core decision engine layer immediately went into operation, achieving approximately 12% basic energy savings compared to the historical energy consumption baseline before system optimization. Simultaneously, the data management center in the platform service layer automatically began data collection, cleaning, and feature repository construction.

[0195] After 14 days of stable operation, the automated training pipeline in the modeling and training center was automatically triggered, completing the training of the cooling load prediction and equipment energy efficiency model. The dynamic decision-maker performed virtual validation on the model, and the calculated virtual energy saving rate (VES) had a stable lower limit of 4.5% in its 95% confidence interval, exceeding the threshold of 2% for three consecutive cycles, successfully passing the first stage of virtual validation.

[0196] Subsequently, the system entered a 30-day gradual handover phase, implementing a cross-operation of "1 day expert / 1 day AI". A dynamic decision-maker scheduling and control mode was used, and based on the collected real data, analysis was conducted through similar day pairing and paired sample t-tests, confirming that the actual energy-saving rate of AI control was 3.8% (consistent with the virtual energy-saving rate trend), and statistically significant (p<0.05). During this period, all AI control commands were reviewed in real time by the security verification module, and no violations were found.

[0197] Once all conditions are met, the dynamic decision-maker switches the system to "AI system-led". In this mode, the overall energy saving rate of the system (compared to the historical benchmark) increases to 21.5%.

[0198] Two months later, the self-learning module detected conceptual drift in the cooling load prediction model due to seasonal changes (its MAPE remained consistently above the 10% threshold) and immediately sent an alert to the dynamic decision-maker. Within one minute, the dynamic decision-maker automatically reverted control to the expert system and simultaneously triggered the automated training pipeline to retrain the model using recent data. A week later, the new model's performance recovered (MAPE < 10%), and it quickly passed virtual validation and a shortened physical validation. The system then automatically and seamlessly switched back to full AI optimization mode, restoring its optimal energy-saving state.

[0199] Figure 10 The schematic diagram of the energy-saving control device for electromechanical equipment provided in this application includes:

[0200] The first control module 21 is used to start the electromechanical equipment and control the electromechanical equipment based on expert strategies during a first time period.

[0201] The second control module 22 is configured to, for a second time following the first time period, acquire first control parameters of the expert strategy for the second time and second control parameters of the second time predicted by the AI ​​model; input the first control parameters and the second control parameters into an energy consumption assessment model, and determine the first simulated energy consumption corresponding to the first control parameters and the second simulated energy consumption corresponding to the second control parameters based on the energy consumption assessment model; determine the virtual energy saving rate corresponding to the AI ​​model based on the first simulated energy consumption and the second simulated energy consumption; if the virtual energy saving rate is greater than a preset energy saving rate threshold, switch from the expert strategy to the AI ​​model to control the electromechanical equipment.

[0202] The second control module 22 is further configured to input historical load data, historical weather data, historical building occupancy density data and historical time series prior to the second time into the load prediction sub-model in the AI ​​model, and predict the load demand for the second time based on the load prediction sub-model; input the load demand, the real-time operating data of the electromechanical equipment and the ambient temperature and humidity data into the energy consumption prediction sub-model in the AI ​​model, and predict the second control parameters when the energy consumption of the electromechanical equipment is at its lowest at the second time based on the energy consumption prediction sub-model.

[0203] The second control module 22 is further configured to determine the virtual energy-saving rate sequence corresponding to the AI ​​model within a preset evaluation period, determine the confidence interval of the virtual energy-saving rate based on a sample statistical algorithm, and determine the virtual energy-saving rate corresponding to the lower limit of the confidence interval according to the virtual energy-saving rate sequence; if the virtual energy-saving rate corresponding to the lower limit of the confidence interval within a first number of consecutive preset evaluation periods is greater than the preset energy-saving rate threshold, the expert strategy is switched to the AI ​​model to control the electromechanical equipment.

[0204] The second control module 22 is also used to control the electromechanical equipment by using a cross-control method of expert strategy and AI model during the second time period; wherein, the cross-control method refers to using the expert strategy and the AI ​​model alternately to control the electromechanical equipment according to a preset time interval.

[0205] The second control module 22 is further configured to: determine the time of expert strategy control that matches the ambient temperature and humidity data for the control time of at least one of the AI ​​models; determine the actual energy consumption of the electromechanical equipment based on the actual energy consumption of the electromechanical equipment during the control time of the matched AI model and the actual energy consumption of the electromechanical equipment during the control time of the expert strategy; determine the average value of the actual energy consumption of the at least one AI model control time and the AI ​​model corresponding to the actual energy consumption to obtain an average energy consumption rate; if the average energy consumption rate is greater than a preset energy consumption rate threshold, proceed with the subsequent process of switching from the expert strategy to the AI ​​model to control the electromechanical equipment.

[0206] The second control module 22 is further configured to determine the average absolute percentage error (MAPE) of load forecasting based on the load demand predicted by the load forecasting sub-model in the AI ​​model at the third time and the actual load value at the third time; wherein the third time includes multiple load forecasting times; if the second consecutive number of MAPE values ​​are greater than the preset error value, the AI ​​model switches to the expert strategy to control the electromechanical equipment.

[0207] The device further includes:

[0208] The first training module 23 is used to take the operating data of the electromechanical equipment during the first time period as the first training set, input the sample historical load data, sample historical weather data, sample historical building personnel density data, and sample historical time series from the first training set into the load prediction sub-model to be trained in the AI ​​model, and predict the sample load demand for the fourth time period based on the load prediction sub-model; determine a first loss value based on the actual load value and the sample load demand for the fourth time period; train the load prediction sub-model based on the first loss value; input the sample load demand, the sample operating data of the electromechanical equipment, and the sample environmental temperature and humidity data into the energy consumption prediction sub-model to be trained in the AI ​​model, and predict the sample energy consumption value for the fourth time period based on the energy consumption prediction sub-model; determine a second loss value based on the actual energy consumption value and the sample energy consumption value for the fourth time period; and train the energy consumption prediction sub-model based on the second loss value.

[0209] The device further includes:

[0210] The second training module 24 is used to input the sample control parameters from the second training set into the energy consumption assessment model to be trained, predict the simulated energy consumption of the sample corresponding to the sample control parameters based on the energy consumption assessment model, determine a third loss value based on the simulated energy consumption of the sample and the label energy consumption corresponding to the sample control parameters, and train the energy consumption assessment model based on the third loss value; wherein, the second training set is different from the sample data stored in the first training set used to train the AI ​​model, and the energy consumption assessment model has a different model structure than the AI ​​model.

[0211] This application also provides an electronic device, such as Figure 11 As shown, it includes: processor 31, communication interface 32, memory 33 and communication bus 34, wherein processor 31, communication interface 32 and memory 33 communicate with each other through communication bus 34;

[0212] The memory 33 stores a computer program, which, when executed by the processor 31, causes the processor 31 to perform any of the above method steps.

[0213] The communication bus mentioned in the above electronic devices can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not mean that there is only one bus or one type of bus.

[0214] Communication interface 32 is used for communication between the above-mentioned electronic device and other devices.

[0215] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0216] The processors mentioned above can be general-purpose processors, including central processing units, network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits, field-programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0217] This application also provides a computer-readable storage medium storing a computer program executable by an electronic device, which, when run on the electronic device, causes the electronic device to perform any of the above method steps.

[0218] This application provides a computer program product, which includes an executable program that, when executed by a processor, implements the method described herein.

[0219] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0220] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for energy-saving control of electromechanical equipment, characterized in that, The method includes: When the electromechanical equipment is started, it is controlled based on expert strategies during the first time period. For a second time period following the first time period, obtain the first control parameters of the expert strategy for the second time period and the second control parameters of the second time period predicted by the AI ​​model; input the first control parameters and the second control parameters into the energy consumption assessment model respectively, and determine the first simulated energy consumption corresponding to the first control parameters and the second simulated energy consumption corresponding to the second control parameters based on the energy consumption assessment model; determine the virtual energy saving rate corresponding to the AI ​​model based on the first simulated energy consumption and the second simulated energy consumption; if the virtual energy saving rate is greater than a preset energy saving rate threshold, switch from the expert strategy to the AI ​​model to control the electromechanical equipment; If the virtual energy saving rate is greater than a preset energy saving rate threshold, the switch from the expert strategy to the AI ​​model to control the electromechanical equipment includes: Within a preset evaluation period, the virtual energy-saving rate sequence corresponding to the AI ​​model is determined, the confidence interval of the virtual energy-saving rate is determined based on the sample statistical algorithm, and the virtual energy-saving rate corresponding to the lower limit of the confidence interval is determined according to the virtual energy-saving rate sequence. If the virtual energy saving rate corresponding to the lower limit of the preset evaluation period built-in information interval is greater than the preset energy saving rate threshold for a first number of consecutive periods, the expert strategy will switch to the AI ​​model to control the electromechanical equipment.

2. The method as described in claim 1, characterized in that, The process of predicting the second control parameter at the second time based on the AI ​​model includes: Historical load data, historical weather data, historical building population density data, and historical time series prior to the second time are input into the load prediction sub-model in the AI ​​model, and the load demand at the second time is predicted based on the load prediction sub-model. The load demand, the real-time operating data of the electromechanical equipment, and the ambient temperature and humidity data are input into the energy consumption prediction sub-model in the AI ​​model. Based on the energy consumption prediction sub-model, the second control parameters are predicted when the energy consumption of the electromechanical equipment is at its lowest at the second time.

3. The method as described in claim 1, characterized in that, If the virtual energy saving rate exceeds a preset energy saving rate threshold, before the expert strategy switches to the AI ​​model to control the electromechanical equipment, the method further includes: During the second time period, the electromechanical equipment is controlled by a combination of expert strategies and AI models; wherein, the combination of expert strategies and AI models is used alternately to control the electromechanical equipment at preset time intervals.

4. The method as described in claim 3, characterized in that, After controlling the electromechanical equipment using a combination of expert strategy and AI model control during the second time period, and before switching from expert strategy to AI model control of the electromechanical equipment, the method further includes: For at least one of the AI ​​models controlling the time, determine the expert strategy control time that matches the ambient temperature and humidity data of the time; based on the actual energy consumption of the electromechanical equipment during the time controlled by the matched AI model and the actual energy consumption of the electromechanical equipment during the time controlled by the expert strategy, determine the actual energy saving rate corresponding to the AI ​​model; Determine the control time of at least one of the AI ​​models, and the average value of the actual energy saving rate corresponding to the AI ​​model to obtain the average energy saving rate; If the average energy saving rate is greater than the preset energy saving rate threshold, the process of switching from the expert strategy to the AI ​​model to control the electromechanical equipment will proceed.

5. The method as described in claim 2, characterized in that, After switching from the expert strategy to the AI ​​model to control the electromechanical equipment, the method further includes: Based on the load demand predicted by the load forecasting sub-model in the AI ​​model at the third time and the actual load value at the third time, the mean absolute percentage error (MAPE) of the load forecast is determined; wherein, the third time includes multiple load forecasting times. If the second consecutive number of MAPE values ​​are greater than the preset error value, the AI ​​model will switch to the expert strategy to control the electromechanical equipment.

6. The method as described in claim 2, characterized in that, The training process of the AI ​​model includes: The operating data of the electromechanical equipment during the first time period is used as the first training set. The sample historical load data, sample historical weather data, sample historical building population density data, and sample historical time series in the first training set are input into the load prediction sub-model to be trained in the AI ​​model. Based on the load prediction sub-model, the sample load demand in the fourth time period is predicted. According to the actual load value in the fourth time period and the sample load demand, a first loss value is determined. The load prediction sub-model is trained based on the first loss value. The sample load demand, the sample operation data of the electromechanical equipment, and the sample environmental temperature and humidity data are input into the energy consumption prediction sub-model to be trained in the AI ​​model. The sample energy consumption value at the fourth time is predicted based on the energy consumption prediction sub-model. A second loss value is determined based on the actual energy consumption value at the fourth time and the sample energy consumption value. The energy consumption prediction sub-model is trained based on the second loss value.

7. The method as described in claim 2, characterized in that, The training process of the energy consumption assessment model includes: The sample control parameters from the second training set are input into the energy consumption assessment model to be trained, and the simulated energy consumption of the sample corresponding to the sample control parameters is predicted based on the energy consumption assessment model; a third loss value is determined based on the simulated energy consumption of the sample and the label energy consumption corresponding to the sample control parameters; and the energy consumption assessment model is trained based on the third loss value. The second training set is different from the sample data stored in the first training set used to train the AI ​​model, and the energy consumption assessment model has a different model structure than the AI ​​model.

8. An energy-saving control system for electromechanical equipment, characterized in that, The system includes: a platform service layer, a core decision engine layer, and a device and control layer; The platform service layer is used to maintain the AI ​​model and the energy consumption assessment model; The core decision engine layer is used for starting up electromechanical equipment. During a first time period, it controls the electromechanical equipment based on an expert strategy. For a second time period following the first, it acquires first control parameters of the expert strategy for the second time and second control parameters predicted by an AI model. It then inputs the first and second control parameters into an energy consumption assessment model, and determines a first simulated energy consumption corresponding to the first control parameter and a second simulated energy consumption corresponding to the second control parameter based on the model. Based on the first and second simulated energy consumptions, it determines the virtual energy saving rate corresponding to the AI ​​model. If the virtual energy saving rate is greater than a preset energy saving rate threshold... The process involves switching from the expert strategy to the AI ​​model to control the electromechanical equipment. Specifically, if the virtual energy-saving rate is greater than a preset energy-saving rate threshold, switching from the expert strategy to the AI ​​model includes: within a preset evaluation period, determining a virtual energy-saving rate sequence corresponding to the AI ​​model; determining a confidence interval for the virtual energy-saving rate based on a sample statistical algorithm; and determining the virtual energy-saving rate corresponding to the lower limit of the confidence interval based on the virtual energy-saving rate sequence. If the virtual energy-saving rate corresponding to the lower limit of the confidence interval is greater than the preset energy-saving rate threshold for a first consecutive number of preset evaluation periods, the process switches from the expert strategy to the AI ​​model to control the electromechanical equipment. The equipment and control layer is used to maintain the operating status and collect operating data of the electromechanical equipment.

9. The system as described in claim 8, characterized in that, The system also includes: an application interaction layer; The application interaction layer provides a visual configuration interface, which offers graphical tools for drag-and-drop configuration of electromechanical equipment topology, data points, control strategies, and basic model algorithm information, as well as display of system operation data, energy-saving effect statistics, model performance indicators, and alarm information.

Images