Energy consumption control method, system and computer program product

Abstract

The present invention relates to an energy consumption control method, system and computer program product, the energy consumption control method comprising: performing, by a processor and system firmware, the following steps: continuously detecting and collecting performance data of the processor, the performance data comprising a first performance parameter, a second performance parameter and a third performance parameter; executing a dual-mode machine learning model to predict the first performance parameter of the processor based on the first table presence data; and implementing a fuzzy feedback control mechanism to adjust the first performance parameter according to the detected second performance parameter and the third performance parameter.

Description

Technical Field

[0001] The present invention relates to an energy consumption control method, system and computer program product, in particular to an energy consumption control method, system and computer program product that integrates long-term and short-term machine learning prediction models into a single prediction model and combines a fuzzy feedback control mechanism. Background Art

[0002] At present, in order to comply with and implement ESG energy efficiency specifications, enterprises are committed to reducing ESG carbon footprints and actively formulating various energy-saving solutions; however, in the industrial and commercial application fields, including the service industry, medical industry, manufacturing industry, retail industry, transportation industry and entertainment industry, etc., the industrial computer devices used still lack an intelligent automatic control method for the purpose of energy saving. Existing industrial computer devices often require users to manually operate to set the idle time to enter the power-saving mode, or manually set the energy-saving mode by the user.

[0003] Figure 1 A time series schematic diagram revealing the processor performance parameters of the internal central processing unit of an existing industrial computer device under normal operation. For industrial computer devices, the performance parameters of the central processing unit (CPU) include power, frequency, load and temperature, etc., which are closely related to each other. Different frequencies have very different effects on power and temperature performance. In traditional settings, the power of the central processing unit is usually set between 7 and 8 watts (Watt), and the working temperature usually falls between 80 and 100 °C.

[0004] Figure 2 A time series schematic diagram revealing the processor performance parameters of the internal central processing unit of an existing industrial computer device under the condition of excessive power limitation. In order to reduce energy consumption, the prior art tends to directly limit the power of the CPU, usually setting the power unconditionally below 6 watts. Although the lower power setting can greatly reduce the frequency and temperature, it causes a large accumulation of work to be processed, resulting in an overloaded CPU, and even being in a 100% full load state for a long time, which instead causes the overall performance of the CPU to decline and affects the user experience. In such a reverse cycle, it will only lead to energy waste.

[0005] In summary, currently common existing industrial computer devices mostly simply use a manual method to limit the power of the CPU, but this method is too simplistic and lacks the evaluation of many parameters, making it difficult to achieve the purpose of optimization and automation, and there are many aspects that need to be improved.

[0006] Therefore, in view of the disadvantages existing in the prior art, the inventor has painstakingly tried and studied, and with perseverance, finally conceived the "energy consumption control method, system and computer program product" of this case, which can overcome the above-mentioned disadvantages. The following is a brief description of the present invention. Summary of the Invention

[0007] The present invention relates to an energy consumption control method, system and computer program product, in particular to an energy consumption control method, system and computer program product that integrates long-term and short-term machine learning prediction models into a single prediction model and combines a fuzzy feedback control mechanism.

[0008] Accordingly, the present invention provides an energy consumption control method, which includes: the processor executes the following steps: continuously detecting and collecting the performance data of the processor, the performance data including a first performance parameter, a second performance parameter and a third performance parameter; executing a dual-mode machine learning model to predict the first performance parameter of the processor based on the first performance data; and implementing a fuzzy feedback control mechanism to adjust the first performance parameter according to the detected second performance parameter and the third performance parameter.

[0009] The present invention further provides an energy consumption control computer program product, which is loaded and executed via a processor and includes: a data collection programming module configured to continuously detect and collect the performance data of the processor, the performance data including a first performance parameter, a second performance parameter and a third performance parameter; a dual-mode machine learning model programming module configured to execute a dual-mode machine learning model to predict the first performance parameter of the processor based on the first performance data; and an automatic control programming module configured to implement a fuzzy feedback control mechanism to adjust the first performance parameter according to the detected second performance parameter and the third performance parameter.

[0010] The present invention further provides an energy consumption control system, which includes: an electronic device including a processor configured to: continuously detect and collect the performance data of the processor, the performance data including a first performance parameter, a second performance parameter and a third performance parameter; execute a dual-mode machine learning model to predict the first performance parameter of the processor based on the first performance data; and implement a fuzzy feedback control mechanism to adjust the first performance parameter according to the detected second performance parameter and the third performance parameter.

[0011] The above summary of the invention aims to provide a simplified summary of the present disclosure to enable readers to have a basic understanding of the present disclosure. This summary of the invention is not a complete description of the present invention, and its intention is not to point out the important / critical elements of the embodiments of the present invention or to define the scope of the present invention. Brief Description of the Drawings

[0012] Figure 1 A time series schematic diagram showing the performance parameters of the central processing unit inside an existing industrial computer device under normal operation;

[0013] Figure 2Disclose a time series schematic diagram of the performance parameters of the central processor inside an existing industrial computer device when the power is overly restricted;

[0014] Figure 3 Disclose a schematic diagram of the system architecture of the energy consumption control system included in the present invention;

[0015] Figure 4 Disclose a schematic diagram of the software architecture of the energy consumption control software included in the present invention;

[0016] Figure 5 Disclose a schematic diagram of the time series of the power change of the processor inside the electronic device included in the present invention for consecutive days;

[0017] Figure 6 Disclose a schematic diagram of the time series of the power change of the processor inside the electronic device included in the present invention within 24 hours;

[0018] Figure 7 Disclose a schematic diagram of the predicted power change of the processor on the next day by the first machine learning model included in the present invention;

[0019] Figure 8 Disclose a schematic diagram of the predicted hourly power change of the processor on the next day by the first machine learning model included in the present invention;

[0020] Figure 9 A schematic diagram of the power change of the processor on the next day predicted by the second machine learning model included in the present invention within the first unit time;

[0021] Figure 10 A schematic diagram of the power change of the processor on the next day predicted by the second machine learning model included in the present invention within the second unit time;

[0022] Figure 11 Disclose a schematic diagram of the fuzzy relationship between the frequency and load of the processor included in the present invention;

[0023] Figure 12 Disclose a schematic diagram of the power - frequency - load relationship after the fuzzy control programming module included in the present invention optimizes the predicted value of TDP;

[0024] Figure 13 Disclose a schematic diagram of the temperature - frequency - load relationship after the fuzzy control programming module included in the present invention optimizes the predicted value of TDP;

[0025] Figure 14 Disclose a time series schematic diagram of the measured power value after the energy consumption control software included in the present invention controls the TDP of the processor;

[0026] Figure 15Schematic diagram of the time series of the measured temperature after the energy consumption control software included in the present invention controls the processor TDP;

[0027] Figure 16 Block diagram showing the operating architecture of the energy consumption control software and the programming modules included in the present invention; and

[0028] Figure 17 Flowchart of the implementation steps of the energy consumption control method included in the present invention.

[0029] Explanation of the reference numerals in the drawings:

[0030] 10 Energy consumption control system

[0031] 20 Electronic device

[0032] 30 Basic input / output system firmware

[0033] 40 Embedded controller

[0034] 50 Processor

[0035] 100 Energy consumption control software

[0036] 110 Data collection programming module

[0037] 120 Dual-mode machine learning model programming module

[0038] 121 First machine learning model

[0039] 122 Second machine learning model

[0040] 130 Automatic control programming module

[0041] 200 Energy consumption control method

[0042] 201 - 205 Implementation steps

[0043] t Time

[0044] M1 Predicted value of the first machine learning model

[0045] T Trend term

[0046] P Period term

[0047] H Holiday term

[0048] R Regression term

[0049] e Noise term Detailed implementation manner

[0050] The present invention can be fully understood from the following embodiments, enabling those skilled in the art to implement it accordingly. However, the implementation of the present invention is not limited to the following embodiments; the drawings of the present invention do not include limitations on size, dimensions, and scale. The actual size, dimensions, and scale during the implementation of the present invention are not limited by the drawings of the present invention.

[0051] The term "preferably" used herein is non-exclusive and should be understood as "preferably but not limited to". Any step described or recited in any specification or claim can be executed in any order, not limited to the order described in the claim. The scope of the present invention should be determined only by the appended claims and their equivalents, and should not be determined by the embodiments of the implementation examples. When the term "comprising" and its variations appear in the specification and claims, it is an open-ended term without a restrictive meaning and does not exclude other features or steps.

[0052] Figure 3 Disclosed is a schematic diagram of the system architecture of the energy consumption control system included in the present invention; the energy consumption control method proposed by the present invention is preferably implemented in the form of the energy consumption control software 100, i.e., the energy consumption control computer program product, by being installed on the energy consumption control system 10. The energy consumption control system 10 at least includes an electronic device 20 and the energy consumption control software 100 that independently executes under the operating system (OS) installed on the electronic device 20. The electronic device 20 internally at least includes the basic input / output system (BIOS) firmware 30, the embedded controller (EC) 40, and the processor 50.

[0053] The electronic device 20 widely covers various devices and apparatuses including electronic components and circuits, at least including but not limited to: industrial computer devices and consumer electronic devices used in industrial and commercial application fields. Among them, industrial computer devices include but are not limited to: point of service (POS), point of sale (POS), touch computer (PPC), multimedia data terminal (Kiosk), embedded computer (box PC), ticket vending machine, vending machine, automated teller machine (ATM), network load balancing device, workstation, industrial computer, medical computer, medical workstation, and cloud server, etc. Consumer electronic devices include but are not limited to: mobile devices, tablet devices, laptop, and personal computer (PC). Operating systems include but are not limited to: WINDOWS, Android, Linux, and iOS, etc.

[0054] Figure 4Schematic diagram of the software architecture of the energy consumption control software included in the present invention; the energy consumption control software 100 proposed by the present invention has control authority superior to that of the BIOS firmware 30 and the embedded controller 40, and obtains control authority over the processor 50 by controlling the BIOS firmware 30 and the embedded controller 40. In an embodiment, the energy consumption control software 100 is preferably configured to include a data collection programming module 110, a dual-mode machine learning model programming module 120, and an automatic control programming module 130.

[0055] The data collection programming module 110 is configured to continuously detect and collect, at a fixed or variable sampling rate, a time series of various performance parameters of the processor 50 during operation under the operation of a specific user of the electronic device 20, such as but not limited to an administrator, on the timeline to form performance data. The performance parameters include but are not limited to: the power of the processor 50, the thermal design power (TDP), the loading, the frequency, and the temperature, etc. Among them, the power and the thermal design power can be regarded as the first performance parameters, the loading can be regarded as the second performance parameter, the frequency can be regarded as the third performance parameter, and the temperature can be regarded as the fourth performance parameter.

[0056] In an embodiment, the loading includes but is not limited to: the processor load rate, which refers to the ratio of the number of programs to be processed and being processed in the processor. The frequency refers to the current clock rate of the processor. The power includes but is not limited to: the graphics processing unit power (GPU power), the kernel power (IA power), and the overall processor power (package power), etc. Among them, the graphics processing unit power refers to the power consumption of the built-in graphics processing unit in the processor, the kernel power refers to the total power consumption of the executing processor core (Intel Architecture, IA), and the overall processor power refers to the current average maximum total power consumption of the processor.

[0057] In an embodiment, for example, under the Windows operating system, the data collection programming module 110 preferably reads the values of various performance parameters of the processor 50 by calling the GetSystemPowerStatus function provided by the Win32 API, records them at a sampling frequency of, for example, 1, 2, 3, 4, or 5 seconds, then averages them at an average period of, for example, but not limited to, 1, 2, 3, 4, or 5 seconds, and stores them as a record to reduce the impact of outlier values or pulse values, ensuring the accuracy and reliability of the sampled performance parameter values.

[0058] In an embodiment, the performance data is preferably divided into first-class performance data and second-class performance data. The first-class performance data refers to the performance data used to train and establish the dual-model machine learning model included in the dual-model machine learning model programming module 120 during the initial modeling stage. The second-class performance data is the performance data recorded by the data collection programming module 110 during the online execution stage for updating, adjusting, and calibrating the dual-model machine learning model, or the second-class performance data refers to the performance data that is newly collected and has not been used to train the dual-model machine learning model and is not included in the first-class performance data.

[0059] The initial modeling stage at least includes the following steps: First, divide the first-class performance data into a training dataset, a validation dataset, and a test dataset, and use the training dataset to train the dual-model machine learning model to perform basic learning on the training dataset. Then, verify and adjust the dual-model machine learning model after basic learning through the validation dataset, and then confirm or fine-tune the performance of the verified dual-model machine learning model with the test dataset.

[0060] The dual-model machine learning model programming module 120 is configured to perform long-term and short-term prediction of the power or TDP of the processor 50. After being trained in the modeling stage, the dual-model machine learning model included in the dual-model machine learning model programming module 120 can learn the performance data composed of the time series of performance parameters to learn and analyze, for example, but not limited to, user habits, usage patterns, or usage cycles, etc., and accordingly formulate long-term and short-term performance control strategies, detect through the system, adjust the optimization parameters, and use interrupt processing to quickly respond to instantaneous system changes, reduce power waste during idle time, and improve the problem of system energy waste to achieve the best optimization of the user system performance.

[0061] Figure 5Schematic diagram revealing the time series of the power change of the internal processor of the electronic device included in the present invention over consecutive days; due to the time series of the performance parameters of the processor 50, when observed from a relatively long time scale, such as weekly, monthly, quarterly, or even annually, it actually exhibits certain regular and logical characteristics, and even has characteristics such as segment characteristics, periodic characteristics, logical characteristics, and local growth trends, as Figure 5 disclosed.

[0062] Figure 6 Schematic diagram revealing the time series of the power change of the internal processor of the electronic device included in the present invention within 24 hours; however, if observed from a relatively short time scale, such as hourly, every 15 minutes, every 10 minutes, or even every 5 minutes, the time series of the performance parameters may exhibit relatively irregular and illogical characteristics full of sudden or irregular events, as Figure 6 disclosed.

[0063] The dual-mode machine learning model programming module 120 is preferably integrated by combining two machine learning models respectively used for time series prediction under different time scales into a single machine learning model, preferably including a first machine learning model 121 responsible for long-term prediction tasks and a second machine learning model 122 responsible for short-term prediction tasks. Among them, the first machine learning model 121 will learn the long-term time series of various performance parameters of the processor 50. The first machine learning model 121 can learn data such as trends, cycles, seasons, holidays, exogenous variables, and noises hidden in the time series of performance parameters. After training, the main task of the first machine learning model 121 is to predict the long-term usage status of the electronic device 20. For example, it predicts the processor load of the next day to configure the best performance parameters for the processor 50.

[0064] The first machine learning model 121 includes but is not limited to: artificial neural network (ANN) model, deep neural network (DNN) model, convolutional neural network (CNN) model, multi-layer perceptron (MLP) model, moving average (MA) model, exponential smoothing (EA) model, autoregressive (AR) model, vector autoregressive (VAR) model, autoregressive moving average (ARMA) model, integrated moving average autoregressive (ARIMA) model, regression tree model, growth model, latent growth curve model, latent growth model, Fourier model, Fourier series model, trend model, prophet model, and any combination thereof.

[0065] In a certain embodiment, the first machine learning model 121 can preferably be presented by formula (1):

[0066] M1(t) = T(t) + P(t) + H(t) + R(t) + e(t) (1)

[0067] where t is time, M1 is the predicted value of the first machine learning model 121 at time t, T is the trend term, P is the periodic term and includes seasonal factors, H is the holiday term, R is the regression term and includes exogenous variables, and e is the noise term. For example, the trend term of T is preferably selected as, for example, a growth model or an ARIMA model, and the periodic term of P is preferably selected as, for example, a Fourier model or an ANN model.

[0068] The first machine learning model 121 can be regarded as a composite time series prediction model, which is formed by adding various time series prediction models, trend prediction models and periodic prediction models. The prediction ability of the first machine learning model 121 for time series is better than that of any single traditional time series prediction model. Moreover, due to the addition of seasonal, holiday and trend terms in the first machine learning model 121, it has good prediction ability for time series with segmented characteristics, logical characteristics and growth trends.

[0069] Figure 7 A schematic diagram showing the power change of the processor predicted by the first machine learning model included in the present invention; preferably, after the first machine learning model 121 included in the present invention is trained, it will be used to predict and evaluate the usage of the processor 50 the next day, as well as the usage of the user, so as to be used as a reference value for setting the basic TDP for the processor 50, as Figure 7 disclosed.

[0070] Furthermore, the first machine learning model 121 can also predict the peak and off-peak periods of the next day, so that the system can formulate different control strategies for the peak and off-peak periods respectively. For example, if the first machine learning model 121 predicts that the period from early morning to noon of the user is the off-peak time, a power-saving strategy can be formulated accordingly, and the parameters can be set to the most power-saving. After noon is the peak period, the parameters can be set to achieve the best performance of the system, as Figure 7 disclosed.

[0071] Figure 8 A schematic diagram showing the hourly power change of the processor predicted by the first machine learning model included in the present invention; the first machine learning model 121 can also predict the hourly power change of the processor 50 the next day, as Figure 8 disclosed.

[0072] However, when observed from a relatively short time scale, the processor 50 often encounters sudden and unknown situations. For example, the user may temporarily need to execute a task that requires a large amount of computation, causing the load on the processor 50 to suddenly increase. However, due to the short cycle of such short-term unexpected events or burst events, their behavior patterns in a relatively long-term time series are similar to short-term pulses. Due to the limitation of the resolution of the time series prediction model, the problem of gradient disappearance will occur. Therefore, it is difficult for the first machine learning model 121 to predict such short-term unexpected events. Therefore, in order to supplement the first machine learning model 121, the second machine learning model 122 is further configured in the dual-mode machine learning model programming module 120 of the present invention.

[0073] The main task of the second machine learning model 122 is to predict the short-term usage status of the electronic device 20. For example, but not limited to, predicting the system usage rate of the next moment every 15 minutes to supplement the prediction deficiency of the first machine learning model 121; the second machine learning model 122 includes but is not limited to: one of a neural network (ANN) model, a deep neural network (DNN) model, a convolutional neural network (CNN) model, a recurrent neural network (RNN) model, a gated recurrent unit (GRU) model, a long short-term memory (LSTM) model, a multi-layer perceptron (MLP) model, and their combinations.

[0074] In an embodiment, the second machine learning model 122 preferably selects a long short-term memory model. The long short-term memory model has special memory cells and gate mechanisms, which can solve the problem of gradient disappearance in the time series prediction model. Therefore, it has the ability to capture and predict sudden or irregular events in short-term usage scenarios. When the first machine learning model 121 and the second machine learning model 122 are combined to form the dual-mode machine learning model programming module 120, complementary effects can be produced, enabling the dual-mode machine learning model programming module 120 to have the ability to make a relatively good comprehensive prediction of the long-term and short-term usage conditions of the processor 50 and calculate appropriate power or TDP prediction values.

[0075] After the second machine learning model 122 is put into operation, it mainly makes a short-term prediction of the processor 50 power when there is a large gap between the power prediction value of the first machine learning model 121 and the actually detected power detection value, so as to enhance the sensitivity of the dual-mode machine learning model programming module 120 to time and power changes and avoid power changes caused by temporary unexpected work.

[0076] Figure 9 Schematic diagram of the power change of the first unit time of the processor predicted by the second machine learning model included in the present invention for the next day; Figure 10Schematic diagram of the second unit-time power change of the processor predicted by the second machine learning model included in the present invention; after the second machine learning model 122 is put into operation, it can predict the power of the processor 50 based on a relatively long unit time, such as every 15 minutes, as Figure 9 disclosed, or it can also predict the power of the processor 50 based on a shorter unit time, such as every 10 minutes, to enhance the sensitivity of the model, as Figure 10 disclosed.

[0077] The trained first and second machine learning models 121 and 122 will be combined into a single dual-mode machine learning model programming module 120 and commanded and executed by the energy consumption control software 100. The first and second machine learning models 121 and 122 also have a self-learning or reinforcement learning mechanism based on self-feedback. After the energy consumption control software 100 is put into operation, during the use of the electronic device 20, through the data collection programming module 110, it can continuously record the second type of performance data corresponding to the electronic device 20 under the operation of the user, and regularly use the recorded second type of performance data to retrain the first and second machine learning models 121 and 122, so as to update, adjust and calibrate the first and second machine learning models 121 and 122.

[0078] After being updated with the second performance data, the first and second machine learning models 121 and 122 can well adapt to the small changes in the user's usage habits of the electronic device 20, continuously learn and dynamically perform self-tuning. Therefore, after the energy consumption control software 100 included in the present invention is executed for a period of time, it will become more and more accurate. It can also calculate the optimal energy-saving configuration for the electronic device 20 for the system to call, or find out the energy configuration of each user and store it as a user profile to establish personalized energy-saving settings for the user to call.

[0079] The dual-mode machine learning model programming module 120 proposed by the present invention integrates two machine learning models respectively used for time series prediction at different time scales, so it has the ability to learn and predict the performance parameters of the processor 50 at longer and shorter time scales respectively, and can quite effectively break through the prediction ability of traditional time series prediction models.

[0080] After the dual-mode machine learning model programming module 120 completes the prediction of the power or TDP of the processor 50, although the TDP prediction value can be directly set for the processor 50, in order to further optimize the power and temperature of the processor 50 and find the optimized power between balancing the load and frequency, the automatic control programming module 130 included in the energy consumption control software 100 will further apply the fuzzy feedback control mechanism to dynamically and real-time adjust the TDP prediction value.

[0081] The automatic control programming module 130 is configured to perform automatic feedback control on the power or TDP of the processor 50 based on the fuzzy feedback control mechanism. When the automatic control programming module 130 actually controls the TDP of the processor 50, it will, based on the detection values of the frequency and load of the processor 50 by the data collection programming module 110, under the condition of balancing the load and frequency, apply the fuzzy feedback control mechanism to fine-tune the TDP prediction value calculated by the dual-mode machine learning model programming module 120 to adjust the optimal TDP value that can balance the load and frequency simultaneously, taking into account both frequency and load while saving energy, without sacrificing performance and affecting the user experience, and at the same time reducing the temperature of the processor 50. Moreover, the fuzzy feedback control can also improve the system control stability.

[0082] Figure 11 Discloses a schematic diagram of the fuzzy relationship between the frequency and load of the processor included in the present invention; in an embodiment, after measurement, it is known that the frequency of the processor 50 ranges from 800 to 2600 MHz, and the load ranges from 0 to 100%, and the relationship between the two is as Figure 11 shown by the columns in. Preferably, if the frequency and load data of the processor are scored, for example but not limited to, into 10 equal parts, and then apply, for example but not limited to, triangular or trapezoidal membership functions, the input-output fuzzy set graph composed of frequency and load and the corresponding function of the graph can be defined, and then, for example but not limited to, the centroid method is used to solve, and the corresponding control point coordinate position on the fuzzy set graph under a certain frequency and a certain load condition can be found.

[0083] According to the actual measurement results, the operating effect is poor when the load is higher and the frequency is lower, and vice versa, unnecessary energy consumption is generated when the load is lower and the frequency is higher. Therefore, the optimal control point preferably should be as close as possible to, for example but not limited to, the centroid position of the fuzzy set graph. In this embodiment, since the membership function is preferably of a symmetric form, the coordinates of the optimal control point preferably fall on the position of the fuzzy set graph where the frequency is 1800 MHz and the load is 50%, but the coordinate position of the optimal control point on the fuzzy set graph will vary depending on the type of membership function used.

[0084] In the fuzzy feedback control process of the present invention, a TDP prediction value is first set for the processor 50, and then subsequent changes in the frequency and load of the processor 50 are detected. Next, the coordinate position of the control point on the fuzzy set graph is solved based on the frequency and load, and the TDP prediction value is finely adjusted again according to the coordinate position of the control point on the fuzzy set graph until the coordinate position of the control point moves to or as close as possible to the optimal control point.

[0085] For example, in the fuzzy feedback control process of the present invention, depending on whether the control point falls on the left or right side of the optimal control point, a fine-tuning value is added to or subtracted from the TDP prediction value respectively until the coordinate position of the control point moves to or as close as possible to the optimal control point. The fine-tuning value is preferably in the range of ±2% of the TDP prediction value.

[0086] For example, under normal circumstances, the TDP is finely adjusted with a fine-tuning value of ±0.5% each time, but in case of an emergency, the TDP is adjusted with a fine-tuning value of ±2% each time to converge to the optimal state faster and quickly respond to emergencies.

[0087] The control point usually fluctuates back and forth on the left and right sides of the optimal control point. Through the continuous intervention of the automatic control programming module 130 of the present invention, more stable system control is obtained, and the processor 50 is automatically maintained to operate at the optimized power.

[0088] The present invention builds a dual-mode machine learning model in the dual-mode machine learning model programming module 120, predicts the usage cycle at each time point through the dual-mode machine learning model, can perform different unit time slicing predictions through data analysis, effectively improves the overall prediction ability, and uses the automatic control parameters based on the fuzzy feedback control mechanism in the automatic control programming module 130, without manual operation, and greatly improves the usage efficiency and convenience.

[0089] Figure 12 Reveal a schematic diagram of the power-frequency-load relationship after optimizing the TDP prediction value by the fuzzy control programming module included in the present invention; Figure 13 Reveal a schematic diagram of the temperature-frequency-load relationship after optimizing the TDP prediction value by the fuzzy control programming module included in the present invention; From Figure 12 And Figure 13 From the revealed relationship, it can be seen that before the automatic control programming module 130 intervenes to optimize the TDP prediction value, the power of the processor 50 is approximately around 7.75 watts, and the temperature is around 100°C. After the automatic control programming module 130 intervenes to optimize the TDP prediction value, the overall power of the processor 50 drops to around 6.5 watts, and the temperature drops to around 80°C, effectively controlling both the power and temperature of the processor 50 within a good range.

[0090] Figure 14 A time series schematic diagram of the measured power value after the energy consumption control software included in the present invention controls the processor TDP for energy consumption; Figure 15 A time series schematic diagram of the measured temperature value after the energy consumption control software included in the present invention controls the processor TDP for energy consumption; When the energy consumption control software 100 included in the present invention intervenes to control the power or TDP of the processor 50 of the electronic device 20, from Figure 14 and Figure 15 the measured time series diagrams of power and temperature, it can be seen that after the electronic device 20 has been used for a period of time, as the temperature rises, the power consumed also gradually increases, resulting in poor system performance. However, when the energy consumption control software 100 intervenes, the processor 50 begins to obtain optimized control. Not only does the temperature performance decrease compared to before, but the power consumption can also achieve better performance. Over time, a considerable amount of power consumption can be saved.

[0091] Figure 16 A block schematic diagram revealing the operation architecture of the energy consumption control software and the programming modules included in the present invention; In summary, after the energy consumption control software 100 of the present invention is executed online, the data collection programming module 110 will actively detect and record various performance parameters of the processor 50, and transmit them to the dual-mode machine learning model programming module 120 trained based on the time series of the performance parameters. The dual-mode machine learning model programming module 120 will actively predict the power or TDP of the processor 50 and generate the predicted value of the power or TDP of the processor. Then, the automatic control programming module 130 will implement fuzzy feedback control based on the power prediction value, and according to the feedback of frequency and load, find the optimized power or TDP that can balance load and frequency, and accordingly actually control and fine-tune the power of the processor 50 or limit the TDP of the processor 50. The present invention accordingly performs full-automatic or semi-automatic energy consumption control on the electronic device 20 to reduce the overall energy consumption of the electronic device.

[0092] Figure 17Flowchart showing the implementation steps of the energy consumption control method included in the present invention; the energy consumption control method 200 included in the present invention preferably includes but is not limited to the following steps: performing the following steps (step 201) by a processor and system firmware: continuously detecting and collecting first-type performance data of the processor, the first-type performance data including a first performance parameter, a second performance parameter, and a third performance parameter (step 202); selectively executing a first machine learning model included in a dual-mode machine learning model to predict a long-term first performance parameter of the processor over a relatively long period as the first performance parameter (step 203); selectively executing a second machine learning model included in the dual-mode machine learning model to predict a short-term first performance parameter of the processor over a relatively short period as the first performance parameter (step 204); and implementing a fuzzy feedback control mechanism to add a fine-tuning value to the first performance parameter according to the detected second performance parameter and third performance parameter to adjust the first performance parameter (step 205).

[0093] The above embodiments of the present invention can be combined or replaced with each other arbitrarily, thereby deriving more implementation manners, but all are within the scope of protection of the present invention. Further, more embodiments of the present invention are provided as follows:

[0094] Embodiment 1: An energy consumption control method, including: performing the following steps by a processor: continuously detecting and collecting performance data of the processor, the performance data including a first performance parameter, a second performance parameter, and a third performance parameter; executing a dual-mode machine learning model to predict the first performance parameter of the processor based on the first-type performance data; and implementing a fuzzy feedback control mechanism to adjust the first performance parameter according to the detected second performance parameter and third performance parameter.

[0095] Embodiment 2: The energy consumption control method according to Embodiment 1, further including: selectively executing a first machine learning model included in the dual-mode machine learning model to predict a first-period first performance parameter of the processor in a first period as the first performance parameter; selectively executing a second machine learning model included in the dual-mode machine learning model to predict a second-period first performance parameter of the processor in a second period as the first performance parameter, where the first period and the second period have different time lengths; and implementing the fuzzy feedback control mechanism to add a fine-tuning value to the first performance parameter according to the detected second performance parameter and third performance parameter to adjust the first performance parameter.

[0096] Embodiment 3: The energy consumption control method as described in Embodiment 2, wherein the first machine learning model includes one of a neural network model, a deep neural network model, a convolutional neural network model, a multi-layer perceptron model, a moving average model, an exponential smoothing model, an autoregressive model, a vector autoregressive model, an autoregressive moving average model, an integrated moving average autoregressive model, a regression tree model, a growth model, a latent growth curve model, a latent growth model, a Fourier model, a Fourier series model, a trend model, a Prophet model, and combinations thereof.

[0097] Embodiment 4: The energy consumption control method as described in Embodiment 2, wherein the second machine learning model is selected from one of a neural network model, a deep neural network model, a convolutional neural network model, a recurrent neural network model, a gated recurrent unit model, a long short-term memory model, a multi-layer perceptron model, and combinations thereof.

[0098] Embodiment 5: The energy consumption control method as described in Embodiment 2, wherein the fine-tuning value is between ±2% of the value of the first performance parameter.

[0099] Embodiment 6: The energy consumption control method as described in Embodiment 1, wherein the first performance parameter includes one of power, thermal design power, and combinations thereof, and the second performance parameter and the third performance parameter are selected from one of load and frequency.

[0100] Embodiment 7: The energy consumption control method as described in Embodiment 1, wherein the performance data is composed of time series of the first performance parameter, the second performance parameter, and the third performance parameter.

[0101] Embodiment 8: An energy consumption control computer program product, which is loaded and executed by a processor and includes: a data collection programming module configured to continuously detect and collect the performance data of the processor, the performance data including a first performance parameter, a second performance parameter, and a third performance parameter; a dual-mode machine learning model programming module configured to execute a dual-mode machine learning model to predict the first performance parameter of the processor based on the first performance data; and an automatic control programming module configured to implement a fuzzy feedback control mechanism to adjust the first performance parameter according to the detected second performance parameter and the third performance parameter.

[0102] Embodiment 9: The energy consumption control computer program product as described in Embodiment 8, wherein the dual-mode machine learning model programming module further includes: a first machine learning model configured to selectively predict the first performance parameter of the first period of the processor in the first period as the first performance parameter; and a second machine learning model configured to selectively predict the first performance parameter of the second period of the processor in the second period as the first performance parameter, wherein the first period and the second period have different time lengths.

[0103] Example 10: The energy consumption control computer program product as described in Example 8, wherein the dual-mode machine learning model programming module further comprises: a first machine learning model configured to predict a first performance parameter of the processor in a first period; and a second machine learning model configured to predict a first performance parameter of the processor in a second period and adjust the first performance parameter in the first period based on the first performance parameter in the second period as the first performance parameter, wherein the first period and the second period have different time lengths.

[0104] Example 11: The energy consumption control computer program product as described in Example 10, wherein the dual-mode machine learning model integrates the first machine learning model and the second machine learning model into a single machine learning model.

[0105] Example 12: An energy consumption control system, comprising: an electronic device including a processor configured to: continuously detect and collect performance data of the processor, the performance data including a first performance parameter, a second performance parameter, and a third performance parameter; execute a dual-mode machine learning model to predict the first performance parameter of the processor based on the first type of performance data; and implement a fuzzy feedback control mechanism to adjust the first performance parameter according to the detected second performance parameter and the third performance parameter.

[0106] Example 13: The energy consumption control system as described in Example 12, wherein the processor is configured to: selectively execute the first machine learning model included in the dual-mode machine learning model to predict a first performance parameter of the processor in a first period as the first performance parameter; selectively execute the second machine learning model included in the dual-mode machine learning model to predict a first performance parameter of the processor in a second period as the first performance parameter, wherein the first period and the second period have different time lengths; and implement the fuzzy feedback control mechanism to add a fine-tuning value to the first performance parameter according to the detected second performance parameter and the third performance parameter to adjust the first performance parameter.

[0107] The embodiments of the present invention can be combined or replaced with each other arbitrarily, thereby deriving more embodiments, but all are within the scope of protection of the present invention. The scope of protection of the present invention is defined by the claims of the present invention application.

Claims

1. A method for controlling energy consumption, comprising: The following steps are performed by the processor: Continuously detecting and collecting performance data of the processor, the performance data comprising a first performance parameter, a second performance parameter, and a third performance parameter; executing a bimodal machine learning model to predict the first performance parameter of the processor based on the first performance data; and A fuzzy feedback control mechanism is implemented to adjust the first performance parameter according to the detected second performance parameter and the third performance parameter.

2. The energy consumption control method according to claim 1, further comprising: selectively executing a first machine learning model included in the dual-mode machine learning model to predict a first performance parameter of the processor in a first period of a first period as the first performance parameter; selectively executing a second machine learning model included in the dual-mode machine learning model to predict a first performance parameter of the processor in a second period as the first performance parameter, wherein the first period and the second period have different time lengths; as well as The fuzzy feedback control mechanism is implemented to add a fine-tuning value to the first performance parameter according to the detected second performance parameter and the third performance parameter to adjust the first performance parameter.

3. The energy consumption control method according to claim 2, wherein the first machine learning model comprises a quasi-neural network model, a deep neural network model, a convolutional neural network model, a multi-layer perception model, a moving average model, an exponential smoothing model, an autoregressive model, a vector autoregressive model, an autoregressive moving average model, an integrated moving average autoregressive model, a regression tree model, a growth model, a potential growth curve model, a potential growth model, a Fourier model, a Fourier series model, a trend model, a prophet model and a combination thereof.

4. The energy consumption control method according to claim 2, wherein the second machine learning model is selected from a quasi-neural network model, a deep neural network model, a convolutional neural network model, a recursive neural network model, a gated recurrent unit model, a long short-term storage model, a multi-layer perception model and a combination thereof. 5 . The energy consumption control method according to claim 2 , wherein the fine-tuning value is between ±2% of the value of the first performance parameter. 6 . The energy consumption control method according to claim 1 , wherein the first performance parameter comprises one of power, thermal design power and a combination thereof, and the second performance parameter and the third performance parameter are selected from one of load and frequency. 7 . The energy consumption control method according to claim 1 , wherein the performance data consists of a time series of the first performance parameter, the second performance parameter and the third performance parameter.

8. An energy consumption control computer program product, which is loaded and executed by a processor and comprises: a data collection programming module configured to continuously detect and collect performance data of the processor, the performance data comprising a first performance parameter, a second performance parameter, and a third performance parameter; a dual-mode machine learning model programming module configured to execute a dual-mode machine learning model to predict the first performance parameter of the processor based on the first representation data; as well as The automatic control programming module is configured to implement a fuzzy feedback control mechanism to adjust the first performance parameter according to the detected second performance parameter and the third performance parameter.

9. The energy consumption control computer program product according to claim 8, wherein the dual-mode machine learning model programming module further comprises: a first machine learning model configured to selectively predict a first performance parameter of the processor at a first period in a first period as the first performance parameter; and A second machine learning model is configured to selectively predict a first performance parameter of the processor in a second period as the first performance parameter, wherein the first period and the second period have different time lengths.

10. The energy consumption control computer program product according to claim 8, wherein the dual-mode machine learning model programming module further comprises: a first machine learning model configured to predict a first performance parameter of the processor at a first time period in a first time period; and A second machine learning model is configured to predict a first performance parameter of the processor in a second period of time, and adjust the first performance parameter of the first period of time based on the first performance parameter of the second period as the first performance parameter, wherein the first period of time and the second period of time have different time lengths.

11. The energy consumption control computer program product according to claim 10, wherein the dual-mode machine learning model integrates the first machine learning model and the second machine learning model into a single machine learning model.

12. An energy consumption control system, comprising: An electronic device comprising a processor configured to: Continuously detecting and collecting performance data of the processor, the performance data comprising a first performance parameter, a second performance parameter, and a third performance parameter; executing a bimodal machine learning model to predict the first performance parameter of the processor based on the first performance data; and A fuzzy feedback control mechanism is implemented to adjust the first performance parameter according to the detected second performance parameter and the third performance parameter.

13. The energy consumption control system according to claim 12, wherein the processor is configured to: selectively executing a first machine learning model included in the dual-mode machine learning model to predict a first performance parameter of the processor in a first period of a first period as the first performance parameter; selectively executing a second machine learning model included in the dual-mode machine learning model to predict a first performance parameter of the processor in a second period as the first performance parameter, wherein the first period and the second period have different time lengths; as well as The fuzzy feedback control mechanism is implemented to add a fine-tuning value to the first performance parameter according to the detected second performance parameter and the third performance parameter to adjust the first performance parameter.

Images