Asset value prediction method and system, electronic equipment and storage medium

Abstract

The invention relates to the technical field of artificial intelligence, and discloses an asset value prediction method and system, electronic equipment and a storage medium. The method comprises the steps that feature time series data influencing the asset value are acquired, and the feature time series data comprise a feature data set and a time series corresponding to the feature data set; performing prediction based on the characteristic time series data to obtain a discrete direction prediction result representing an asset value fluctuation direction and a continuous amplitude prediction result representing an asset value fluctuation amplitude; and taking the discrete direction prediction result as a gating signal, and performing gating routing processing on the continuous amplitude prediction result to obtain an asset value prediction result. According to the method, the prediction task is decoupled into two independent prediction stages in the discrete direction and the continuous amplitude, so that the stability of asset value prediction and the definition of decision making are remarkably improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to an asset value prediction method, system, electronic device, and storage medium. Background Technology

[0002] In the field of traditional asset valuation, related technologies generally rely on a single regression model for direct prediction. Because regression models overfit to minor fluctuations and market noise in the training data, they become extremely sensitive to small changes in the input data. This frequent minor adjustments further lead to unstable predictions. Such unstable predictions lack operational feasibility at the business level and are difficult to support robust decision-making. Summary of the Invention

[0003] The embodiments of this application aim to at least partially solve one of the technical problems in the related art. To this end, embodiments of this application propose an asset value prediction method, system, electronic device, and storage medium.

[0004] The embodiments of this application provide an asset value prediction method, the method comprising: acquiring characteristic time-series data that affects asset value, wherein the characteristic time-series data includes a characteristic dataset and a time series corresponding to the characteristic dataset; performing prediction based on the characteristic time-series data to obtain a discrete direction prediction result representing the direction of asset value fluctuation and a continuous amplitude prediction result representing the amplitude of asset value fluctuation; using the discrete direction prediction result as a gating signal to perform gating routing processing on the continuous amplitude prediction result to obtain the asset value prediction result.

[0005] In some implementations, the discrete direction prediction result is used as a gating signal to perform gating routing processing on the continuous amplitude prediction result to obtain the asset value prediction result. This includes: when the discrete direction prediction result is in a stable state, filtering the continuous amplitude prediction result and using a preset threshold as the asset value prediction result; when the discrete direction prediction result is in a rising or falling state, calculating the continuous amplitude prediction result to obtain the asset value prediction result.

[0006] In some implementations, the rise / fall status includes both an upward status and a downward status; when the discrete direction prediction result is a rise / fall status, the continuous amplitude prediction result is calculated to obtain the asset value prediction result, including: when the discrete direction prediction result is an upward status, the absolute value of the continuous amplitude prediction result is used as the asset value prediction result; when the discrete direction prediction result is a downward status, the negative value of the absolute value of the continuous amplitude prediction result is used as the asset value prediction result.

[0007] In some implementations, the discrete direction prediction result is obtained by a direction prediction model, which is trained as follows: acquiring historical feature time-series sample data, wherein the historical feature time-series sample data includes a feature sample dataset and the corresponding time sample sequence; obtaining three-class direction labels based on the feature sample dataset and the time sample sequence; inputting the feature sample dataset and the time sample sequence into the direction prediction model for prediction to obtain the discrete direction prediction training result; obtaining the discrete direction prediction loss based on the discrete direction prediction training result and the three-class direction labels; and updating the model parameters of the direction prediction model in reverse based on the discrete direction prediction loss.

[0008] In some implementations, a three-class directional label is obtained based on a feature sample dataset and a time sample sequence, including: obtaining the sample mean and current sample data value within a preset statistical period corresponding to each feature sample data in the feature sample dataset based on the feature sample dataset and the time sample sequence; obtaining the sample data fluctuation value based on the sample mean and current sample data value within the preset statistical period; obtaining a personalized absolute tolerance threshold corresponding to each feature sample data based on the current sample data value and a preset stability percentage threshold, wherein the stability percentage threshold represents the percentage of fluctuation in the sample data fluctuation value that has not changed during the training of the directional prediction model; and obtaining the three-class directional label based on the personalized absolute tolerance threshold and the sample data fluctuation value.

[0009] In some implementations, the three-category directional labels include an upward directional label, a downward directional label, and a stable directional label; the personalized absolute tolerance threshold includes a first personalized absolute tolerance interval, a second personalized absolute tolerance interval, and a third personalized absolute tolerance interval; based on the personalized absolute tolerance threshold and the sample data fluctuation value, the three-category directional labels are calculated, including: if the sample data fluctuation value is in the first personalized absolute tolerance interval, an upward directional label is generated; if the sample data fluctuation value is in the second personalized absolute tolerance interval, a downward directional label is generated; if the sample data fluctuation value is in the third personalized absolute tolerance interval, a stable directional label is generated.

[0010] In some implementations, the continuous amplitude prediction result is obtained by an amplitude prediction model, which is trained in the following way: obtaining amplitude prediction training labels for the amplitude prediction model based on a feature sample dataset and a time sample sequence; inputting the feature sample dataset and the time sample sequence into the amplitude prediction model for prediction to obtain the continuous amplitude prediction training result; obtaining the continuous amplitude prediction loss based on the continuous amplitude prediction training result and the amplitude prediction training labels; and updating the model parameters of the amplitude prediction model in reverse based on the continuous amplitude prediction loss.

[0011] The embodiments of this application provide an asset value prediction system, the system comprising: an acquisition module for acquiring characteristic time-series data that affects asset value, wherein the characteristic time-series data includes a characteristic dataset and a time series corresponding to the characteristic dataset; a prediction module for making predictions based on the characteristic time-series data to obtain discrete direction prediction results representing the direction of asset value fluctuations and continuous amplitude prediction results representing the magnitude of asset value fluctuations; and a processing module for using the discrete direction prediction results as a gating signal to perform gating routing processing on the continuous amplitude prediction results to obtain asset value prediction results.

[0012] An embodiment of this application provides an electronic device, which includes: a memory, and one or more processors communicatively connected to the memory; the memory stores instructions executable by one or more processors, which are executed by one or more processors to cause the one or more processors to implement the steps of the method of any of the above embodiments.

[0013] Embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method of any of the above embodiments.

[0014] The embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the steps of the method according to any of the above embodiments.

[0015] The solution provided in this application can achieve accurate and multi-dimensional prediction of asset value based on feature time-series data that integrates multiple characteristics. By combining macro trend judgment (discrete direction prediction results) with micro amplitude quantification (continuous amplitude prediction results) and through an innovative gating routing mechanism, the asset value prediction results are made more targeted and robust. This hierarchical and conditional prediction approach effectively improves the model's adaptability to different market conditions and avoids the bias that may arise from a single model.

[0016] The solution provided in this application transforms complex market information into clear and reliable decision support tools, which helps optimize asset returns and enhance overall competitiveness in volatile markets. Attached Figure Description

[0017] Figure 1 A flowchart illustrating the asset value prediction method provided in this application embodiment; Figure 2 A schematic diagram of the offline training phase of a model provided in an embodiment of this application; Figure 3 A flowchart illustrating the online prediction and decision-making stages of a model provided in an embodiment of this application; Figure 4A schematic diagram of the asset valuation system provided in the embodiments of this application; Figure 5 A block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0018] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0019] In the field of traditional asset valuation, related technologies generally rely on a single regression model for direct prediction. However, current asset valuation strategies based on a single regression model suffer from the following problems. First, the model's prediction results lack stability. Because the model attempts to directly predict a precise continuous value, it is highly sensitive to minor fluctuations in the data or market noise, leading to frequent and small adjustment suggestions. Such frequent changes lack practical operability and have poor stability. Second, the model struggles to clearly identify key business decision points and cannot effectively distinguish between two completely different business states: "market stable, no price adjustment needed" and "market trend emerging, price adjustment required." When the model outputs a prediction of a small market fluctuation, decision-makers will be confused by such minor fluctuations, unable to determine whether it is market noise or a weak upward signal, thus reducing the clarity and effectiveness of decision-making. Finally, the model also needs to learn the direction and magnitude of market changes simultaneously in one task. These two sub-task objectives may have inherent conflicts, thus limiting the model's accuracy in both direction judgment and magnitude prediction, making it difficult to achieve optimal prediction results.

[0020] These problems collectively restrict the reliability and practicality of asset valuation technology in actual business operations, and a more stable and clearer solution is urgently needed. Therefore, this application provides an asset valuation method that can effectively solve the above problems.

[0021] Figure 1 This is a flowchart illustrating the asset value prediction method provided in the embodiments of this application.

[0022] like Figure 1 As shown, the asset value prediction method 100 provided in this application embodiment includes steps S110-S130.

[0023] Step S110: Obtain feature time series data that affects asset value, wherein the feature time series data includes feature datasets and time series corresponding to feature datasets.

[0024] For example, the asset value mentioned in this solution is, for instance, the market value of real estate. When the market value of real estate fluctuates, it is necessary to accurately predict the market value of real estate. This can be achieved by first obtaining the characteristic time-series data related to the real estate value from each branch responsible for real estate business. This characteristic time-series data can be the latest available characteristic time-series data.

[0025] Feature time series data includes feature datasets and the corresponding time series. Feature datasets are rich, multi-dimensional data that can reflect the status of the real estate market or property. Specifically, they can include both static and dynamic attributes of the real estate itself (such as store locations, unit types, historical rents, etc.) and data on changes in the external market environment (such as competitor prices, economic indicators, etc.). Time series, on the other hand, is continuous time series data that can be used to describe the dynamic patterns of changes in each feature over time.

[0026] For example, the solution provided in this application can also provide decision support for the leasing business of apartment assets. Therefore, the asset value forecast mentioned in the solution can specifically refer to the forecast of the rental level of the apartment.

[0027] Step S120: Based on the characteristic time series data, make predictions to obtain discrete direction prediction results representing the direction of asset value fluctuations and continuous amplitude prediction results representing the magnitude of asset value fluctuations.

[0028] For example, discrete direction prediction results can be obtained by a direction prediction model, and continuous magnitude prediction results can be obtained by a magnitude prediction model. In the application scenario of apartment asset leasing, the discrete direction prediction results obtained based on the solution provided in this application can, for example, reflect the trend of rent changes, while the continuous magnitude prediction results can, for example, be used to quantify the degree of rent changes. The discrete direction prediction results and the continuous magnitude prediction results can serve as the core basis for rent adjustment and revenue management of apartment assets.

[0029] Step S130: Using the discrete direction prediction result as a gating signal, the continuous amplitude prediction result is subjected to gating routing processing to obtain the asset value prediction result.

[0030] For example, the solution provided in this application establishes a strict, asymmetric set of judgment rules. By using discrete direction prediction results as gating signals, it controls the flow and usage of continuous amplitude prediction results. Through the state reflected by the gating signal, it controls the acquisition of optimized asset value prediction results that conform to that state.

[0031] The solution provided in this application can achieve accurate and multi-dimensional prediction of asset value based on feature time-series data that integrates multiple characteristics. By decoupling the asset value prediction task into two independent sub-tasks, namely "trend judgment" and "magnitude estimation," and through a gating routing mechanism, the system can only provide specific adjustment ranges when it identifies clear trends that exceed business tolerance. This hierarchical and conditional prediction approach effectively improves the model's adaptability to different market conditions and avoids the biases that may arise from a single model.

[0032] The solution provided in this application transforms complex market information into clear and reliable decision support tools, which helps optimize asset returns and enhance overall competitiveness in volatile markets.

[0033] In the scheme provided in this application, the discrete direction prediction result can be predicted by a trained direction prediction model, and the continuous amplitude prediction result can be predicted by a trained amplitude prediction model. The training process of the direction prediction model and the amplitude prediction model will be described below.

[0034] Figure 2 This is a schematic diagram of the offline training phase of a model provided in an embodiment of this application.

[0035] In one embodiment, the orientation prediction model can be trained in the following manner.

[0036] First, historical feature time-series sample data is acquired, including a feature sample dataset and its corresponding time sample sequences. Then, based on the feature sample dataset and the time sample sequences, three-class orientation labels are obtained. Next, the feature sample dataset and the time sample sequences are input into the orientation prediction model for prediction, yielding discrete orientation prediction training results. Subsequently, based on the discrete orientation prediction training results and the three-class orientation labels, a discrete orientation prediction loss is obtained. Finally, based on the discrete orientation prediction loss, the model parameters of the orientation prediction model are updated in reverse.

[0037] For example, before making predictions, the system can first acquire historical feature time-series sample data. This data covers different apartment types in multiple city stores, recording a multi-dimensional feature sample dataset F at consecutive time points t (time sample series). This dataset F contains multiple feature sample data corresponding to time points t. The feature sample data includes, for example, the core indicators of the asset itself, such as historical rent and occupancy rate, and can also integrate external market dynamics, such as competitor rents and regional economic indices.

[0038] Based on the feature sample dataset and the time sample sequence, three-class directional labels can be obtained. These three-class directional labels are, for example, discrete category labels containing three categories: "rising," "stable," and "falling." Through this process, the system transforms the raw time-series data into a training sample set with clear trend labels. Then, the system uses the feature sample dataset F as input to the model, along with the three-class directional labels generated in the previous step, to train a directional prediction model.

[0039] Specifically, the feature sample dataset and time sample sequence can be input into the orientation prediction model for prediction to obtain discrete orientation prediction training results. Then, based on the discrete orientation prediction training results and the three-class orientation labels, the discrete orientation prediction loss can be obtained.

[0040] Finally, based on the discrete direction prediction loss, the model parameters of the direction prediction model are updated in reverse, resulting in a well-trained direction prediction model. This direction prediction model can learn the complex mapping relationship between multi-dimensional features and the future direction of rent changes, providing a reliable judgment signal for the subsequent gated routing mechanism.

[0041] In one embodiment, three-class directional labels are obtained based on a feature sample dataset and a time sample sequence.

[0042] For example, the solution provided in this application will be described in detail below in the context of apartment operation and management. In apartment operation and management, one of the core forecasting tasks is to predict price fluctuations over a future period in order to make reasonable price adjustment decisions and thus balance apartment occupancy rates and rental income. The price fluctuations over a future period are, for example, the price fluctuations over the next 7 days.

[0043] Specifically, such as Figure 2 As shown, based on the feature sample dataset and the time sample sequence, the sample mean and the current sample data value within a preset statistical period corresponding to each feature sample data in the feature sample dataset can be obtained. Subsequently, based on the sample mean and the current sample data value within the preset statistical period, the sample data fluctuation value is obtained.

[0044] Then, based on the current sample data value and the pre-set stable percentage threshold, the personalized absolute tolerance threshold corresponding to each feature sample data is obtained. The stable percentage threshold represents the percentage of fluctuation in the sample data fluctuation value that has not changed during the training of the direction prediction model.

[0045] Finally, based on the personalized absolute tolerance threshold and the sample data fluctuation value, the three-class directional labels are obtained.

[0046] For example, the preset statistical period described in this embodiment is, for example, 7 days. The sample mean within the preset statistical period is, for example, Figure 2 The 7-day average price shown Here, "future" refers to the time point t relative to each training sample (feature sample data). In other words, the target for predicting the average price over the next 7 days for each sample is the average price over the following 7 consecutive days, starting from its time point t. For example, the current sample data value is... Figure 2 The current selling price shown The sample data fluctuation value is, for example, Figure 2 The price fluctuation values ​​shown .

[0047] For example, such as Figure 2 As shown, the system can perform relevant calculations on the feature sample data corresponding to each time point t based on its internal target variable calculation unit. Specifically, based on the feature sample dataset F and the time sample sequence t, it obtains the average price over the next 7 days corresponding to each feature sample data in the feature sample dataset F. and current selling price .

[0048] Continue to base on the 7-day moving average and current selling price Calculate price fluctuation value The following formula can be used:

[0049] The system also introduces a pre-defined stability percentage threshold δ, for example, δ=0.009. This parameter represents the maximum percentage of fluctuation that is acceptable in business and not considered a price change. The system dynamically calculates the personalized absolute tolerance threshold τ for each feature sample data according to the following formula:

[0050] The personalized absolute tolerance threshold τ is dynamically calculated for each feature sample data.

[0051] In the embodiments of this application, a preset stable percentage threshold δ is introduced, and it is compared with the current sales price of each sample. By combining these methods, a personalized absolute tolerance threshold τ is dynamically calculated, thereby generating three-class directional labels that include "stationary" states. This is a significant improvement over purely data-driven labeling methods in related technologies, allowing the model training objective to be highly aligned with business decision-making needs from the outset.

[0052] By introducing a preset stability percentage threshold δ, the judgment of the market's "stable" state can be adapted to apartments at different price levels. Compared to using a uniform fixed threshold, this personalized threshold mechanism significantly improves the model's rationality and robustness, ensuring that business decision-making criteria can flexibly adapt to diverse actual price scenarios.

[0053] Finally, the system, based on relevant judgment logic, sets the price fluctuation value... The data is mapped to one of three categories: "rising," "falling," or "stable," thus generating a three-category directional label. .

[0054] In another embodiment, the stability percentage threshold δ provided in this embodiment can also be a dynamic threshold. For example, a personalized absolute tolerance threshold τ can be dynamically calculated based on the volatility of historical prices for each store (e.g., the standard deviation σ of the historical price series), such as τ = k × σ (where k is a constant coefficient), enabling the scheme to better adapt to the price volatility characteristics of different markets.

[0055] In one embodiment, the three-category directional labels include an upward directional label, a downward directional label, and a stable directional label. The personalized absolute tolerance threshold includes a first personalized absolute tolerance range, a second personalized absolute tolerance range, and a third personalized absolute tolerance range.

[0056] Based on the personalized absolute tolerance threshold and sample data fluctuation values, the three-class directional labels are calculated, specifically including: If the fluctuation value of the sample data is in the first personalized absolute tolerance range, an upward direction label is generated; if the fluctuation value of the sample data is in the second personalized absolute tolerance range, a downward direction label is generated; if the fluctuation value of the sample data is in the third personalized absolute tolerance range, a stable direction label is generated.

[0057] For example, the following conditional judgment logic can be used to obtain three-category directional labels based on three personalized absolute tolerance intervals. :

[0058] In the embodiments of this application, by mapping continuous price fluctuation values ​​to three distinct business classification labels (three-category directional labels) of "rising", "falling" and "stable", the abstract mathematical values ​​are transformed into business classification labels that can directly support decision-making, laying the foundation for subsequent noise filtering and stabilization decisions.

[0059] In one embodiment, the continuous amplitude prediction result is predicted by an amplitude prediction model, which is trained in the following manner.

[0060] Specifically, first, training labels for amplitude prediction are obtained based on the feature sample dataset and the time sample sequence. Then, the feature sample dataset and the time sample sequence are input into the amplitude prediction model for prediction, yielding continuous amplitude prediction training results. Next, based on the continuous amplitude prediction training results and the amplitude prediction training labels, the continuous amplitude prediction loss is obtained. Finally, the model parameters of the amplitude prediction model are updated in reverse based on the continuous amplitude prediction loss.

[0061] For example, compared to the training process of the direction prediction model described above, the data used to train the amplitude prediction model is also historical feature time-series sample data. Figure 2 As shown, the system takes the same feature sample dataset F as input and uses the original, continuous price fluctuation values. The objective is to train the magnitude prediction model.

[0062] This magnitude prediction model can be specifically used to quantify the degree of price changes, providing accurate numerical basis for refined decision-making, thereby enabling precise assessment of the degree of market changes in business decisions.

[0063] After training the direction prediction model and the magnitude prediction model, the following section describes the usage of these two models.

[0064] Figure 3 This is a flowchart illustrating the online prediction and decision-making stage of a model provided in an embodiment of this application.

[0065] like Figure 3 As shown, when the system receives the latest time-series data of features that will affect asset value, it will execute two prediction tasks in parallel. The system calls the pre-trained direction prediction model. This yields a discrete direction prediction result (discrete direction prediction result). The prediction result in this direction For example, it could be one of "rising", "falling", or "stable".

[0066] At the same time, the system calls the pre-trained amplitude prediction model in parallel. The model will output a continuous numerical value as the amplitude prediction value (continuous amplitude prediction result). .

[0067] The parallel processing mechanism proposed in this application is not simply a combination of two models, but rather a complete set of interconnected technical solutions to address the specific technical problem of unstable output in dynamic pricing models. This parallel processing mechanism can efficiently and simultaneously obtain the direction of price changes and the specific magnitude estimate, providing effective support for accurate decision-making.

[0068] In one embodiment, the discrete direction prediction result is used as a gating signal to perform gating routing processing on the continuous amplitude prediction result to obtain the asset value prediction result. This includes: filtering the continuous amplitude prediction result when the discrete direction prediction result is in a stable state and using a preset threshold as the asset value prediction result; and calculating the continuous amplitude prediction result when the discrete direction prediction result is in a rising or falling state to obtain the asset value prediction result.

[0069] For example, such as Figure 3 As shown, the system also includes a decision integration unit, which can provide gating logic. This application embodiment provides a direction-priority gating logic that does not employ simple weighting or combination, but instead establishes a strict, asymmetric set of judgment rules.

[0070] In one embodiment, the rise and fall status includes an upward status and a downward status; when the discrete direction prediction result is an upward or downward status, the continuous amplitude prediction result is calculated to obtain the asset value prediction result, including: when the discrete direction prediction result is an upward status, the absolute value of the continuous amplitude prediction result is used as the asset value prediction result; when the discrete direction prediction result is a downward status, the negative value of the absolute value of the continuous amplitude prediction result is used as the asset value prediction result.

[0071] Specifically, the prediction results in discrete directions As a gating signal, it determines whether to adopt the continuous amplitude prediction result. The information can be integrated based on the following judgment rules to obtain the final price adjustment value (asset value prediction result). .

[0072]

[0073] in, This represents the absolute value of the continuous amplitude prediction result.

[0074] According to the judgment rules above, when the discrete direction prediction result... When the value is "stationary," the output of the amplitude prediction model can be completely blocked, forcing the final result to be 0. This "gate" effectively filters out all tiny fluctuations that the model identifies as business noise, ensuring the stability of the decision.

[0075] When the discrete direction prediction result This "door" is only opened when the price is "rising" or "falling". At this time, only the information about the "magnitude" from the magnitude prediction model is adopted (achieved by taking the absolute value), while the direction is strictly determined by the direction prediction model.

[0076] In the embodiments of this application, an integrated architecture is designed that uses the direction prediction result as a "gating signal" to decide whether to adopt the amplitude prediction result. When the direction prediction is "stationary," the architecture completely blocks the output of the amplitude prediction model, forcing the final adjustment value to be 0. This "one-vote veto" asymmetric gating logic is a novel technical solution specifically designed to suppress the "jitter" of the prediction model output. It fundamentally guarantees the stability and executability of the final decision, effectively solving the technical problems that existing single regression models cannot solve.

[0077] In the embodiments of this application, both the direction prediction model and the magnitude prediction model employ the gradient boosting decision tree algorithm. These two models can be replaced by any other machine learning algorithm capable of performing the corresponding tasks. For example, the classification model can be replaced with logistic regression; the regression model can be replaced with linear regression. Alternatively, both models can be uniformly implemented by a multi-task learning neural network with a shared low-level feature extraction layer and two independent output heads (a Softmax output head for classification and a linear output head for regression).

[0078] Figure 4 This is a schematic diagram of an asset value prediction system provided in an embodiment of this application.

[0079] Embodiments of this application provide an asset value prediction system 400, the system 400 comprising: The acquisition module 410 is used to acquire feature time series data that affects the value of assets, wherein the feature time series data includes feature datasets and time series corresponding to feature datasets.

[0080] The prediction module 420 is used to make predictions based on characteristic time series data, and to obtain discrete direction prediction results representing the direction of asset value fluctuations and continuous amplitude prediction results representing the magnitude of asset value fluctuations.

[0081] The processing module 430 is used to perform gated routing processing on the continuous amplitude prediction results using the discrete direction prediction results as gate signals to obtain the asset value prediction results.

[0082] In some embodiments, the processing module 430 is further configured to: filter the continuous amplitude prediction results when the discrete direction prediction result is in a stable state, and use a preset threshold as the asset value prediction result; the prediction module 420 is further configured to: calculate the continuous amplitude prediction results when the discrete direction prediction result is in a rising or falling state, and obtain the asset value prediction result.

[0083] In some embodiments, the prediction module 420 is further configured to: when the discrete direction prediction result is in an upward trend, use the absolute value of the continuous amplitude prediction result as the asset value prediction result; and when the discrete direction prediction result is in a downward trend, use the negative value of the absolute value of the continuous amplitude prediction result as the asset value prediction result.

[0084] In some embodiments, the system 400 further includes a training module, which is configured to: acquire historical feature time-series sample data, wherein the historical feature time-series sample data includes a feature sample dataset and a time sample sequence corresponding to the feature sample dataset; obtain three-class orientation labels based on the feature sample dataset and the time sample sequence; input the feature sample dataset and the time sample sequence into the orientation prediction model for prediction to obtain discrete orientation prediction training results; obtain discrete orientation prediction loss based on the discrete orientation prediction training results and the three-class orientation labels; and update the model parameters of the orientation prediction model in reverse based on the discrete orientation prediction loss.

[0085] In some embodiments, the training module is further configured to: obtain the sample mean and current sample data value within a preset statistical period corresponding to each feature sample data in the feature sample dataset based on the feature sample dataset and the time sample sequence; obtain the sample data fluctuation value based on the sample mean and current sample data value within the preset statistical period; obtain the personalized absolute tolerance threshold corresponding to each feature sample data based on the current sample data value and a preset stability percentage threshold, wherein the stability percentage threshold represents the percentage of fluctuation in the sample data fluctuation value that has not changed during the training of the direction prediction model; the training module includes a label module, which is configured to: obtain three-class direction labels based on the personalized absolute tolerance threshold and the sample data fluctuation value.

[0086] In some embodiments, the labeling module is further configured to: generate an upward direction label if the sample data fluctuation value is in the first personalized absolute tolerance range; generate a downward direction label if the sample data fluctuation value is in the second personalized absolute tolerance range; and generate a stable direction label if the sample data fluctuation value is in the third personalized absolute tolerance range.

[0087] In some embodiments, the training module is further configured to: obtain amplitude prediction training labels for the amplitude prediction model based on the feature sample dataset and the time sample sequence; input the feature sample dataset and the time sample sequence into the amplitude prediction model for prediction to obtain continuous amplitude prediction training results; obtain continuous amplitude prediction loss based on the continuous amplitude prediction training results and the amplitude prediction training labels; and update the model parameters of the amplitude prediction model in reverse based on the continuous amplitude prediction loss.

[0088] It is understandable that for a detailed description of the asset value prediction system 400, please refer to the description of the asset value prediction method 100 above.

[0089] This application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the method in any of the above embodiments.

[0090] This application provides a computer program product that includes instructions that, when executed by a processor of a computer device, enable the computer device to perform the steps of the method described in any of the above embodiments.

[0091] Figure 5 A block diagram of an electronic device provided in an embodiment of this application.

[0092] This application provides an electronic device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method in any of the above embodiments.

[0093] like Figure 5 As shown, for ease of understanding, embodiments of this application illustrate a specific electronic device 500.

[0094] Electronic device 500 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0095] like Figure 5 As shown, device 500 includes a computing unit 501, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 502 or a computer program loaded from storage unit 508 into random access memory (RAM) 503. RAM 503 may also store various programs and data required for the operation of electronic device 500. The computing unit 501, ROM 502, and RAM 503 are interconnected via bus 504. Input / output (I / O) interface 505 is also connected to bus 504.

[0096] Multiple components in electronic device 500 are connected to I / O interface 505. These components include: input unit 506, such as a keyboard or mouse; output unit 507, such as various types of displays or speakers; storage unit 508, such as a disk or optical disk; and communication unit 509, such as a network interface card (NIC), modem, or wireless transceiver. Communication unit 509 allows electronic device 500 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0097] The computing unit 501 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 501 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 501 performs the various methods described above. For example, in some embodiments, any one or more of the methods described above can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 508. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 500 via ROM 502 and / or communication unit 509. When the computer program is loaded into RAM 503 and executed by the computing unit 501, one or more steps of any one or more of the methods described above can be performed. Alternatively, in other embodiments, the computing unit 501 can be configured to perform any one or more of the methods described above by any other suitable means (e.g., by means of firmware).

[0098] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this application, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0099] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0100] In the description of this application, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this application, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0101] In the description of this application, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "radial", "circumferential", etc., indicating the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application.

[0102] Furthermore, the terms "first," "second," etc., used in the embodiments of this application are for descriptive purposes only and should not be construed as indicating or implying relative importance, or implicitly specifying the number of technical features indicated in this embodiment. Therefore, features defined with terms such as "first" and "second" in the embodiments of this application can explicitly or implicitly indicate that the embodiment includes at least one of those features. In the description of this application, the word "multiple" means at least two or more, such as two, three, four, etc., unless otherwise explicitly and specifically defined in the embodiments.

[0103] In this application, unless otherwise explicitly specified or limited in the embodiments, the terms "installation," "connection," "joining," and "fixing" appearing in the embodiments should be interpreted broadly. For example, a connection can be a fixed connection, a detachable connection, or an integral part; it can also be a mechanical connection, an electrical connection, etc. Of course, it can also be a direct connection, or an indirect connection through an intermediate medium, or it can be the internal communication between two components, or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific implementation.

[0104] In this application, unless otherwise expressly specified and limited, "above" or "below" the second feature can mean that the first feature is in direct contact with the second feature, or that the first feature is in indirect contact with the second feature through an intermediate medium. Furthermore, "above," "on top of," and "over" the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature.

Claims

1. A method for predicting asset value, characterized in that, The method includes: Acquire feature time-series data that affect asset value, wherein the feature time-series data includes a feature dataset and the time series corresponding to the feature dataset; Based on the aforementioned characteristic time-series data, predictions are made to obtain discrete direction prediction results representing the direction of asset value fluctuations and continuous amplitude prediction results representing the magnitude of asset value fluctuations. Using the discrete direction prediction result as a gating signal, the continuous amplitude prediction result is subjected to gating routing processing to obtain the asset value prediction result.

2. The method according to claim 1, characterized in that, The step of using the discrete direction prediction result as a gate signal to perform gated routing processing on the continuous amplitude prediction result to obtain the asset value prediction result includes: When the discrete direction prediction result is in a stationary state, the continuous amplitude prediction result is filtered, and a preset threshold is used as the asset value prediction result. If the discrete direction prediction result indicates a rise or fall, the continuous amplitude prediction result is calculated to obtain the asset value prediction result.

3. The method according to claim 2, characterized in that, The rise / fall status includes an upward status and a downward status; when the discrete direction prediction result is a rise / fall status, the continuous amplitude prediction result is calculated to obtain the asset value prediction result, including: If the discrete direction prediction result is in the upward state, the absolute value of the continuous amplitude prediction result is used as the asset value prediction result. If the discrete direction prediction result indicates a downward trend, the negative value of the absolute value of the continuous magnitude prediction result is taken as the asset value prediction result.

4. The method according to claim 1, characterized in that, The discrete direction prediction result is obtained by a direction prediction model, which is trained in the following way: Obtain historical feature time series sample data, wherein the historical feature time series sample data includes a feature sample dataset and a time sample sequence corresponding to the feature sample dataset; Based on the feature sample dataset and the time sample sequence, three-class directional labels are obtained; The feature sample dataset and the time sample sequence are input into the direction prediction model for prediction to obtain discrete direction prediction training results; Based on the discrete direction prediction training results and the three-class direction labels, the discrete direction prediction loss is obtained; Based on the discrete direction prediction loss, the model parameters of the direction prediction model are updated in reverse.

5. The method according to claim 4, characterized in that, The process of obtaining three-class directional labels based on the feature sample dataset and the time sample sequence includes: Based on the feature sample dataset and the time sample sequence, the sample mean and the current sample data value within a preset statistical period corresponding to each feature sample data in the feature sample dataset are obtained; Based on the sample mean within the preset statistical period and the current sample data value, the sample data fluctuation value is obtained; Based on the current sample data value and the preset stability percentage threshold, a personalized absolute tolerance threshold is obtained for each of the feature sample data, wherein the stability percentage threshold represents the percentage of fluctuation in the sample data fluctuation value that has not changed during the training of the direction prediction model. The three-category directional labels are obtained based on the personalized absolute tolerance threshold and the sample data fluctuation value.

6. The method according to claim 5, characterized in that, The three-category directional labels include an upward directional label, a downward directional label, and a stable directional label; the personalized absolute tolerance threshold includes a first personalized absolute tolerance range, a second personalized absolute tolerance range, and a third personalized absolute tolerance range. The calculation of the three-class directional labels based on the personalized absolute tolerance threshold and the sample data fluctuation value includes: If the fluctuation value of the sample data is within the first personalized absolute tolerance range, then the upward direction label is generated; If the fluctuation value of the sample data is within the second personalized absolute tolerance range, then the downward direction label is generated; If the fluctuation value of the sample data falls within the third personalized absolute tolerance range, then the stable direction label is generated.

7. The method according to claim 4, characterized in that, The continuous amplitude prediction result is obtained by an amplitude prediction model, which is trained in the following way: Based on the feature sample dataset and the time sample sequence, the amplitude prediction training labels of the amplitude prediction model are obtained; The feature sample dataset and the time sample sequence are input into the amplitude prediction model for prediction to obtain continuous amplitude prediction training results; Based on the continuous amplitude prediction training results and amplitude prediction training labels, the continuous amplitude prediction loss is obtained; Based on the continuous amplitude prediction loss, the model parameters of the amplitude prediction model are updated in reverse.

8. An asset valuation system, characterized in that, The system includes: The acquisition module is used to acquire feature time-series data that affects asset value, wherein the feature time-series data includes a feature dataset and the time series corresponding to the feature dataset; The prediction module is used to make predictions based on the characteristic time series data to obtain discrete direction prediction results representing the direction of asset value fluctuations and continuous amplitude prediction results representing the magnitude of asset value fluctuations. The processing module is used to perform gated routing processing on the continuous amplitude prediction results using the discrete direction prediction results as a gate signal to obtain asset value prediction results.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-7.

Images