Flash memory bad block prediction method, system, medium and device based on deep forest

Abstract

The application provides a flash memory bad block prediction method and system based on a deep forest, a medium and equipment, and the method comprises the following steps: obtaining historical data of a flash memory bad block, classifying and arranging the historical data to obtain a data set, and dividing the data set into a training set, a verification set and a test set; training an improved deep forest network model by using the training set to obtain a trained deep forest network model, wherein the improved deep forest network model comprises a multi-granularity scanning structure and a cascaded extreme random tree structure; optimizing the trained deep forest network model by using the verification set to obtain an optimal deep forest network model; and inputting data of the test set into the optimal deep forest network model to obtain a prediction result for the flash memory bad block. The improved deep forest network model is used to accurately predict the flash memory bad block, which is helpful for early warning before the data block becomes a bad block and prediction of the proportion of bad blocks of the entire solid state disk.

Description

Technical Field

[0001] This invention relates to the field of storage technology, and in particular to a method, system, medium, and device for predicting bad blocks in flash memory based on deep forest. Background Technology

[0002] With the rapid development of internet technology, human society has entered a big data era characterized by an exponential explosion in data volume. The big data era demands greater support from storage technologies, and various big data applications pose greater challenges to the performance and reliability of storage systems. With the emergence and widespread use of NAND flash memory (a storage device superior to hard disk drives), SSDs (Solid State Disks) based on NAND flash memory have become a hot topic in the storage field due to their higher reliability, better performance, and lower power consumption.

[0003] Due to the manufacturing process of NAND flash memory, bad blocks will inevitably appear during manufacturing and use; these are known as factory-made bad blocks and newly added bad blocks. When NAND flash memory leaves the factory, manufacturers allow a certain number of bad blocks, typically between 2% and 5%, and mark them as factory-made bad blocks. Simultaneously, flash memory devices have a limited number of erase / write cycles. After exceeding these cycles, some bits will fail to flip (from "0" to "1"), resulting in an error that prevents erasure. This block then becomes an invalid block, i.e., a newly added bad block. NAND flash memory datasheets specify that a bad block cannot be erased or written to. Therefore, if it's possible to provide early warning before a block becomes a bad block, or to predict the overall percentage of bad blocks on a solid-state drive (SSD) in advance, and to transfer the data in advance, the safety of stored data can be guaranteed, and the reliability of SSD storage can be improved. This leads to the issue of bad block prediction.

[0004] In recent years, deep learning has achieved remarkable results in outcome prediction. Built upon multi-layered neural networks—specifically, parameterized, multi-layered, differentiable nonlinear modules—it discovers better features through end-to-end training to perform predictions (or other tasks). However, deep learning models suffer from complexity, numerous hyperparameters, high training difficulty, and the need for sufficiently large datasets. Deep forest models, developed in response to deep learning models, are a novel tree-based ensemble learning method. They integrate and connect trees forming a forest, enabling the classifier to learn representations and thus improve classification performance. Deep forest models have fewer hyperparameters, are easier to train, offer better performance, are more efficient and scalable, and support small-scale training data.

[0005] Therefore, deep forest models are more suitable for solving the problem of bad block prediction in NAND flash memory. Summary of the Invention

[0006] In view of this, the purpose of this invention is to propose a flash memory bad block prediction method, system, medium and device based on deep forest, for predicting flash memory bad blocks, thereby ensuring the security of stored data.

[0007] To achieve the above objectives, this invention provides a flash memory bad block prediction method based on deep forests, comprising the following steps:

[0008] Historical data on bad blocks in flash memory is obtained, and the historical data is classified and organized to obtain a dataset, which is then divided into a training set, a validation set, and a test set.

[0009] The improved deep forest network model is trained using the training set to obtain the trained deep forest network model, which includes a multi-granularity scanning structure and a cascaded extreme random tree structure.

[0010] The trained deep forest network model is optimized using the validation set to obtain the optimal deep forest network model;

[0011] The test set data is input into the optimal deep forest network model to obtain the prediction results for bad blocks in flash memory.

[0012] In some embodiments, the improved deep forest network model is trained using a training set to obtain a trained deep forest network model, including:

[0013] The training set data is input into a multi-granularity scanning structure and sampled through multiple sliding windows to obtain multiple sample counts.

[0014] Multiple sets of sample data are input into the corresponding random forest and fully random forest respectively, and the outputs of the random forest and fully random forest corresponding to each sample size are superimposed to obtain the corresponding class vector, so as to obtain multiple sets of sample data;

[0015] Multiple sets of sample data are input into a cascaded extreme random tree structure for training, resulting in a trained deep forest network model.

[0016] In some embodiments, multiple sets of sample data are input into a cascaded extreme random tree structure for training to obtain a trained deep forest network model, including:

[0017] Multiple sets of sample data are input into the first layer of the cascaded extreme random tree structure to obtain the feature vector of the first layer of the network.

[0018] For the current layer of the network, the feature vector output by the previous layer is combined with multiple sets of sample data input to the first layer of the network, and used as the input data for the current layer of the network.

[0019] Determine whether the feature vector output by the current layer of the network is an improvement compared to the feature vector output by the previous layer;

[0020] If there is no improvement, training is stopped, and the trained deep forest network model is obtained.

[0021] In some embodiments, the method further includes:

[0022] In response to training stopping, output an array of prediction results for each sample for each feature of each bad block, and calculate the prediction accuracy and weighting for each feature;

[0023] The prediction results for each feature of each bad block are weighted according to the weighted weights to obtain the final prediction results for the training set.

[0024] In some embodiments, optimizing the trained deep forest network model using a validation set to obtain the optimal deep forest network model includes:

[0025] The trained deep forest network model is optimized by adjusting the parameters of the trained model, selecting the optimal parameters corresponding to the best prediction result, and obtaining the optimal deep forest network model based on the optimal parameters.

[0026] The parameters include the size of the sliding window in the multi-granularity scan structure, the number of extreme random trees in each layer of the cascaded extreme random tree structure, and the number of decision trees in each extreme random tree.

[0027] In some embodiments, historical data on bad blocks in flash memory is acquired, and the historical data is categorized and organized to obtain a dataset including:

[0028] Run flash memory of the same batch and type in a business simulation environment to promote the generation of bad blocks in the flash memory;

[0029] Log monitoring is performed during operation to obtain relevant data on bad blocks in the flash memory, and abnormal data is removed to obtain the data to be used;

[0030] The data to be used is classified by features to obtain multiple feature data;

[0031] Dimensionless quantization is performed on multiple feature data to give each feature data a uniform data specification, thereby obtaining a dataset.

[0032] In some embodiments, multiple feature data include: the total number of PE cycles for flash memory bad blocks, the number of PE cycles when a bad block first enters the error handling process, the total data retention time of the bad block, and the number of read interferences caused by the bad block.

[0033] Another aspect of the present invention provides a flash memory bad block prediction system based on deep forest, comprising:

[0034] The dataset module is configured to acquire historical data on bad blocks in flash memory, classify and organize the historical data to obtain a dataset, and divide the dataset into a training set, a validation set, and a test set.

[0035] The training module is configured to train the improved deep forest network model using the training set to obtain the trained deep forest network model, wherein the improved deep forest network model includes a multi-granularity scanning structure and a cascaded extreme random tree structure.

[0036] The optimization module is configured to optimize the trained deep forest network model using the validation set to obtain the optimal deep forest network model; and

[0037] The prediction module is configured to input test set data into the optimal deep forest network model to obtain prediction results for flash memory bad blocks.

[0038] In another aspect, the present invention provides a computer-readable storage medium storing computer program instructions that, when executed by a processor, implement the above-described method.

[0039] In another aspect, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program that, when executed by the processor, performs the above-described method.

[0040] The present invention has at least the following beneficial technical effects:

[0041] This invention modifies the deep forest network model by combining multi-granularity scanning with cascaded extreme random trees. The improved deep forest network model enables accurate prediction of bad blocks in flash memory, which helps to provide early warning before data blocks become bad blocks and to predict the proportion of bad blocks in the entire solid-state drive. This facilitates the early transfer of the stored data, ensures the security of the stored data, and improves the reliability of solid-state drive storage. Attached Figure Description

[0042] 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 only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a schematic diagram of a flash memory bad block prediction method based on deep forest according to an embodiment of the present invention;

[0044] Figure 2 This is a schematic diagram of the structure of a classic deep forest network model based on existing technology;

[0045] Figure 3 This is a schematic diagram of the structure of the improved deep forest network model provided in an embodiment of the present invention;

[0046] Figure 4 This is a schematic diagram of a flash memory bad block prediction system based on deep forest according to an embodiment of the present invention;

[0047] Figure 5 This is a schematic diagram of a computer-readable storage medium that implements a deep forest-based flash memory bad block prediction method according to an embodiment of the present invention.

[0048] Figure 6 This is a schematic diagram of the hardware structure of a computer device for performing a deep forest-based flash memory bad block prediction method according to an embodiment of the present invention. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to specific examples and the accompanying drawings.

[0050] It should be noted that all uses of "first" and "second" in the embodiments of the present invention are for the purpose of distinguishing two different entities or different parameters with the same name. Therefore, "first" and "second" are merely for convenience of expression and should not be construed as limiting the embodiments of the present invention. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, such as other steps or units inherent in a process, method, system, product, or device that includes a series of steps or units.

[0051] Based on the above objectives, a first aspect of the present invention provides an embodiment of a flash memory bad block prediction method based on deep forest. Figure 1 The diagram shown is a schematic representation of an embodiment of the flash memory bad block prediction method based on deep forest provided by the present invention. Figure 1As shown, the embodiments of the present invention include the following steps:

[0052] Step S10: Obtain historical data of bad blocks in flash memory, classify and organize the historical data to obtain a dataset, and divide the dataset into a training set, a validation set and a test set;

[0053] Step S20: Train the improved deep forest network model using the training set to obtain the trained deep forest network model, wherein the improved deep forest network model includes a multi-granularity scanning structure and a cascaded extreme random tree structure.

[0054] Step S30: Optimize the trained deep forest network model using the validation set to obtain the optimal deep forest network model;

[0055] Step S40: Input the test set data into the optimal deep forest network model to obtain the prediction results for flash memory bad blocks.

[0056] Figure 2 This is a schematic diagram of the structure of a classic deep forest network model based on existing technology. (Example:) Figure 2 As shown, the classic deep forest network model structure mainly includes multi-granularity scanning structure and cascaded forest structure. Multi-granularity scanning enhances the input samples of cascaded forest by sampling through sliding windows of various sizes to obtain more feature subsets with different dimensions, thereby mining more latent feature information and enhancing the feature learning ability of deep forest, thus achieving the effect of multi-granularity scanning. For a specific sliding window, the original input features are divided into examples of the same size and class, and then the examples from the same window are used to form two different random forests. Therefore, the output of each window sliding after the random forest is a class probability vector. All examples from the same input are sent to these random forests, and the outputs are superimposed to create a new class vector for a given input.

[0057] Each layer of a cascaded forest structure contains several ensemble learning random forests. This ensemble structure enables deep forests to learn representations. Each layer of a cascaded forest consists of two extreme random trees and two random forests. The cascaded forest combines the input and output data of the previous layer as the input to the next layer, allowing each layer to receive both feature information from the previous layer and the original feature information. After multiple cascades, the category corresponding to the maximum value in the prediction results of the last layer is taken as the final predicted category.

[0058] Given the characteristics of bad block data in NAND flash memory (a storage device that is better than hard disk drives), data collection is difficult and it is hard to form a large sample size; the data has many dimensions and it is difficult to determine the quantitative relationship between the data in each dimension. Therefore, the embodiments of the present invention improve the classic deep forest. Figure 3 This is a schematic diagram of the structure of the improved deep forest network model provided according to an embodiment of the present invention. Figure 3 As shown, the improved deep forest network model includes a multi-granularity scanning structure and a cascaded extreme random tree structure.

[0059] Existing technologies have not been able to accurately predict bad blocks in NAND flash memory. This invention improves the deep forest network model by combining multi-granularity scanning with cascaded extreme random trees. This improved deep forest network model enables accurate prediction of bad blocks in NAND flash memory, facilitating early warning before data blocks become bad and predicting the overall bad block percentage of the solid-state drive (SSD). This allows for the early transfer of stored data, ensuring data security and improving the reliability of SSD storage.

[0060] In some embodiments, training the improved deep forest network model using a training set to obtain the trained deep forest network model includes: inputting the training set data into a multi-granularity scanning structure and sampling through multiple sliding windows to obtain multiple sample numbers; inputting the sample data of multiple sample numbers into corresponding random forests and fully random forests respectively, and superimposing the output results of the random forests and fully random forests corresponding to each sample number to obtain corresponding category vectors, thereby obtaining multiple sets of sample data; inputting the multiple sets of sample data into a cascaded extreme random tree structure for training to obtain the trained deep forest network model.

[0061] In some embodiments, inputting multiple sets of sample data into a cascaded extreme random tree structure for training to obtain a trained deep forest network model includes: inputting multiple sets of sample data into the first layer of the cascaded extreme random tree structure to obtain the feature vector of the first layer; for the current layer, combining the feature vector output by the previous layer with the multiple sets of sample data input to the first layer as the input data of the current layer; determining whether the feature vector output by the current layer is improved compared to the feature vector output by the previous layer; and stopping training in response to no improvement to obtain a trained deep forest network model.

[0062] In some embodiments, the method further includes: in response to training stopping, outputting an array of prediction results for each feature of each sample for each bad block, and calculating the prediction accuracy and weight of each feature; and weighting the prediction results of each feature of each bad block according to the weight to obtain the final prediction results of the training set.

[0063] In some embodiments, optimizing the trained deep forest network model using a validation set to obtain the optimal deep forest network model includes: optimizing the trained deep forest network model by adjusting the parameters of the trained deep forest network model to select the optimal parameters corresponding to the optimal prediction result, and obtaining the optimal deep forest network model based on the optimal parameters; wherein the parameters include the size of the sliding window of the multi-granularity scan structure, and the number of extreme random trees in each layer of the cascaded extreme random tree structure and the number of decision trees in each extreme random tree.

[0064] In some embodiments, acquiring historical data on flash memory bad blocks and classifying and organizing the historical data to obtain a dataset includes: running flash memory of the same batch and type in a business simulation environment to promote the generation of flash memory bad blocks; monitoring logs during the operation to acquire relevant data on flash memory bad blocks and removing abnormal data to obtain data to be used; classifying the data to be used by features to obtain multiple feature data; and performing non-dimensional quantization on the multiple feature data to ensure that each feature data has a uniform data specification, thereby obtaining a dataset.

[0065] In some embodiments, multiple feature data include: the total number of PE cycles for flash memory bad blocks, the number of PE cycles when a bad block first enters the error handling process, the total data retention time of the bad block, and the number of read interferences caused by the bad block.

[0066] In this embodiment, PE (Program / Erase) represents programming and erasing / writing.

[0067] The following is an exemplary embodiment of the flash memory bad block prediction method based on deep forest of the present invention:

[0068] For NAND flash memory of the same batch and type from the same manufacturer, a self-developed SSD (Solid State Disk) experimental platform was used to simulate real-world SSD operations and run workloads under real-world SSD scenarios, thus promoting the generation of bad blocks. During this process, the SSD firmware runtime logs were monitored, and data related to NAND flash memory bad blocks was updated and statistically analyzed. The statistical data related to NAND flash memory bad blocks are as follows:

[0069]

[0070] After acquiring 1000 NAND flash memory bad block operation data sets, a four-dimensional array [Block_PE, Block_Err_PE, Block_DR, Block_RD] was formed, with each array representing a NAND flash memory bad block data set. First, data with potential outliers or missing values ​​was cleaned and removed to reduce the impact of abnormal data. Then, the data was standardized using non-dimensional quantization to ensure uniform data specifications for all features. Finally, the 1000 collected NAND flash memory bad block data sets were randomly divided into three parts: 60% as the training set, 20% as the validation set, and 20% as the test set.

[0071] like Figure 2 and Figure 3 As shown, the classic deep forest network model uses random forests in its cascaded layers, while the improved deep forest network model uses extreme random trees in its cascaded layers. The difference between extreme random trees and random forests is that extreme random trees select branching features completely randomly, and each tree in an extreme random tree is trained using all samples, meaning each tree has the same sample set. This improves the model's ability to prevent overfitting and results in excellent performance on the test set. This embodiment fully utilizes the superior predictive performance of extreme random trees by replacing them entirely with extreme random trees.

[0072] Specifically, the improved deep forest network model is trained using the training set. First, 600 data points from the training set are used as input data for multi-granularity scanning. It's important to note that although the feature data for each bad block is a multi-dimensional array, during multi-granularity scanning, the feature array of each bad block is treated as a whole to meticulously divide the samples for multi-granularity scanning. When inputting the random forest for training, each dimension of the feature data is trained separately to maintain the integrity of each bad block's data. The training results for each feature of each bad block are also output as the same result array.

[0073] After inputting 600 training set data into the multi-granularity scanning structure, the sample data is obtained by sampling through sliding windows of 100, 200, and 300 respectively (i.e., multiple sliding windows). The sample data can be obtained by referring to the following formula (1):

[0074] N samples =(Sum train -Window Size ) / λ+1 (1)

[0075] In the above formula (1), N samples Sum represents the number of samples after the sliding window sampling. train λ represents the total number of samples in the training set, and λ represents the window sliding step size. In order to acquire as many samples as possible, λ is set to 1 in this embodiment.Size This represents the size of the sliding window.

[0076] According to the formula, the number of samples after sampling through sliding windows of 100, 200, and 300 are 501, 401, and 301, respectively. These samples are then input into a random forest and a fully random forest (both are random forests containing 1000 random trees, the difference being: when selecting split nodes, the fully random forest generally selects specific features randomly, and then continues to grow until a single leaf node has only one category; the ordinary random forest selects sqrt(k) features on each tree, also following the random sampling principle, and then uses the Gini index as the selection criterion for split nodes). The output of each random forest is the same as the number of input samples (the result is singular). The outputs of the two random forests are superimposed to obtain a new category vector, with corresponding category vector sizes of 1002, 802, and 602, respectively, finally obtaining 2406 sets of sample data.

[0077] The 2406 sets of sample data obtained above were input into the first layer of the cascaded extreme random trees. Each subsequent layer's input was a combination of the feature vector obtained from the previous layer and the original input (i.e., the input to the first layer). The first layer used default parameters, and the training process terminated if no significant improvement was observed when expanding to a new level. Therefore, the number of cascaded layers was automatically determined. In the cascaded extreme random trees part, this training setup included 4 extreme random trees per layer, and each extreme random tree contained 100 decision trees.

[0078] The cascaded layers of the classic deep forest network model use an averaging method in the final result processing. This method is not suitable for predicting multi-scale features of NAND flash memory bad blocks. This paper improves upon this method by calculating the influence weight of each feature on NAND flash memory bad blocks by calculating the prediction accuracy. The prediction results of each scale feature are then weighted according to the weights to obtain the final prediction result.

[0079] After the cascaded extreme random tree automatically stops training, output the prediction result array for each sample with respect to each feature of each bad block, and calculate the prediction accuracy and weighting of each feature using the following formulas (2) and (3):

[0080]

[0081]

[0082] Among them, P true S represents the prediction accuracy, n represents the nth feature, and S represents the prediction accuracy. true Sum represents the number of correct predictions, W represents the total number of predictions. kThis represents the weighted weights. The prediction results for each feature of each bad block are weighted according to these weights to obtain the final prediction results for the training set. Multiple attempts can be made by modifying the size of the multi-granularity scanning sliding window, the number of layers in the cascaded extreme random trees, and the number of decision trees contained in each random tree, and the prediction results can be compared to obtain the optimal prediction result.

[0083] Next, the trained deep forest network model is optimized using a validation set to obtain the optimal deep forest network model. Specifically, adjustable parameters such as the size of the multi-granularity scanning sliding window, the number of layers in the cascaded extreme random trees, and the number of decision trees contained in each random tree are fine-tuned. The above steps are repeated, and the prediction results are compared. The parameters with the best prediction results are then selected to obtain the optimal deep forest network model.

[0084] Finally, the test set is input into the optimal deep forest network model to obtain the bad block prediction results, thus testing the accuracy of the deep forest network model.

[0085] A second aspect of this invention provides a flash memory bad block prediction system based on deep forests. Figure 4 The diagram shown is a schematic representation of an embodiment of the deep forest-based flash memory bad block prediction system provided by the present invention. Figure 4 As shown, a flash memory bad block prediction system based on deep forest includes: a dataset module 10, configured to acquire historical data of flash memory bad blocks, classify and organize the historical data to obtain a dataset, and divide the dataset into a training set, a validation set, and a test set; a training module 20, configured to train an improved deep forest network model using the training set to obtain a trained deep forest network model, wherein the improved deep forest network model includes a multi-granularity scan structure and a cascaded extreme random tree structure; an optimization module 30, configured to optimize the trained deep forest network model using the validation set to obtain an optimal deep forest network model; and a prediction module 40, configured to input the test set data into the optimal deep forest network model to obtain prediction results for flash memory bad blocks.

[0086] A third aspect of the present invention also provides a computer-readable storage medium. Figure 5 A schematic diagram of a computer-readable storage medium implementing a deep forest-based flash memory bad block prediction method according to an embodiment of the present invention is shown. Figure 5 As shown, the computer-readable storage medium 3 stores computer program instructions 31. When executed by a processor, the computer program instructions 31 implement the method of any of the above embodiments.

[0087] It should be understood that, where there is no conflict, all the embodiments, features and advantages described above for the flash memory bad block prediction method based on deep forest according to the present invention are equally applicable to the flash memory bad block prediction system and storage medium based on deep forest according to the present invention.

[0088] A fourth aspect of the present invention also provides a computer device, including as follows: Figure 6 The memory 402 and processor 401 shown are provided. The memory 402 stores a computer program that, when executed by the processor 401, implements the method of any of the above embodiments.

[0089] like Figure 6 The diagram shown is a hardware structure schematic of an embodiment of a computer device for implementing a deep forest-based flash memory bad block prediction method provided by the present invention. Figure 6 Taking the computer device shown as an example, this computer device includes a processor 401 and a memory 402, and may also include an input device 403 and an output device 404. The processor 401, memory 402, input device 403, and output device 404 can be connected via a bus or other means. Figure 6 Taking a bus connection as an example, input device 403 can receive input digital or character information, and generate key signal inputs related to user settings and function control of the deep forest-based flash memory bad block prediction system. Output device 404 may include display devices such as a display screen.

[0090] Memory 402, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions / modules corresponding to the deep forest-based flash memory bad block prediction method in this embodiment. Memory 402 may include a program storage area and a data storage area, wherein the program storage area may store the operating system and application programs required for at least one function; the data storage area may store data created by using the deep forest-based flash memory bad block prediction method, etc. In addition, memory 402 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, memory 402 may optionally include memory remotely located relative to processor 401, and these remote memories can be connected to the local module via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0091] The processor 401 executes various server functions and data processing by running non-volatile software programs, instructions, and modules stored in the memory 402, thereby implementing the deep forest-based flash memory bad block prediction method of the above method embodiment.

[0092] Finally, it should be noted that the computer-readable storage medium (e.g., memory) described herein can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. By way of example, and not limitation, non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM), which can act as external cache memory. By way of example, and not limitation, RAM can be obtained in various forms, such as synchronous RAM (DRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct Rambus RAM (DRRAM). The storage devices disclosed herein are intended to include, but are not limited to, these and other suitable types of memory.

[0093] Those skilled in the art will also understand that the various exemplary logic blocks, modules, circuits, and algorithm steps described in conjunction with the disclosure herein can be implemented as electronic hardware, computer software, or a combination of both. To clearly illustrate this interchangeability between hardware and software, the functionality of various illustrative components, blocks, modules, circuits, and steps has been generally described. Whether this functionality is implemented as software or as hardware depends on the specific application and the design constraints imposed on the system as a whole. Those skilled in the art can implement the functionality in various ways for each specific application, but such implementation decisions should not be construed as departing from the scope of the embodiments disclosed herein.

[0094] The above are exemplary embodiments disclosed in this invention. However, it should be noted that various changes and modifications can be made without departing from the scope of the embodiments of this invention as defined by the claims. The functions, steps, and / or actions of the methods according to the disclosed embodiments described herein do not need to be performed in any particular order. Furthermore, although the elements disclosed in the embodiments of this invention may be described or claimed individually, they may be understood as multiple unless explicitly limited to a singular number.

[0095] It should be understood that, as used herein, the singular form "a" is intended to include the plural form as well, unless the context clearly supports an exception. It should also be understood that, as used herein, "and / or" refers to any and all possible combinations of one or more of the associatedly listed items. The embodiment numbers disclosed above are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0096] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of the invention (including the claims) is limited to these examples. Within the framework of the invention, technical features of the above embodiments or different embodiments can be combined, and many other variations of different aspects of the invention exist, which are not provided in the details for the sake of brevity. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the protection scope of the invention.

Claims

1. A flash memory bad block prediction method based on deep forest, characterized in that, Includes the following steps: Historical data on bad blocks in flash memory is obtained, and the historical data is classified and organized to obtain a dataset, which is then divided into a training set, a validation set, and a test set. The improved deep forest network model is trained using the training set to obtain a trained deep forest network model, wherein the improved deep forest network model includes a multi-granularity scanning structure and a cascaded extreme random tree structure. The trained deep forest network model is optimized using the validation set to obtain the optimal deep forest network model. The data from the test set is input into the optimal deep forest network model to obtain prediction results for bad blocks in flash memory; The process involves acquiring historical data on bad blocks in flash memory and classifying and organizing this historical data to obtain a dataset. This includes: running flash memory of the same batch and type in a business simulation environment to promote the generation of bad blocks; monitoring logs during the operation to acquire relevant data on bad blocks and removing abnormal data to obtain data to be used; classifying the data to be used by features to obtain multiple feature data; and performing dimensionless quantization on the multiple feature data to ensure that each feature data has a uniform data specification, thereby obtaining the dataset.

2. The method according to claim 1, characterized in that, The improved deep forest network model is trained using the training set to obtain the trained deep forest network model, which includes: The training set data is input into the multi-granularity scanning structure and sampled through multiple sliding windows to obtain multiple sample counts. The sample data of the multiple sample sizes are respectively input into the corresponding random forest and fully random forest, and the output results of the random forest and fully random forest corresponding to each sample size are superimposed to obtain the corresponding category vector, so as to obtain multiple sets of sample data; The multiple sets of sample data are input into the cascaded extreme random tree structure for training, resulting in a trained deep forest network model.

3. The method according to claim 2, characterized in that, The multiple sets of sample data are input into the cascaded extreme random tree structure for training, resulting in a trained deep forest network model including: The multiple sets of sample data are input into the first layer network of the cascaded extreme random tree structure to obtain the feature vector of the first layer network. For the current layer network, the feature vector output by the previous layer network is combined with the multiple sets of sample data input to the first layer network to serve as the input data for the current layer network. Determine whether the feature vector output by the current layer of the network is improved compared to the feature vector output by the previous layer of the network; If there is no improvement, training is stopped, and the trained deep forest network model is obtained.

4. The method according to claim 3, characterized in that, Also includes: In response to training halt, the system outputs an array of prediction results for each sample for each feature of each bad block, and calculates the prediction accuracy and weighting for each feature. The calculation formula includes... ； ； Among them, P true S represents the prediction accuracy, n represents the nth feature, and S represents the prediction accuracy. true Sum represents the number of correct predictions, and W represents the total number of predictions. k Represents the weighted average; The prediction results of each feature of each bad block are weighted according to the weighted weights to obtain the final prediction results of the training set.

5. The method according to claim 4, characterized in that, Optimizing the trained deep forest network model using the validation set to obtain the optimal deep forest network model includes: The trained deep forest network model is optimized by adjusting its parameters to select the optimal parameters corresponding to the optimal prediction result, and the optimal deep forest network model is obtained based on the optimal parameters. The parameters include the size of the sliding window of the multi-granularity scanning structure, the number of extreme random trees in each layer of the cascaded extreme random tree structure, and the number of decision trees in each extreme random tree.

6. The method according to claim 1, characterized in that, The multiple characteristic data include: the total number of PE cycles for bad blocks in the flash memory, the number of PE cycles when a bad block first enters the error handling process, the total data retention time of the bad block, and the number of read interferences caused by the bad block.

7. A flash memory bad block prediction system based on deep forest, characterized in that, include: The dataset module is configured to acquire historical data on bad blocks in flash memory, classify and organize the historical data to obtain a dataset, and divide the dataset into a training set, a validation set, and a test set. The training module is configured to train the improved deep forest network model using the training set to obtain the trained deep forest network model, wherein the improved deep forest network model includes a multi-granularity scanning structure and a cascaded extreme random tree structure. An optimization module is configured to optimize the trained deep forest network model using the validation set to obtain an optimal deep forest network model; and The prediction module is configured to input the data from the test set into the optimal deep forest network model to obtain prediction results for bad blocks in flash memory. The dataset module is further configured to run flash memory of the same batch and type in a business simulation environment to promote the generation of bad blocks in the flash memory; to monitor logs during the operation, to obtain relevant data on bad blocks in the flash memory, and to remove abnormal data to obtain data to be used; to classify the data to be used by features to obtain multiple feature data; and to perform non-dimensional quantization on the multiple feature data to make each feature data have a uniform data specification, thereby obtaining a dataset.

8. A computer-readable storage medium, characterized in that, The system stores computer program instructions that, when executed by a processor, implement the method as described in any one of claims 1-6.

9. A computer device, comprising a memory and a processor, characterized in that, The memory stores a computer program, which, when executed by the processor, performs the method as described in any one of claims 1-6.

Images