Hard disk fault prediction method and device, equipment and storage medium

Abstract

The embodiment of the invention provides a hard disk fault prediction method and device, equipment and a storage medium. The method comprises the steps of obtaining S.M.A.R.T. Data of a local to-be-predicted hard disk of a data center; performing dimension reduction processing on the S.M.A.R.T. Data of the to-be-predicted hard disk based on a principal component analysis method to obtain preprocessed S.M.A.R.T. Data of the to-be-predicted hard disk; the preprocessed S.M.A.R.T. Data of the to-be-predicted hard disk is input into a predetermined hard disk fault prediction model, a hard disk prediction result for the S.M.A.R.T. Data of the to-be-predicted hard disk is obtained, the hard disk fault prediction model is obtained through training of a long-short-term memory-converter model, and the S.M.A.R.T. Data of the to-be-predicted hard disk is obtained through training of the long-short-term memory-converter model. The hard disk prediction result comprises the health state and the remaining available life of the hard disk. By using the method, the long and short term memory-converter model can analyze the time domain features and sequence features of the hard disk S.M.A.R.T. Data in parallel under the condition of no mutual influence, the purpose of improving the hard disk fault prediction precision is achieved, the calculation efficiency is improved, the production operation and maintenance work is assisted, and the operation and maintenance cost is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of hard disk fault detection, and in particular to a hard disk fault prediction method, device, equipment and storage medium. Background Art

[0002] The main task of a data center in an enterprise is data storage and business maintenance. The larger the scale of data storage, the greater the demand for storage devices, and the frequency of storage device failures will also increase greatly. Compared with storage devices such as floppy disks and tapes, hard disks provide data centers with high-density, large-capacity, and cost-effective storage space, which also makes hard disks bear the most important storage tasks. After investigation, it was found that the probability of hard disk failure is far greater than that of redundant disk array (RAID) cards, memory and other components.

[0003] In order to prevent data loss or damage due to hard disk failure, data centers usually adopt a redundancy strategy. The so-called redundancy strategy is to use a large amount of redundant resources to back up data, and immediately enable the backup hard disk when a failure occurs, so as to ensure data security and stable business operation. The resources required for the redundancy strategy are large. For example, if a certain business requires N storage resources, it is usually necessary to provide 2N or 2N+1 resources for the business to meet the main and standby switching and ensure data security, backup and non-loss. It is not difficult to see that the use of redundancy strategies has led to a significant increase in data center resource costs and operation and maintenance costs, which is especially unacceptable for small enterprises. If hard disk failures can be predicted, the number of redundant hard disks can be reduced, and the saved resources can meet other business needs and improve resource utilization efficiency. Therefore, hard disk failure prediction technology has gradually become a key topic of concern to relevant academic and business circles.

[0004] Existing hard disk failure prediction technology relies on public hard disk SMART data to establish a neural network model for prediction. However, in the actual data center production environment, this method has obvious defects. First, the public data set is significantly different from the local data. The feasible public data set in academic research cannot provide sufficient learning of local hard disk data for machine learning models in actual production, resulting in the inability to accurately predict hard disk failures. Secondly, it is difficult to improve the accuracy of applying the neural network model trained based on the public data set to the production environment through transfer learning, which increases the equipment maintenance cost and the work pressure of operation and maintenance personnel, which runs counter to the safe and stable operation and maintenance policy of the data center. Furthermore, the traditional neural network model has a simple structure and poor generalization ability. Faced with the large and complex hard disk data in the data center, its simple structure cannot effectively train these data, and it is very easy to have underfitting phenomenon, which seriously affects the prediction accuracy. Summary of the invention

[0005] The present invention provides a hard disk fault prediction method, device, equipment and storage medium, which improve the hard disk fault prediction accuracy and improve the calculation efficiency.

[0006] In a first aspect, an embodiment of the present invention provides a hard disk failure prediction method, comprising:

[0007] Obtain the hard disk SMART data to be predicted in the data center;

[0008] Performing dimensionality reduction processing on the hard disk SMART data to be predicted based on principal component analysis to obtain preprocessed hard disk SMART data to be predicted;

[0009] The preprocessed hard disk SMART data to be predicted is input into a predetermined hard disk failure prediction model to obtain a hard disk prediction result for the hard disk SMART data to be predicted, wherein the hard disk failure prediction model is obtained by training a long short-term memory-converter model, and the hard disk prediction result includes the health status and remaining usable life of the hard disk.

[0010] In a second aspect, an embodiment of the present invention provides a hard disk failure prediction device, comprising:

[0011] A data acquisition module is used to obtain the hard disk SMART data to be predicted in the data center;

[0012] A preprocessing module, used for performing dimensionality reduction processing on the hard disk SMART data to be predicted based on a principal component analysis method to obtain preprocessed hard disk SMART data to be predicted;

[0013] The result prediction module is used to input the preprocessed hard disk SMART data to be predicted into a predetermined hard disk failure prediction model to obtain a hard disk prediction result for the hard disk SMART data to be predicted, wherein the hard disk failure prediction model is obtained by training a long short-term memory-converter model, and the hard disk prediction result includes the health status and remaining usable life of the hard disk.

[0014] In a third aspect, an embodiment of the present invention provides an electronic device, including:

[0015] at least one processor; and

[0016] a memory communicatively connected to the at least one processor; wherein,

[0017] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the hard disk failure prediction method as described in any embodiment of the present invention.

[0018] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the hard disk failure prediction method as described in any embodiment of the present invention when executed.

[0019] The embodiment of the present invention provides a hard disk fault prediction method, device, equipment and storage medium, the method comprising: first obtaining the hard disk SMART data to be predicted in the local data center; performing dimensionality reduction processing on the hard disk SMART data to be predicted based on the principal component analysis method to obtain the preprocessed hard disk SMART data to be predicted; inputting the preprocessed hard disk SMART data to be predicted into a predetermined hard disk fault prediction model to obtain a hard disk prediction result for the hard disk SMART data to be predicted, wherein the hard disk fault prediction model is obtained by training the long short-term memory-converter model, and the hard disk prediction result includes the health status and remaining usable life of the hard disk. The above technical scheme uses the hard disk SMART data set built by the local data center, avoids the limitations of the public data set in the production environment and the use of transfer learning, and ensures that the data set contains local unique features. At the same time, the long short-term memory-converter model is more flexible and easier to mine deep feature information. It analyzes the time domain features and sequence features of the hard disk SMART data in parallel under the condition of no mutual influence, achieves the purpose of improving the accuracy of hard disk fault prediction, improves the calculation efficiency, assists production and operation and maintenance work, and reduces the operation and maintenance cost, reduces the consumption of storage and computing resources, and provides more protection for the hard disk storage system.

[0020] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0022] Figure 1A flowchart of a hard disk failure prediction method provided in Embodiment 1 of the present invention;

[0023] Figure 2 A schematic diagram of a flow chart of determining a hard disk failure prediction model in another hard disk failure prediction method provided in Embodiment 2 of the present invention;

[0024] Figure 3 A schematic diagram of a long short-term memory neuron in the execution of a hard disk failure prediction method provided in Embodiment 2 of the present invention;

[0025] Figure 4 A schematic diagram of a converter in the execution of a hard disk fault prediction method provided in Embodiment 2 of the present invention;

[0026] Figure 5 An example diagram of a process for determining a long short-term memory-converter model provided in the second embodiment of the present invention;

[0027] Figure 6 A schematic diagram of the structure of a hard disk failure prediction device provided in Embodiment 3 of the present invention;

[0028] Figure 7 A schematic diagram of the structure of an electronic device provided in Embodiment 4 of the present invention. DETAILED DESCRIPTION

[0029] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0030] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0031] Embodiment 1

[0032] Figure 1 This is a flowchart of a hard disk failure prediction method provided in the first embodiment of the present invention. The method is applicable to the case of predicting hard disk failures. The method can be executed by a hard disk failure prediction device, which can be implemented in the form of hardware and / or software and can be configured in an electronic device. Figure 1 As shown, the hard disk failure prediction method provided in this embodiment may specifically include the following steps:

[0033] S101. Obtain SMART data of a hard disk to be predicted in a local data center.

[0034] Among them, the Self-Monitoring, Analysis and Reporting Technology (SMART) data refers to the various attribute data presented in the analysis and reporting technology. The SMART standard is formulated by the hard disk interface technology, which can actively monitor the operation and use status of the hard disk. SMART includes parameters such as read and write flux information, motor start-up time, and relocation sector count. When the SMART attribute value exceeds the safety threshold, the system will alarm to complete the self-test. However, the detection rate of faulty hard disks is only 3-10%, and the self-test accuracy is too low, which is not practical in production operation and maintenance.

[0035] In this embodiment, the hard disk health status is predicted by combining the local hardware SMART data of the data center with the pre-trained hard disk failure prediction model. The hard disk SMART data currently obtained from the local data center is recorded as the hard disk SMART data to be predicted.

[0036] S102 , performing dimensionality reduction processing on the hard disk SMART data to be predicted based on the principal component analysis method to obtain pre-processed hard disk SMART data to be predicted.

[0037] Among them, principal component analysis is a data dimensionality reduction algorithm that can further discover easy-to-understand features through data dimensionality reduction, speed up data processing, and improve model training efficiency. Its basic principle is to project the original data into a new coordinate system through linear transformation so that the variance of the projected data is maximized. This calculation method can ensure that the amount of information in the data after dimensionality reduction will not be lost too much, resulting in the disappearance of data features.

[0038] Specifically, before inputting the hard disk SMART data to be predicted into the hard disk failure prediction model, the hard disk SMART data to be predicted is first subjected to dimensionality reduction processing, thereby improving the computational efficiency of the long short-term memory-converter model while retaining sufficient information and reducing training time. In this embodiment, the data after dimensionality reduction processing is recorded as the pre-processed hard disk SMART data to be predicted.

[0039] S103: input the preprocessed hard disk SMART data to be predicted into a predetermined hard disk failure prediction model to obtain a hard disk prediction result for the hard disk SMART data to be predicted.

[0040] Among them, the hard disk failure prediction model is obtained by training the long short-term memory-transformer model, and the hard disk prediction results include the health status and remaining usable life of the hard disk.

[0041] It is important to know that the LSTM model is an improvement of the RNN. The emergence of the LSTM model solves the problem of gradient vanishing or exploding as the time series lengthens, and can stably and effectively learn the time domain features of the data during the training process. Its basic unit is the LSTM neuron, which contains 3 gates (forget gate, input gate, output gate) and 2 states (cell state, hidden state). The application of the gate structure makes the training effect of long time series far exceed that of the RNN. The transformer model is the main model used in the field of natural language processing. Unlike the RNN and the convolutional neural network, the basic structure of the transformer model consists of an encoder and a decoder. The encoder is responsible for encoding the input sequence into an intermediate representation, and the decoder generates the output sequence from this intermediate representation. The core idea of ​​the transformer model is the attention mechanism, which allows the model to process the input sequence in parallel, which helps to interpret long sequences and can effectively reduce training time and improve computational efficiency.

[0042] In this embodiment, the structure of the long short-term memory-transformer model includes two parts, the long short-term memory model and the transformer, which are in a cascade relationship. First, the long short-term memory model is used to preliminarily extract time domain features, and then the time domain encoder and the sensor encoder in the transformer model are used to process the time domain features and sequence features in parallel. The two directions do not interfere with each other, and better feature extraction effect and prediction accuracy are obtained.

[0043] In this step, the preprocessed hard disk SMART data to be predicted is input into a predetermined hard disk fault prediction model to obtain the hard disk health status and remaining usable life for the hard disk SMART data to be predicted, and this information is recorded as the hard disk prediction result. For example, the health status of the hard disk can be undamaged or damaged, and can also include how long the hard disk has a usable life.

[0044] The above technical solution uses the hard disk SMART data set built by the local data center, avoiding the limitations of public data sets in production environments and the use of transfer learning, and ensuring that the data set contains local unique features. At the same time, the long short-term memory-transformer model is more flexible and easier to mine deep feature information. It analyzes the time domain features and sequence features of the hard disk SMART data in parallel without affecting each other, achieving the purpose of improving the accuracy of hard disk failure prediction, improving computing efficiency, assisting production and operation and maintenance work, and reducing operation and maintenance costs, reducing storage and computing resource consumption, and providing more protection for hard disk storage systems.

[0045] Embodiment 2

[0046] Figure 2 A flowchart of determining a hard disk failure prediction model in another hard disk failure prediction method provided in Embodiment 2 of the present invention is provided. This embodiment is a further optimization of the above-mentioned embodiment. In this embodiment, the "step of determining the hard disk failure prediction model" is further limited and optimized.

[0047] like Figure 2 As shown, the second embodiment provides a hard disk failure prediction method, wherein the steps of determining a hard disk failure prediction model include:

[0048] S201. Collect SMART data of local hard disks in the data center within a set time period and build a local data set.

[0049] In this embodiment, a SMART data set of a local data center is established, and the data labels are marked as two parts: health status and remaining usable life. The data center has a large amount of hard disk data, which not only meets the training and testing needs of the long short-term memory-converter model, but also can obtain a model suitable for the local data center, which is more suitable for application in a production environment. If the equipment is upgraded, new data can be collected in time to train the model and upgrade the model version. The local data set overcomes the limitations of the public data set, making the training and test data more specific and directional, avoiding the impact of transfer learning on the prediction accuracy, and effectively ensuring the accuracy of the model prediction. The hard disk SMART data with health status and remaining usable life labels constitutes a local data set.

[0050] As a specific implementation method, the step of collecting the local hard disk SMART data of the data center within the set time period and constructing the local data set can be optimized, including:

[0051] a1) Collect SMART data of local hard disks in the data center within a set time period.

[0052] In this embodiment, the first step is to prepare the data set. It is necessary to collect the local hard disk SMART data of the data center and establish a local data set to overcome the limitations caused by the mismatch between the hard disk type of the public data set and the hard disk type of the local data center. The data set includes data on broken hard disks and data on intact hard disks. The set time period can be set according to actual conditions. For example, according to production and operation and maintenance experience, there is usually a one-month time interval from the time servers and other equipment are put on the shelf and powered on to the time they are put into production and use. Therefore, extracting SMART data from the previous 30 days can better predict hard disk failures.

[0053] b1) Mark the SMART data of the undamaged hard disk as undamaged.

[0054] In this embodiment, the health status and remaining available time of each hard disk SMART data need to be marked. Specifically, the SMART data of the undamaged hard disk is marked as an undamaged label. Continuing with the above example, the SMART data of the undamaged hard disk is marked with a label A.

[0055] c1) The damaged hard disk SMART data is marked as a damaged label, and the remaining usable life of the hard disk SMART data of each day within a set time period is numbered in reverse order, starting from the day of damage.

[0056] Specifically, the damaged hard disk SMART data is marked as a damaged label, and the hard disk SMART data of each day within the set time period is numbered in reverse order based on the day of damage. Continuing with the above example, the damaged data is marked as B, and at the same time, the data of the 30 days before the hard disk was damaged is extracted in reverse order based on the day of damage, and the data of each day is labeled as B-0, B-1, B-2,..., B-29, where B-0 represents the day of damage, B-1 represents 1 day before damage, and so on. It can be understood that the following number can be understood as the remaining usable life. For example, B-29 means that the remaining usable life of the hard disk is 29 days.

[0057] d1) The SMART data of the hard disk with labels is used to form a local data set.

[0058] In this embodiment, the SMART data of various hard disks with undamaged labels and damaged labels are combined into a data set, which is recorded as a local data set. Then, the data preprocessing stage is entered.

[0059] The above technical solution is based on the establishment of hard disk SMART data sets in the local data center, overcoming the limitations of using public data sets in production and operation, and highlighting the characteristics of local hard disks. At the same time, the SMART data set can be updated in time according to the specific changes in the data center, strengthening the relevance of the model and the data center, and ensuring the accuracy of hard disk failure prediction.

[0060] S202: Perform dimensionality reduction processing on the data in the local data set based on the principal component analysis method to obtain a preprocessed target data set.

[0061] In this embodiment, in the data preprocessing stage, the principal component analysis method is used to reduce the dimension of the data, while retaining as much data information as possible, the data dimension is reduced, and the training efficiency is improved. At the same time, the data after dimension reduction is normalized and the data format is standardized. The principal component analysis method is used to reduce the dimension of the SMART data, while retaining enough information, the computational efficiency of the long short-term memory-converter model is improved, and the training time is reduced. At present, most studies on hard disk failure prediction rarely use data dimension reduction. This is because the data will inevitably lose some information after dimension reduction, and the traditional neural network model cannot extract effective features from the data after the information is lost for data analysis, resulting in a decrease in prediction accuracy. In this embodiment, the converter model used is more flexible and has a stronger ability to obtain information than the traditional neural network model due to the uniqueness of the structure and algorithm. Therefore, it can still extract enough features from the dimension reduction data to obtain a satisfactory prediction accuracy. The data set obtained after preprocessing the local data using the principal component analysis method is recorded as the target data set.

[0062] As a specific implementation method, the steps of performing dimensionality reduction processing on local data based on principal component analysis to obtain the preprocessed target data set can be optimized, including:

[0063] a2) Based on the principal component analysis method, the SMART data of various hard disks in the local data set are reduced in dimension to obtain the local data set after dimensionality reduction.

[0064] In the data preprocessing stage, we first use the principal component analysis method to reduce the dimension of various hard disk data in the data set, while retaining the main features of the SMART data, reducing the data dimension and accelerating the training efficiency of the model. The data set obtained by reducing the dimension of various hard disk SMART data in the local data set based on the principal component analysis method is recorded as the local data set after dimension reduction.

[0065] b2) Normalizing the hard disk SMART data in the local data set after dimension reduction to obtain a normalized local data set.

[0066] In this embodiment, the data after dimensionality reduction is then normalized to ensure that all data are in [0, 1], thereby accelerating network learning efficiency. The data set obtained in this step is recorded as the normalized local data set.

[0067] c2) Divide the normalized local dataset into a training dataset and a test dataset as the target dataset.

[0068] In this embodiment, the normalized local data set is randomly divided into a training data set and a test data set. For example, all the data in the local data set is randomly divided into a training data set and a test data set in a ratio of 7:3, and the two parts are recorded as the target data set, and then enter the training phase.

[0069] S203, training and testing the constructed long short-term memory-converter model according to the target data set to obtain a hard disk failure prediction model that meets the prediction conditions.

[0070] In this embodiment, after obtaining the target data set, a long short-term memory-converter model is built for training and testing to obtain a model with high prediction accuracy for the health status and remaining usable life of the hard disk, which is recorded as a hard disk failure prediction model. The prediction condition may include that the prediction evaluation index meets the set evaluation index.

[0071] In this embodiment, the model used focuses on both the time domain and sequence characteristics of the SMART data set, extracts the time domain features through the long short-term memory module, and uses the transformer model to process the time domain features and sequence features in parallel, which not only avoids the mutual influence between the two, but also can deeply mine the information features, while improving the prediction accuracy, reducing the probability of underfitting, and overcoming the limitations of traditional neural networks such as inflexibility and weak generalization ability.

[0072] The LSTM-Transformer model is rarely used in the field of hard disk failure prediction. Its structure includes two parts: the LSTM model and the transformer, which are in a cascade relationship. First, the LSTM model is used to preliminarily extract time domain features, and then the time domain encoder and sensor encoder in the transformer model are used to process time domain features and sequence features in parallel. The two directions do not interfere with each other, and better feature extraction effect and prediction accuracy are obtained.

[0073] As a specific implementation method, the steps of training and testing the constructed long short-term memory-transformer model according to the target data set to obtain a hard disk failure prediction model that meets the prediction conditions can be optimized, including:

[0074] a3) Training the long short-term memory-converter model based on the hard disk SMART data in the training data set of the target data set to obtain a trained long short-term memory-converter model.

[0075] In this embodiment, the long short-term memory-converter model is trained based on the hard disk SMART data in the training data set, that is, the hard disk SMART data in the training data set is input into the long short-term memory-converter model, and the predicted health status and remaining service life of the hard disk are output. Then the predicted health status and remaining service life of the hard disk are compared with the label marked by the hard disk SMART data, the loss value is calculated, and the parameters of the long short-term memory-converter model are adjusted based on the loss value. Through the above repeated adjustments, a model that can better predict the health status and remaining service life of the hard disk is obtained, which is recorded as the trained long short-term memory-converter model.

[0076] Further, the step of training the long short-term memory-converter model based on the hard disk SMART data in the training data set to obtain the trained long short-term memory-converter model can be optimized, including:

[0077] a31) Inputting the data in the training data set into the long short-term memory model of the long short-term memory-transformer model to extract the first time domain features of the hard disk SMART data.

[0078] In this embodiment, in the training phase, the training set data is first fed into the long short-term memory model for training and learning. The long short-term memory model is a time recursive neural network that can decide which information to keep and which information to discard during the training process, and has excellent performance in obtaining time features. Therefore, the long short-term memory model is used in the training phase to extract the effective features of the hard disk SMART data within 30 days. For example, Figure 3 A schematic diagram of a long short-term memory neuron in a hard disk failure prediction method according to a second embodiment of the present invention is shown in FIG. Figure 3 As shown in the figure, several long short-term memory neurons are connected to form a long short-term memory model, which is mainly composed of a forget gate, an input gate, an output gate, a cell state Ct and a hidden state ht. ① represents the forget gate; ② represents the input gate; ③ represents the output gate. The forget gate determines the forgetting of the cell state Ct of the previous neuron through the activation function σ. t-1 The input is the hidden state h of the previous neuron. t-1 and the current input information x t , the output is f t :

[0079] f t =σ(Wf ·[h t-1 ,x t ]+b f ) (1)

[0080] Among them, W f and b f is the parameter of the forget gate. The input gate determines the information i to be added t , and the information obtained after calculation by the tanh function

[0081]

[0082] Among them, W i , b i , W C , b C is the parameter of the input gate. The output gate is used to output the input parameters of the next neuron, mainly the cell state C t and the hidden state h t :

[0083]

[0084] After training and learning, the hard disk SMART data with time domain features extracted by the long short-term memory model is obtained, and it is input into the converter model for further training. In this embodiment, the time domain features of the hard disk SMART data are extracted from the long short-term memory model of the long short-term memory-converter model, which is recorded as the first time domain feature.

[0085] a32) Inputting the first time domain feature of the hard disk SMART data into the converter of the long short-term memory-converter model to obtain the predicted health status and predicted remaining usable life of the hard disk.

[0086] In this embodiment, the first time domain feature of the hard disk SMART data is input into the converter of the long short-term memory-converter model to obtain output information including the predicted health status and predicted remaining usable life of the hard disk. The predicted health status of the hard disk includes whether it is damaged or not damaged.

[0087] Further, the step of optimizing the input of the first time domain feature of the hard disk SMART data into the converter of the long short-term memory-converter model to obtain the predicted health state and the predicted remaining usable life of the hard disk includes:

[0088] a321) Inputting the first time domain feature of the hard disk SMART data into the converter, processing the first time domain feature based on the time domain encoder in the converter, and extracting the second time series feature of the hard disk SMART data.

[0089] In this embodiment, the core structure of the converter model is a combination of an encoder and a decoder. Figure 4 A schematic diagram of a converter in the execution of a hard disk fault prediction method provided in the second embodiment of the present invention, such as Figure 4 As shown in the figure, the time domain encoder and sensor encoder are applied in the converter model to further analyze the time domain characteristics and attribute characteristics of the SMART data. The attention mechanism is a very important mechanism of the converter model. It can flexibly focus on and select the required information from the input, greatly improving the flexibility of the model. Specifically, the attention mechanism can effectively transfer the implicit state to the decoder, allowing the decoder to calculate the correlation between the demodulated sequences and adjust the weight of the hard disk failure prediction based on the correlation, thereby achieving effective allocation of the model's attention. The expression of the attention mechanism Attention can be expressed as:

[0090]

[0091] Among them, softmax() is an activation function, d model Represents the length of the input vector, Q represents the information vector to be queried, K represents the vector of the correlation between the queried information and other information, and V represents the vector of the queried information. The concept of multi-head attention mechanism can be derived from the attention mechanism. The multi-head attention mechanism is composed of multiple single attention mechanisms, which can learn the characteristics of different information dimensions, so that the model can better understand the data from multiple dimensions. In addition, the calculation method of the multi-head attention mechanism is parallel, which can greatly improve the calculation efficiency and training efficiency, and better grasp the deep characteristics of the information. The multi-head attention mechanism MultiHead can be expressed as:

[0092] MultiHead(Q,K,V)=Concat(head1,head2,...,head h )·W O (5)

[0093] Among them, W oRepresents a learnable weight matrix, which is used to linearly combine the outputs of all individual attention mechanisms, headi=Attention(Q,K,V), and Cancat() is the concatenation function. Since the transformer model embeds all output results into the decoder at once during training, the values ​​of some attention mechanisms are obtained through known future information (i.e., output results). However, the output of the decoder is obtained through continuous iteration of the previous results and should not be affected by future information, so a mask needs to be applied to mask future information. Therefore, a masked multi-head attention layer is introduced in the decoder to avoid the use of future information. The feedforward network layer is an important component for obtaining complex information in the input data. It consists of several fully connected layers and nonlinear activation functions, which can process the information obtained by the attention mechanism and improve the ability to learn complex relationships in the sequence. The residual connection & normalization module appears in each sublayer. The residual connection can increase the depth without losing the initial features, and the role of normalization is to reduce operations and avoid gradient disappearance. In this embodiment, the sensor encoder and the time domain encoder are used in the encoder. The two encoders are parallel calculations and can focus on the extraction of SMART data attribute features and time features respectively, avoiding the mutual influence of the two aspects of information. The sensor encoder is designed from an ordinary encoder. For the i-th hard disk data, the input of T time steps is defined as Xi = {X 1i ,X 2i ,...,X Ti}, X S ' is the transpose of the input matrix passed to the sensor encoder. The values ​​of Q, K, and V can be obtained by the following formula:

[0094] Q s =X' s W s q , K s =X' s W s k , V s =X' s W s v (6)

[0095] Among them, W S q , W S k , W S v is a trainable weight. For the attention mechanism of the sensor encoder S The expression can be expressed as:

[0096]

[0097] The expression of the multi-head attention mechanism can be expressed as:

[0098]

[0099] Among them, head i =Attention S (Q S ,K S ,V S ). The structural operation method of the time domain encoder is similar to that of the sensor encoder. During the training process, it can adaptively learn the time domain features of various types of information to achieve effective and flexible feature acquisition. The number of time domain encoders and sensor encoders is the same, and their outputs are O t and O S , can be directly spliced ​​to get the total output O r :

[0100] O r =Concat(O S ,O t ) (9)

[0101] The number of decoders is the same as that of encoders, and their output is the remaining usable life.

[0102] In addition to the functions and parameters mentioned above, the LSTM-Transformer model contains 30 LSTM neurons, 8 encoders, and 8 decoders. The LSTM neurons use the sigmoid activation function, and the normalization uses the MinMaxScaler function. The feedforward network layer in the Transformer model uses the RELU function, the normalization function also uses the MinMaxScaler function, and the loss function is the cross entropy function.

[0103] a322) Processing the first time domain feature based on the sensor encoder in the converter to extract the sequence feature of the hard disk SMART data.

[0104] In this embodiment, the first time domain feature is processed based on the sensor encoder in the converter to extract the attribute feature of the hard disk SMART data, which is recorded as the sequence feature of the hard disk SMART data.

[0105] a323) Based on the decoder in the converter, the second timing feature and the sequence feature are processed to determine the predicted health status and the predicted remaining usable life of the hard disk.

[0106] In this embodiment, the attention mechanism can effectively transmit the implicit state to the decoder, allowing the decoder to calculate the correlation between the demodulated sequences and adjust the weight of the hard disk failure prediction according to the correlation, thereby realizing the effective allocation of the model's attention. Through processing, the predicted health status and remaining usable life of the hard disk are output, which are recorded as the predicted health status and predicted remaining usable life of the hard disk, respectively.

[0107] a33) Based on the predicted health status and the predicted remaining usable life of the hard disk and the label annotated by the hard disk SMART data, the long short-term memory-converter model is adjusted to obtain a trained long short-term memory-converter model.

[0108] In this embodiment, the predicted health status and remaining service life of the hard disk are compared with the label marked by the hard disk SMART data, the loss value is calculated, and the parameters of the long short-term memory-converter model are adjusted based on the loss value. Through the above repeated adjustments, a model that can better predict the health status and remaining service life of the hard disk is obtained, which is recorded as the trained long short-term memory-converter model.

[0109] The above technical solution concretizes the training steps of the long short-term memory-transformer model.

[0110] b3) Input the hard disk SMART data in the test data set of the target data set into the trained LSTM-converter model to obtain the test output result, determine the evaluation index according to the test output result and the corresponding marked label, and use the LSTM-converter model whose evaluation index meets the evaluation requirements as the hard disk failure prediction model.

[0111] In this embodiment, the last stage is the test stage. In this stage, the data in the test data set is imported into the trained long short-term memory model and the converter model to obtain the output health status and remaining useful life data, which are compared with the test set data labels to evaluate the model performance. The main evaluation index is the root mean square error RMSE, which is expressed as:

[0112]

[0113] Among them, n represents the number of test set data, Y j , Y j 'represent the labels of the test set data and the model output values ​​respectively.

[0114] The above technical solution specifies the steps for determining the hard disk failure prediction model, and provides a basis for subsequent hard disk prediction.

[0115] Exemplarily, in order to more clearly describe the determination process of the long short-term memory-converter model provided by the embodiment of the present invention, Figure 5The flowchart of the long short-term memory-converter model determination provided in the second embodiment of the present invention is as follows: Figure 5 As shown in the figure, the process of determining the long short-term memory-transformer model includes: 1) In the data collection stage, the local hard disk SMART data set of the data center is constructed, and the data labels are marked as two parts: health status and remaining usable life. 2) In the data preprocessing stage, the principal component analysis method is used to reduce the dimension of the data, while retaining as much data information as possible, reducing the data dimension and improving the training efficiency. At the same time, the data after dimension reduction is normalized, the data format is standardized, and all the data in the data set is divided into two parts: training data set and test data set according to the proportion, and then enters the training stage. 3) Build a long short-term memory-transformer model for training and testing to obtain a model with high prediction accuracy. Its structure includes two parts: long short-term memory model and transformer, and the two are in a cascade relationship. First, the long short-term memory model is used to preliminarily extract time domain features, and then the time domain encoder and sensor encoder in the transformer model are used to process the time domain features and sequence features in parallel. The two directions do not interfere with each other, and better feature extraction effect and prediction accuracy are obtained. In the training stage, a model that meets the training termination conditions is obtained; in the testing stage, the model that meets the training termination conditions is tested, and the model that meets the test termination conditions is used as the final prediction model for subsequent use. This model can provide information support for data center staff and assist in production and operation and maintenance work. While reducing operation and maintenance costs, it can also reduce the work pressure of operation and maintenance personnel and provide a foundation for the information and intelligent construction of data centers.

[0116] Embodiment 3

[0117] Figure 6 Schematic diagram of the structure of a hard disk failure prediction device provided by the third embodiment of the present invention. The device can be applied to the situation of predicting hard disk failure. The hard disk failure prediction device can be configured in an electronic device, such as Figure 6 As shown, the device includes: a data acquisition module 31, a preprocessing module 32 and a result prediction module 33; wherein,

[0118] The data acquisition module 31 is used to acquire the hard disk SMART data to be predicted in the local data center;

[0119] A preprocessing module 32 is used to perform dimensionality reduction processing on the hard disk SMART data to be predicted based on the principal component analysis method to obtain the preprocessed hard disk SMART data to be predicted;

[0120] The result prediction module 33 is used to input the preprocessed hard disk SMART data to be predicted into a predetermined hard disk failure prediction model to obtain a hard disk prediction result for the hard disk SMART data to be predicted, wherein the hard disk failure prediction model is obtained by training the long short-term memory-converter model, and the hard disk prediction result includes the health status and remaining usable life of the hard disk.

[0121] The above technical solution uses the hard disk SMART data set built by the local data center, avoiding the limitations of public data sets in production environments and the use of transfer learning, and ensuring that the data set contains local unique features. At the same time, the long short-term memory-transformer model is more flexible and easier to mine deep feature information. It analyzes the time domain features and sequence features of the hard disk SMART data in parallel without affecting each other, achieving the purpose of improving the accuracy of hard disk failure prediction, improving computing efficiency, assisting production and operation and maintenance work, and reducing operation and maintenance costs, reducing storage and computing resource consumption, and providing more protection for hard disk storage systems.

[0122] Optionally, the device further includes a model determination module, and the model determination module includes:

[0123] The historical data collection unit is used to collect the local hard disk SMART data of the data center within a set time period and build a local data set;

[0124] A historical preprocessing unit, used to perform dimensionality reduction processing on the data in the local data set based on the principal component analysis method to obtain a preprocessed target data set;

[0125] The model determination unit is used to train and test the constructed long short-term memory-converter model according to the target data set to obtain a hard disk failure prediction model that meets the prediction conditions.

[0126] Optionally, the historical data collection unit is specifically used for:

[0127] Collect the SMART data of the local hard disk in the data center within the set time period;

[0128] Mark the undamaged hard disk SMART data as undamaged;

[0129] The damaged hard disk SMART data is marked as damaged, and the remaining usable life of the hard disk SMART data of each day within the set time period is numbered in reverse order, starting from the day of damage.

[0130] The SMART data of the hard disk with labels is used to form a local data set.

[0131] Optionally, the history preprocessing unit is specifically used to:

[0132] Based on the principal component analysis method, the SMART data of various hard disks in the local data set are reduced in dimension to obtain the local data set after dimension reduction;

[0133] Normalize the hard disk SMART data in the local data set after dimension reduction to obtain a normalized local data set;

[0134] The normalized local dataset is divided into a training dataset and a test dataset as the target dataset.

[0135] Optionally, the model determination unit includes:

[0136] A training subunit, used for training the long short-term memory-converter model based on the hard disk SMART data in the training data set of the target data set to obtain a trained long short-term memory-converter model;

[0137] The test subunit is used to input the hard disk SMART data in the test data set of the target data set into the trained long short-term memory-converter model to obtain the test output result, determine the evaluation index according to the test output result and the corresponding marked label, and use the long short-term memory-converter model whose evaluation index meets the evaluation requirements as the hard disk failure prediction model.

[0138] Optionally, the training subunit is specifically configured to:

[0139] Input the data in the training data set into the long short-term memory model of the long short-term memory-transformer model to extract the first time domain features of the hard disk SMART data;

[0140] Inputting the first time domain feature of the hard disk SMART data into the converter of the long short-term memory-converter model to obtain the predicted health status and predicted remaining usable life of the hard disk;

[0141] Based on the predicted health status and the predicted remaining usable life of the hard disk and the labels annotated by the hard disk SMART data, the long short-term memory-converter model is adjusted to obtain a trained long short-term memory-converter model.

[0142] Optionally, the first time domain feature of the hard disk SMART data is input into a converter of the long short-term memory-converter model to obtain a predicted health state and a predicted remaining usable life of the hard disk, including:

[0143] Inputting the first time domain feature of the hard disk SMART data into the converter, processing the first time domain feature based on the time domain encoder in the converter, and extracting the second time series feature of the hard disk SMART data;

[0144] The first time domain feature is processed based on the sensor encoder in the converter to extract the sequence feature of the hard disk SMART data;

[0145] The decoder in the converter processes the second timing feature and the sequence feature to determine the predicted health state and the predicted remaining usable life of the hard disk.

[0146] The hard disk failure prediction device provided in the embodiment of the present invention can execute the hard disk failure prediction method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0147] Embodiment 4

[0148] Figure 7 A schematic diagram of the structure of an electronic device provided for Embodiment 4 of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0149] like Figure 7 As shown, the electronic device 40 includes at least one processor 41, and a memory connected to the at least one processor 41, such as a read-only memory (ROM) 42, a random access memory (RAM) 43, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 41 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 42 or the computer program loaded from the storage unit 48 to the random access memory (RAM) 43. In the RAM 43, various programs and data required for the operation of the electronic device 40 can also be stored. The processor 41, the ROM 42, and the RAM 43 are connected to each other through a bus 44. An input / output (I / O) interface 45 is also connected to the bus 44.

[0150] A number of components in the electronic device 40 are connected to the I / O interface 45, including: an input unit 46, such as a keyboard, a mouse, etc.; an output unit 47, such as various types of displays, speakers, etc.; a storage unit 48, such as a disk, an optical disk, etc.; and a communication unit 49, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 49 allows the electronic device 40 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0151] The processor 41 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 41 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 41 executes the various methods and processes described above, such as a hard disk failure prediction method.

[0152] In some embodiments, the hard disk failure prediction method may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 48. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 40 via the ROM 42 and / or the communication unit 49. When the computer program is loaded into the RAM 43 and executed by the processor 41, one or more steps of the hard disk failure prediction method described above may be performed. Alternatively, in other embodiments, the processor 41 may be configured to execute the hard disk failure prediction method in any other appropriate manner (e.g., by means of firmware).

[0153] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0154] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0155] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0156] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0157] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0158] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0159] An embodiment of the present invention further provides a computer program product, including a computer program, which, when executed by a processor, implements a hard disk failure prediction method as provided in any embodiment of the present invention.

[0160] In the process of implementation, the computer program product can be written in one or more programming languages ​​or a combination thereof to perform the computer program code for the operation of the present disclosure, including but not limited to object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, using an Internet service provider to connect through the Internet).

[0161] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0162] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A hard disk failure prediction method, characterized in that: include: Obtain the hard disk SMART data to be predicted in the data center; Performing dimensionality reduction processing on the hard disk SMART data to be predicted based on principal component analysis to obtain preprocessed hard disk SMART data to be predicted; The preprocessed hard disk SMART data to be predicted is input into a predetermined hard disk failure prediction model to obtain a hard disk prediction result for the hard disk SMART data to be predicted, wherein the hard disk failure prediction model is obtained by training a long short-term memory-converter model, and the hard disk prediction result includes the health status and remaining usable life of the hard disk.

2. The method according to claim 1, characterized in that The step of determining the hard disk failure prediction model includes: Collect the local hard disk SMART data of the data center within the set time period and build a local data set; Performing dimensionality reduction processing on the data in the local data set based on the principal component analysis method to obtain a preprocessed target data set; The constructed long short-term memory-converter model is trained and tested according to the target data set to obtain a hard disk failure prediction model that meets the prediction conditions.

3. The method according to claim 2, characterized in that The collecting of the local hard disk SMART data of the data center within the set time period and constructing the local data set includes: Collect the SMART data of the local hard disk in the data center within the set time period; Mark the undamaged hard disk SMART data as undamaged; The damaged hard disk SMART data is marked as damaged, and the remaining usable life of the hard disk SMART data of each day within the set time period is numbered in reverse order, starting from the day of damage. The SMART data of the hard disk with the label constitutes the local data set.

4. The method according to claim 2, characterized in that: The performing dimensionality reduction processing on the local data based on the principal component analysis method to obtain a preprocessed target data set includes: Based on the principal component analysis method, dimensionality reduction processing is performed on various types of hard disk SMART data in the local data set to obtain a local data set after dimensionality reduction; Normalize the hard disk SMART data in the local data set after dimension reduction to obtain a normalized local data set; The normalized local data set is divided into a training data set and a test data set as the target data set.

5. The method according to claim 2, characterized in that: The step of training and testing the constructed long short-term memory-converter model according to the target data set to obtain a hard disk failure prediction model that meets the prediction conditions includes: Training a long short-term memory-converter model based on the hard disk SMART data in the training data set of the target data set to obtain a trained long short-term memory-converter model; The hard disk SMART data in the test data set of the target data set is input into the trained long short-term memory-converter model to obtain the test output result, and the evaluation index is determined according to the test output result and the corresponding marked label, and the long short-term memory-converter model whose evaluation index meets the evaluation requirements is used as the hard disk failure prediction model.

6. The method according to claim 5, characterized in that The step of training the long short-term memory-converter model based on the hard disk SMART data in the training data set to obtain a trained long short-term memory-converter model includes: Inputting the data in the training data set into the long short-term memory model of the long short-term memory-converter model to extract the first time domain feature of the hard disk SMART data; Inputting the first time domain feature of the hard disk SMART data into the converter of the long short-term memory-converter model to obtain the predicted health status and predicted remaining usable life of the hard disk; Based on the predicted health status and predicted remaining usable life of the hard disk and the label marked by the hard disk SMART data, the long short-term memory-converter model is adjusted to obtain a trained long short-term memory-converter model.

7. The method according to claim 6, characterized in that The step of inputting the first time domain feature of the hard disk SMART data into the converter of the long short-term memory-converter model to obtain the predicted health status and the predicted remaining usable life of the hard disk comprises: Inputting the first time domain feature of the hard disk SMART data into the converter, processing the first time domain feature based on the time domain encoder in the converter, and extracting the second time series feature of the hard disk SMART data; Processing the first time domain feature based on the sensor encoder in the converter to extract the sequence feature of the hard disk SMART data; The decoder in the converter processes the second timing feature and the sequence feature to determine a predicted health state and a predicted remaining usable life of the hard disk.

8. A hard disk failure prediction device, characterized in that: include: A data acquisition module is used to obtain the hard disk SMART data to be predicted in the data center; A preprocessing module, used for performing dimensionality reduction processing on the hard disk SMART data to be predicted based on a principal component analysis method to obtain preprocessed hard disk SMART data to be predicted; The result prediction module is used to input the preprocessed hard disk SMART data to be predicted into a predetermined hard disk failure prediction model to obtain a hard disk prediction result for the hard disk SMART data to be predicted, wherein the hard disk failure prediction model is obtained by training a long short-term memory-converter model, and the hard disk prediction result includes the health status and remaining usable life of the hard disk.

9. An electronic device, characterized in that: include: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the hard disk failure prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the hard disk failure prediction method according to any one of claims 1 to 7 when executed.

Images