CPU power distribution method for PC based on AI cluster system

Abstract

The invention relates to the technical field of power distribution and power saving of a power system, in particular to an AI cluster system-based CPU (central processing unit) power distribution method for a PC (personal computer), which comprises the following steps of: step 1, collecting long-short period data, step 2, analyzing long-period and short-period habits, step 3, dynamically adjusting habit weights, step 4, detecting and responding to abnormity, and step 5, distributing power to the PC. Continuously optimizing and updating; according to the scheme, long-period and short-period data are comprehensively collected and finely analyzed, so that the accuracy and the dynamic performance of CPU power distribution are realized. Long-period data can capture long-term trends, short-period data can reflect short-term changes, strategies are adjusted in time to adapt to dynamic requirements, time sequence analysis, clustering algorithms and the like are used for deep mining and classification, a simulation environment is constructed to determine an initial scheme, CPU operation and energy utilization efficiency is improved in the mode, different user scene requirements are met, and user experience is improved. And smooth, stable and efficient use experience is provided.

Description

Technical Field

[0001] The present invention relates to the technical field of power distribution and power saving in power systems, and in particular to a method for distributing power to a CPU for a PC based on an AI cluster system. Background Art

[0002] With the rapid development of information technology, personal computers (PCs) are playing an increasingly important role in people's work, study, and entertainment. At the same time, PC performance requirements are becoming increasingly diverse and complex, especially in scenarios such as processing large amounts of data, running complex algorithms, and multitasking in parallel. The central processing unit (CPU), as a core component in the PC's hardware architecture, has a performance and energy consumption that directly impacts the overall system's operating efficiency and power consumption.

[0003] However, the existing AI-based power distribution methods used in the PC field have difficulty in accurately collecting and analyzing data related to user usage habits. This may cause game users to experience game lag and frame drops during entertainment, affecting the gaming experience. When used by office users, the CPU power consumption is too high, resulting in unnecessary waste of energy.

[0004] Therefore, the present invention proposes a method for allocating CPU power for PCs based on an AI cluster system. Summary of the Invention

[0005] Technical Problem Solved: Existing AI-based power distribution methods in the PC field have difficulty accurately collecting and analyzing data related to user usage habits. This may cause game users to experience game lag and frame drops during entertainment, affecting the gaming experience. When used by office users, the CPU power consumption is too high, resulting in unnecessary energy waste.

[0006] In view of the shortcomings of the existing technology, the present invention provides a CPU power distribution method for PCs based on an AI cluster system, thereby solving the technical problems mentioned in the background technology.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for distributing power to a CPU for a PC based on an AI cluster system includes the following steps: Step 1: Collect long- and short-cycle data. Long-cycle data is collected to record users' monthly usage of various applications at a granularity of minutes and stored in a monthly data warehouse. This data captures long-term usage patterns and overall trends, providing a basis for formulating basic power allocation strategies and is suitable for addressing fixed routines. Short-cycle data is collected to record the same type of data every day at a granularity of seconds and stored in a daily data warehouse. This data can reflect short-term real-time changes and special circumstances, allowing for timely adjustments to power allocation strategies based on long-cycle data. Step 2: Long- and short-term habit analysis. Combining long- and short-term habit analysis enables comprehensive optimization of power distribution, ensuring system stability and efficiency in the long term while flexibly adapting to changes and special needs in the short term. S21: Long-term habit analysis By collecting various application usage data every month, we first carefully categorize them by application type, usage time period, and other dimensions. Then, we use time series analysis to observe the trend changes of various application usage within the same time period in different months. The formula is: ; in, is the value at the current time point, is the number of periods of the moving average. Months with similar usage patterns are grouped together using a clustering algorithm. K data points are randomly selected as initial cluster centers. The distance from each data point to the K cluster centers is calculated. Each data point is assigned to the cluster to which the nearest cluster center belongs. The center of each cluster is recalculated. The above steps are repeated until the cluster center no longer changes or the specified number of iterations is reached. When calculating the distance from a data point to a cluster center, the Euclidean distance formula is often used: ; in, and There are two data points. and They are in the The values ​​in the dimensions, is the dimension of the data; A simulation environment is constructed to simulate different CPU power allocation schemes. The optimal initial scheme is determined through performance testing and energy consumption evaluation. The performance evaluation formula is: response time = task completion time - task submission time. The energy consumption evaluation formula is: total energy consumption = CPU energy consumption + energy consumption of other hardware (such as memory, hard disk, etc.). CPU energy consumption is calculated based on parameters such as the CPU's operating frequency and voltage.

[0008] S22: Short-term habit analysis Compare the usage data of various applications on the current day with the data of the previous few days one by one, carefully calculate the difference in usage time and resource usage of the same application on the current day and the previous few days, and closely observe the trend of the difference to determine whether there is any abnormal fluctuation. The resource usage change rate of an application on the current day is , the resource usage on that day is , the resource usage the previous day was , and its calculation formula is: ; Based on these difference data, the short-term usage trend is calculated using the following formula: Slope (Trend) = ; If the usage time of an application shows an upward trend for several consecutive days, it is considered to be an increasing usage trend. Based on the calculated short-term trend, the current CPU power allocation plan is fine-tuned. If usage is increasing, the CPU power allocation for the corresponding application or time period is appropriately increased. Conversely, if usage decreases, the power allocation is reduced. After fine-tuning the power allocation, changes in system performance and energy consumption are monitored in real time. The power adjustment ratio is calculated as follows: When usage time increases, the increased power allocation ratio = (usage time on the current day - usage time on the previous day) / usage time on the previous day * preset ratio coefficient. This feedback information is used to further optimize short-term habit analysis and power allocation adjustment strategies. Step 3: Dynamically adjust habit weights. Changes in usage habits can be promptly captured, ensuring that electricity distribution aligns with actual demand. Long-term data reflects stable habits, while short-term data reflects recent, temporary demand. Dynamic weighting is used to balance the two. Initially, due to insufficient long-term data, the short-term weight is set to 100%. As usage increases, the long-term weight is introduced and adjusted weekly according to the rules until the long-term weight is reduced to 30%. During the transition phase, electricity distribution will be guided primarily by short-term data, while long-term data is accumulated to improve the weighting mechanism. Step 4: Anomaly detection and response can promptly detect significant changes in user habits and avoid allocating power based on inaccurate historical data. The data for each day of the previous week must be stored and averaged. The difference between the average data for the current day and the previous week is calculated using the formula (Difference = (Daily App Usage Duration - Average App Usage Duration for the Previous Week) / Average App Usage Duration for the Previous Week). If the difference exceeds 50%, anomaly detection is triggered. New data is then quickly collected and analyzed within three days to re-determine the CPU power allocation plan. Emergency relearning and weight adjustments are then used to quickly adapt to the new model, ensuring system performance and power efficiency. Step 5: Continuously optimize and update to adapt to hardware changes. As hardware updates, adjust parameters and thresholds to leverage the advantages of new hardware and meet new user needs. Establish a quarterly evaluation mechanism, selecting system operational data (such as CPU power usage and application usage), and evaluate using the performance improvement ratio = (post-optimization performance index - pre-optimization performance index) / pre-optimization performance index. User needs are constantly evolving, and continuous optimization ensures that power allocation plans meet the latest requirements, improves system performance and efficiency, and ensures more accurate and reasonable power allocation.

[0009] In one possible implementation, the collection of long- and short-term data captures users' long-term usage patterns and overall trends. By recording users' monthly usage of various applications at a minute-by-minute granularity, including application name, start and end time, and resource usage, and storing this data in a monthly data warehouse, it is possible to identify the overall usage frequency and duration of certain applications within each month, providing an important basis for formulating a basic and relatively stable power allocation strategy. By recording the same type of data at a granularity of seconds every day and storing it in a daily data warehouse, we can promptly detect temporary changes, such as sudden increases in user demand for specific applications on a given day. This is crucial for timely adjustments to power distribution strategies based on long-term data, enabling power distribution to better adapt to short-term dynamic demands.

[0010] In one possible implementation, in the dynamic adjustment of the habit weight, preliminary CPU power allocation is first performed based on short-term data. The short-term weight can be set to 100%. As the usage time increases, the long-term weight is gradually introduced. When the long-term data is available, the weight is dynamically adjusted according to the set rules (every week, if the short-term data is stable and different from the long-term trend, the short-term weight will be increased by 10% and the long-term weight will be reduced by 10% accordingly) until the long-term weight is reduced to 30%.

[0011] In one possible implementation, the anomaly detection and response can quickly adapt to new usage patterns by initiating emergency relearning and adjusting weights, ensuring system performance and power utilization efficiency, collecting new data and redetermining allocation plans in a short period of time, which helps to improve the flexibility and adaptability of the system.

[0012] In one possible implementation, in the anomaly detection and response, after obtaining the data for each day, the difference between the data and the average data of the previous week is calculated. The calculation formula is: Difference = (the duration of use of the application on that day - the average duration of use of the application in the previous week) / the average duration of use of the application in the previous week. If the difference exceeds 50%, anomaly detection is triggered. Within the next three days, new data is quickly collected for analysis and the CPU power allocation plan is re-determined.

[0013] In one possible implementation, a regular evaluation mechanism is established during the continuous optimization and update, and system operation data over a period of time is selected every quarter, including CPU power usage and application usage. The calculation formula of the evaluation mechanism is: performance improvement rate = (performance index after optimization - performance index before optimization) / performance index before optimization.

[0014] In a possible implementation, during the continuous optimization and updating, continuous optimization can ensure that the power distribution plan always meets the latest needs of users, improve system performance and efficiency, and continuous improvement can make power distribution more accurate and reasonable.

[0015] Beneficial effects compared with existing technologies: 1. In this solution, the CPU power distribution is made precise and dynamic through the comprehensive collection and detailed analysis of long-cycle and short-cycle data. Long-cycle data collection can capture users' long-term usage patterns and overall trends, providing a basis for formulating a basic and stable power distribution strategy, especially for dealing with fixed routine habits. Short-cycle data collection accurately reflects users' real-time changes and special circumstances in the short term, and timely adjusts the power distribution strategy to adapt to short-term dynamic needs. Combining long-cycle and short-cycle habit analysis, using time series analysis, clustering algorithms and other means, the data is deeply mined and classified, a simulation environment is constructed to determine the optimal initial plan, and power distribution is fine-tuned according to short-term trends. At the same time, short-cycle data is relied upon in the initial stage of use, and long-cycle weights are introduced and dynamically adjusted as time accumulates until a reasonable proportion is reached. This comprehensive and dynamic power distribution method not only improves the CPU operating efficiency and energy utilization efficiency, but also meets the needs of different users in various scenarios, providing users with a smoother, more stable and efficient user experience; 2. In this solution, the stability, flexibility, and adaptability of the system are guaranteed by establishing a comprehensive anomaly detection and response mechanism and a continuous optimization and update system. The anomaly detection and response mechanism can promptly detect significant changes in user habits and avoid allocating power based on inaccurate historical data. By storing and calculating the average value of the previous week's data and comparing it with the current day's data, anomaly detection is triggered when the difference exceeds the threshold. The data is then recollected and analyzed in a short period of time to determine a new CPU power allocation plan, thereby improving the flexibility and adaptability of the system. The continuous optimization and update mechanism adapts to hardware changes and meets new user needs. System operation data, including CPU power usage and application usage, is regularly evaluated every quarter. Parameters and thresholds are adjusted based on indicators such as performance improvement rate. The accuracy and rationality of power allocation are continuously improved, the overall system performance and power utilization efficiency are improved, and the power allocation plan is ensured to always meet the latest user needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention and implement it according to the contents of the specification, the following is a detailed description of the preferred embodiments of the present invention with reference to the accompanying drawings.

[0017] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0018] Preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, the present invention can be implemented in various forms, so the present invention is not limited to the embodiments described below. The technical solution in the embodiments of the present application is to solve the problems of the above-mentioned background technology, and the overall idea is as follows: Example: Please refer to Figure 1 As shown, this embodiment introduces a method for allocating CPU power for PCs based on an AI cluster system, including the following steps: Step 1: Collection of long and short cycle data S11: Long-term data collection It can comprehensively capture the long-term usage patterns and overall trends of users. By recording the monthly usage of various applications by users at the minute level, including application name, usage start time, end time, and resource usage, and storing it in the monthly data warehouse, it can be found that the overall usage frequency and duration of certain applications in each month are regular, providing an important basis for formulating a basic and relatively stable power allocation strategy. It is particularly suitable for dealing with users' relatively fixed and regular usage habits. For example, the monthly usage frequency of an application is , the app has been used for , the number of days per month is , and its calculation formula is: ; S12: Short-cycle data collection It can accurately reflect the real-time changes and special situations of users in the short term. It records the same type of data with the granularity of seconds every day and stores it in the daily data warehouse. It can promptly detect temporary changes such as sudden high demand for specific applications on a certain day. This is very important for timely adjustment of power distribution strategies formulated based on long-term data, so that power distribution can better adapt to dynamic needs in the short term. The change rate of the usage time of an application on that day is , the usage time on that day was , the usage time the previous day was , and its calculation formula is: ; Step 2: Long-term and short-term habit analysis S21: Long-term habit analysis By collecting various application usage data every month, we first carefully categorize them by application type, usage time period, and other dimensions. Then, we use time series analysis to observe the trend changes of various application usage within the same time period in different months. The formula is: ; in, is the value at the current time point, is the number of periods of the moving average. Months with similar usage patterns are grouped together using a clustering algorithm. K data points are randomly selected as initial cluster centers. The distance from each data point to the K cluster centers is calculated. Each data point is assigned to the cluster to which the nearest cluster center belongs. The center of each cluster is recalculated. The above steps are repeated until the cluster center no longer changes or the specified number of iterations is reached. When calculating the distance from a data point to a cluster center, the Euclidean distance formula is often used: ; in, and There are two data points. and They are in the The values ​​in the dimensions, is the dimension of the data; A simulation environment is constructed to simulate different CPU power allocation schemes. The optimal initial scheme is determined through performance testing and energy consumption evaluation. The performance evaluation formula is: response time = task completion time - task submission time. The energy consumption evaluation formula is: total energy consumption = CPU energy consumption + energy consumption of other hardware (such as memory, hard disk, etc.). CPU energy consumption is calculated based on parameters such as the CPU's operating frequency and voltage.

[0019] S22: Short-term habit analysis Compare the usage data of various applications on the current day with the data of the previous few days one by one, carefully calculate the difference in usage time and resource usage of the same application on the current day and the previous few days, and closely observe the trend of the difference to determine whether there is any abnormal fluctuation. The resource usage change rate of an application on the current day is , the resource usage on that day is , the resource usage the previous day was , and its calculation formula is: ; Based on these difference data, the short-term usage trend is calculated using the following formula: Slope (Trend) =

[0020] If the usage duration of an application shows an upward trend for several consecutive days, it is considered that there is a trend of increasing usage. Based on the calculated short-term trend, the current CPU power allocation plan is fine-tuned. If usage is increasing, the CPU power allocation for the corresponding application or time period is appropriately increased. Conversely, if usage decreases, the power allocation is reduced. After fine-tuning the power allocation, changes in system performance and energy consumption are monitored in real time. The power adjustment ratio is calculated as follows: When usage duration increases, the increased power allocation ratio = (usage duration on the current day - usage duration on the previous day) / usage duration on the previous day * preset ratio coefficient. This feedback information is used to further optimize short-term habit analysis and power allocation adjustment strategies.

[0021] Step 3: Dynamically adjust habit weight S31: Usage habits change over time, and new demands, changes in work modes, or special circumstances may arise. It can promptly capture these changes, aligning power allocation with actual needs. It also balances long-term and short-term information, with long-term data reflecting stable habits and short-term data reflecting recent temporary needs, dynamically adjusting to achieve a balance between the two. It improves accuracy and efficiency, making power allocation more accurately adaptable to scenarios, improving CPU operating efficiency and energy utilization, and providing a better performance experience. It enables personalized customization to meet the unique needs of different users. It also addresses uncertainties, such as sudden changes in user behavior caused by system updates or new application installations, to ensure reasonable power allocation. S32: When it is first used, since there is not enough long-term data accumulated to determine the long-term weight, only short-term data is relied upon for preliminary CPU power allocation. The short-term weight can be set to 100%. As the usage time increases, the long-term weight is gradually introduced. When the long-term data is available, the weight is dynamically adjusted according to the set rules (every week, if the short-term data is stable and different from the long-term trend, the short-term weight is increased by 10% and the long-term weight is reduced by 10%) until the long-term weight is reduced to 30%. In this transition stage, the power allocation is mainly guided by the analysis of short-term data, while long-term data is continuously accumulated to improve the weight allocation mechanism.

[0022] Step 4: Anomaly Detection and Response S41: Promptly detect significant changes in user habits and avoid power allocation based on inaccurate historical data. By initiating emergency relearning and adjusting weights, it can quickly adapt to new usage patterns, ensuring system performance and power efficiency. Collecting new data and re-determining allocation plans in a short period of time helps improve system flexibility and adaptability. S42: It is necessary to store the data of each day of the previous week and calculate its average value. After obtaining the data of each day, calculate the difference between the data of the previous week and the average data of the previous week. The calculation formula is: Difference = (Application usage time of the current day - Average application usage time of the previous week) / Average application usage time of the previous week. If the difference exceeds 50%, anomaly detection is triggered. Within the next three days, new data is quickly collected for analysis and the CPU power allocation plan is re-determined.

[0023] Step 5: Continuous optimization and updating S51: Adapt to hardware changes. As hardware is updated, parameters and thresholds can be adjusted to better leverage the advantages of new hardware and meet new user needs. A regular evaluation mechanism will be established. System operation data over a period of time will be selected every quarter, including CPU power usage and application usage. The evaluation mechanism is calculated as follows: Performance Improvement Rate = (Performance Index after Optimization - Performance Index before Optimization) / Performance Index before Optimization. S52: User needs are constantly changing. Continuous optimization can ensure that the power distribution plan always meets the latest user needs, improve system performance and efficiency, and continuous improvement can make power distribution more accurate and reasonable, thereby improving the overall performance of the system and power utilization efficiency.

[0024] Finally, it should be noted that the above embodiments are merely examples for the purpose of illustrating the present invention and are not intended to limit the embodiments. Those skilled in the art will readily appreciate that other variations or modifications based on the above description are possible. It is not necessary and impossible to provide an exhaustive list of all embodiments. However, obvious variations or modifications arising therefrom remain within the scope of protection of the present invention.

Claims

1. A method for distributing CPU power for a PC based on an AI cluster system, characterized in that: The following steps are involved: Step 1: Collect long- and short-cycle data. Long-cycle data is collected to record users' monthly usage of various applications at a granularity of minutes and stored in a monthly data warehouse. This captures long-term usage patterns and overall trends, providing a basis for formulating basic power distribution strategies and is suitable for addressing fixed routines. Short-cycle data is collected to record the same type of data every day at a granularity of seconds and stored in a daily data warehouse to reflect short-term real-time changes and special circumstances. Step 2: Long-term and short-term habit analysis: Combine long-term and short-term habit analysis to achieve comprehensive optimization of power distribution; S21: Long-term habit analysis By collecting various application usage data every month, we first carefully categorize them by application type, usage time period, and other dimensions. Then, we use time series analysis to observe the trend changes of various application usage within the same time period in different months. The formula is: ； in, is the value at the current time point, is the number of periods of the moving average. Months with similar usage patterns are grouped together using a clustering algorithm. K data points are randomly selected as initial cluster centers. The distance from each data point to the K cluster centers is calculated. Each data point is assigned to the cluster to which the nearest cluster center belongs. The center of each cluster is recalculated. The above steps are repeated until the cluster center no longer changes or the specified number of iterations is reached. When calculating the distance from a data point to a cluster center, the Euclidean distance formula is often used: ； in, and There are two data points. and They are in the The values ​​in the dimensions, is the dimension of the data; A simulation environment was built to simulate different CPU power allocation schemes. The optimal initial scheme was determined through performance testing and energy consumption evaluation. The performance evaluation formula was: response time = task completion time - task submission time. The energy consumption evaluation formula was: total energy consumption = CPU energy consumption + other hardware energy consumption. CPU energy consumption was calculated based on parameters such as CPU operating frequency and voltage. S22: Short-term habit analysis Compare the usage data of various applications on the current day with the data of the previous few days one by one, carefully calculate the difference in usage time and resource usage of the same application on the current day and the previous few days, and closely observe the trend of the difference to determine whether there is any abnormal fluctuation. The resource usage change rate of an application on the current day is , the resource usage on that day is , the resource usage the previous day was , and its calculation formula is: ； Based on these difference data, the short-term usage trend is calculated using the following formula: Trend Slope = ; in, is the current time point, is the distance between data points, If the usage time of an application shows an upward trend for several consecutive days, it is considered that there is an increasing usage trend. Based on the calculated short-term trend, the current CPU power allocation plan is fine-tuned. If usage is increasing, the CPU power allocation of the corresponding application or time period is appropriately increased. Conversely, if usage decreases, the power allocation is reduced. After fine-tuning the power allocation, the changes in system performance and energy consumption are monitored in real time. The power adjustment ratio is calculated as follows: When usage time increases, the increased power allocation ratio = (usage time on the current day - usage time on the previous day) / usage time on the previous day * preset ratio coefficient; Step 3: Dynamically adjust habit weights. Changes in usage habits can be promptly captured, ensuring that electricity distribution meets actual demand. Long-term data reflects stable habits, while short-term data reflects recent temporary demand. Dynamic weighting is used to balance the two. Initially, due to insufficient long-term data, the short-term weight is set to 100%. As usage increases, the long-term weight is introduced and adjusted weekly according to the rules until the long-term weight is reduced to 30%. Step 4: Anomaly detection and response: This can promptly identify significant changes in user habits and avoid allocating power based on inaccurate historical data. This involves storing daily data from the previous week and calculating its average. The formula used is: Difference = (Daily app usage time - Average app usage time from the previous week) / Average app usage time from the previous week. The difference between the daily and weekly averages is calculated. If the difference exceeds 50%, anomaly detection is triggered. Step 5: Continuously optimize and update to adapt to hardware changes. As hardware updates, adjust parameters and thresholds to leverage the advantages of new hardware and meet new user needs. Establish a quarterly evaluation mechanism, select system operation data, and evaluate using the performance improvement rate = (performance indicator after optimization - performance indicator before optimization) / performance indicator before optimization.

2. The method for distributing CPU power for a PC based on an AI cluster system according to claim 1, wherein: The collection of long- and short-term data captures users' long-term usage patterns and overall trends. By recording users' monthly usage of various applications at a minute-by-minute granularity, including application name, start and end time, and resource usage, and storing this data in a monthly data warehouse, we can discover the overall usage frequency and duration of certain applications within each month. By recording the same type of data at the granularity of seconds every day and storing it in a daily data warehouse, we can promptly detect temporary changes such as sudden high demand for specific applications on a certain day. This is critical for timely adjustment of power distribution strategies formulated based on long-term data, enabling power distribution to better adapt to dynamic demands in the short term. Timely adjustment of power distribution strategies formulated based on long-term data is critical, enabling power distribution to better adapt to dynamic demands in the short term.

3. The method for distributing CPU power for a PC based on an AI cluster system according to claim 1, wherein: In the dynamic adjustment of habit weights, preliminary CPU power allocation is first performed based on short-term data. The short-term weight can be set to 100%. As usage time increases, the long-term weight is gradually introduced. After the long-term data is obtained, according to the set rules, every week, if the short-term data is stable and different from the long-term trend, the short-term weight is increased by 10% and the long-term weight is reduced by 10% accordingly, to dynamically adjust the weight until the long-term weight is reduced to 30%.

4. The method for distributing CPU power for a PC based on an AI cluster system according to claim 1, wherein: In the anomaly detection and response, by initiating emergency relearning and adjusting weights, it is possible to quickly adapt to new usage patterns, ensure system performance and power utilization efficiency, collect new data and redetermine allocation plans in a short period of time.

5. The method for distributing CPU power for a PC based on an AI cluster system according to claim 4, wherein: In the anomaly detection and response, after obtaining the data for each day, the difference between the data for the day and the average data for the previous week is calculated. The calculation formula is: Difference = (Daily App Usage Time - Average App Usage Time for the Previous Week) / Average App Usage Time for the Previous Week.

6. The method for distributing CPU power for a PC based on an AI cluster system according to claim 1, wherein: During the continuous optimization and update, a regular evaluation mechanism is established. System operation data over a period of time is selected every quarter, including CPU power usage and application usage. The calculation formula of the evaluation mechanism is: Performance Improvement Rate = (Performance Index after Optimization - Performance Index before Optimization) / Performance Index before Optimization.

Images