Transform-based equipment-oriented energy-saving optimization method

Abstract

The invention provides an equipment-oriented energy-saving optimization method based on Transform, and relates to the technical field of energy-saving optimization, and the method comprises the following steps: S1, building an equipment operation condition model; s2, introducing an energy efficiency model to calculate equipment energy consumption under different working conditions; s3, introducing a multi-target minimization optimization algorithm, combining short-term and long-term energy consumption prediction models, and considering energy consumption and operation state stability; s4, a fine-grained real-time parameter adjustment mechanism is introduced, and equipment parameters are adjusted in real time under different working conditions and environment disturbances; according to the method, short-term and long-term energy consumption prediction is integrated, the energy consumption of the current equipment state and the future multi-step energy consumption trend are combined, the short-term energy saving effect and the long-term operation effect are effectively balanced through a multi-objective function, real-time energy saving can be achieved, meanwhile, it is ensured that the future overall energy consumption is kept at a low level, and the real-time energy saving effect is achieved. And the adverse effect of the local optimal solution on the global energy consumption is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of energy-saving optimization, and particularly to an energy-saving optimization method for devices based on Transformer. Background Art

[0002] With the rapid growth of global energy consumption, energy-saving optimization has become one of the core tasks in the control of industrial and household devices. In recent years, facing the pressure of rising energy costs and environmental protection, the demand for energy efficiency optimization in various industries has increased sharply. However, due to the wide variety of devices and complex operating environments, traditional energy-saving methods often struggle to meet the energy-saving requirements of multiple devices under different working conditions. In the existing technology, energy-saving optimization means are usually limited to the energy consumption control of a single device or a single target, lacking consideration of complex coupling relationships. For example, many optimization algorithms only focus on minimizing power consumption, while failing to take into account other key objectives such as operating stability and comfort, resulting in optimization results that are difficult to meet the diverse needs in practical applications.

[0003] In addition, during the operation of devices, they are dynamically affected by environmental factors (such as temperature, humidity, load changes, etc.) and internal states (such as power, rotational speed, etc.), leading to large fluctuations in the energy efficiency curve. Such fluctuations directly affect the stability of device operation and the user's comfort experience, and also significantly increase the complexity of energy consumption management. In multi-device collaboration or source-network-load integration application scenarios, there are often strong coupling relationships between the operating parameters of different devices, further increasing the difficulty of energy-saving optimization. Existing energy-saving control methods are difficult to achieve coordinated optimization among multiple devices, so they cannot effectively reduce the overall energy consumption while ensuring device stability. Therefore, the present invention proposes an energy-saving optimization method for devices based on Transformer to solve the problems existing in the prior art. Summary of the Invention

[0004] To achieve the objectives of the present invention, the present invention is realized through the following technical solutions: An energy-saving optimization method for devices based on Transformer, comprising the following steps:

[0005] To achieve the object of the present invention, the present invention is implemented by the following technical solutions: The energy-saving optimization method for devices based on Transformer includes the following steps:

[0006] S1: Establish a device operating condition model;

[0007] S2: Introduce an energy efficiency model to calculate the energy consumption of the device under different operating conditions;

[0008] S3: Introduce a multi-objective minimization optimization algorithm, combine short-term and long-term energy consumption prediction models, and take into account both energy consumption and the stability of the operating state;

[0009] S4: Introduce a fine-grained real-time parameter adjustment mechanism to adjust the device parameters in real time under different operating conditions and environmental disturbances.

[0010] The further improvement lies in: S1 includes the following steps:

[0011] Use a Transformer model to establish a device operating condition model. By introducing the device state }, that is, the state vector composed of the operating parameters of the device, to capture the non-linear relationship and the high-order interaction of each parameter in the device operating state. The mathematical description of the device operating state is:

[0012]

[0013] Among them, is the state of the device at the current moment, is the state of the device at the previous moment, is the external disturbance (such as environmental parameters like temperature and humidity), is the specific parameter of the device, including fixed attributes and adjustable parameters;

[0014] Adopt a multi-head attention mechanism, set the number of attention heads to 8, the hidden layer dimension to 512, each Transformer encoder layer contains a feed-forward neural network layer with a dimension of 2048, and the Transformer model contains 4 encoder layers to effectively capture the complex interaction relationship between the device state and the external disturbance.

[0015] The further improvement lies in: The training data of the model comes from the real device operation log data and the data generated by the simulation platform, including the historical operation parameters, environmental variables, and device state records of the device under different operating conditions and environmental conditions. By combining real data and simulation data, the Transformer model can be fully trained to achieve accurate modeling of the device operating state;

[0016] For , use a time series formula based on weighted moving average to smooth short-term fluctuations and consider historical trends at the same time. It is a simple and commonly used environmental parameter prediction method. The specific formula is as follows:

[0017]

[0018] Among them represents the historical environmental parameter values of the previous time steps, is the weighting coefficient, usually satisfying , is the historical number of steps of observation;

[0019] By training with the Transformer model, an effective modeling of the device operating state can be obtained, providing data support for further energy efficiency prediction.

[0020] A further improvement lies in that: S2 includes the following steps:

[0021] After constructing the device operating state model, an energy efficiency model is introduced to calculate the device energy consumption under different working conditions, and according to the device operating state and environmental factors calculate the energy efficiency score of the device:

[0022]

[0023] Among them is the energy efficiency score based on the current operating state of the device, is the influence factor based on the external environment, is the coupling term between the operating state and external factors. The energy consumption prediction is based on this energy efficiency model, combining short-term and long-term data to predict the energy consumption situation in the future for a period of time;

[0024] The short-term energy consumption prediction uses the historical time series and predicts the short-term energy consumption recursively:

[0025]

[0026] The long-term energy consumption prediction is carried out through a multi-step forward Transformer model:

[0027] .

[0028] A further improvement lies in that: S3 includes the following steps:

[0029] The multi-objective minimization optimization algorithm takes energy consumption and stability as the core objectives, and the objective function is defined as:

[0030]

[0031] Among them, is the short-term energy consumption target, that is, the total energy consumption within the current prediction period, calculated as the accumulation of the instantaneous energy efficiency model of the device under the current state and environmental parameters, that is

[0032]

[0033] For the long-term energy consumption target, i.e., the prediction of the energy consumption trend in the future for a period of time, it is predicted by the Transformer model and represents the total energy consumption of the device in the next cycle, that is

[0034]

[0035] D is the stability target, which measures the amplitude of the state jitter of the device during operation, that is

[0036]

[0037] and are the weight parameters of short-term energy consumption, stability, and long-term energy consumption respectively;

[0038] To solve this multi-objective optimization problem, the Particle Swarm Optimization algorithm PSO or the Genetic Algorithm GA is used as the optimization solver;

[0039] During the optimization process, the Pareto front screening method combining non-dominated sorting and crowding distance is used to stratify and retain the solution space, so as to obtain the optimal solution set under multiple objectives;

[0040] For the above objective function, by continuously iteratively optimizing it, the optimal parameter combination can be solved under multiple variables and multiple constraint conditions, so that the device can reach the best balance point of energy saving and stability under the established constraint conditions.

[0041] The further improvement lies in: using the Particle Swarm Optimization algorithm PSO or the Genetic Algorithm GA as the optimization solver, and the specific parameter settings are as follows:

[0042] When using PSO, set the particle swarm size to 50, the maximum number of iterations to 200, the inertia weight to 0.7, and both the individual learning factor and the social learning factor to 1.5;

[0043] When using GA, set the population size to 100, the maximum number of generations to 300, the crossover probability to 0.8, the mutation probability to 0.1, and use the tournament selection strategy for individual selection.

[0044] The further improvement lies in: S4 includes the following steps:

[0045] By detecting the change of the device state to determine whether adjustment is needed, and making fine adjustments to the parameters when specific conditions are met, and setting the adjustment threshold This is used to prevent unnecessary frequent adjustments. When the status change exceeds this threshold, parameter adjustment is triggered. The specific adjustment rules are as follows:

[0046]

[0047] in Indicates that at the current moment The parameter adjustment range is is the adjustment factor, which determines the size of the adjustment range. is the threshold for adjustment.

[0048] Further improvements are: threshold Specifically: by statistically analyzing the response data of the equipment under stable working conditions, calculating the standard deviation range of the natural fluctuation of the state vector, and setting is the upper limit of the fluctuation range to ensure that parameter adjustment is triggered only when the system status changes significantly.

[0049] Regulatory Factors Specifically: Based on the dynamic response characteristics of different devices, the relationship between the system's hysteresis time and response intensity for parameter adjustment is determined through experiments, and a coefficient value that strikes a balance between response efficiency and energy consumption improvement is selected. Offline parameter adjustment is performed within the interval through grid search or Bayesian optimization.

[0050] The beneficial effects of the present invention are:

[0051] 1. The present invention integrates short-term and long-term energy consumption forecasts, combines the energy consumption of the current equipment state with the energy consumption trend of multiple steps in the future, and effectively balances short-term energy saving and long-term operation effects through multi-objective functions. While achieving real-time energy saving, it can ensure that the overall energy consumption in the future remains at a low level, avoiding the adverse effects of local optimal solutions on global energy consumption.

[0052] 2. The present invention introduces the stability target D and extends it to the long-term prediction step, so that the algorithm can simultaneously optimize energy consumption fluctuations and avoid equipment wear and efficiency loss caused by frequent parameter adjustments or changes in operating status.

[0053] 3. Through the dynamic parameter adjustment mechanism, the algorithm of the present invention can make subtle adjustments to device parameters in response to real-time environmental changes, so that the operating status of the device can quickly respond to external disturbances. The fine-grained adjustment mechanism can not only prevent excessive adjustment, but also achieve more accurate energy-saving effects, further improving the precision and stability of energy-saving optimization.

[0054] 4. The general device condition model constructed based on the Transformer model in the present invention is applicable to the operation scenarios of various devices, with strong adaptability and generality. This model can flexibly adapt to different device types and operating environments, and is widely applicable to various general devices in the source-network-load system. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a processing flow chart of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0056] To deepen the understanding of the present invention, the following will further elaborate on the present invention in combination with embodiments. These embodiments are only used to explain the present invention and do not constitute a limitation on the protection scope of the present invention.

[0057] Embodiment 1

[0058] According to Figure 1 shown, this embodiment proposes an energy-saving optimization method for devices based on Transformer, including the following steps:

[0059] Device data preprocessing

[0060] The device operation status and environmental parameter data obtained from the data acquisition module need to be systematically preprocessed to improve the quality of the model input and enhance the prediction stability and robustness of the model. The preprocessing process includes the following key steps:

[0061] Outlier identification and processing.

[0062] Use statistical analysis and rule-based detection methods to identify outliers.

[0063] Statistical discrimination method: Based on the statistical distribution of historical data, if the data at a certain moment deviates from the mean by more than 3 times the standard deviation (3 ) it is regarded as an outlier.

[0064] Boundary rule method: Combine the actual working parameter range of the device (such as temperature ∈ [10°C, 45°C]) to set reasonable upper and lower limits, and if it exceeds, it is determined as incorrect data.

[0065] For outlier processing, use linear interpolation or neighborhood mean filling methods for repair to maintain the continuity of the time series.

[0066] Normalization processing.

[0067] All input variables (especially environmental parameters ) need to be normalized before entering the model to eliminate the dimension difference and improve the model convergence speed. The normalization method uses Z-Score standardization, and the specific formula is as follows:

[0068]

[0069] Among them, is the mean value in the training dataset, is the standard deviation, is the normalized environmental parameter value. This method can ensure that the data satisfies the zero-mean and unit-variance distribution, and is suitable for the input requirements of various machine learning models.

[0070] Training and Prediction of Equipment Condition Model

[0071] Transformer Model Training

[0072] Use the historical operation data of the equipment as the training set to input into the Transformer model for training the condition model. Through the model, learn the complex coupling relationships among the various parameters of the equipment to obtain the energy efficiency performance under different working conditions.

[0073] Short-Term Energy Consumption Prediction

[0074] Use the above-trained Transformer model and the time series formula based on weighted moving average to predict the equipment state at future time steps and external disturbance parameters , that is

[0075] [[ID=...]] [[ID=...]] [[ID=...]]

[0076] [[ID=...]] [[ID=...]] [[ID=...]]

[0077] Weights can be set according to the influence degree at different time steps. For example, if we hope that the most recent historical data has a greater influence, it can be set as is the decreasing coefficient:

[0078]

[0079] Based on this, we conduct short-term energy consumption prediction for the next T steps, and the calculation formula is as follows, where is the energy consumption prediction value of the equipment under the current working condition.

[0080]

[0081] Long-Term Energy Consumption Prediction

[0082] Through the Transformer model, predict the equipment state and environmental parameters for the next steps to obtain the long-term energy consumption target .

[0083]

[0084] Multi-Objective Optimization Process Note: There are some consecutive tags with ellipsis in the original text ( - ), and the translation is presented in a way that tries to maintain the overall structure and flow while leaving those parts as is for now. If there are specific requirements or corrections for those parts, please provide more details.

[0085] Construction of the objective function

[0086] Combine the short - term and long - term energy consumption prediction results to construct the objective function for multi - objective optimization:

[0087]

[0088] Among them, the stability objective is defined as the energy consumption fluctuation within the future steps:

[0089]

[0090] Optimization solution

[0091] Through the genetic algorithm (GA) or particle swarm optimization (PSO), iteratively optimize the device operation parameters. The optimization process includes steps such as individual selection, crossover mutation, fitness evaluation, and long - term feedback adjustment until the objective function converges or reaches the set iteration upper limit.

[0092] Fine - grained real - time parameter adjustment

[0093] Based on the multi - objective optimization results, dynamically adjust the key parameters of the device (such as temperature setpoint, wind speed, etc.). When the change range of the device state exceeds the threshold trigger the adjustment mechanism:

[0094]

[0095] The adjusted state will be re - input into the working condition model for energy consumption prediction and optimization, so as to realize the real - time energy - saving control of the device in different operating environments.

[0096] Real - time guarantee and deployment plan

[0097] To ensure the real - time performance and deployability of the energy - saving optimization strategy in the actual device operation process, the present invention has been optimized in terms of algorithm design and system architecture to ensure the response speed and deployment efficiency, specifically including the following two aspects:

[0098] Optimize the algorithm response time

[0099] Adopt a lightweight Transformer architecture and an efficient multi - objective optimizer (such as PSO / GA), and significantly reduce the online calculation delay through the following means:

[0100] (1) Model pruning and quantization: Prune the parameters and quantize the weights of the Transformer model to reduce the computational burden during inference.

[0101] (2) Pre - calculation window sliding strategy: For short - term energy efficiency prediction and adjustment of control parameters, a sliding window mechanism is adopted to calculate some results in advance, reducing the real - time calculation pressure.

[0102] (3) Optimizer iteration control: Under the premise of ensuring optimization accuracy, limit the number of iterations per round (e.g., ≤20 rounds), and combine with heuristic initial solution generation to shorten the convergence time.

[0103] Through actual measurement, under typical configurations (Intel i7 CPU, 16GB RAM), the average response time of the complete decision - making cycle of this method is less than 50 ms, which can meet the requirements of real - time scheduling within 100 ms for most industrial equipment control systems.

[0104] Edge computing deployment architecture

[0105] To achieve high - reliability and low - latency online control, this method supports deployment on edge computing nodes, such as industrial gateways, intelligent edge boxes (Edge Box), etc. The system architecture is as follows:

[0106] (1) Cloud - edge collaboration architecture: Model training and parameter optimization can be carried out in the cloud, while prediction and scheduling decisions in the running stage are deployed on edge devices, reducing dependence on the cloud.

[0107] (2) Hardware compatibility: The optimization algorithm is adapted to mainstream embedded devices (such as ARM Cortex - A72 chips) and industrial edge servers (such as NVIDIA Jetson, Raspberry Pi, etc.) to ensure low power consumption and high performance.

[0108] (3) Communication protocol adaptation: Support communication with standard protocols such as OPC UA, Modbus, MQTT on the device side to achieve seamless integration.

[0109] This architecture not only improves the response speed and system security of algorithm decisions, but also enables off - network autonomous operation, which is suitable for scenarios where the plant network is unstable or edge - independent control is required.

[0110] Example 2

[0111] According to Figure 1 As shown, this example proposes an energy - saving optimization method for devices based on Transformer:

[0112] To verify the effectiveness of the multi - objective energy - saving optimization method based on Transformer proposed in the present invention, a comparative experiment was designed and carried out. An industrial ventilation system was selected as the object of the experiment, and under the same environmental disturbances and equipment task conditions, the following two methods were respectively used for energy - saving control:

[0113] (1)Comparison method A: LSTM + single-objective optimization, with minimizing energy consumption as the only optimization objective, using Long Short-Term Memory (LSTM) network for energy consumption prediction;

[0114] (2)The method B of the present invention: Transformer + multi-objective optimization, introducing the Transformer model in energy consumption prediction, and conducting multi-objective optimization control by combining the stationarity objective with short-term / long-term energy efficiency.

[0115] The experiment was run for 24 hours respectively on a simulation platform and a real device log dataset, with a sampling interval of 5 seconds. The following two key indicators were mainly evaluated:

[0116] (1)Average energy consumption per unit time (kWh).

[0117] (2)State fluctuation index (standard deviation), used to measure the stationarity of the operation process.

[0118] The experimental results are as follows:

[0119]

[0120] It can be seen from the experimental results that compared with the traditional optimization method combining LSTM with a single energy consumption target, the method of the present invention significantly improves the overall energy-saving effect while maintaining the operation stability of the device. Among them, the average energy consumption is reduced by about 20%, and the state fluctuation amplitude is reduced by about 30%, fully verifying the dual optimization ability of this method in terms of energy saving and stationarity.

[0121] To further verify the adaptability of the multi-objective energy-saving optimization method of the present invention in complex and dynamic environments, a cross-scenario test was designed, introducing extreme working conditions for simulation, including the following two typical situations:

[0122] (1)Sudden load: The load of the device suddenly rises to 150% of the rated power within a short time, and then remains in the high-load state for 5 minutes;

[0123] (2)Environmental mutation: The environmental temperature suddenly changes from normal temperature (25°C) to high temperature (40°C), and the humidity rises from 40% to 80%, simulating the scenario of rapid deterioration of the external climate.

[0124] The test was run on a simulation platform, comparing the performance of the method of the present invention with the traditional LSTM + single-objective optimization strategy under extreme working conditions, and mainly evaluating:

[0125] (1)System response time (the duration from the appearance of the disturbance to the restoration of stability);

[0126] (2)Peak energy consumption (kW);

[0127] (3)State fluctuation intensity (maximum deviation amplitude);

[0128] The experimental results are shown in the following table:

[0129]

[0130] The above tests show that in extreme scenarios such as sudden loads and drastic environmental changes, the method of the present invention can quickly identify changes in working conditions, dynamically adjust parameters, achieve smoother responses and faster energy consumption regression, and is significantly superior to traditional strategies. This ability is particularly important for application scenarios with high requirements for stability and comfort, such as industrial equipment and smart homes.

[0131] The present invention combines short-term and long-term energy consumption predictions, combines the energy consumption of the current device state with the energy consumption trends of multiple future steps, and effectively balances short-term energy saving and long-term operation effects through a multi-objective function. It can achieve real-time energy saving while ensuring that the overall future energy consumption remains at a low level, avoiding the adverse effects of local optimal solutions on the global energy consumption. Moreover, the present invention introduces the stationarity objective D and extends it to the long-term prediction step, enabling the algorithm to optimize energy consumption fluctuations simultaneously, avoiding equipment wear and efficiency losses caused by frequent parameter adjustments or changes in operating states. At the same time, through the dynamic parameter adjustment mechanism of the present invention, the algorithm can finely adjust the device parameters according to real-time environmental changes, enabling the operating state of the device to quickly respond to external disturbances. The fine-grained adjustment mechanism can not only prevent over-adjustment but also achieve a more precise energy-saving effect, further improving the fineness and stability of energy-saving optimization. In addition, the general device condition model constructed based on the Transformer model in the present invention is applicable to the operating scenarios of various devices, has strong self-adaptability and generality, and can flexibly adapt to different device types and operating environments, and is widely applicable to various general devices in the source-network-load system.

[0132] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification only illustrate the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A Transformer-based device-oriented energy-saving optimization method, characterized in that It includes the following steps: S1: Establish a device operating condition model; S2: Introduce an energy efficiency model to calculate the device energy consumption under different operating conditions; S3: Introduce a multi-objective minimization optimization algorithm, combine short-term and long-term energy consumption prediction models, and take into account both energy consumption and the stability of the operating state; S4: Introduce a fine-grained real-time parameter adjustment mechanism to adjust device parameters in real time under different operating conditions and environmental disturbances.

2. The energy-saving optimization method for devices based on Transformer according to claim 1, wherein: The S1 includes the following steps: Use the Transformer model to establish a device operating condition model. By introducing the device state }, that is, the state vector composed of the operating parameters of the device, to capture the non-linear relationship in the device operating state and the high-order interaction of each parameter. The mathematical description of the device operating state is as follows: ， Among them, is the state of the device at the current moment, is the state of the device at the previous moment, is the external disturbance (such as environmental parameters like temperature, humidity, etc.), is the specific parameter of the device, including fixed attributes and adjustable parameters; Adopt a multi-head attention mechanism, set the number of attention heads to 8, the hidden layer dimension to 512, each layer of the Transformer encoder contains a feed-forward neural network layer with a dimension of 2048, and the Transformer model contains 4 encoder layers to effectively capture the complex interaction relationship between the device state and external disturbances.

3. The energy-saving optimization method for devices based on Transformer according to claim 2, characterized in that: The training data of the model comes from real device operation log data and data generated by the simulation platform, including the historical operation parameters, environmental variables, and device state records of the device under different operating conditions and environmental conditions. By combining real data and simulation data, the Transformer model can be fully trained to achieve accurate modeling of the device operating state; For , a time series formula based on weighted moving average is adopted to smooth short-term fluctuations and consider historical trends at the same time. It is a simple and commonly used environmental parameter prediction method, and the specific formula is as follows: ， Among them represents the historical environmental parameter values of the previous time steps, is the weighting coefficient, usually satisfying , is the historical number of steps of the observation; Through training with the Transformer model, an effective modeling of the device operating state can be obtained, providing data support for further energy efficiency prediction.

4. The energy-saving optimization method for devices based on Transformer according to claim 1, characterized in that: The S2 includes the following steps: After constructing the device operating state model, introduce the energy efficiency model to calculate the energy consumption of the device under different working conditions, based on the operating state of the device and environmental factors calculate the energy efficiency score of the device: ， Among them is the energy efficiency score based on the current operating state of the device, is the influence factor based on the external environment, is the coupling term between the operating state and external factors. The energy consumption prediction is based on this energy efficiency model, combined with short-term and long-term data, to predict the energy consumption situation in a future period of time; Short-term energy consumption prediction uses historical time series to predict short-term energy consumption recursively: ， Long-term energy consumption prediction is carried out through a multi-step forward Transformer model: 。 5. The energy-saving optimization method for devices based on Transformer according to claim 1, characterized in that: The S3 includes the following steps: The multi-objective minimization optimization algorithm takes energy consumption and stability as the core objectives, and the objective function is defined as: ， Among them, is the short-term energy consumption target, that is, the total energy consumption within the current prediction period, which is calculated as the accumulation of the instantaneous energy efficiency model of the equipment under the current state and environmental parameters, that is ， For the long-term energy consumption target, which is the predicted energy consumption trend over a period of time in the future, obtained by the Transformer model, representing the total energy consumption of the equipment in the next cycle, that is ， D is the stability objective, which measures the state jitter amplitude of the device during operation, that is ， and are the weight parameters of short-term energy consumption, stability, and long-term energy consumption, respectively; To solve this multi-objective optimization problem, use the particle swarm optimization algorithm PSO or the genetic algorithm GA as the optimization solver; During the optimization process, adopt a Pareto front screening method that combines non-dominated sorting and crowding distance to hierarchically screen and retain the solution space, so as to obtain the optimal solution set under multiple objectives; For the above objective function, continuously iterate and optimize it, and the optimal parameter combination can be solved under multiple variables and multiple constraints, so that the device reaches the best balance point of energy saving and stability under the established constraints.

6. The energy-saving optimization method for devices based on Transformer according to claim 5, characterized in that: Use the particle swarm optimization algorithm PSO or the genetic algorithm GA as the optimization solver, and the specific parameter settings are as follows: When using PSO, set the particle swarm size to 50, the maximum number of iterations to 200, the inertia weight to 0.7, and both the individual learning factor and the social learning factor to 1.5; When using GA, set the population size to 100, the maximum number of generations to 300, the crossover probability to 0.8, the mutation probability to 0.1, and adopt the tournament selection strategy for individual selection.

7. The energy-saving optimization method for devices based on Transformer according to claim 1, characterized in that: The S4 includes the following steps: By detecting the device status to determine whether adjustment is needed, and making fine adjustments to the parameters when specific conditions are met, set the adjustment threshold to prevent unnecessary frequent adjustments. When the status change exceeds this threshold, trigger parameter adjustment. The specific adjustment rules are as follows: ， Among them represents the parameter adjustment range at the current moment, is the adjustment factor that determines the size of the adjustment range, is the threshold for adjustment.

8. The energy-saving optimization method for devices based on Transformer according to claim 7, characterized in that: Threshold Specifically: by statistically analyzing the response data of the device under stable operating conditions, calculating the standard deviation range of the natural fluctuations of the state vector, and setting as the upper limit of this fluctuation range to ensure that parameter adjustment is only triggered when there is a significant change in the system state; Adjustment factor Specifically: according to the dynamic response characteristics of different devices, the relationship between the hysteresis time and response intensity of the system to parameter adjustment is determined through experiments, and the coefficient value that balances response efficiency and energy consumption improvement is selected. Offline parameter tuning is performed through grid search or Bayesian optimization within the interval.

Images