An automatic learning method and system for time series data prediction

Abstract

The present invention relates to the field of time series data prediction and provides an automatic learning method and system for time series data prediction. The method includes using Spark to acquire time series data and preprocess the time series data; setting a hyperparameter search space and a corresponding hyperparameter search algorithm for a neural network; and distributing tasks to a cluster so that the cluster uses the neural network to perform hyperparameter tuning and model training on the tasks to obtain the optimal hyperparameter combination. Ray Serve is then used to deploy the neural network with the optimal hyperparameters on each node for use in predicting time series data. The present invention can efficiently, quickly, and conveniently automatically complete model training, hyperparameter search optimization, and model deployment for time series data prediction.

Description

Technical Field

[0001] The present invention relates to the field of time series data prediction, and in particular to an automatic learning method and system for time series data prediction. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Current time series data prediction faces the following problems:

[0004] 1. When time series data is of poor quality (e.g., disordered timestamps, missing values, outliers, and noise in the data), the feature engineering used in automated machine learning can be less than ideal for data processing, resulting in lower-than-expected prediction accuracy. Therefore, this process still requires human involvement in several important steps, including understanding the properties of domain-specific data, creating a suitable training dataset, and selecting a promising machine learning technique. Even within the same dataset, different approaches may be required. These processes often require lengthy, iterative testing, making the process inefficient for domain experts and data scientists and preventing so-called automated machine learning systems from achieving true automation.

[0005] 2. In large-scale time series data prediction scenarios, models need to be trained quickly to handle the prediction needs of hundreds of thousands of data points. For example, in an ocean area, numerous ocean sensors are distributed, collecting data on a large number of indicators. Training a single model for this data would not yield good prediction results for every single data point. Therefore, a set of specialized hyperparameter combinations is required for each indicator. This requires automated machine learning tools to simultaneously identify numerous hyperparameter combinations. Even for experienced algorithm engineers, this workload is enormous. Therefore, for those without parameter tuning experience, making effective predictions based on the available data is undoubtedly difficult.

[0006] 3. To address the significant increase in learning costs for non-algorithm engineers seeking to predict time series data, automated machine learning can be used. When processing large-scale time series data, automated machine learning alone is no longer effective. To rapidly process large-scale time series data, an architecture that seamlessly integrates data processing tools with automated machine learning is employed. Currently, the mainstream approach to integrating big data and AI is to combine Spark with AI training frameworks. However, Spark and AI tools operate in separate clusters, requiring significant data flow between them. Therefore, achieving seamless integration between Spark and AI tools can still present challenges, such as data format incompatibility, communication overhead, and resource management. Summary of the Invention

[0007] In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides an automatic learning method and system for time series data prediction. The present invention can automatically complete model training, hyperparameter search optimization and model deployment in an efficient, fast and convenient manner when predicting time series data.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions:

[0009] A first aspect of the present invention provides an automatic learning method for time series data prediction.

[0010] An automatic learning method for time series data prediction, comprising:

[0011] Use Spark to obtain time series data and preprocess the time series data;

[0012] Set the hyperparameter search space and corresponding hyperparameter search algorithm for the neural network, and distribute tasks to the cluster so that the cluster can use the neural network to perform hyperparameter tuning and model training for the tasks to obtain the optimal hyperparameter combination. Use Ray Serve to deploy the neural network with the optimal hyperparameters on each node for predicting time series data.

[0013] Furthermore, after the preprocessing, the data is divided into large object data and small object data according to the size of the preprocessed time series data; the small object data is converted into binary format through serialization and then stored in Plasma, and an ID of the small object data is set; the large object data is split into several data blocks, and the data blocks are divided into multiple shards, each of which has a reasonable and relatively uniform size. After being converted into binary format through serialization, it is stored on several nodes of Plasma, and each data block is named with a different ID.

[0014] Furthermore, when retrieving data from Plasma, it includes:

[0015] A list of data IDs corresponding to the computing task request;

[0016] Locate the node where the data object is located according to the data ID list;

[0017] Find the node where the data object is located and determine whether the data object is on the current computing task node;

[0018] If so, get a copy of the data object from Plasma;

[0019] Otherwise, the IPC mechanism is used to communicate with other nodes to obtain reference information corresponding to the data from other nodes; through the reference information, the computing task locates the node where the data object is located, and uses the reference information to access the data object in the local shared memory.

[0020] Furthermore, when retrieving data objects in Plasma, it also includes: when transmitting data, the data object is first serialized and converted into binary format at the original node to facilitate high-speed transmission between nodes, and then restored to the original data through deserialization at the computing task node.

[0021] Furthermore, when the hyperparameter search space is too large to be suitable for training every set of hyperparameter configurations, a hyperparameter search algorithm is introduced to perform different forms of search for hyperparameters.

[0022] Furthermore, when calling the hyperparameter algorithm, the hyperparameter search algorithm, the criteria to be measured, and the criteria for whether the metric is good or bad are passed as parameters to the search_alg, metric, and mode parameters.

[0023] Furthermore, during the training process, a task scheduler is used to set metrics and control the process of hyperparameter training of the running tasks.

[0024] A second aspect of the present invention provides an automatic learning system for time series data prediction.

[0025] An automatic learning system for time series data prediction, comprising:

[0026] The data acquisition and processing module is configured to: use Spark to acquire time series data and preprocess the time series data;

[0027] The hyperparameter tuning and prediction module is configured to: set the hyperparameter search space of the neural network and the corresponding hyperparameter search algorithm, and distribute tasks to the cluster so that the cluster uses the neural network to perform hyperparameter tuning and model training on the tasks to obtain the optimal hyperparameter combination, and use Ray Serve to deploy the neural network with the optimal hyperparameters on each node for predicting time series data.

[0028] A third aspect of the present invention provides a computer-readable storage medium.

[0029] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the automatic learning method for time series data prediction as described in the first aspect above.

[0030] A fourth aspect of the present invention provides a computer device.

[0031] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the automatic learning method for time series data prediction as described in the first aspect above are implemented.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] Compared to traditional model deployment tools, Ray Serve eliminates the need to reload models on every request. Instead, it loads the model into memory once and calls it directly when needed. This approach significantly reduces model deployment latency and resource consumption. Furthermore, Ray Serve's API and plugin mechanism enable users to quickly build and deploy custom services to meet the needs of various application scenarios.

[0034] The present invention can automatically complete model training, hyperparameter search optimization, and model deployment in an efficient, rapid, and convenient manner when predicting time series data.

[0035] When targeted predictions are required for different data, the automatic learning solution designed in the present invention can quickly find the optimal hyperparameter combination and can process multiple prediction tasks simultaneously in a parallel manner, greatly improving efficiency compared to manual parameter adjustment and a single automatic machine learning tool.

[0036] By caching the time series data to be processed into an object storage based on shared memory, the present invention enables big data processing tools and AI tools to read data conveniently and efficiently for calculation at the same time, thus achieving seamless integration of big data processing tools and AI tools and solving the dilemma that a single automatic machine learning framework cannot efficiently cope with large-scale time series data prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0038] Figure 1 is a flow chart of an automatic learning method for time series data prediction shown in the present invention;

[0039] Figure 2 is a diagram of a data transmission architecture shown in the present invention;

[0040] Figure 3 It is a system architecture diagram shown in the present invention. DETAILED DESCRIPTION

[0041] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0042] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0043] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0044] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the methods and systems according to the various embodiments of the present disclosure. It should be noted that each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the module, program segment, or a part of code can include one or more executable instructions for implementing the logical functions specified in the various embodiments. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, or they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the flowchart and / or block diagram, and the combination of the boxes in the flowchart and / or block diagram, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or can be implemented using a combination of dedicated hardware and computer instructions.

[0045] Explanation of terms:

[0046] Ray: Ray is a high-performance, distributed computing framework that provides fast object storage and task scheduling. Ray provides a flexible way to perform computations in a cluster that can easily scale to thousands of nodes.

[0047] Spark: A fast, general-purpose, and scalable distributed computing system for handling large-scale data processing tasks. It supports multiple programming languages ​​(such as Scala, Python, and Java) and can run on Hadoop clusters, standalone servers, or in the cloud. Spark is characterized by its speed, ease of use, high scalability, and flexibility.

[0048] Time series forecasting: This method predicts future data based on historical trends, cycles, seasonality, and other characteristics to help people make more accurate decisions. Common time series forecasting methods include statistical ARIMA and SARIMA models, and neural network-based LSTM and GRU models. These methods can handle different types of time series data and significantly improve the accuracy and stability of time series forecasting.

[0049] Automatic machine learning: uses computer algorithms and technologies to automate the entire machine learning process. The purpose is to enable non-machine learning experts (domain experts) to use machine learning, improve the efficiency of machine learning, reduce the time and cost of manual intervention, and accelerate machine learning research.

[0050] Example 1

[0051] like Figure 1 As shown, this embodiment provides an automatic learning method for time series data prediction. This embodiment uses the method applied to the server as an example for illustration. It is understandable that the method can also be applied to terminals, and can also be applied to a system including terminals, servers, and implemented through the interaction between terminals and servers. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network servers, cloud communications, middleware services, domain name services, security services CDN, and big data and artificial intelligence platforms. The terminal can be a smart phone, tablet computer, laptop computer, desktop computer, smart speaker, smart watch, etc., but is not limited to this. The terminal and the server can be directly or indirectly connected by wired or wireless communication, which is not limited in this application. In this embodiment, the method includes the following steps:

[0052] Step 1: Time series data collected by sensors received through Spark.

[0053] Step 2: Use Ray as the underlying scheduler. This allows Spark tasks to run more efficiently in a distributed environment because Ray has faster task scheduling and data transfer mechanisms. Perform data preprocessing on time series data in Spark (such as out-of-order timestamps, missing values, missing timestamps, outliers, and noise in the data).

[0054] The architecture of this invention uses Plasma as its distributed object storage system, which provides efficient data storage and access capabilities.

[0055] When data is stored in the storage system, it is divided into large objects and small objects based on the size of the data. Small object data is directly converted into binary format through serialization, transmitted to the storage system in binary format, and then restored to the original data through deserialization. When large objects are stored, they are split into smaller data blocks, each of which is stored on a node in the Plasma object storage system. These data blocks can be distributed across multiple nodes, thereby achieving distributed storage and redundant backup of data. Plasma stores data blocks in shared memory, which allows high-speed access through shared memory and avoids additional copying of data. In Plasma, data blocks are organized into a structure called "objects", and each object has a unique ID for identification. Based on the ID, computing tasks can easily obtain data.

[0056] When a computing task within the system needs to access data, the system works as follows:

[0057] (1) Sending data request: The computing task sends a data request to the system, specifying the ID list of the data objects to be obtained.

[0058] (2) Data Location: After receiving a data request, the system locates the node where the data object is located based on the unique ID of the data object. The system searches for the node where the data object is located and determines whether the data object is on the current node.

[0059] (3) Data transfer (same node): If the data object is on the current node (i.e., the requested object is already in the shared memory storage), the system will directly obtain a copy of the data object from the shared memory storage, achieving zero-copy data access. In this way, computing tasks can access data efficiently, such as Figure 2 shown.

[0060] (4) Cross-node data transmission (different nodes): If the requested data object is not on the current node, it is necessary to communicate with other nodes through the IPC (Inter-Process Communication) mechanism to obtain reference information of the data object from other nodes. Through the reference information, the computing task can locate the node where the data object is located and use the reference to access the data object in the local shared memory. When transmitting data, the data object is first serialized and converted into binary format at the original node for high-speed transmission between nodes. It is then deserialized and restored to the original data at the computing task node.

[0061] (5) Data access: Once the reference information of a data object is passed to the current node, the computing task can use this reference to access the content of the data object without the need for additional data transmission. In this way, the computing task can access data on the remote node locally, achieving efficient data sharing and access.

[0062] Step 3: After data preprocessing, the original data has been processed into the format of LSTM input, ready for the subsequent hyperparameter search.

[0063] Step 4: By analyzing the data, we set a reasonable hyperparameter search space for the LSTM (such as epochs, number of hidden layers, learning rate, batch size, etc.). While setting the hyperparameter range, we also use a corresponding hyperparameter search algorithm to determine how the hyperparameters vary within that range, such as grid_search() (grid search within a given set of values) or lograndint() (randomly sampling integers in logarithmic space). Once the hyperparameter search space is determined, we package it into a dictionary and pass it as a parameter to the training function.

[0064] Select a hyperparameter search algorithm (such as grid search, random search, Bayesian optimization, etc.) that evaluates the defined hyperparameter search space, pass it to the hyperparameter search engine through the param_space parameter, and perform reasonable evaluation of the hyperparameter search space.

[0065] The training function (Trainable) is a callable object passed to the hyperparameter search engine. It defines the LSTM training logic. The training function takes as input a dictionary containing key-value pairs of hyperparameters to be tuned, other special parameters, and a processed time series dataset. The training function also defines metrics for evaluating model performance, such as mean squared error (MSE) and mean absolute error (MAE). The goal of the training function is to execute the training process based on the given hyperparameter configuration and time series dataset, and return a metric that evaluates the model's performance as the training result. A dictionary containing various metrics and other relevant information can be passed.

[0066] When the hyperparameter range is too large to train every set of hyperparameter configurations, we can introduce hyperparameter search algorithms (such as random search and Bayesian optimization) to perform different forms of hyperparameter search. When calling a hyperparameter algorithm, we simply pass the hyperparameter search algorithm, the metric to be measured, and the metric's quality as parameters to the search_alg, metric, and mode parameters.

[0067] To make the training process more efficient, the module provides a Trial Scheduler feature. The Trial Scheduler can stop, pause, or adjust the hyperparameters of a running task, making the hyperparameter tuning process faster. Unlike search algorithms, the Trial Scheduler does not select the hyperparameter configuration to evaluate. All Trial Schedulers require only a metric, which is the value reported by the training function and the mode in which that metric will be maximized or minimized.

[0068] Once the hyperparameter search space, training function, hyperparameter search algorithm, and task scheduler are defined, they can be submitted to the hyperparameter search engine for task distribution.

[0069] Step 5: In the hyperparameter search engine of the present invention, an Actor class for training tasks is defined, usually called Trainable. Each Trainable instance represents a training task. The Trainable instance needs to implement a specific training method, set to the train() method.

[0070] The hyperparameter search engine distributes these Trainable instances across different nodes in the cluster using the specified distributed scheduling algorithm. For example, multiple Trainable instances can run on each node, enabling parallel training. Each Trainable instance runs on its own execution thread and executes the defined train() method to perform LSTM prediction training. These Trainable instances are independent, allowing training with different hyperparameter configurations to be performed in parallel.

[0071] When Trainable instances are distributed across different nodes, the hyperparameter search engine needs to ensure that these nodes can communicate with each other. This is to enable real-time collection of training progress and results, as well as to transfer data between nodes as needed. We chose to implement inter-node communication using an IPC mechanism (inter-process communication), which is fast, efficient, low-latency, resource-efficient, and suitable for multi-process communication.

[0072] Step 6: After all tasks are completed, the hyperparameter search engine will return a set of hyperparameter combinations with the best performance based on the set indicators. We can use the checkpoint to restore the best-performing prediction model to predict subsequent ocean sensor data.

[0073] Step 7: After obtaining the best-performing prediction model through checkpointing, use Ray Server to deploy the model.

[0074] Example 2

[0075] like Figure 3 As shown, this embodiment provides an automatic learning system for time series data prediction, including:

[0076] Data acquisition and processing module:

[0077] The main function of this module is to preprocess the raw ocean sensor data into a format that can be input into the LSTM model.

[0078] First, Spark is used to preprocess the poor quality time series data collected from sensors (such as unordered timestamps, missing values, outliers, and noise in the data) to address the impact of poor time series data quality on model performance. Ray is used as the foundation to help Spark better utilize distributed resources, combining Spark's computing power with Ray's efficiency and scalability to achieve efficient distributed computing and improve Spark's computing performance. After the processed time series data is stored in the Ray Object Store, automatic machine learning can use Ray to easily parallelize hyperparameter tuning tasks and efficiently share data in the cluster, thereby achieving faster model training and optimization on large-scale time series data sets.

[0079] Hyperparameter tuning module:

[0080] The main function of this module is to search for hyperparameters of the LSTM model and adjust the model structure of the LSTM through different hyperparameter combinations to adapt to effective prediction of different ocean sensor data.

[0081] The hyperparameter tuning module mainly consists of hyperparameter search space, training function, hyperparameter search algorithm, and task scheduler.

[0082] After completing all the time series data processing, select the corresponding time series prediction algorithm (such as statistical methods, machine learning, deep learning) according to the needs. In this invention, we choose the LSTM algorithm.

[0083] LSTM (Long Short-Term Memory) is an improved recurrent neural network (RNN) model specifically designed to process sequential data. Compared to traditional RNNs, LSTM overcomes the problems of vanishing and exploding gradients, enabling it to effectively capture long-term dependencies.

[0084] LSTM introduces three key gating units: the input gate, the forget gate, and the output gate. These gating units help determine which information in a sequence should be stored, forgotten, or output. The cell state is the core of the LSTM, persisting throughout the entire sequence and responsible for long-term memory, enabling the model to better process long sequences.

[0085] In the hyperparameter tuning module, we first set hyperparameters such as epoch, number of hidden layers, learning rate, and batch size. Once the hyperparameter search space is determined, we package it into a dictionary and pass it as a parameter to the training function.

[0086] The trainable function is an object passed to the hyperparameter search engine. It is a callable function that defines the model training logic. This function receives the prediction algorithm to be tuned, the hyperparameter search space, the model training logic, and the process for evaluating the model.

[0087] To make the training process more efficient, the module provides a Trial Scheduler feature. The Task Scheduler can stop, pause, or adjust the hyperparameters of a running trial, making the hyperparameter tuning process faster. Unlike search algorithms, the Task Scheduler does not select the hyperparameter configuration to evaluate. All Task Schedulers require only a metric, which is the value reported by the training function and the mode in which that metric will be maximized or minimized.

[0088] After all tasks are completed in the cluster, the hyperparameter search engine selects the best performing model based on pre-set metrics. We can return the best performing model through the checkpoint provided by Ray and deploy it using Ray Serve.

[0089] Prediction module: When the hyperparameter search engine obtains the defined content, it distributes the task to the cluster, processes large-scale ocean sensor data in parallel, selects the best hyperparameter combination for different data, and makes predictions.

[0090] Example 3

[0091] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps of the automatic learning method for time series data prediction as described in the first embodiment above are implemented.

[0092] Example 4

[0093] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the automatic learning method for time series data prediction as described in the first embodiment above are implemented.

[0094] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage and optical storage, etc.) containing computer-usable program code.

[0095] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0096] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0097] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0098] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0099] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. An automatic learning method for time series data prediction, characterized in that: include: Use Spark to obtain time series data and preprocess the time series data; Using Spark to obtain time series data is as follows: Receive time series data collected by ocean sensors through Spark; After the preprocessing, the data is divided into large object data and small object data according to the size of the preprocessed time series data; Small object data is serialized into binary format and then stored in Plasma, with an ID set for the small object data. Large object data is split into several data blocks, and each data block is divided into multiple shards. After serialization and conversion into binary format, the shards are stored on several Plasma nodes, and each data block is named with a different ID. When retrieving data from Plasma, it includes: A list of data IDs corresponding to the computing task request; Locate the node where the data object is located according to the data ID list; Find the node where the data object is located and determine whether the data object is on the current computing task node; If so, get a copy of the data object from Plasma; Otherwise, it uses the IPC mechanism to communicate with other nodes and obtain the reference information corresponding to the data from other nodes. Through the reference information, the computing task locates the node where the data object is located and uses the reference information to access the data object in the local shared memory. Set the hyperparameter search space and corresponding hyperparameter search algorithm for the neural network, and distribute tasks to the cluster so that the cluster can use the neural network to perform hyperparameter tuning and model training for the tasks to obtain the optimal hyperparameter combination. Use Ray Serve to deploy the neural network with the optimal hyperparameters on each node for predicting time series data.

2. The automatic learning method for time series data prediction according to claim 1, characterized in that: When retrieving data objects in Plasma, it also includes: when transmitting data, the data object is first serialized and converted into binary format at the original node for high-speed transmission between nodes, and then restored to the original data through deserialization at the computing task node.

3. The automatic learning method for time series data prediction according to claim 1, characterized in that: When the hyperparameter search space is too large to be suitable for training every set of hyperparameter configurations, a hyperparameter search algorithm is introduced to perform different forms of search for hyperparameters.

4. The automatic learning method for time series data prediction according to claim 3, characterized in that: When calling the hyperparameter algorithm, pass the hyperparameter search algorithm, the standard to be measured, and the standard for whether the indicator is good or bad as parameters, and pass in the search_alg, metric, and mode parameters.

5. The automatic learning method for time series data prediction according to claim 1, characterized in that: During the training process, a task scheduler is used to set metrics and control the hyperparameter training process of the running tasks.

6. An automatic learning system for time series data prediction, characterized by: include: The data acquisition and processing module is configured to: use Spark to acquire time series data and preprocess the time series data; Using Spark to obtain time series data is as follows: Receive time series data collected by ocean sensors through Spark; After the preprocessing, the data is divided into large object data and small object data according to the size of the preprocessed time series data; Small object data is serialized into binary format and then stored in Plasma, with an ID set for the small object data. Large object data is split into several data blocks, and each data block is divided into multiple shards. After serialization and conversion into binary format, the shards are stored on several Plasma nodes, and each data block is named with a different ID. When retrieving data from Plasma, it includes: A list of data IDs corresponding to the computing task request; Locate the node where the data object is located according to the data ID list; Find the node where the data object is located and determine whether the data object is on the current computing task node; If so, get a copy of the data object from Plasma; Otherwise, it uses the IPC mechanism to communicate with other nodes and obtain the reference information corresponding to the data from other nodes. Through the reference information, the computing task locates the node where the data object is located and uses the reference information to access the data object in the local shared memory. The hyperparameter tuning and prediction module is configured to set the hyperparameter search space and corresponding hyperparameter search algorithm of the neural network, and distribute tasks to the cluster so that the cluster uses the neural network to perform hyperparameter tuning and model training on the tasks to obtain the optimal hyperparameter combination, and use Ray Serve to deploy the neural network with the optimal hyperparameters on each node for predicting the preprocessed time series data.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in the automatic learning method for time series data prediction as described in any one of claims 1 to 5 are implemented.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps in the automatic learning method for time series data prediction as described in any one of claims 1-5 are implemented.

Images