Multi-target prediction method, system and device for financial time sequence and storage medium

Abstract

The invention discloses a multi-target prediction method, system and device for a financial time sequence and a storage medium, and relates to the technical field of artificial intelligence, and the method comprises the steps: firstly obtaining financial time sequence data and auxiliary information, and carrying out the vector conversion and fusion to form a unified input representation; extracting features through a shared encoder, and generating a global feature vector; and inputting the data to a multi-task prediction output head to realize parallel prediction of the prediction tasks. According to the scheme, cross-task feature sharing is realized by using the shared encoder, the modeling collaboration is improved, and resource waste and prediction conflicts are avoided; through data fusion, the depiction capability of complex financial dynamics is enhanced; the generalization and robustness of the model are improved by adopting a lightweight output head, the method is suitable for diversified financial prediction scenes, and the prediction efficiency and accuracy are improved while the model complexity is reduced.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a multi-objective prediction method, system, device, and storage medium for financial time series. Background Art

[0002] For multi-objective forecasting tasks involving monetary time series data, independent modeling is often used to predict different tasks separately. However, this approach has several significant shortcomings in practical applications and cannot fully meet the requirements for efficiency, robustness, and multi-task coordination in complex financial scenarios.

[0003] Specifically, first, most existing methods are single-task modeling, which not only lacks a unified multi-objective prediction framework, but also lacks a collaborative optimization mechanism between tasks, which makes the model generalization ability weak; second, the support capability for continuous prediction of multiple time points is insufficient, and it is not suitable for financial scenarios that require continuous prediction of trends or risk changes; at the same time, in actual financial scenarios, in addition to time series data, there is also a large amount of unstructured and heterogeneous information. Traditional models are mostly based on numerical features, and have limited modeling capabilities for text or knowledge information, making it difficult to play the role of these external data; since multiple prediction tasks rely on different models to run independently, the system needs to repeatedly load the model and calculation process when executing the prediction, resulting in low overall prediction efficiency and difficulty in meeting real-time requirements. Summary of the Invention

[0004] The main purpose of this application is to provide a multi-objective prediction method, system, device and storage medium for financial time series, aiming to solve the technical problems of task separation, low efficiency and weak generalization ability in current financial series analysis.

[0005] To achieve the above objectives, the present application proposes a multi-objective forecasting method for financial time series, which includes:

[0006] Obtain financial time series data and auxiliary information;

[0007] Performing vector conversion and fusion based on the financial time series data and the auxiliary information to form a unified input representation;

[0008] Performing feature extraction on the input representation through a preset shared encoder, outputting a global feature vector, and inputting the global feature vector into a preset multi-task prediction output head;

[0009] The preset multi-task prediction output head is used to predict the prediction task according to the global feature vector to generate a corresponding target prediction result.

[0010] In one embodiment, the step of performing vector conversion and fusion based on the financial time series data and the auxiliary information to form a unified input representation includes:

[0011] Converting the financial time series data into a time embedding vector;

[0012] Mapping the auxiliary information to a separate embedding space based on a pre-trained language model or embedding model to obtain a corresponding knowledge embedding vector;

[0013] The time embedding vector and the knowledge embedding vector are fused to form a unified input representation.

[0014] In one embodiment, the step of extracting features from the input representation using a preset shared encoder and outputting a global feature vector includes:

[0015] In the preset shared encoder, interactively modeling the feature dimensions in the input representation through a multi-head attention mechanism to obtain an attention-weighted feature vector;

[0016] Performing a residual connection on the attention weighted feature vector and the input representation to obtain a superposition result;

[0017] The superposition results are normalized through the layer normalization mechanism to obtain the global feature vector.

[0018] In one embodiment, the prediction task includes a single-time-point task and a multi-time-point task, and the single-time-point task includes a regression task and a classification task;

[0019] The step of predicting the prediction task according to the global feature vector by the preset multi-task prediction output head to generate a corresponding target prediction result includes:

[0020] In the preset multi-task prediction output header, identifying the prediction task type;

[0021] If the prediction task type is the regression task, linear mapping is applied to the global feature vector to output a continuous value;

[0022] If the prediction task type is the classification task, the global feature vector is converted into a raw score through a fully connected layer in a preset multi-task prediction output head, and the raw score is converted into a category probability distribution through a normalized exponential function;

[0023] If the prediction task type is the multi-time point task, a time-expanded multi-output structure is constructed to generate a target prediction result of a preset time step.

[0024] In one embodiment, after the step of predicting the prediction task according to the global feature vector by the multi-task prediction output head and generating the corresponding target prediction result, the method further includes:

[0025] Combining the target prediction result with a preset loss function, jointly training and optimizing the financial time series prediction model to obtain a pre-trained financial time series prediction model;

[0026] Use the pre-trained financial time series forecasting model for online or batch forecasting.

[0027] In one embodiment, the step of combining the target prediction result with a preset loss function to jointly train and optimize the financial time series prediction model to obtain a pre-trained financial time series prediction model includes:

[0028] According to each task type, extracting corresponding prediction tasks from the target prediction results;

[0029] Calculating the corresponding subtask loss value for each prediction task;

[0030] Perform weighted summation based on the loss value of each subtask and the corresponding loss weight to obtain a total loss value;

[0031] Based on the total loss value, back-propagation optimization and parameter update are performed on the financial time series forecasting model to obtain a pre-trained financial time series forecasting model.

[0032] In one embodiment, the step of performing online or batch prediction using the pre-trained financial time series prediction model includes:

[0033] Based on a preset single forward propagation strategy, the pre-trained financial time series prediction model is used to perform synchronous online or batch prediction on multiple tasks.

[0034] In addition, to achieve the above objectives, the present application also proposes a multi-objective prediction system for financial time series, which includes:

[0035] Acquisition module, used to obtain financial time series data and auxiliary information;

[0036] A data processing module, configured to perform vector conversion and fusion based on the financial time series data and the auxiliary information to form a unified input representation;

[0037] A feature extraction module, configured to extract features from the input representation using a preset shared encoder, output a global feature vector, and input the global feature vector into a preset multi-task prediction output head;

[0038] The task prediction module is used to predict the prediction task according to the global feature vector through the multi-task prediction output head and generate a corresponding target prediction result.

[0039] In addition, to achieve the above-mentioned purpose, the present application also proposes a multi-objective prediction device for financial time series, which includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the computer program is configured to implement the steps of the multi-objective prediction method for financial time series as described above.

[0040] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and stores a computer program on the storage medium. When the computer program is executed by the processor, it implements the steps of the multi-objective prediction method of financial time series as described above.

[0041] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the steps of the multi-objective forecasting method of financial time series as described above.

[0042] This application proposes a multi-target prediction method, system, device, and storage medium for financial time series. The method comprises: obtaining financial time series data and auxiliary information; performing vector conversion and fusion based on the financial time series data and auxiliary information to form a unified input representation; extracting features from the input representation using a preset shared encoder, outputting a global feature vector, and inputting the global feature vector into a preset multi-task prediction output head; and predicting the prediction task based on the global feature vector using the preset multi-task prediction output head to generate the corresponding target prediction result. By introducing a shared encoder to extract features from the input data, this solution enables parallel collaborative modeling of multiple types of tasks, avoiding resource waste and prediction conflicts caused by task separation. Simultaneously, the fusion of financial time series data and auxiliary information significantly enhances the model's ability to model complex financial dynamics. Furthermore, outputting prediction results through the preset multi-task prediction output head ensures the independence and interpretability of each task's output logic while enabling cross-task knowledge transfer and regularization through a shared underlying feature representation, thereby improving the generalization and robustness of the overall model. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0044] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0045] Figure 1 A flowchart of the first embodiment of the multi-objective prediction method for financial time series provided in this application;

[0046] Figure 2 A flowchart of the second embodiment of the multi-objective prediction method for financial time series provided in this application;

[0047] Figure 3 A schematic diagram of a simplified flow chart of the multi-objective forecasting method for financial time series provided in Example 1 of the present application;

[0048] Figure 4 This is a schematic diagram of the module structure of the multi-objective prediction system for financial time series according to an embodiment of the present application;

[0049] Figure 5 Schematic diagram of the device structure of the hardware operating environment involved in the multi-objective prediction method for financial time series in the embodiment of the present application.

[0050] The purpose, features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0051] It should be understood that the specific embodiments described herein are merely used to explain the technical solutions of the present application and are not intended to limit the present application.

[0052] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0053] The main solutions of the embodiments of the present application are: obtaining financial time series data and auxiliary information; performing vector conversion and fusion based on the financial time series data and auxiliary information to form a unified input representation; extracting features from the input representation through a preset shared encoder, outputting a global feature vector, and inputting the global feature vector into a preset multi-task prediction output head; predicting the prediction task based on the global feature vector through the preset multi-task prediction output head to generate the corresponding target prediction result.

[0054] In this embodiment, for ease of description, the following description is made with the financial time series target prediction system as the execution subject.

[0055] Existing technologies often use independent modeling to predict different tasks. This approach suffers from the following shortcomings in practical applications: First, different prediction tasks are typically trained and deployed using separate models, such as regression models for trend prediction and classification models for risk identification. This approach not only results in a large number of models and complex system maintenance, but also prevents different tasks from sharing underlying feature information, wasting computing resources. Second, traditional modeling methods often focus on a single task objective, ignoring the potential connections between multiple related tasks. This fragmented modeling approach makes the model prone to overfitting and makes it difficult to benefit from related tasks to improve overall performance. Many current forecasting systems only support forecasting data for a specific future point in time and lack the ability to model continuous time series trends. This makes them unsuitable for financial scenarios that require continuous prediction of trends or risk changes. Furthermore, in real-world financial scenarios, in addition to time series data, there is also a large amount of unstructured and heterogeneous information, such as macroeconomic indicators, sentiment factors, industry reports, and market text. Traditional models, which primarily rely on numerical features, have limited ability to model textual or knowledge-based information, making it difficult to leverage this external data. In addition, since multiple prediction tasks rely on different models to run independently, the system needs to repeatedly load the model and calculation process when executing the prediction, resulting in long response time and high computing resource overhead, making it difficult to adapt to the real-time monitoring and rapid response needs of financial business.

[0056] This application provides a solution. By introducing a task-sharing coding structure and a task-specific decoupling mechanism, the present invention realizes parallel collaborative modeling of multiple tasks such as regression, classification, and multi-time point prediction, avoiding resource waste and prediction conflicts caused by task separation. At the same time, the present invention innovatively introduces a knowledge embedding module to effectively integrate financial time series data and unstructured knowledge information, significantly improving the model's ability to model complex financial dynamics. In addition, the system design supports the joint optimization of multi-task loss functions, combined with the task weight adjustment mechanism, to achieve efficient collaboration and adaptive balance between different tasks, and enhance the overall generalization ability and stability of the system.

[0057] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, mobile phone, etc., or an electronic device capable of performing the above functions. The following uses a personal computer as an example to illustrate this embodiment and the following embodiments.

[0058] Based on this, the embodiment of the present application provides a multi-objective prediction method for financial time series, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the multi-objective forecasting method for financial time series of this application.

[0059] In this embodiment, the multi-objective prediction method for financial time series includes steps S10 to S40:

[0060] Step S10, obtaining financial time series data and auxiliary information;

[0061] It should be noted that the financial time series data refers to numerical sequence data collected at continuous or discrete time points that reflects the dynamic changes of financial assets or financial markets, including but not limited to: historical price data of financial products such as stocks, bonds, futures, and foreign exchange (such as opening price, closing price, highest price, and lowest price); trading behavior data such as trading volume, turnover rate, and buy and sell order data; and derivative time series indicators such as volatility, yield, and technical indicators (such as moving average, MACD, and RSI).

[0062] Auxiliary information refers to supplementary information that can influence or explain financial market behavior in addition to original financial time series data. It includes two types of data:

[0063] Auxiliary data: refers to structured or semi-structured numerical or categorical data;

[0064] Auxiliary information: refers to unstructured or rule-based knowledge information.

[0065] In this example, the acquired raw financial time series data is first standardized, including but not limited to preprocessing operations such as missing value filling, outlier detection, data smoothing, and normalization, to improve data quality and eliminate dimensional differences. For example, linear normalization or Z-score normalization is used to map indicators of different magnitudes, such as price and trading volume, to a unified numerical range, facilitating the model's effective learning of feature representations.

[0066] At the same time, the system incorporates auxiliary information to enhance the model's understanding of the complex dynamics of financial markets, addressing information blind spots caused by relying solely on historical price data and improving the accuracy, robustness, and interpretability of forecasts. Specifically, auxiliary information supplements the external drivers and underlying logical relationships that influence asset price fluctuations from multiple dimensions. This auxiliary information includes auxiliary data such as macroeconomic indicators and market sentiment factors, as well as text and knowledge-based information such as industry knowledge, market reports, and user conversations.

[0067] Step S20: performing vector conversion and fusion based on the financial time series data and the auxiliary information to form a unified input representation;

[0068] It is understandable that due to the significant differences in data modality, structural form and semantic granularity between financial time series data and auxiliary information - the former is a numerical time series with strong temporal dependence; the latter contains a variety of heterogeneous information such as text, events, rules, etc., and it is difficult to achieve effective modeling by direct splicing or simple weighting. Therefore, executing step S20 can uniformly map multi-source heterogeneous information into a computable representation in a high-dimensional semantic vector space, and construct a unified input representation rich in contextual semantics through a fusion mechanism, providing a structurally consistent and semantically rich input basis for the subsequent shared encoder feature extraction.

[0069] In this embodiment, the system uses natural language processing (NLP) technology to perform semantic analysis and key information extraction on text-based auxiliary knowledge, and converts it into structured features or embedded vectors in combination with a rule engine or a pre-trained language model; for financial time series data, it uses a learnable time encoding module (such as a linear projection layer, a convolutional network, or a recurrent structure) to convert it into a high-dimensional vector representation with time perception, namely a time embedding vector.

[0070] In a feasible embodiment, step S20 may further include steps S21 to S23:

[0071] Step S21, converting the financial time series data into a time embedding vector;

[0072] In this embodiment, the financial time series data is input into a time series embedding module, which extracts the temporal dynamic characteristics of the sequence and generates a time embedding vector. Specifically, the system performs segmented encoding or point-by-point embedding on the raw time series data (such as price and trading volume), mapping it into a fixed-dimensional vector sequence through linear transformation or nonlinear neural network. Each time step corresponds to a time embedding vector, which represents the local and global time series patterns at that moment.

[0073] Step S22: mapping the auxiliary information to a separate embedding space based on a pre-trained language model or embedding model to obtain a corresponding knowledge embedding vector;

[0074] In this embodiment, the non-time series part of the auxiliary information (including auxiliary data such as macroeconomic indicators and market sentiment factors, as well as text information such as industry knowledge, market reports, and user conversations) is sent to the knowledge embedding module for processing.

[0075] For structured auxiliary data, an embedding lookup table or a fully connected layer is used to convert it into a numerical vector of fixed dimension; for unstructured text information, a pre-trained language model is used for context encoding to output its semantic embedding vector; further, the entity relationships in the domain knowledge are modeled through knowledge graph embedding technology to generate relationship-aware knowledge embedding vectors.

[0076] Ultimately, all auxiliary information is uniformly mapped into an independent but fusible knowledge embedding space, outputting one or more knowledge embedding vectors to represent the external environment, event impact, and domain prior knowledge.

[0077] Step S23: Fusing the time embedding vector with the knowledge embedding vector to form a unified input representation.

[0078] In this embodiment, the time embedding vector and the knowledge embedding vector are cross-modally fused to generate a unified input representation that combines time series dynamics and external knowledge. This unified input representation is a high-dimensional vector sequence that combines financial time series features with multi-source auxiliary knowledge. This serves as the input to the subsequent shared encoder, supporting multi-task joint modeling.

[0079] Through the above steps, the knowledge embedding module is introduced to support the utilization of external structured or unstructured knowledge. Information fusion is achieved through pre-trained models or rule-based modeling, improving the model's understanding and prediction capabilities for complex financial scenarios. Furthermore, through the fusion of time embedding and knowledge embedding, multi-source information is mapped into a unified feature space, constructing a context-aware representation of sequential inputs, effectively improving the representation capabilities of the model input.

[0080] Step S30, performing feature extraction on the input representation through a preset shared encoder, outputting a global feature vector, and inputting the global feature vector into a preset multi-task prediction output head;

[0081] It should be noted that the preset shared encoder in this embodiment refers to the Transformer encoder, which is a neural network structure based on the self-attention mechanism. It was originally used for natural language processing tasks and has been widely used in time series modeling tasks in recent years. It has good long-term dependency modeling capabilities and parallel computing efficiency.

[0082] In this example, the Transformer encoder consists of multiple stacked modules, each of which incorporates a multi-head self-attention mechanism, a feedforward network, residual connections, and layer normalization. The multi-head attention mechanism allows the model to simultaneously model long-term dependencies and local fluctuations in the sequence from different perspectives, thereby fully capturing the diverse patterns in financial time series.

[0083] The global feature vector refers to a high-dimensional feature representation that is generated after multi-layer nonlinear transformation of the shared encoder and integrates the semantic information of the entire input sequence.

[0084] Understandably, traditional modeling methods typically focus on a single task objective, ignoring the potential connections between multiple related tasks. This fragmented modeling approach makes the model prone to overfitting and makes it difficult to benefit from related tasks to improve overall performance. Therefore, executing step S30 can uniformly extract and abstract the underlying features shared by multiple tasks through a shared encoder, achieving cross-task instruction transfer and feature sharing, thereby achieving unified multi-task modeling and improving the sharing of feature expressions and the ability to collaborate between tasks.

[0085] In a feasible embodiment, step S30 may include steps S31 to S33:

[0086] Step S31, in the preset shared encoder, interactively modeling the feature dimensions in the input representation through a multi-head attention mechanism to obtain an attention-weighted feature vector;

[0087] In this embodiment, the unified input representation outputted in step S20 is inputted into a preset shared encoder, wherein the shared encoder is based on the Transformer architecture, and its core module is a multi-head self-attention mechanism.

[0088] The multi-head attention mechanism is used to model the long-range dependencies between time steps in the input sequence and the dynamic angles between feature dimensions.

[0089] Specifically, the input vector at each time step is first linearly transformed to generate the corresponding query vector, key vector, and value vector. Then, the above vectors are used to calculate the attention weights and weighted aggregate information to achieve dynamic modeling of key dependencies in the sequence.

[0090] It's also important to note that the above process is completed within a single "attention head." To enable the model to capture different types of dependencies (such as trends, volatility, and event responses) from different subspaces, the system executes multiple attention heads in parallel, each using an independent parameter matrix for mapping and calculation.

[0091] After the outputs of all attention heads are concatenated, they are integrated through a linear layer to output a sequence of attention-weighted feature vectors.

[0092] Among them, each vector in the attention-weighted feature vector sequence integrates the contextual information of the entire input sequence, realizes dynamic focus on key time points, and effectively enhances the model's ability to perceive complex financial dynamics.

[0093] Through the above-mentioned multi-head attention mechanism, the model is allowed to simultaneously model long-term dependencies and local fluctuations in the sequence from different perspectives, thereby fully capturing the diverse patterns in financial time series.

[0094] Step S32, performing a residual connection on the attention weighted feature vector and the input representation to obtain a superposition result;

[0095] After completing the attention calculation, in order to alleviate the gradient vanishing problem during deep network training and improve model stability and convergence speed, this embodiment introduces a residual connection mechanism.

[0096] Specifically, the attention weighted feature vector outputted in step S31 is added element-by-element to the input representation obtained in step S20 to form a superposition result.

[0097] This step establishes a "straight-through information path," ensuring that the original input signal is not excessively distorted or lost during nonlinear transformations. This helps preserve important temporal structural information. Furthermore, residual connections effectively support the stacking of deep networks and are a key technology for achieving high-performance Transformer models.

[0098] Step S33: normalize the superposition result through a layer normalization mechanism to obtain a global feature vector.

[0099] To further improve the stability of model training and reduce distribution fluctuations caused by internal covariate shift, this embodiment adopts a layer normalization mechanism after the residual connection.

[0100] The layer normalization performs normalization on each sample in its feature dimension, and the formula is as follows:

[0101]

[0102] Where μ represents the mean of the current sample in the feature dimension; σ represents the variance of the current sample in the feature dimension; γ represents the learnable scaling parameter; β represents the offset parameter; and ε represents the smoothing term.

[0103] The standardized data has a stable distribution, which is conducive to efficient learning and parameter updating of the subsequent feedforward neural network module. Subsequently, residual connections and layer normalization are performed again through the feedforward neural network module.

[0104] Finally, after processing through all encoder layers, the system extracts the global feature vector from the output of the last layer and inputs the global feature vector output by the Transformer encoder into multiple task-specific output heads respectively.

[0105] Through the above steps, based on the unified modeling structure of the shared Transformer encoder, parallel modeling and collaborative optimization of multiple tasks such as regression, classification, and multi-time point prediction are achieved in the same network, which significantly simplifies the system structure and improves the consistency and robustness of predictions.

[0106] Step S40 , performing single-time-point task prediction and / or multi-time-point task prediction on the prediction task according to the global feature vector through the preset multi-task prediction output head, and generating a corresponding target prediction result.

[0107] It should be noted that the multi-task prediction output head refers to a task-specific prediction submodule (i.e., "output head") mounted on the global feature vector output by the shared encoder. Each output head is structured for a specific type of prediction task and independently completes its decoding and output processes, but shares the same set of high-level semantic features extracted by the shared encoder;

[0108] The prediction tasks include but are not limited to single time point tasks and multi-time point tasks;

[0109] The single-time-point task refers to a task type that performs prediction at a specific target time point, including but not limited to regression tasks (such as trend change amplitude prediction) and classification tasks (such as risk level judgment).

[0110] The multi-time point task prediction refers to the task of jointly modeling and serializing the target variables of multiple continuous or discontinuous time steps in the future, aiming to capture the dynamic evolution path of financial variables rather than focusing only on the endpoint value.

[0111] Understandably, many current financial forecasting systems only support forecasting data at a single point in time and lack the ability to model trends in continuous time series, making them unsuitable for financial scenarios requiring continuous prediction of trends or risk changes. Furthermore, traditional systems typically rely on independent operation of different models, requiring repeated loading of models and computational processes when executing forecasts. This results in long response times and high computational resource overhead, making it difficult to adapt to the real-time monitoring and rapid response requirements of financial services. Therefore, executing step S40 enables parallel collaborative modeling of multiple forecasting tasks in a unified feature space. Only one feature extraction is required to simultaneously output regression, classification, and multi-time point series forecast results, significantly improving system reasoning efficiency and resource utilization. Furthermore, by sharing the underlying semantic representation, the logical consistency and overall robustness of the forecast results across tasks are enhanced.

[0112] In a feasible implementation, step S40 may include steps S41 to S44:

[0113] Step S41, identifying the prediction task type in the preset multi-task prediction output header;

[0114] In this embodiment, the system first identifies the category of the current prediction task to be executed. This identification process is completed through configuration files, API parameters, or task scheduling instructions, and specifically includes determining whether the current prediction task is a regression task, a classification task, or a multi-time point task.

[0115] The task type identification result is used to activate the corresponding task prediction output header, ensuring that the model only calls the decoding structure related to the current task, avoiding redundant calculations. For example, when receiving a "yield prediction" request, it is identified as a regression task; when receiving a "risk level assessment" request, it is identified as a classification task; when receiving a "price trend prediction for the next 5 days" request, it is identified as a multi-time point task.

[0116] Step S42: If the prediction task type is the regression task, linear mapping is applied to the global feature vector to output a continuous value;

[0117] When step S41 identifies that the current task is a regression task (such as trend change amplitude prediction, volatility prediction, return prediction, etc.), the system calls the preset regression output head.

[0118] The preset regression output head is composed of a lightweight fully connected layer (i.e., a linear layer), and its mathematical expression is:

[0119] y=w T h+b

[0120] Where h is the global feature vector output by the shared encoder; w is the science weight vector; b is the bias term; and y is the continuous prediction of the final output.

[0121] The final output results can be directly used in financial applications such as quantitative trading signal generation and risk exposure calculation.

[0122] Step S43: If the prediction task type is the classification task, the global feature vector is converted into a raw score through a fully connected layer in a preset multi-task prediction output head, and the raw score is converted into a category probability distribution through a normalized exponential function;

[0123] When the task type is identified as a classification task (such as risk level judgment, market status classification, credit rating, etc.), the system enables the classification output header.

[0124] The classification output head consists of two stages:

[0125] The first stage is to map the global feature vector h into a raw score vector of the same dimension as the number of categories C:

[0126] Z=Wh+b,z∈R c

[0127] Wherein, W is the science system parameter matrix; b is the bias vector; and h is the global eigenvector.

[0128] The second stage is to perform normalized exponential transformation on the original scores to generate category probability distribution.

[0129] The final output result is a probability vector in the form of [0.12, 0.25, 0.63], which respectively represents the confidence level of the sample belonging to "low risk", "medium risk" and "high risk", which facilitates the subsequent decision-making system to perform threshold judgment or risk weighting processing.

[0130] Step S44: If the prediction task type is the multi-time point task, a time-expanded multi-output structure is constructed to generate a target prediction result of a preset time step.

[0131] When the task type is a multi-time point task (such as future multi-day price, volatility, and risk score series prediction), the system activates the sequence prediction output head to generate future time series prediction results of length T.

[0132] The sequence prediction output head uses one or more fully connected layers to directly map the global feature vector to a T-dimensional output vector. Alternatively, the global feature vector is used as the initial hidden state to drive a shallow recurrent network to gradually decode and generate the output sequence for the next T steps.

[0133] At the same time, the combination of position encoding and self-attention mechanism can achieve high-precision sequence generation, which is suitable for long sequence or multi-variable joint prediction.

[0134] Regardless of the structure adopted, the final output is a target prediction sequence with a preset time step, such as the closing price forecast for the next 5 days [P1, P2, P3, P4, P5], which can be used for trend visualization, path simulation or portfolio optimization.

[0135] Through the above steps, task-specific output heads are mounted on the shared encoder. Each output head receives the same global feature vector and independently completes the prediction output for its corresponding task. This design ensures unified sharing of the underlying feature representation while supporting the parallel generation of multi-task prediction results. Compared to the traditional serial processing mode in which each task relies on an independent model and requires repeated data loading and feature extraction, this method avoids multiple model calls and redundant calculations, significantly reducing system computational overhead and response latency.

[0136] Through the above-described embodiment method, financial time series data and auxiliary information are obtained; vector conversion and fusion are performed based on the financial time series data and auxiliary information to form a unified input representation; features are extracted from the input representation through a preset shared encoder, a global feature vector is output, and the global feature vector is input into a preset multi-task prediction output head; the prediction task is predicted based on the global feature vector through the preset multi-task prediction output head to generate the corresponding target prediction result. By introducing a shared encoder to extract features from the input data, this solution realizes parallel collaborative modeling of multiple types of tasks, avoiding resource waste and prediction conflicts caused by task separation. At the same time, the fusion of financial time series data and auxiliary information significantly improves the model's ability to model complex financial dynamics. In addition, outputting prediction results through the task prediction output head can not only ensure the independence and interpretability of the output logic of each task, but also realize cross-task knowledge transfer and regularization through the shared underlying feature representation, thereby improving the generalization ability and robustness of the overall model.

[0137] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above embodiment 1 can be referred to the above introduction and will not be described in detail later. Figure 2 After step S40, the multi-objective forecasting method for financial time series further includes steps S401 to S402:

[0138] Step S401: Combine the target prediction result and a preset loss function to jointly train and optimize the financial time series prediction model to obtain a pre-trained financial time series prediction model;

[0139] In this embodiment, a unified multi-task loss function is used to measure the difference between the target prediction results output by the model and the corresponding true labels, and the loss terms of multiple task objectives are integrated to perform end-to-end joint training and parameter optimization.

[0140] Specifically, the financial time series forecasting model comprises a shared encoder and multiple task-specific output heads. By sharing the underlying encoder parameters, different tasks can interact and regularize at the feature representation layer during training, effectively leveraging inter-task correlation information and improving generalization. This also prevents overfitting of a single task from impacting overall performance. Furthermore, to balance the contributions of each task in gradient updates, this embodiment sets an adjustable loss weight for each task, enabling the financial time series forecasting model to adaptively adjust its learning strategy based on task difficulty and target priority.

[0141] In a feasible embodiment, step S401 includes steps A1 to A4:

[0142] Step A1, extracting corresponding prediction tasks from the target prediction results according to each task type;

[0143] During the training process, the prediction task type involved in the current batch of data is first analyzed, including but not limited to regression tasks, classification tasks, and multi-time point sequence prediction tasks.

[0144] Based on the task configuration or input label structure, the system identifies and extracts the model output corresponding to each task, preparing for the subsequent independent calculation of subtask losses. This mechanism supports flexible task combination and dynamic scheduling, facilitating multi-objective learning in different scenarios.

[0145] Step A2: Calculate the corresponding subtask loss value for each prediction task;

[0146] For each identified prediction task, the loss value of its subtask is calculated using the loss function that matches its output form.

[0147] For example, for regression tasks, the mean squared error or mean absolute error is used:

[0148]

[0149] Where N is the total number of samples involved in the loss calculation in the current batch or the entire dataset; i is the true label of the i-th sample; is the predicted value of the i-th sample.

[0150] For classification tasks, cross entropy loss is used:

[0151]

[0152] Where N is the total number of samples involved in the loss calculation in the current batch or the entire dataset; C is the total number of categories for each sample; y i,c is the true label of the i-th sample belonging to the c-th category; is the predicted probability that the i-th sample belongs to the c-th class.

[0153] For multi-time point prediction tasks, a sequence loss function such as average MSE or dynamic regularization in the time dimension is used:

[0154]

[0155] Wherein, T is the number of time help singing; y t is the true label of the t-th time step; is the predicted value at the t-th time step.

[0156] Step A3, performing weighted summation based on the loss value of each subtask and the corresponding loss weight to obtain a total loss value;

[0157] Then, the loss values ​​of all subtasks are weighted and summed according to the preset weights to obtain the overall joint loss function:

[0158]

[0159] Step A4: performing back-propagation optimization and parameter updating on the financial time series forecasting model based on the total loss value to obtain a pre-trained financial time series forecasting model.

[0160] In this embodiment, the total loss value is used as the optimization target, the back propagation algorithm is executed, the gradients of all learnable parameters of the model are calculated, and the parameter update is completed through the optimizer.

[0161] Through multiple rounds of iterative training, the model gradually converged, ultimately obtaining a pre-trained financial time series forecasting model that has completed multi-task joint training. This model has the ability to share features across tasks and has good generalization performance.

[0162] Through the above steps, a unified loss function is adopted for joint optimization, and a dynamic adjustment mechanism of task weights is introduced, so that the model can achieve effective collaborative training and learning strategy adjustment between different tasks, thereby improving the overall stability and generalization ability of the system.

[0163] Step S402: Use the pre-trained financial time series prediction model to perform online or batch prediction.

[0164] After completing the training and optimization of the financial time series prediction model, the pre-trained financial time series prediction model is deployed in the financial prediction system to achieve online or batch prediction.

[0165] Since this embodiment adopts a unified input, shared encoder and parallel multi-task output structure, a single forward propagation mechanism can complete multiple task predictions simultaneously, significantly reducing computing resource consumption and improving prediction efficiency. It is suitable for financial analysis scenarios with high real-time requirements.

[0166] During the model deployment and prediction phase, a single forward propagation mechanism (with a single, unified, fused data input) allows all task-specific output heads to simultaneously generate their own predictions. Compared to traditional approaches (where each task's independent model must be loaded and run separately), this eliminates the computational overhead of multiple model loading and repeated feature extraction, significantly improving efficiency.

[0167] Through the above-mentioned embodiment method, the loss functions of multiple prediction tasks are weightedly fused, and the joint loss is calculated based on the true label and the target prediction result. The shared encoder and the output head of each task are end-to-end backpropagated and the parameters are updated, which effectively utilizes the correlation information between tasks and improves the overall generalization ability of the model.

[0168] For example, to help understand the implementation process of the multi-objective forecasting method for financial time series obtained by combining this embodiment with the above-mentioned embodiment 1, please refer to Figure 3 , Figure 3 This paper provides a brief flowchart of a multi-objective forecasting method for financial time series, specifically:

[0169] First, the original financial time series data obtained is standardized, and auxiliary information is introduced, such as auxiliary data such as macroeconomic indicators and market sentiment factors, as well as text and knowledge information such as industry knowledge, market reports, and user conversations.

[0170] Next, the financial time series data and auxiliary data are transferred to the time series embedding module and the non-time series embedding module respectively, and converted into time embedding vectors in different forms; the auxiliary knowledge information is mapped to a separate embedding space through the knowledge embedding module based on the pre-trained language model or embedding model to obtain the knowledge embedding vector.

[0171] The time embedding vector is then fused with the knowledge embedding vector to form a unified input representation, which is then passed to a multi-layer Transformer encoder for deep feature extraction. The Transformer encoder incorporates a multi-head self-attention mechanism, a feedforward network, residual connections, and layer normalization. This multi-head attention mechanism allows the model to simultaneously model long-term dependencies and local fluctuations in the sequence from different perspectives, fully capturing the diverse patterns in financial time series.

[0172] The global feature vector output by the Transformer encoder is then fed into multiple task-specific output heads. Each output head is responsible for completing a specific prediction task. Specifically, for regression tasks (such as trend change magnitude prediction), linear mapping is used to achieve continuous numerical output; for classification tasks (such as risk level determination), a fully connected layer plus a normalized exponential function is used to output the category probability distribution; for multi-time point prediction tasks, a multi-output structure with time expansion is constructed to generate prediction results for multiple future time steps.

[0173] In addition, this embodiment also adopts a unified multi-task loss function, realizes cross-task interaction and regularization of the feature layer through a shared encoder, and effectively suppresses overfitting; at the same time, it introduces adjustable task loss weights, so that the financial time series prediction model can adaptively optimize the learning process according to task difficulty and business priority.

[0174] Finally, the financial time series forecasting model is directly deployed in the financial forecasting system, and combined with the single forward propagation mechanism to achieve online or batch forecasting.

[0175] In summary, this embodiment provides a time series prediction framework with a multi-task shared encoder. Through a shared attention mechanism and multi-task loss design, it effectively learns inter-task relationships and improves the accuracy, efficiency, and stability of financial time series prediction. This complete multi-target prediction solution addresses the needs of mixed tasks such as classification and regression in financial time series scenarios.

[0176] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the multi-objective forecasting method of financial time series of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0177] This application also provides a multi-objective forecasting system for financial time series, please refer to Figure 4 , the multi-objective forecasting system for financial time series includes:

[0178] Acquisition module 10, used to acquire financial time series data and auxiliary information;

[0179] A data processing module 20 is configured to perform vector conversion and fusion based on the financial time series data and the auxiliary information to form a unified input representation;

[0180] A feature extraction module 30 is configured to extract features from the input representation using a preset shared encoder, output a global feature vector, and input the global feature vector into a preset multi-task prediction output head;

[0181] The task prediction module 40 is configured to predict the prediction task according to the global feature vector through the multi-task prediction output head and generate a corresponding target prediction result.

[0182] The multi-objective forecasting system for financial time series provided in this application, employing the multi-objective forecasting method for financial time series described in the aforementioned embodiments, can address the technical issues of task separation, low efficiency, and weak generalization capabilities currently encountered in financial series analysis. Compared to the prior art, the multi-objective forecasting system for financial time series provided in this application offers the same beneficial effects as the multi-objective forecasting method for financial time series described in the aforementioned embodiments. Other technical features of the multi-objective forecasting system for financial time series are the same as those disclosed in the aforementioned embodiments and are not further elaborated upon here.

[0183] The present application provides a multi-objective prediction device for financial time series, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the multi-objective prediction method for financial time series in the above-mentioned embodiment 1.

[0184] Reference below Figure 5 , which shows a schematic diagram of the structure of a multi-objective prediction device for financial time series suitable for implementing embodiments of the present application. The multi-objective prediction device for financial time series in embodiments of the present application may include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), and fixed terminals such as digital TVs and desktop computers. Figure 5 The multi-objective prediction device for financial time series shown is merely an example and should not limit the functions and scope of use of the embodiments of the present application.

[0185] like Figure 5 As shown, the multi-objective prediction device for financial time series may include a processing device 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes based on programs stored in a read-only memory 1002 or programs loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the multi-objective prediction device for financial time series. The processing device 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: an input device 1007 including, for example, a touchscreen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; a storage device 1003 including, for example, a magnetic tape or hard disk; and a communication device 1009. Communication device 1009 can allow the multi-objective forecasting device for financial time series to communicate wirelessly or wired with other devices to exchange data. Although the figure shows a multi-objective forecasting device for financial time series with various systems, it should be understood that implementation or presence of all the illustrated systems is not required. More or fewer systems may alternatively be implemented or present.

[0186] In particular, according to the embodiments disclosed in the present application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, the embodiments disclosed in the present application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program comprising program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via a communication device, or installed from a storage device 1003, or installed from a read-only memory 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are performed.

[0187] The multi-objective forecasting device for financial time series provided in this application, employing the multi-objective forecasting method for financial time series described in the aforementioned embodiments, can address the technical issues of task separation, low efficiency, and weak generalization capabilities currently encountered in financial series analysis. Compared to the prior art, the multi-objective forecasting device for financial time series provided in this application offers the same beneficial effects as the multi-objective forecasting method for financial time series described in the aforementioned embodiments. Other technical features of this multi-objective forecasting device for financial time series are the same as those disclosed in the aforementioned embodiments and are not further elaborated upon here.

[0188] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0189] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0190] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, a computer program) stored thereon, wherein the computer-readable program instructions are used to execute the multi-objective prediction method for financial time series in the above-mentioned embodiment.

[0191] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, 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 thereof. In this embodiment, the computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0192] The computer-readable storage medium may be included in the multi-objective prediction device for financial time series; or it may exist independently without being assembled into the multi-objective prediction device for financial time series.

[0193] The above-mentioned computer-readable storage medium carries one or more programs. When the above-mentioned one or more programs are executed by the multi-objective prediction device of financial time series, the multi-objective prediction device of financial time series: obtains financial time series data and auxiliary information; performs vector conversion and fusion based on the financial time series data and auxiliary information to form a unified input representation; extracts features of the input representation through a preset shared encoder, outputs a global feature vector, and inputs the global feature vector into a preset multi-task prediction output head; predicts the prediction task according to the global feature vector through the preset multi-task prediction output head to generate a corresponding target prediction result.

[0194] Computer program code for performing the operations of the present application may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may 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 may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0195] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. In this regard, each box in the flow chart or block diagram can represent a module, program segment or a part of code, and the module, program segment or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented by a dedicated hardware-based system that performs the specified function or operation, or can be implemented by a combination of dedicated hardware and computer instructions.

[0196] The modules described in the embodiments of the present application may be implemented in software or hardware, wherein the name of a module does not necessarily limit the unit itself.

[0197] The computer-readable storage medium provided in this application stores computer-readable program instructions (i.e., a computer program) for executing the multi-objective forecasting method for financial time series described above. This computer-readable storage medium can address the technical issues of task separation, low efficiency, and weak generalization capabilities currently encountered in financial series analysis. Compared to the prior art, the beneficial effects of the computer-readable storage medium provided in this application are similar to those of the multi-objective forecasting method for financial time series provided in the aforementioned embodiments, and are not further elaborated here.

[0198] The present application also provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of the multi-objective forecasting method for financial time series as described above.

[0199] The computer program product provided in this application can address the current technical issues of task separation, low efficiency, and weak generalization in financial time series analysis. Compared to existing technologies, the beneficial effects of the computer program product provided in this application are similar to those of the multi-objective forecasting method for financial time series provided in the aforementioned embodiments, and are not further elaborated here.

[0200] The above description is only part of the embodiments of the present application and does not limit the patent scope of the present application. All equivalent structural transformations made by using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect application in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A multi-objective forecasting method for financial time series, characterized by: The multi-objective forecasting method for financial time series includes: Obtain financial time series data and auxiliary information; Performing vector conversion and fusion based on the financial time series data and the auxiliary information to form a unified input representation; Performing feature extraction on the input representation through a preset shared encoder, outputting a global feature vector, and inputting the global feature vector into a preset multi-task prediction output head; The preset multi-task prediction output head is used to predict the prediction task according to the global feature vector to generate a corresponding target prediction result.

2. The multi-target prediction method according to claim 1, wherein: The step of performing vector conversion and fusion based on the financial time series data and the auxiliary information to form a unified input representation includes: Converting the financial time series data into a time embedding vector; Mapping the auxiliary information to a separate embedding space based on a pre-trained language model or embedding model to obtain a corresponding knowledge embedding vector; The time embedding vector and the knowledge embedding vector are fused to form a unified input representation.

3. The multi-objective forecasting method for financial time series according to claim 1, characterized in that: The step of extracting features from the input representation using a preset shared encoder and outputting a global feature vector comprises: In the preset shared encoder, interactively modeling the feature dimensions in the input representation through a multi-head attention mechanism to obtain an attention-weighted feature vector; Performing a residual connection on the attention weighted feature vector and the input representation to obtain a superposition result; The superposition results are normalized through the layer normalization mechanism to obtain the global feature vector.

4. The multi-objective prediction method for financial time series according to claim 1, characterized in that: The prediction tasks include single-time-point tasks and multi-time-point tasks, and the single-time-point tasks include regression tasks and classification tasks; The step of predicting the prediction task according to the global feature vector by the preset multi-task prediction output head to generate a corresponding target prediction result includes: In the preset multi-task prediction output header, identifying the prediction task type; If the prediction task type is the regression task, linear mapping is applied to the global feature vector to output a continuous value; If the prediction task type is the classification task, the global feature vector is converted into a raw score through a fully connected layer in a preset multi-task prediction output head, and the raw score is converted into a category probability distribution through a normalized exponential function; If the prediction task type is the multi-time point task, a time-expanded multi-output structure is constructed to generate a target prediction result of a preset time step.

5. The multi-objective prediction method for financial time series according to claim 1, characterized in that: After the step of predicting the prediction task according to the global feature vector through the multi-task prediction output head and generating the corresponding target prediction result, the method includes: Combining the target prediction result with a preset loss function, jointly training and optimizing the financial time series prediction model to obtain a pre-trained financial time series prediction model; Use the pre-trained financial time series forecasting model for online or batch forecasting.

6. The multi-objective forecasting method for financial time series according to claim 5, characterized in that: The step of combining the target prediction result with a preset loss function to jointly train and optimize the financial time series prediction model to obtain a pre-trained financial time series prediction model includes: According to each task type, extracting the corresponding prediction task from the target prediction result; Calculating the corresponding subtask loss value for each prediction task; Perform weighted summation based on the loss value of each subtask and the corresponding loss weight to obtain a total loss value; Based on the total loss value, back-propagation optimization and parameter update are performed on the financial time series forecasting model to obtain a pre-trained financial time series forecasting model.

7. The multi-objective prediction method for financial time series according to claim 5, characterized in that: The step of using the pre-trained financial time series prediction model to perform online or batch prediction includes: Based on a preset single forward propagation strategy, the pre-trained financial time series prediction model is used to perform synchronous online or batch prediction on multiple tasks.

8. A multi-objective forecasting system for financial time series, characterized by: The multi-objective forecasting system for financial time series includes: Acquisition module, used to obtain financial time series data and auxiliary information; a data processing module, configured to perform vector conversion and fusion based on the financial time series data and the auxiliary information to form a unified input representation; A feature extraction module, configured to extract features from the input representation using a preset shared encoder, output a global feature vector, and input the global feature vector into a preset multi-task prediction output head; The task prediction module is used to predict the prediction task according to the global feature vector through the multi-task prediction output head and generate a corresponding target prediction result.

9. A multi-objective prediction device for financial time series, characterized in that: The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the multi-objective forecasting method for financial time series according to any one of claims 1 to 7.

10. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the multi-objective forecasting method for financial time series according to any one of claims 1 to 7 are implemented.

Images