Artificial intelligence-based decision optimization method, device, equipment, medium and product

Abstract

The application provides an artificial intelligence-based decision optimization method and device, equipment, medium and product, and relates to the field of artificial intelligence. The method comprises: analyzing and processing an input / output request to obtain an analysis request; obtaining a load prediction result according to a request feature, a preset prediction strategy and a linear regression model; obtaining a cache prediction result according to the analysis request, a neural network model and the preset prediction strategy; performing optimization fusion processing according to the analysis request, the cache prediction result and a preset optimization fusion strategy to obtain an optimization decision; optimizing the analysis request through the optimization decision to obtain an optimization request; executing the optimization request and mapping the optimization request to a correct physical storage location. The application solves the technical problem of low efficiency of artificial intelligence-based decision acceleration caused by the fact that the prior art is difficult to adapt to dynamically changing workload scenarios and often leads to resource waste or performance bottlenecks due to the lack of intelligent decision-making ability.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, and in particular to a decision optimization method, apparatus, device, medium and product based on artificial intelligence. Background Technology

[0002] With the rapid development of cloud computing, big data and artificial intelligence technologies, the scale of data centers and enterprise-level storage systems is constantly expanding, and the demand for performance optimization of storage devices is becoming increasingly urgent.

[0003] In traditional storage architectures, block device mapping and cache management mainly rely on static algorithms or simple heuristics.

[0004] However, existing technologies are difficult to adapt to dynamically changing workload scenarios, and often lead to resource waste or performance bottlenecks due to a lack of intelligent decision-making capabilities, resulting in low efficiency of AI-based decision optimization. Summary of the Invention

[0005] This application provides a decision optimization method, apparatus, device, medium, and product based on artificial intelligence, which addresses the problem that existing technologies are difficult to adapt to dynamically changing workload scenarios, often resulting in resource waste or performance bottlenecks due to a lack of intelligent decision-making capabilities, thus leading to low efficiency in decision optimization based on artificial intelligence.

[0006] Firstly, this application provides a decision optimization method based on artificial intelligence, including:

[0007] The system acquires input / output requests, parses and processes these requests to obtain parsed requests; these parsed requests include multiple request characteristics.

[0008] Based on the request characteristics, the preset prediction strategy and the linear regression model, the first prediction process is performed to obtain the load prediction result.

[0009] Based on the parsing request, the neural network model, and the preset prediction strategy, a second prediction process is performed to obtain the cached prediction results;

[0010] Based on the parsing request, cache prediction results, and preset optimization fusion strategy, optimization fusion processing is performed to obtain optimization decisions;

[0011] The parsing request is optimized by optimizing the decision-making process to obtain an optimized request;

[0012] Execute optimization requests and map them to the correct physical storage locations.

[0013] In one possible design, the input / output request is parsed to obtain a parsed request, including:

[0014] Feature extraction processing is performed on the input / output requests to obtain a first request containing multiple request features;

[0015] The request features are classified using a classification algorithm to obtain classification labels.

[0016] The first request is tagged according to the category label and the preset tagging strategy to obtain the parsed request.

[0017] In one possible design, request characteristics include request size, access mode, priority, operation type, and starting sector, etc.

[0018] Category labels include sequential access labels, random access labels, and mixed access labels;

[0019] The preset labeling strategies include:

[0020] If the request size is less than the value of the first parameter, then the first request is tagged with a random access label;

[0021] If the request size is greater than the value of the second parameter and the access mode is sequential, then the first request will access the tags in the order they are labeled.

[0022] If the first request does not match the random access tag or the sequential access tag, then the first request is tagged with the mixed access tag.

[0023] In one possible design, based on request characteristics, a preset prediction strategy, and a linear regression model, a first prediction process is performed to obtain load prediction results, including:

[0024] Obtain historical request datasets by using a sliding window and request features;

[0025] Input the historical request dataset into the linear regression model to calculate the load prediction value;

[0026] Collect real-time load values ​​and calculate the load error value based on the real-time load values ​​and the load prediction values;

[0027] The weight array of the linear regression model is updated based on the load error value to obtain an optimized linear regression model.

[0028] In one possible design, based on the parsing request, the neural network model, and a preset prediction strategy, a second prediction process is performed to obtain the cached prediction result, including:

[0029] Based on the request characteristics, the temporal locality feature value, spatial locality feature value, and access frequency feature value are calculated.

[0030] The temporal locality feature value, spatial locality feature value, access frequency feature value and preset weight value are input into the neural network model to calculate the comprehensive prediction score;

[0031] The comprehensive prediction score is compared with the first threshold.

[0032] If the overall predicted score is greater than the first threshold, then the output cache hit is achieved.

[0033] If the overall predicted score is not greater than the first threshold, then the output cache is not hit.

[0034] In one possible design, the preset prediction strategy includes a low-latency optimization strategy;

[0035] Low-latency optimization strategies include:

[0036] If the request size is less than the second threshold, then the parsed request is tagged with a fast path label;

[0037] If the parsing request has a high priority, then the parsing request will be tagged and processed first.

[0038] In one possible design, after executing the optimization request and mapping it to the correct physical storage location, the following is also included:

[0039] Real-time performance data is collected, and the linear regression model, neural network model, and preset prediction strategy are updated based on the performance data.

[0040] Secondly, this application provides a decision optimization device based on artificial intelligence, comprising:

[0041] The parsing module is used to acquire input / output requests, parse the input / output requests to obtain parsed requests; the parsed requests include multiple request features;

[0042] The first prediction module is used to perform the first prediction process based on the request characteristics, the preset prediction strategy and the linear regression model to obtain the load prediction result.

[0043] The second prediction module is used to perform a second prediction process based on the parsing request, the neural network model, and the preset prediction strategy to obtain the cached prediction results.

[0044] The fusion module is used to perform optimization fusion processing based on the parsed request, cached prediction results and preset optimization fusion strategy to obtain optimization decisions;

[0045] The optimization module is used to optimize the parsing request through optimization decisions to obtain an optimized request;

[0046] The execution module is used to execute optimization requests and map them to the correct physical storage locations.

[0047] Thirdly, this application provides a decision optimization device based on artificial intelligence, including: a memory and a processor;

[0048] The memory stores the instructions that the computer executes;

[0049] The processor executes computer execution instructions stored in memory, causing the processor to perform an artificial intelligence-based decision optimization method as described in the first aspect of the invention.

[0050] Fourthly, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the artificial intelligence-based decision optimization method as described in the first aspect of the invention.

[0051] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the artificial intelligence-based decision optimization method described in the first aspect of the invention.

[0052] This application provides an AI-based decision optimization method, apparatus, device, medium, and product, comprising: acquiring input / output requests; parsing the input / output requests to obtain parsed requests; performing a first prediction process based on request features, a preset prediction strategy, and a linear regression model to obtain a load prediction result; performing a second prediction process based on the parsed requests, a neural network model, and a preset prediction strategy to obtain cached prediction results; performing an optimization fusion process based on the parsed requests, cached prediction results, and a preset optimization fusion strategy to obtain an optimization decision; optimizing the parsed requests using the optimization decision to obtain an optimized request; executing the optimized request and mapping the optimized request to the correct physical storage location. Compared to existing technologies that struggle to adapt to dynamically changing workload scenarios and often suffer from resource waste or performance bottlenecks due to a lack of intelligent decision-making capabilities, AI-based decision acceleration efficiency is relatively low. This application improves the efficiency of AI-based decision optimization by directly embedding artificial intelligence (AI) algorithms (such as machine learning and neural network inference) into the kernel space and achieving intelligent processing of storage input / output (I / O) requests through steps such as feature extraction, workload prediction, and dynamic adjustment of caching strategies. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 A schematic diagram of a system architecture for an artificial intelligence-based decision optimization method provided in an embodiment of this application;

[0055] Figure 2 A schematic diagram of an AI-based decision optimization system architecture provided for embodiments of this application;

[0056] Figure 3 A schematic diagram of a decision optimization method based on artificial intelligence provided in an embodiment of this application;

[0057] Figure 4 This application provides a schematic diagram of the collaborative workflow of each module in its embodiments;

[0058] Figure 5 A schematic diagram illustrating the operation flow of an AI-based decision optimization system provided in this application embodiment;

[0059] Figure 6 A schematic diagram of the structure of the AI-based decision optimization device provided in the embodiments of this application;

[0060] Figure 7 This is a schematic diagram of the structure of an artificial intelligence-based decision optimization device provided in an embodiment of this application. Detailed Implementation

[0061] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0062] In the embodiments of this application, the terms "first" and "second" are used to distinguish identical or similar items with substantially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, nor do they necessarily imply difference. It should be noted that in the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design scheme described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner. In the embodiments of this application, "at least one" refers to one or more, and "more than one" refers to two or more.

[0063] It should be noted that the phrase "at...time" in the embodiments of this application can refer to the instant at which a certain situation occurs, or to a period of time after the occurrence of a certain situation; the embodiments of this application do not specifically limit this. Furthermore, the AI-based decision optimization method provided in the embodiments of this application is merely an example; AI-based decision optimization methods may also include more or less content.

[0064] With the rapid development of cloud computing, big data, and artificial intelligence technologies, the scale of data centers and enterprise-level storage systems is constantly expanding, making the need for performance optimization of storage devices increasingly urgent. In traditional storage architectures, block device mapping and cache management mainly rely on static algorithms or simple heuristic rules, which are difficult to adapt to dynamically changing workload scenarios.

[0065] Specifically, the device mapper is an important component in the Linux kernel, providing a general way to create virtual block devices. Traditionally, device mappers are primarily used to implement RAID, LVM (Logical Volume Management), snapshots, and other functionalities. However, with the continuous expansion of data centers and the increasing complexity of application workloads, traditional device mappers face numerous challenges in performance optimization.

[0066] Current storage device mapping and caching optimization technologies primarily rely on traditional algorithms and fixed rules. For example, the Linux kernel's device-mapper component manages block device mapping through statically configured caching strategies (such as LRU or FIFO), but it lacks deep analysis capabilities of application access patterns and cannot dynamically adjust the strategy. At the caching algorithm level, LRU evicts data based on the principle of temporal locality, but performs poorly with scanning access patterns; the ARC algorithm, combining LRU and LFU, is more adaptable, but relies on heuristic rules and has complex parameter tuning. Furthermore, some commercial solutions implement intelligent prefetching and cache management through user-space cache libraries, but suffer from high context switching overhead and inability to deeply integrate with the kernel storage stack. In recent years, AI-driven storage optimization solutions have gradually emerged, such as cache prediction models based on LSTM networks and cloud AI storage services, but these solutions mostly run in user space, resulting in high kernel-user space data copy latency and inability to optimize in real time. While prototype systems such as "Neural Cache" in academic research attempt to introduce deep learning, they have not yet solved the technical bottlenecks of kernel-level integration and low-latency decision-making.

[0067] The existing technical solutions have the following limitations:

[0068] Optional, technical architecture limitations: Most AI storage solutions run in user space, incurring kernel-user space switching overhead, lacking deep integration with the Linux kernel device-mapper, and thus failing to achieve true zero-copy AI processing.

[0069] Alternatively, algorithmic limitations exist. Traditional algorithms, based on heuristic rules, cannot handle complex non-linear access patterns. AI algorithms are mostly trained offline, lacking real-time online learning capabilities, resulting in limited prediction accuracy (usually <80%) and long model update cycles.

[0070] Optional limitations in performance and real-time performance: AI decision-making latency is relatively high (milliseconds), making it unsuitable for high IOPS scenarios; it consumes a lot of memory, making it unsuitable for resource-constrained kernel environments; and it cannot achieve microsecond-level real-time intelligent decision-making.

[0071] Optionally, system integration limitations: lack of collaborative optimization with kernel components such as the I / O scheduler, inability to achieve cross-level global performance optimization, limited scalability and compatibility, and difficulty in adapting to different hardware platforms.

[0072] To address the aforementioned issues, the inventors, during their research on the low efficiency of AI-based decision optimization, discovered that existing technologies struggle to adapt to dynamically changing workload scenarios. They often suffer from resource waste or performance bottlenecks due to a lack of intelligent decision-making capabilities, resulting in low efficiency for AI-based decision acceleration. Therefore, the inventors considered embedding Artificial Intelligence (AI) algorithms (such as machine learning and neural network inference) directly into the kernel space. Through steps such as feature extraction, workload prediction, and dynamic adjustment of caching strategies, they aimed to achieve intelligent processing of storage input / output (I / O) requests. Based on this, embodiments of this application provide an AI-based decision optimization method, apparatus, device, medium, and product, applicable to the field of artificial intelligence, aiming to solve the problem of low efficiency in existing AI-based decision optimization technologies.

[0073] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0074] Figure 1 This is a schematic diagram of a system architecture for an artificial intelligence-based decision optimization method provided in an embodiment of this application. The artificial intelligence-based decision optimization system is a computer device. Figure 1 In the above architecture, at least one of data acquisition device 101, processing device 102 and display device 103 is included.

[0075] It is understood that the structures illustrated in the embodiments of this application do not constitute a specific limitation on the processing system architecture of the AI-based decision optimization method. In other feasible embodiments of this application, the above architecture may include more or fewer components than illustrated, or combine some components, or divide some components, or arrange different components, which can be determined according to the actual application scenario and is not limited here. Figure 1 The components shown can be implemented in hardware, software, or a combination of both.

[0076] In the specific implementation process, the data acquisition device 101 may include an input / output interface or a communication interface. The data acquisition device 101 can connect to the processing device through the input / output interface or the communication interface to acquire relevant data.

[0077] The processing device 102 can obtain the search results corresponding to the search request information based on the relevant data.

[0078] The display device 103 can also be a touch screen or the screen of a terminal device, used to receive user commands while displaying the above-mentioned content, so as to realize interaction with the user.

[0079] It should be understood that the aforementioned processing device can be implemented by a processor reading instructions from memory and executing those instructions, or it can be implemented by a chip circuit.

[0080] Furthermore, the network architecture and business scenarios described in the embodiments of this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided in the embodiments of this application. As those skilled in the art will know, with the evolution of network architecture and the emergence of new business scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0081] The technical solution of this application will be described in detail below with reference to specific embodiments:

[0082] This application also provides a possible embodiment. Figure 2 A schematic diagram of an AI-based decision optimization system architecture provided in this application embodiment is shown below. Figure 2 As shown, the system includes: an AI core processing module (device-mapper deviceai core), an intelligent cache management module (device-mapper device cache), an AI inference engine (device-mapper device ai inference), a machine learning module (device-mapper device machine learn), and a device mapping optimization module (device-mapper device target).

[0083] The system's operation is based on a layered architecture, with data flow exhibiting closed-loop optimization characteristics. The AI ​​core layer serves as the starting point, with its core engine generating initial decisions through application classifiers, workload predictors, and intelligent schedulers. The machine learning module utilizes online learning algorithms, pattern recognizers, and quantification mechanisms to achieve dynamic learning and pattern recognition. The AI ​​inference engine executes specific inference tasks through neural network predictors, decision trees, and cache predictors. Data, after undergoing feature extraction, AI decision application, strategy implementation, and performance monitoring via the intelligent_map() function in the intelligent mapping interface layer, flows into the Device-Mapper framework. The cache management module executes intelligent caching strategies, the I / O mapping module implements intelligent block device mapping, and the device management module monitors device status. Finally, through information sharing, strategy coordination, performance feedback, and closed-loop optimization mechanisms via collaborative interfaces, the data collaborates with the AI-accelerated I / O scheduler to form a complete closed-loop process from data input to decision execution and feedback optimization.

[0084] The system's AI core layer achieves end-to-end intelligent decision-making, from data classification and prediction to scheduling, through vertical integration of engines and modules. The intelligent mapping interface layer acts as an intermediary bridge, transforming the abstract decisions of the AI ​​core layer into concrete operational instructions and executing them at the physical device level through the Device-Mapper framework. The collaborative interface, through horizontal linkage with the AI-accelerated I / O scheduler, forms a two-way channel for strategy coordination and performance feedback, ensuring that the decision execution effect is quantifiable and optimizable. A closed-loop feedback mechanism is maintained throughout—the collaborative interface and the AI-accelerated I / O scheduler input the execution results back into the AI ​​core layer, driving the online learning algorithm of the machine learning module to continuously update the model. The quantification mechanism dynamically adjusts the decision threshold, ultimately achieving a full-cycle intelligent operation logic from data collection, decision generation, execution monitoring to feedback optimization.

[0085] In this embodiment, the synergy between the AI ​​core engine and the inference engine enhances decision-making accuracy and response speed, while the online learning capability of the machine learning module ensures the system adapts to dynamic environments. The feature extraction and strategy application functions of the intelligent mapping interface layer optimize resource allocation efficiency, and the cache management and I / O mapping of the Device-Mapper framework enhance device performance and storage efficiency. The closed-loop optimization mechanism of the collaborative interface achieves continuous performance improvement, and the quantification mechanism provides a quantitative assessment of decision quality. The overall layered architecture reduces system coupling, improves modular scalability, and forms a self-optimizing intelligent ecosystem through closed-loop feedback, ultimately generating significant comprehensive benefits in terms of improving decision-making efficiency, optimizing resource allocation, and enhancing system adaptability.

[0086] Figure 3 A schematic diagram of an artificial intelligence-based decision optimization method provided in this application embodiment is shown below. Figure 3 As shown, the method includes:

[0087] S301. Obtain input / output requests, parse and process the input / output requests to obtain parsed requests.

[0088] The parsing request includes multiple request characteristics.

[0089] The request characteristics include request size, access mode, priority, operation type, and starting sector.

[0090] Specifically, feature extraction processing is performed on the input / output requests to obtain a first request containing multiple request features.

[0091] Furthermore, the request features are classified using a classification algorithm to obtain classification labels.

[0092] The classification labels include sequential access labels, random access labels, and mixed access labels.

[0093] Optionally, the first request can be tagged according to the category label and the preset tagging strategy to obtain the parsed request.

[0094] The preset marking strategies include:

[0095] Optionally, if the request size is less than the value of the first parameter, the first request is labeled with a random access tag.

[0096] Optionally, if the request size is greater than the second parameter value and the access mode is sequential, then the first request will access the tags in the order they are tagged.

[0097] Optionally, if the first request does not match the random access label or the sequential access label, then the first request is labeled with a mixed access label.

[0098] In this embodiment, by parsing input / output requests and extracting multi-dimensional features such as request size, access mode, priority, operation type, and starting sector, and combining classification algorithms to automatically classify the features into sequential / random / mixed access labels, and accurately labeling requests based on preset labeling strategies (such as request size threshold matching and access mode determination), refined feature modeling and dynamic access mode recognition of complex I / O requests are achieved, thereby optimizing the storage system resource allocation strategy and improving request processing efficiency and overall system performance.

[0099] S302. Based on the request characteristics, the preset prediction strategy and the linear regression model, perform the first prediction process to obtain the load prediction result.

[0100] Specifically, historical request datasets are obtained by using a sliding window and request features.

[0101] Furthermore, the historical request dataset is input into the linear regression model to calculate the load prediction value.

[0102] Furthermore, real-time load values ​​are collected, and the load error value is calculated based on the real-time load values ​​and the load prediction values.

[0103] Furthermore, the weight array of the linear regression model is updated based on the load error value to obtain an optimized linear regression model.

[0104] In this embodiment, historical request datasets are dynamically collected through a sliding window. The linear regression model is driven by combining request features and a preset prediction strategy to predict the load. The model weights are continuously optimized iteratively based on the error between the real-time load value and the predicted value. This achieves dynamic adaptation and high-precision modeling of load prediction, thereby improving the storage system's responsiveness to load changes, optimizing resource allocation strategies, and ultimately enhancing the overall operating efficiency and stability of the system.

[0105] In one possible embodiment, the machine learning module implements lightweight online learning, with the core algorithm principle as follows:

[0106] / / Simplified gradient descent learning (core logic)

[0107] int device_mapper_dev_ml_online_learn(struct feature_vector features,u32 target) {

[0108] / / Calculate the prediction value: prediction = Σ(feature[i] weight[i])

[0109] u32 prediction = 0;

[0110] for (int i = 0; i < FEATURE_COUNT; i++) {

[0111] prediction += features->values[i] weights[i] / 1000;

[0112] }

[0113] / / Calculate the error and update the weights

[0114] int error = target - prediction;

[0115] for (int i = 0; i < FEATURE_COUNT; i++) {

[0116] int gradient = (features->values[i] error) / 1000;

[0117] weights[i] += (learning_rate gradient) / 1000;

[0118] }

[0119] return 0;

[0120] }

[0121] S303. Based on the parsing request, the neural network model, and the preset prediction strategy, perform a second prediction process to obtain the cached prediction result.

[0122] Specifically, temporal locality feature values, spatial locality feature values, and access frequency feature values ​​are calculated based on the request characteristics.

[0123] Furthermore, the temporal locality feature value, spatial locality feature value, access frequency feature value, and preset weight values ​​are input into the neural network model to calculate the comprehensive prediction score.

[0124] Furthermore, the comprehensive predicted score is compared with the first threshold.

[0125] Furthermore, if the overall predicted score is greater than the first threshold, then the output cache is hit.

[0126] Furthermore, if the overall predicted score is not greater than the first threshold, then the output cache is not hit.

[0127] In this embodiment, by extracting the three-dimensional feature values ​​of temporal locality, spatial locality, and access frequency of the request, and combining them with a neural network model to perform nonlinear fusion calculations on multiple features, and by using a dynamic threshold comparison mechanism to accurately determine the cache hit result, the cache prediction capability is upgraded from single feature to multi-dimensional feature comprehensive modeling, thereby significantly improving the cache hit rate, reducing invalid cache access, and ultimately optimizing the response latency and resource utilization efficiency of the storage system.

[0128] In addition, the preset prediction strategies include low-latency optimization strategies.

[0129] Specifically, low-latency optimization strategies include:

[0130] Optionally, if the request size is less than the second threshold, a fast path tag is applied to the parsed request.

[0131] Optionally, if the parsing request has a high priority, then the parsing request should be tagged and processed first.

[0132] Additionally, the delay target is dynamically adjusted (default 10ms).

[0133] In one possible implementation, the AI ​​inference engine provides real-time prediction and decision support:

[0134] / / Smart cache prediction (core logic)

[0135] int device_mapper_dev_cache_predict(u64 block_id) {

[0136] / / Neural Network Prediction (Simplified Version)

[0137] u16 score = 0;

[0138] score += temporal_locality_score(block_id) 40; / / Temporal locality weight 40%

[0139] score += spatial_locality_score(block_id) 30; / / Spatial locality weight 30%

[0140] score += frequency_score(block_id) 30; / / Frequency score weighted at 30%

[0141] / / Prediction threshold determination

[0142] return (score > CACHE_THRESHOLD) ? CACHE_HIT_PREDICTED : CACHE_MISS_PREDICTED;

[0143] }

[0144] S304. Based on the parsing request, cache prediction results, and preset optimization fusion strategy, perform optimization fusion processing to obtain an optimization decision.

[0145] S305. Optimize the parsing request through optimization decision to obtain an optimized request.

[0146] S306. Execute the optimization request and map the optimization request to the correct physical storage location.

[0147] In one possible implementation, deep integration of AI and device-mapper is achieved through the device_mapper_dev_intelligent_map function:

[0148] Core mapping process

[0149] / / AI-enhanced device mapping (core process)

[0150] static int device_mapper_dev_intelligent_map(struct dm_target ti,struct bio bio) {

[0151] / / Step 1: AI Feature Extraction and Classification

[0152] enum app_type type = device_mapper_dev_app_classify_bio(bio);

[0153] / / Step 2: Machine Learning Prediction

[0154] struct ml_prediction pred;

[0155] device_mapper_dev_ml_predict(bio, &pred);

[0156] / / Step 3: AI Inference Optimization

[0157] if (pred.confidence > 80) {

[0158] device_mapper_dev_ai_optimize_io(bio, &pred);

[0159] }

[0160] / / Step 4: Execute optimized I / O mapping

[0161] return dm_map_bio(ti, bio);

[0162] }

[0163] The process after step S306 also includes real-time acquisition of performance data, and updating of the linear regression model, neural network model, and preset prediction strategy based on the performance data.

[0164] In one possible implementation, when the system detects that the workload has changed from random access to sequential access: First, pattern recognition: the ML module identifies the continuous sector access pattern; second, policy switching: the prefetch size is automatically adjusted from 8KB to 64KB; third, performance monitoring: the cache hit rate is monitored in real time; and finally, feedback optimization: the prefetch policy parameters are adjusted according to the effect.

[0165] In one possible implementation, the AI-accelerated device-mapper and the AI ​​scheduler work together through a shared AI intelligence layer: First, information sharing: sharing workload prediction and application classification results; second, strategy coordination: a unified AI decision-making framework coordinates and optimizes strategies; and finally, performance feedback: establishing a closed-loop feedback mechanism to continuously optimize the collaborative effect.

[0166] In one possible embodiment, specific optimizations for AI inference applications include: first, pattern recognition: identifying feature access patterns for AI inference; second, dedicated caching: allocating dedicated cache space for model data; third, latency optimization: using fast paths for small requests to reduce latency; and finally, concurrency optimization: optimizing the concurrent processing of multiple inference requests.

[0167] It should be noted that this embodiment also includes a confidence mechanism:

[0168] Optionally, the confidence level reaches 80% when the number of input / output request samples processed reaches 100.

[0169] Optionally, a caching strategy adjustment can be triggered when the confidence level exceeds a threshold.

[0170] Optionally, the learning rate can be dynamically adjusted to improve convergence speed.

[0171] This embodiment provides an AI-based decision optimization method, comprising: acquiring input / output requests; parsing the input / output requests to obtain parsed requests; performing a first prediction process based on request features, a preset prediction strategy, and a linear regression model to obtain a load prediction result; performing a second prediction process based on the parsed requests, a neural network model, and a preset prediction strategy to obtain a cache prediction result; performing an optimization fusion process based on the parsed requests, the cache prediction results, and a preset optimization fusion strategy to obtain an optimization decision; optimizing the parsed requests using the optimization decision to obtain an optimized request; executing the optimized request and mapping the optimized request to the correct physical storage location. Compared to existing technologies that struggle to adapt to dynamically changing workload scenarios, often lacking intelligent decision-making capabilities leading to resource waste or performance bottlenecks, AI-based decision acceleration efficiency is relatively low. This application directly embeds artificial intelligence (AI) algorithms (such as machine learning and neural network inference) into the kernel space, achieving intelligent processing of storage input / output (I / O) requests through steps such as feature extraction, workload prediction, and dynamic adjustment of caching strategies, thereby improving the efficiency of AI-based decision optimization.

[0172] This application also provides a possible embodiment, which firstly uses a reinforcement learning algorithm to replace the current supervised learning method; secondly, performs policy optimization through Q-learning or Actor-Critic algorithms; and finally, establishes a reward function and performs policy learning based on I / O performance metrics.

[0173] Specifically, the implementation method is as follows:

[0174] struct device_mapper_dev_rl_agent {

[0175] struct q_table q_table; / Q value table /

[0176] struct policy_network policy; / Policy network /

[0177] struct value_network value; / Value Network /

[0178] float learning_rate; / learning rate /

[0179] float discount_factor; / Discount factor /

[0180] };

[0181] int device_mapper_dev_rl_select_action(

[0182] struct device_mapper_dev_rl_agent agent,

[0183] struct io_state state)

[0184] {

[0185] / Select the optimal action based on the current state /

[0186] return epsilon_greedy_action_selection(agent, state);

[0187] }

[0188] In this embodiment, it is able to handle more complex decision-making problems and is more adaptable.

[0189] This application also provides a possible embodiment, which firstly optimizes the caching strategy parameters using a genetic algorithm; secondly, finds the optimal parameter combination through evolutionary computation; and finally, supports multi-objective optimization (latency, throughput, hit rate, etc.).

[0190] Specifically, the implementation method is as follows:

[0191] struct device_mapper_dev_ga_optimizer {

[0192] struct individual population; / population /

[0193] int population_size; / Population size /

[0194] float mutation_rate; / mutation rate /

[0195] float crossover_rate; / Crossover rate /

[0196] struct fitness_function fitness; / Fitness function /

[0197] };

[0198] int device_mapper_dev_ga_evolve_parameters(struct device_mapper_dev_ga_optimizer ga)

[0199] {

[0200] / Selection, crossover, and mutation operations /

[0201] selection(ga);

[0202] crossover(ga);

[0203] mutation(ga);

[0204] evaluate_fitness(ga);

[0205] return 0;

[0206] }

[0207] In this embodiment, the global search capability is strong, and a better combination of parameters can be found.

[0208] This application also provides a possible embodiment, which firstly uses a fuzzy logic system for caching decisions; secondly, establishes a fuzzy rule base to handle uncertain information; and finally, performs intelligent control through fuzzy inference.

[0209] Specifically, the implementation method is as follows:

[0210] struct device_mapper_dev_fuzzy_controller {

[0211] struct fuzzy_rule rule_base; / rule base /

[0212] struct membership_function mf; / Membership function /

[0213] struct defuzzifier defuzz; / Defuzzifier /

[0214] };

[0215] int device_mapper_dev_fuzzy_decision(struct device_mapper_dev_fuzzy_controller fuzzy,

[0216] struct io_features features)

[0217] {

[0218] / Fuzzification → Reasoning → Defuzzification /

[0219] float fuzzy_input = fuzzification(fuzzy, features);

[0220] float fuzzy_output = inference(fuzzy, fuzzy_input);

[0221] return defuzzification(fuzzy, fuzzy_output);

[0222] }

[0223] In this embodiment, the ability to handle uncertain information is strong, and the rules are interpretable.

[0224] This application also provides a possible embodiment, which firstly combines a hybrid architecture of multiple AI technologies; secondly, selects the most suitable AI algorithm according to different scenarios; and finally, realizes dynamic switching and combination of algorithms.

[0225] Specifically, the implementation method is as follows:

[0226] struct device_mapper_dev_hybrid_ai {

[0227] struct ml_module ml; / Machine Learning Module /

[0228] struct rl_module rl; / Reinforcement learning module /

[0229] struct fuzzy_module fuzzy; / Fuzzy logic module /

[0230] struct algorithm_selector selector; / Algorithm selector /

[0231] };

[0232] int device_mapper_dev_hybrid_decision(struct device_mapper_dev_hybrid_ai hybrid,

[0233] struct io_context context)

[0234] {

[0235] Choose the most suitable algorithm based on the context.

[0236] enum ai_algorithm algo = select_algorithm(hybrid->selector,context);

[0237] switch (algo) {

[0238] case AI_ALGORITHM_ML:

[0239] return ml_decision(hybrid->ml, context);

[0240] case AI_ALGORITHM_RL:

[0241] return rl_decision(hybrid->rl, context);

[0242] case AI_ALGORITHM_FUZZY:

[0243] return fuzzy_decision(hybrid->fuzzy, context);

[0244] }

[0245] }

[0246] In this embodiment, the advantages of multiple algorithms are combined, resulting in the strongest adaptability.

[0247] This application also provides a possible embodiment. Figure 4 This application provides a schematic diagram of the collaborative workflow of each module in its embodiments, such as... Figure 4 As shown, the AI ​​scheduler and AI acceleration device-mapper collaborate through a shared AI intelligence layer to achieve end-to-end performance optimization. The shared AI intelligence layer is responsible for global workload analysis, collaborative decision-making engine, performance prediction model, and feedback learning mechanism to ensure the coordination and consistency of the two components.

[0248] Specifically, data generates I / O requests from the application layer (database / file system / virtualization). The AI-accelerated I / O scheduler manages request queues (priority / merging / sorting), performs load prediction and schedule policy optimization (CFQ / Deadline / BFQ), and interacts with the shared AI intelligence layer through a collaborative interface. The shared AI intelligence layer performs global workload analysis, collaborative decision engine calculations, and performance prediction model derivation, driving scheduler policy adjustments and device-mapper intelligent mapping. The AI-accelerated device-mapper layer uses neural prediction and decision tree inference to achieve storage prefetching / replacement / tiering, path selection, and load balancing, ultimately mapping the optimized requests to the storage device layer (SSD / HDD / NVMe / network storage / distributed storage). In the collaborative workflow, I / O requests undergo preliminary AI analysis by the scheduler, global decision-making by the shared AI, policy coordination parameter synchronization, scheduler optimization, and device-mapper intelligent mapping in the order of nodes 1 to 7. Finally, performance feedback triggers model updates, forming a complete closed-loop process.

[0249] Specifically, the application layer and storage device layer serve as data input / output endpoints, while the three middle layers achieve intelligent collaboration through bidirectional connections: the AI-accelerated I / O scheduler performs local optimization through request queue management and the AI ​​prediction engine; the shared AI intelligence layer integrates cross-layer strategies through global workload analysis and a collaborative decision-making engine; and the AI-accelerated device-mapper completes physical execution through intelligent storage management and device mapping optimization. A cyclical logic runs throughout—the performance monitoring module (throughput / latency / IOPS) collects execution data and feeds it back to the shared AI intelligence layer's feedback learning mechanism via the collaborative interface, driving continuous iteration of the performance prediction model. Ultimately, closed-loop optimization achieves a full-cycle intelligent operation logic from request generation to strategy updates.

[0250] In this embodiment, the application layer supports diverse workload types (database / virtualization, etc.), the AI-accelerated I / O scheduler improves I / O efficiency through intelligent scheduling strategies (CFQ / Deadline, etc.), the shared AI intelligence layer achieves cross-layer optimization through global analysis and collaborative decision-making, the AI-accelerated device-mapper optimizes device utilization through intelligent storage management (prefetch / tiering) and load balancing, and the storage device layer supports flexible adaptation to multiple storage media types (SSD / NVMe, etc.). A closed-loop feedback mechanism achieves continuous self-optimization through performance monitoring and model updates, while a feedback learning mechanism ensures the system adapts to dynamic load changes. The overall layered architecture reduces module coupling, improves system scalability and maintainability, and ultimately produces significant synergistic effects in improving I / O performance, optimizing resource allocation, and enhancing adaptive capabilities.

[0251] This application also provides a possible embodiment. Figure 5 This is a schematic diagram of the operation flow of an artificial intelligence-based decision optimization system provided in an embodiment of this application, such as... Figure 5 As shown, the system adopts a multi-branch parallel processing architecture, with AI core processing, ML module processing and AI inference processing performed simultaneously. Decision quality control is carried out through a confidence threshold (80%) to ensure the reliability and accuracy of AI decisions.

[0252] Specifically, from the moment an I / O request arrives, the feature extraction module parses characteristic information such as sector address, size, access mode, and time, and synchronously inputs it into three major processing units: AI core processing (application classification / workload prediction / intelligent scheduling), ML module processing (online learning / pattern recognition / caching strategy), and AI inference processing (neural prediction / caching decision / latency optimization). The processing results converge at the confidence decision node (threshold 80%). If the confidence level > 800, policy adjustment and dynamic optimization are triggered; otherwise, information exchange directly enters the collaborative optimization module. The collaborative optimization module integrates cache operations (intelligent cache management), I / O mapping (device mapping), and performance monitoring (real-time monitoring) functions. Finally, through the feedback learning module, it completes model updates, parameter adjustments, and policy optimization, forming a complete closed-loop process from data input to decision execution and feedback iteration.

[0253] Specifically, after feature extraction, the underlying I / O requests are processed in parallel by three major processing units: the AI ​​core, the ML module, and the AI ​​inference module, forming multi-dimensional decision support. The mid-level confidence decision node selects the strategy path based on threshold judgment, ensuring efficient execution and resource optimization. The upper-level collaborative optimization module achieves precise control at the physical execution level through caching, mapping, and monitoring. A cyclical logic runs throughout—the feedback learning module feeds the execution results back to the front-end modules such as the AI ​​core processing, driving continuous model iteration and forming a full-cycle intelligent operation logic of "processing-decision-execution-feedback," ultimately achieving key performance indicators such as decision latency ≤50ms, prediction accuracy ≥95%, and cache hit rate ≥85%.

[0254] In this embodiment, the feature extraction module enhances the targeting of data processing, while the three processing units—AI core, ML, and inference—collaborate to improve decision accuracy. The confidence-based decision node ensures efficient execution path selection, and the collaborative optimization module integrates caching, mapping, and monitoring functions to achieve precise control at the physical execution level. The feedback learning module achieves continuous self-optimization through model updates and parameter adjustments, and the closed-loop feedback mechanism ensures the system adapts to dynamic load changes. The overall architecture supports modular expansion, and key performance indicators (decision latency, prediction accuracy, and cache hit rate) quantify system performance. Ultimately, it produces significant synergistic effects in improving I / O processing efficiency, optimizing resource allocation, and enhancing adaptive capabilities, forming a highly reliable and scalable intelligent decision optimization system.

[0255] Figure 6 A schematic diagram of the structure of the artificial intelligence-based decision optimization device provided in the embodiments of this application is shown below. Figure 6 As shown, the device includes: a parsing module 61, a first prediction module 62, a second prediction module 63, a fusion module 64, an optimization module 65, and an execution module 66.

[0256] The parsing module 61 is used to acquire input / output requests and parse the input / output requests to obtain parsed requests; wherein, the parsed requests include multiple request features;

[0257] The first prediction module 62 is used to perform a first prediction process based on the request characteristics, a preset prediction strategy and a linear regression model to obtain the load prediction result.

[0258] The second prediction module 63 is used to perform a second prediction process based on the parsing request, the neural network model and the preset prediction strategy to obtain the cached prediction result.

[0259] The fusion module 64 is used to perform optimization fusion processing based on the parsed request, cached prediction results and preset optimization fusion strategy to obtain optimization decisions;

[0260] Optimization module 65 is used to optimize the parsing request through optimization decisions to obtain an optimized request;

[0261] Execution module 66 is used to execute optimization requests and map optimization requests to the correct physical storage location.

[0262] In one possible design, the input / output request is parsed to obtain a parsed request, including:

[0263] The parsing module 61 is also used to perform feature extraction processing on the input / output requests to obtain a first request containing multiple request features;

[0264] The request features are classified using a classification algorithm to obtain classification labels.

[0265] The first request is tagged according to the category label and the preset tagging strategy to obtain the parsed request.

[0266] In one possible design, request characteristics include request size, access mode, priority, operation type, and starting sector, etc.

[0267] Category labels include sequential access labels, random access labels, and mixed access labels;

[0268] The preset labeling strategies include:

[0269] If the request size is less than the value of the first parameter, then the first request is tagged with a random access label;

[0270] If the request size is greater than the value of the second parameter and the access mode is sequential, then the first request will access the tags in the order they are labeled.

[0271] If the first request does not match the random access tag or the sequential access tag, then the first request is tagged with the mixed access tag.

[0272] In one possible design, based on request characteristics, a preset prediction strategy, and a linear regression model, a first prediction process is performed to obtain load prediction results, including:

[0273] The first prediction module 62 is also used to obtain historical request datasets through a sliding window and request features;

[0274] Input the historical request dataset into the linear regression model to calculate the load prediction value;

[0275] Collect real-time load values ​​and calculate the load error value based on the real-time load values ​​and the load prediction values;

[0276] The weight array of the linear regression model is updated based on the load error value to obtain an optimized linear regression model.

[0277] In one possible design, based on the parsing request, the neural network model, and a preset prediction strategy, a second prediction process is performed to obtain the cached prediction result, including:

[0278] The second prediction module 63 is also used to calculate the temporal locality feature value, spatial locality feature value and access frequency feature value based on the request characteristics;

[0279] The temporal locality feature value, spatial locality feature value, access frequency feature value and preset weight value are input into the neural network model to calculate the comprehensive prediction score;

[0280] The comprehensive prediction score is compared with the first threshold.

[0281] If the overall predicted score is greater than the first threshold, then the output cache hit is achieved.

[0282] If the overall predicted score is not greater than the first threshold, then the output cache is not hit.

[0283] In one possible design, the preset prediction strategy includes a low-latency optimization strategy;

[0284] Low-latency optimization strategies include:

[0285] If the request size is less than the second threshold, then the parsed request is tagged with a fast path label;

[0286] If the parsing request has a high priority, then the parsing request will be tagged and processed first.

[0287] In one possible design, after executing the optimization request and mapping it to the correct physical storage location, the following is also included:

[0288] Real-time performance data is collected, and the linear regression model, neural network model, and preset prediction strategy are updated based on the performance data.

[0289] This embodiment provides an artificial intelligence-based decision optimization device that can execute an artificial intelligence-based decision optimization method described in the above embodiment. Its implementation principle and technical effects are similar, and will not be repeated here.

[0290] In a specific implementation of the aforementioned AI-based decision optimization method, each module can be implemented as a processor. The processor can execute computer execution instructions stored in the memory, thereby enabling the processor to execute the aforementioned AI-based decision optimization method.

[0291] Figure 7 This is a schematic diagram of the structure of a decision optimization device based on artificial intelligence, provided as an embodiment of this application. Figure 7 As shown, the AI-based decision optimization device 70 includes at least one processor 71 and a memory 72. The AI-based decision optimization device 70 also includes a communication component 73. The processor 71, memory 72, and communication component 73 are connected via a bus 74.

[0292] In the specific implementation process, at least one processor 71 executes computer execution instructions stored in memory 72, causing at least one processor 71 to execute a method in the field of artificial intelligence as executed by the above-mentioned artificial intelligence-based decision optimization device.

[0293] The specific implementation process of processor 71 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0294] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0295] The memory may include high-speed RAM, and may also include non-volatile storage (NVM), such as at least one disk storage.

[0296] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0297] The above describes the solutions provided by the embodiments of the present invention regarding the functions implemented by the AI-based decision optimization device and the main control device. It is understood that, in order to achieve the above functions, the AI-based decision optimization device or main control device includes hardware structures and / or software modules corresponding to the execution of each function. By combining the units and algorithm steps of the various examples described in the embodiments of the present invention, the embodiments of the present invention can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed by hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the technical solutions of the embodiments of the present invention.

[0298] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the above-described method in the field of artificial intelligence.

[0299] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0300] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in an AI-based decision optimization device or master control device.

[0301] This application also provides a computer program product, comprising: a computer program stored in a readable storage medium, wherein at least one processor of an artificial intelligence-based decision optimization device can read the computer program from the readable storage medium, and the at least one processor executes the computer program to cause the artificial intelligence-based decision optimization device to perform the scheme provided in any of the above embodiments.

[0302] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disk, or optical disk.

[0303] The technical solutions of this application have been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it is readily understood by those skilled in the art that the scope of protection of this application is obviously not limited to these specific embodiments. The above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A decision optimization method based on artificial intelligence, characterized in that, include: The system acquires input / output requests and parses them to obtain parsed requests; wherein the parsed requests include multiple request features. Based on the request characteristics, the preset prediction strategy and the linear regression model, the first prediction process is performed to obtain the load prediction result; Based on the parsing request, the neural network model, and the preset prediction strategy, a second prediction process is performed to obtain the cached prediction result; Based on the parsing request, the cache prediction result, and the preset optimization fusion strategy, optimization fusion processing is performed to obtain an optimization decision; The parsing request is optimized using the optimization decision to obtain an optimized request; Execute the optimization request and map the optimization request to the correct physical storage location.

2. The method according to claim 1, characterized in that, The parsing process of the input / output request to obtain the parsed request includes: The input / output request is subjected to feature extraction processing to obtain a first request containing multiple request features; The request features are classified using a classification algorithm to obtain classification labels; The first request is tagged according to the classification label and the preset tagging strategy to obtain the parsing request.

3. The method according to claim 2, characterized in that, The request characteristics include request size, access mode, priority, operation type, and starting sector, etc. The classification labels include sequential access labels, random access labels, and mixed access labels; The preset marking strategy includes: If the request size is less than the first parameter value, then the first request is tagged with the random access label; If the request size is greater than the second parameter value and the access mode is sequential, then the first request is tagged with the sequential access label; If the first request does not match the random access tag or the sequential access tag, then the first request is tagged with the mixed access tag.

4. The method according to claim 3, characterized in that, The first prediction process, based on the request characteristics, a preset prediction strategy, and a linear regression model, to obtain the load prediction result includes: By using a sliding window and the aforementioned request features, a historical request dataset can be obtained. Input the historical request dataset into the linear regression model described above to calculate the load prediction value; Collect real-time load values, and calculate the load error value based on the real-time load values ​​and the predicted load values; The weight array of the linear regression model is updated based on the load error value to obtain an optimized linear regression model.

5. The method according to claim 4, characterized in that, The second prediction process, based on the parsing request, the neural network model, and the preset prediction strategy, to obtain the cached prediction result includes: Based on the request characteristics, temporal locality feature values, spatial locality feature values, and access frequency feature values ​​are calculated. The temporal locality feature value, the spatial locality feature value, the access frequency feature value, and the preset weight value are input into the neural network model to calculate the comprehensive prediction score; The comprehensive prediction score is compared with the first threshold. If the comprehensive prediction score is greater than the first threshold, then output a cache hit. If the comprehensive prediction score is not greater than the first threshold, then the output cache is missed.

6. The method according to claim 5, characterized in that, The preset prediction strategy includes a low-latency optimization strategy; The low-latency optimization strategy includes: If the request size is less than the second threshold, then the parsing request is tagged with a fast path label; If the priority of the parsing request is high priority, then the parsing request is tagged with a priority processing label.

7. The method according to claim 6, characterized in that, After executing the optimization request and mapping the optimization request to the correct physical storage location, the process further includes: Real-time performance data is collected, and the linear regression model, the neural network model, and the preset prediction strategy are updated based on the performance data.

8. A decision optimization device based on artificial intelligence, characterized in that, include: The parsing module is used to acquire input / output requests and parse the input / output requests to obtain parsed requests; wherein, the parsed requests include multiple request features; The first prediction module is used to perform a first prediction process based on the request characteristics, a preset prediction strategy and a linear regression model to obtain the load prediction result. The second prediction module is used to perform a second prediction process based on the parsing request, the neural network model, and the preset prediction strategy to obtain a cached prediction result. The fusion module is used to perform optimization fusion processing based on the parsing request, the cache prediction result, and the preset optimization fusion strategy to obtain an optimization decision; An optimization module is used to optimize the parsing request based on the optimization decision to obtain an optimized request; An execution module is used to execute the optimization request and map the optimization request to the correct physical storage location.

9. A decision optimization device based on artificial intelligence, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-7.

11. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method described in any one of claims 1-7.

Images