Solid state disk power consumption optimization method and system based on load prediction

Abstract

The invention discloses a load prediction-based solid state disk power consumption optimization method and system, and particularly relates to the technical field of solid state disks, and the method comprises the following steps: S1, data acquisition; s2, constructing and training a load prediction model; s3, load prediction; s4, power consumption optimization: optimizing the power consumption of the solid state disk by adopting a multi-objective optimized dynamic power consumption adjustment algorithm according to a load prediction result; and S5, feedback adjustment: monitoring actual power consumption and performance parameters of the solid state disk after optimization in real time, and adjusting the load prediction model and the dynamic power consumption adjustment algorithm in combination with a reinforcement learning algorithm. According to the solid state disk power consumption optimization method and system based on load prediction, a set of complete and efficient solid state disk power consumption optimization scheme is formed from load prediction model construction, dynamic power consumption adjustment algorithm design and feedback adjustment mechanism optimization; and a new technical thought and a new practice direction are expected to be provided for reducing energy consumption and improving performance of the solid state disk.

Description

Technical Field

[0001] The present application relates to the technical field of solid-state hard disks, and more specifically, to a method and system for optimizing the power consumption of solid-state hard disks based on load prediction. Background Art

[0002] With the rapid development of information technology, solid-state drives (SSDs) have gained widespread application in computers, mobile devices, data centers, and other fields due to their high read / write speeds, low power consumption, and vibration resistance. However, in large-scale data storage and processing scenarios, the power consumption of SSDs has become increasingly prominent. Excessive power consumption not only increases device operating costs but also causes increased heat generation, affecting device stability and lifespan.

[0003] Currently, most existing SSD power consumption optimization methods rely on passive adjustments based on real-time load. This means that power consumption is adjusted only after the load changes. This method has a certain lag and cannot make power consumption adjustments in advance of load changes, resulting in suboptimal power consumption optimization. Furthermore, existing load prediction models have low prediction accuracy, making it difficult to accurately grasp load trends, further impacting the effectiveness of power consumption optimization. Therefore, to address the above issues, a method and system for optimizing the power consumption of solid-state drives based on load prediction are proposed. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present application provides a method and system for optimizing the power consumption of a solid-state drive based on load prediction to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above objectives, the present application provides the following technical solution: a method for optimizing the power consumption of a solid-state drive based on load prediction, the method comprising the following steps: S1. Data Collection: Collect historical load data and corresponding power consumption data of the solid-state drive, including the number of I / O requests, data transfer volume, request response time, and request type; S2. Load Forecasting Model Construction and Training: Based on the collected historical load data, a bidirectional LSTM-Transformer load forecasting model integrating an attention mechanism is constructed. This model is then trained using the historical load data to obtain a trained load forecasting model. S3. Load Forecasting: Input the real-time collected SSD load data into the trained load forecasting model to obtain load forecast results for a preset time period in the future. S4. Power consumption optimization: Based on the load prediction results, a multi-objective optimization dynamic power consumption adjustment algorithm is used to optimize the power consumption of the solid-state drive. The dynamic power consumption adjustment algorithm includes a voltage adjustment sub-algorithm based on the load prediction results, a frequency adjustment sub-algorithm, a storage unit activation number adjustment sub-algorithm, and a cache policy adjustment sub-algorithm; S5. Feedback adjustment: Real-time monitoring of the actual power consumption and performance parameters of the SSD after optimization, and adjustment of the load prediction model and dynamic power consumption adjustment algorithm using reinforcement learning algorithms.

[0006] Furthermore, the specific process of constructing the bidirectional LSTM-Transforme load prediction model integrating the attention mechanism in S2 is as follows: S21. Use a bidirectional LSTM network to extract features from historical load data and capture the bidirectional temporal dependencies of the data. S22. Input the features output by the bidirectional LSTM network into the encoder of the Transformer model and perform weighted processing on the features through the attention mechanism to highlight important features. S23. Input the features output by the encoder into the decoder of the Transformer model to obtain the load prediction result.

[0007] Furthermore, the forward propagation and back propagation formulas of the bidirectional LSTM network are as follows: Forward propagation formula: , , , , ; Back propagation formula: , , , , ; in, 、 、 They are input gate, forget gate, and output gate respectively. is the cell state, is hidden state, 、 is the weight matrix, is the bias, superscript Indicates positive direction, superscript Indicates the reverse direction, is the sigmoid function, is element-wise multiplication, For the moment Input data of; bidirectional LSTM output features ; The attention mechanism formula is: ,in, They are query matrix, key matrix, and value matrix respectively. is the feature dimension, is the feature importance weight matrix; the decoder output layer uses linear transformation and softmax function to obtain the final prediction value: ,in, is the output weight matrix, is the output bias, is the decoder output; during the model training process, an improved loss function is used, combining the mean square error and prediction deviation penalty terms: ,in is the predicted value, is the actual value, is the penalty coefficient, and the model parameters are adjusted through the back propagation algorithm.

[0008] Furthermore, the specific steps of the voltage regulation sub-algorithm based on the load prediction result in S4 are as follows: S411. Based on the load prediction value and performance constraints, the target voltage value is calculated using the voltage optimization formula, and the voltage optimization formula is: ,in, is the load forecast value, For performance constraints, is the reference voltage, is the voltage regulation coefficient, is the performance compensation coefficient, For current performance; S412. Adjust the operating voltage of the solid-state drive to the target voltage value according to the preset voltage adjustment step size so that the voltage change rate does not exceed the voltage change rate threshold. The adjustment formula is: ,in, is the preset voltage adjustment step size, is the voltage change rate threshold, and They are Moment and The voltage at the moment, is a symbolic function.

[0009] Furthermore, the specific steps of the frequency adjustment sub-algorithm based on the load prediction result in S4 are: S421. Based on the load prediction results, the target operating frequency is calculated using the frequency optimization formula. The frequency optimization formula is: ,in, is the load prediction result, is the target operating frequency, is the reference frequency, is the reference load, is the frequency adjustment coefficient, is the energy consumption coefficient, is the current power consumption, For optimal power consumption; S422. Use a smooth transition method to adjust the current operating frequency of the solid-state drive to the target operating frequency, and the frequency change formula is: ,in, is the smooth transition coefficient, and They are Moment and The frequency of the moment.

[0010] Furthermore, the specific steps of the storage unit activation quantity adjustment sub-algorithm based on the load prediction result in S4 are: S431. According to the load prediction result and the storage unit performance parameters, the number of storage units to be activated is calculated using the storage unit number calculation formula. The storage unit number calculation formula is: ,in, is the number of storage units, For preset time, The processing capacity of a single storage unit, is the redundancy coefficient; S432. Activate a corresponding number of storage units and close the remaining storage units.

[0011] Furthermore, the specific steps of the cache policy adjustment sub-algorithm in S4 are: S441. Determine data access hotspots based on load prediction results; S442. Based on data access hotspots and cache capacity, the cache allocation ratio of different data is calculated using the cache allocation formula. The cache allocation formula is: ,in: For the The cache allocation ratio of the data, For the The access probability of the data, is the sum of all data access probabilities, is the cache capacity; S443. Adjust the cache strategy according to the calculated cache allocation ratio, giving priority to caching data with high access probability.

[0012] Furthermore, the specific steps of the feedback adjustment in S5 are: S51. Build a real-time monitoring parameter system; The real-time monitoring parameter system adopts a three-layer monitoring architecture, collecting core parameters every 5ms, including: Power consumption parameters: core chip supply voltage, operating current, and flash memory array power consumption ratio; Performance parameters: random read and write IOPS, continuous read and write bandwidth, average response time, and read and write latency jitter; Prediction deviation parameter: the instantaneous error between the load prediction value and the actual value, i.e. ; The sliding window error mean, that is ,in, express The instantaneous error between the load prediction value and the actual value at the moment, To show The load forecast value at that moment, for Actual load value at any moment, Represents the sliding window error mean, window size , that is, the current moment The instantaneous error of the first 20 moments including Take the average value; All parameters are transmitted to the monitoring database through an encrypted communication channel and stored in a time series database, retaining the last 72 hours of high-frequency data and 90 days of minute-level summary data. S52. Reinforcement learning algorithm deployment; The deep Q network (DQN) is used as the core framework of reinforcement learning, specifically: State space definition: ,in, express The state vector at the moment, which integrates multiple key parameters related to the power consumption and operating status of the solid-state drive, is the power consumption at the current moment, The reference power consumption is a preset benchmark value, which is used to calculate the power consumption at the current moment. Perform normalization processing, is the historical average load, is the frequency adjustment amount, is the voltage regulation amount; Action Space Design: Discretizing Action Sets , the strategy corresponding to each action is: Load forecasting model parameter adjustment: Dynamically adjust the learning rate, support ±10% or ±20% gradient update amplitude adjustment, and optimize model training efficiency; Dynamic power consumption algorithm parameter calibration: adjusts key coefficients, supports ±5% or ±10% parameter fluctuations, and achieves a dynamic balance between power consumption and performance; Reward function optimization: , , is the weight coefficient; Experience replay mechanism: maintain an experience pool with a capacity of 10,000, randomly extract 32 samples for training in each iteration, and use - Greedy strategy, and Decay linearly from 0.9 to 0.1; S53. Dynamic adjustment of load forecasting model; When there are 5 consecutive monitoring cycles, i.e. 25ms When , model fine-tuning is triggered: Attention weight matrix The update formula is: , The learning rate output for reinforcement learning; LSTM gating parameter correction: forget gate bias Increase to: ; S54. Adaptive adjustment of dynamic power consumption algorithm; Voltage regulation sub-algorithm correction: Real-time calculation of voltage regulation error , is the actual voltage, when hour, ; Frequency adjustment coefficient optimization: Based on energy consumption-performance ratio , dynamic adjustment: ,in To preset the optimal energy consumption-performance ratio, For The input and output operation rate of the solid-state drive at any time.

[0013] Furthermore, a multimedia equipment operation and maintenance monitoring system based on big data includes: Data acquisition module: used to collect historical load data and corresponding power consumption data of the solid-state drive, as well as real-time collection of the solid-state drive's current load data, optimized actual power consumption and performance parameters; Load Forecasting Module: This module builds and trains a bidirectional LSTM-Transformer load forecasting model that incorporates an attention mechanism based on historical load data collected by the data collection module. It also inputs real-time load data into the trained model to obtain load forecast results. Power consumption optimization module: used to optimize the power consumption of the solid-state drive using a dynamic power consumption adjustment algorithm with multi-objective optimization based on the load prediction results obtained by the load prediction module; Feedback adjustment module: used to adjust the load prediction model in the load prediction module and the dynamic power consumption adjustment algorithm in the power consumption optimization module based on the optimized actual power consumption and performance parameters collected by the data acquisition module and combined with the reinforcement learning algorithm.

[0014] The technical effects and advantages of this application are: Compared with existing technologies, this load-forecasting-based SSD power consumption optimization method and system, from load prediction model construction, dynamic power consumption adjustment algorithm design, to feedback adjustment mechanism optimization, forms a complete and efficient SSD power consumption optimization solution. It is expected to provide new technical ideas and practical directions for reducing energy consumption and improving performance of SSDs, and promote the further development and application of SSD technology in the field of low power consumption. The LSTM-Transformer load prediction model with integrated attention mechanism is used for load prediction. This model combines the advantages of LSTM in capturing temporal dependencies and the Transformer self-attention mechanism in highlighting important features, improving the accuracy of load prediction and being able to more accurately grasp the changing trend of SSD load.

[0015] Based on load prediction results, a dynamic power adjustment algorithm is used to optimize power consumption, including voltage, frequency, and the number of memory cell activations, achieving multi-dimensional power optimization. Furthermore, power consumption can be adjusted in advance based on prediction results, avoiding the lag of traditional passive adjustment and improving the timeliness and effectiveness of power optimization.

[0016] Setting up a feedback adjustment module can adjust the load prediction model and dynamic power consumption adjustment algorithm according to actual operating conditions, ensuring the stability and continuity of the power consumption optimization effect, and enabling the system to adapt to different application scenarios and load changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of the method flow of this application; Figure 2 This is a schematic diagram of the system flow of this application; Figure 3 This is a flowchart of the bidirectional LSTM-Transforme load prediction model that integrates the attention mechanism in this application; Figure 4 Schematic diagram of the flow of the voltage regulation sub-algorithm based on load prediction results in this application; Figure 5 Schematic diagram of the flow of the frequency adjustment sub-algorithm based on load prediction results in this application; Figure 6 Schematic diagram of the flow of the storage unit activation quantity adjustment sub-algorithm based on the load prediction result in this application; Figure 7 This is a flow chart of the cache policy adjustment sub-algorithm in this application; Figure 8 Schematic diagram of the specific steps of feedback adjustment in this application.

[0018] The accompanying drawings are marked as: 1. Data acquisition module; 2. Load prediction module; 3. Power consumption optimization module; 4. Feedback adjustment module. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0020] Example 1 like Figure 1 and Figure 3-Figure 8 The method for optimizing the power consumption of a solid-state hard disk based on load prediction shown in the figure comprises the following steps: S1. Data Collection: Collect historical load data and corresponding power consumption data of the solid-state drive, including the number of I / O requests, data transfer volume, request response time, and request type; S2. Load Forecasting Model Construction and Training: Based on the collected historical load data, a bidirectional LSTM-Transformer load forecasting model integrating an attention mechanism is constructed. This model is then trained using the historical load data to obtain a trained load forecasting model. The specific process of constructing a bidirectional LSTM-Transforme load prediction model integrating the attention mechanism in S2 is as follows: S21. Use a bidirectional LSTM network to extract features from historical load data and capture the bidirectional temporal dependencies of the data. S22. Input the features output by the bidirectional LSTM network into the encoder of the Transformer model and perform weighted processing on the features through the attention mechanism to highlight important features. S23. Input the features output by the encoder into the decoder of the Transformer model to obtain the load prediction result.

[0021] The forward propagation and back propagation formulas of the bidirectional LSTM network are as follows: Forward propagation formula: , , , , ; Back propagation formula: , , , , ; in, 、 、 They are input gate, forget gate, and output gate respectively. is the cell state, is hidden state, 、 is the weight matrix, is the bias, superscript Indicates positive direction, superscript Indicates the reverse direction, is the sigmoid function, is element-wise multiplication, For the moment Input data of; bidirectional LSTM output features ; The attention mechanism formula is: ,in, They are query matrix, key matrix, and value matrix respectively. is the feature dimension, is the feature importance weight matrix; the decoder output layer uses linear transformation and softmax function to obtain the final prediction value: ,in, is the output weight matrix, is the output bias, is the decoder output; during the model training process, an improved loss function is used, combining the mean square error and prediction deviation penalty terms: ,in is the predicted value, is the actual value, is the penalty coefficient, and the model parameters are adjusted through the back propagation algorithm.

[0022] S3. Load Forecasting: Input the real-time collected SSD load data into the trained load forecasting model to obtain load forecast results for a preset time period in the future. S4. Power consumption optimization: Based on the load prediction results, a multi-objective optimization dynamic power consumption adjustment algorithm is used to optimize the power consumption of the solid-state drive. The dynamic power consumption adjustment algorithm includes a voltage adjustment sub-algorithm based on the load prediction results, a frequency adjustment sub-algorithm, a storage unit activation number adjustment sub-algorithm, and a cache policy adjustment sub-algorithm; The specific steps of the voltage regulation sub-algorithm based on the load prediction results in S4 are: S411. Based on the load prediction value and performance constraints, the target voltage value is calculated using the voltage optimization formula. The voltage optimization formula is: ,in, is the load forecast value, For performance constraints, is the reference voltage, is the voltage regulation coefficient, is the performance compensation coefficient, For current performance; S412. Adjust the operating voltage of the solid-state drive to the target voltage value according to the preset voltage adjustment step size so that the voltage change rate does not exceed the voltage change rate threshold. The adjustment formula is: ,in, is the preset voltage adjustment step size, is the voltage change rate threshold, and They are Moment and The voltage at the moment, is a symbolic function.

[0023] The specific steps of the frequency adjustment sub-algorithm based on the load prediction results in S4 are: S421. Based on the load prediction results, the target operating frequency is calculated using the frequency optimization formula. The frequency optimization formula is: ,in, is the load prediction result, is the target operating frequency, is the reference frequency, is the reference load, is the frequency adjustment coefficient, is the energy consumption coefficient, is the current power consumption, For optimal power consumption; S422. Use a smooth transition method to adjust the current operating frequency of the solid-state drive to the target operating frequency, and the frequency change formula is: ,in, is the smooth transition coefficient, and They are Moment and The frequency of the moment.

[0024] The specific steps of the storage unit activation quantity adjustment sub-algorithm based on the load prediction result in S4 are as follows: S431. According to the load prediction result and the storage unit performance parameters, the number of storage units to be activated is calculated using the storage unit quantity calculation formula. The storage unit quantity calculation formula is: ,in, is the number of storage units, For preset time, The processing capacity of a single storage unit, is the redundancy coefficient; S432. Activate a corresponding number of storage units and close the remaining storage units.

[0025] The specific steps of the cache policy adjustment sub-algorithm in S4 are: S441. Determine data access hotspots based on load prediction results; S442. Based on data access hotspots and cache capacity, the cache allocation ratio of different data is calculated using the cache allocation formula. The cache allocation formula is: ,in: For the The cache allocation ratio of the data, For the The access probability of the data, is the sum of all data access probabilities, is the cache capacity; S443. Adjust the cache strategy according to the calculated cache allocation ratio, giving priority to caching high-access probability data; S5. Feedback adjustment: Real-time monitoring of the actual power consumption and performance parameters of the SSD after optimization, and adjustment of the load prediction model and dynamic power consumption adjustment algorithm using reinforcement learning algorithms.

[0026] The specific steps of feedback regulation in S5 are: S51. Build a real-time monitoring parameter system; The real-time monitoring parameter system adopts a three-layer monitoring architecture, collecting core parameters every 5ms, including: Power consumption parameters: core chip supply voltage, operating current, and flash memory array power consumption ratio; Performance parameters: random read and write IOPS, continuous read and write bandwidth, average response time, and read and write latency jitter; Prediction deviation parameter: the instantaneous error between the load prediction value and the actual value, i.e. ; The sliding window error mean, that is ,in, express The instantaneous error between the load prediction value and the actual value at the moment, To show The load forecast value at that moment, for Actual load value at any moment, Represents the sliding window error mean, window size , that is, the current moment The instantaneous error of the first 20 moments including Take the average value; All parameters are transmitted to the monitoring database through an encrypted communication channel and stored in a time series database, retaining the last 72 hours of high-frequency data and 90 days of minute-level summary data. S52. Reinforcement learning algorithm deployment; The deep Q network (DQN) is used as the core framework of reinforcement learning, specifically: State space definition: ,in, express The state vector at the moment, which integrates multiple key parameters related to the power consumption and operating status of the solid-state drive, is the power consumption at the current moment, The reference power consumption is a preset benchmark value, which is used to calculate the power consumption at the current moment. Perform normalization processing, is the historical average load, is the frequency adjustment amount, is the voltage regulation amount; Action Space Design: Discretizing Action Sets , the strategy corresponding to each action is: Load forecasting model parameter adjustment: Dynamically adjust the learning rate, support ±10% or ±20% gradient update amplitude adjustment, and optimize model training efficiency; Dynamic power consumption algorithm parameter calibration: adjusts key coefficients, supports ±5% or ±10% parameter fluctuations, and achieves a dynamic balance between power consumption and performance; Reward function optimization: , , is the weight coefficient; Experience replay mechanism: maintain an experience pool with a capacity of 10,000, randomly extract 32 samples for training in each iteration, and use - Greedy strategy, and Decay linearly from 0.9 to 0.1; S53. Dynamic adjustment of load forecasting model; When there are 5 consecutive monitoring cycles, i.e. 25ms When , model fine-tuning is triggered: Attention weight matrix The update formula is: , The learning rate output for reinforcement learning; LSTM gating parameter correction: forget gate bias Increase to: ; S54. Adaptive adjustment of dynamic power consumption algorithm; Voltage regulation sub-algorithm correction: Real-time calculation of voltage regulation error , is the actual voltage, when hour, ; Frequency adjustment coefficient optimization: Based on energy consumption-performance ratio , dynamic adjustment: ,in To preset the optimal energy consumption-performance ratio, For The input and output operation rate of the solid-state drive at any time.

[0027] Example 2 Please refer to Figure 2 , a multimedia equipment operation and maintenance monitoring system based on big data, the system includes: Data acquisition module 1: used to collect historical load data and corresponding power consumption data of the solid-state drive, as well as real-time collection of the solid-state drive's current load data, optimized actual power consumption and performance parameters; Historical load and power consumption data is collected every 10ms and stored in a local cache. The cached data is aggregated into a database every hour for long-term storage. During the collection process, data is preprocessed in real time, including outlier detection and correction. The 3σ criterion is used to identify outliers, and data exceeding ±3 standard deviations of the mean is corrected through linear interpolation. For real-time load data, actual power consumption, and performance parameters, the collection frequency is increased to every 5ms to ensure data timeliness and provide accurate data support for subsequent load forecasting and feedback adjustments. Load Forecasting Module 2: This module builds and trains a bidirectional LSTM-Transformer load forecasting model that incorporates an attention mechanism based on the historical load data collected by Data Collection Module 1. It also inputs the current load data collected in real time into the trained model to obtain load forecast results. The bidirectional LSTM-Transformer load forecasting model first performs feature normalization on the input historical load data and maps the data to the [0, 1] interval. Power consumption optimization module 3: used to optimize the power consumption of the solid-state drive using a dynamic power consumption adjustment algorithm with multi-objective optimization based on the load prediction result obtained by load prediction module 2; Feedback adjustment module 4: used to adjust the load prediction model in load prediction module 2 and the dynamic power consumption adjustment algorithm in power consumption optimization module 3 according to the optimized actual power consumption and performance parameters collected by data acquisition module 1 in combination with reinforcement learning algorithm.

[0028] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0029] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for optimizing the power consumption of a solid-state drive based on load prediction, characterized in that: The method comprises the following steps: S1. Data Collection: Collect historical load data and corresponding power consumption data of the solid-state drive, including the number of I / O requests, data transfer volume, request response time, and request type; S2. Load Forecasting Model Construction and Training: Based on the collected historical load data, a bidirectional LSTM-Transformer load forecasting model integrating an attention mechanism is constructed. This model is then trained using the historical load data to obtain a trained load forecasting model. S3. Load Forecasting: Input the real-time collected SSD load data into the trained load forecasting model to obtain load forecast results for a preset time period in the future. S4. Power consumption optimization: Based on the load prediction results, a multi-objective optimization dynamic power consumption adjustment algorithm is used to optimize the power consumption of the solid-state drive. The dynamic power consumption adjustment algorithm includes a voltage adjustment sub-algorithm based on the load prediction results, a frequency adjustment sub-algorithm, a storage unit activation number adjustment sub-algorithm, and a cache policy adjustment sub-algorithm; S5. Feedback adjustment: Real-time monitoring of the actual power consumption and performance parameters of the SSD after optimization, and adjustment of the load prediction model and dynamic power consumption adjustment algorithm using reinforcement learning algorithms.

2. The method for optimizing power consumption of a solid-state drive based on load prediction according to claim 1, wherein: The specific process of constructing the bidirectional LSTM-Transforme load prediction model integrating the attention mechanism in S2 is as follows: S21. Use a bidirectional LSTM network to extract features from historical load data and capture the bidirectional temporal dependencies of the data. S22. Input the features output by the bidirectional LSTM network into the encoder of the Transformer model and perform weighted processing on the features through the attention mechanism to highlight important features. S23. Input the features output by the encoder into the decoder of the Transformer model to obtain the load prediction result.

3. The method for optimizing power consumption of a solid-state drive based on load prediction according to claim 2, wherein: The forward propagation and back propagation formulas of the bidirectional LSTM network are as follows: Forward propagation formula: ， ， ， ， ； Back propagation formula: ， ， ， ， ； in, 、 、 They are input gate, forget gate, and output gate respectively. is the cell state, is hidden state, 、 is the weight matrix, is the bias, superscript Indicates positive direction, superscript Indicates the reverse direction, is the sigmoid function, is element-wise multiplication, For the moment Input data of; bidirectional LSTM output features ; The attention mechanism formula is: ,in, They are query matrix, key matrix, and value matrix respectively. is the feature dimension, is the feature importance weight matrix; the decoder output layer uses linear transformation and softmax function to obtain the final prediction value: ,in, is the output weight matrix, is the output bias, is the decoder output; during the model training process, an improved loss function is used, combining the mean square error and prediction deviation penalty terms: ,in is the predicted value, is the actual value, is the penalty coefficient, and the model parameters are adjusted through the back propagation algorithm.

4. The method for optimizing power consumption of a solid-state drive based on load prediction according to claim 3, wherein: The specific steps of the voltage regulation sub-algorithm based on the load prediction result in S4 are: S411. Based on the load prediction value and performance constraints, the target voltage value is calculated using the voltage optimization formula. The voltage optimization formula is: ,in, is the load forecast value, For performance constraints, is the reference voltage, is the voltage regulation coefficient, is the performance compensation coefficient, For current performance; S412. Adjust the operating voltage of the solid-state drive to the target voltage value according to the preset voltage adjustment step size so that the voltage change rate does not exceed the voltage change rate threshold. The adjustment formula is: ,in, is the preset voltage adjustment step size, is the voltage change rate threshold, and They are Moment and The voltage at the moment, is a symbolic function.

5. The method for optimizing power consumption of a solid-state drive based on load prediction according to claim 4, wherein: The specific steps of the frequency adjustment sub-algorithm based on the load prediction result in S4 are: S421. Based on the load prediction results, the target operating frequency is calculated using the frequency optimization formula. The frequency optimization formula is: ,in, is the load prediction result, is the target operating frequency, is the reference frequency, is the reference load, is the frequency adjustment coefficient, is the energy consumption coefficient, is the current power consumption, For optimal power consumption; S422. Use a smooth transition method to adjust the current operating frequency of the solid-state drive to the target operating frequency, and the frequency change formula is: ,in, is the smooth transition coefficient, and They are Moment and The frequency of the moment.

6. The method for optimizing power consumption of a solid-state drive based on load prediction according to claim 5, wherein: The specific steps of the storage unit activation quantity adjustment sub-algorithm based on the load prediction result in S4 are: S431. According to the load prediction result and the storage unit performance parameters, the number of storage units to be activated is calculated using the storage unit number calculation formula. The storage unit number calculation formula is: ,in, is the number of storage units, For preset time, The processing capacity of a single storage unit, is the redundancy coefficient; S432. Activate a corresponding number of storage units and close the remaining storage units.

7. The method for optimizing power consumption of a solid-state drive based on load prediction according to claim 6, wherein: The specific steps of the cache strategy adjustment sub-algorithm in S4 are: S441. Determine data access hotspots based on load prediction results; S442. Based on data access hotspots and cache capacity, the cache allocation ratio of different data is calculated using the cache allocation formula. The cache allocation formula is: ,in: For the The cache allocation ratio of the data, For the The access probability of the data, is the sum of all data access probabilities, is the cache capacity; S443. Adjust the cache strategy according to the calculated cache allocation ratio, giving priority to caching data with high access probability.

8. The method for optimizing power consumption of a solid-state drive based on load prediction according to claim 1, wherein: The specific steps of feedback regulation in S5 are: S51. Build a real-time monitoring parameter system; The real-time monitoring parameter system adopts a three-layer monitoring architecture, collecting core parameters every 5ms, including: Power consumption parameters: core chip supply voltage, operating current, and flash memory array power consumption ratio; Performance parameters: random read and write IOPS, continuous read and write bandwidth, average response time, and read and write latency jitter; Prediction deviation parameter: the instantaneous error between the load prediction value and the actual value, i.e. ; The sliding window error mean, that is ,in, express The instantaneous error between the load prediction value and the actual value at the moment, To show The load forecast value at that moment, for Actual load value at any moment, Represents the sliding window error mean, window size , that is, the current moment The instantaneous error of the first 20 moments including Take the average value; All parameters are transmitted to the monitoring database through an encrypted communication channel and stored in a time series database, retaining the last 72 hours of high-frequency data and 90 days of minute-level summary data. S52. Reinforcement learning algorithm deployment; The deep Q network (DQN) is used as the core framework of reinforcement learning, specifically: State space definition: ,in, express The state vector at the moment, which integrates multiple key parameters related to the power consumption and operating status of the solid-state drive, is the power consumption at the current moment, The reference power consumption is a preset benchmark value, which is used to calculate the power consumption at the current moment. Perform normalization processing, is the historical average load, is the frequency adjustment amount, is the voltage regulation amount; Action Space Design: Discretizing Action Sets , the strategy corresponding to each action is: Load forecasting model parameter adjustment: Dynamically adjust the learning rate, support ±10% or ±20% gradient update amplitude adjustment, and optimize model training efficiency; Dynamic power consumption algorithm parameter calibration: adjusts key coefficients, supports ±5% or ±10% parameter fluctuations, and achieves a dynamic balance between power consumption and performance; Reward function optimization: , , is the weight coefficient; Experience replay mechanism: maintain an experience pool with a capacity of 10,000, randomly extract 32 samples for training in each iteration, and use - Greedy strategy, and Decay linearly from 0.9 to 0.1; S53. Dynamic adjustment of load forecasting model; When there are 5 consecutive monitoring cycles, i.e. 25ms When , model fine-tuning is triggered: Attention weight matrix The update formula is: , The learning rate output for reinforcement learning; LSTM gating parameter correction: forget gate bias Increase to: ; S54. Adaptive adjustment of dynamic power consumption algorithm; Voltage regulation sub-algorithm correction: Real-time calculation of voltage regulation error , is the actual voltage, when hour, ; Frequency adjustment coefficient optimization: Based on energy consumption-performance ratio , dynamic adjustment: ,in To preset the optimal energy consumption-performance ratio, For The input and output operation rate of the solid-state drive at any time.

9. A multimedia equipment operation and maintenance monitoring system based on big data, characterized in that: The system includes: Data acquisition module (1): used to collect historical load data and corresponding power consumption data of the solid-state hard disk, as well as real-time current load data, optimized actual power consumption and performance parameters of the solid-state hard disk; load prediction module (2): used to build a bidirectional LSTM-Transformer load prediction model integrating an attention mechanism based on the historical load data collected by the data acquisition module (1), and to train it, and also used to input the current load data collected in real time into the trained model to obtain a load prediction result; power consumption optimization module (3): used to optimize the power consumption of the solid-state hard disk using a dynamic power consumption adjustment algorithm with multi-objective optimization based on the load prediction result obtained by the load prediction module (2); Feedback adjustment module (4): used to adjust the load prediction model in the load prediction module (2) and the dynamic power consumption adjustment algorithm in the power consumption optimization module based on the optimized actual power consumption and performance parameters collected by the data collection module (1) in combination with the reinforcement learning algorithm.

Images