Equipment asset depreciation prediction method based on machine learning

Abstract

The invention discloses an equipment asset depreciation prediction method based on machine learning, and the method comprises the following steps: S1, collecting an asset original value, a depreciation code, an age limit, a starting date and a historical net value, and generating an asset state input sequence in a unified time domain through normalization; s2, inputting the normalized asset state into a Shenchang differential equation model, and constructing a differential system of the asset state evolving along with time; s3, identifying parameter changes in the asset depreciation period, extracting change time and amplitude, and constructing a disturbance event sequence; s4, constructing a time gating function and a disturbance coupling function according to the disturbance event sequence; s5, injecting the disturbance coupling function into a differential system, and resolving to obtain a depreciation evolution path; and S6, mapping the prediction path to an actual time interval, and outputting a depreciation trend prediction result. According to the invention, fine prediction of the depreciation trend of the equipment and dynamic response of strategy change are realized, and the intelligent level of asset management is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment asset management, and in particular to an equipment asset depreciation prediction method based on machine learning. Background Art

[0002] In the process of fixed asset management in an enterprise, depreciation forecasting for equipment assets is a critical foundation for asset valuation, financial budgeting, retirement planning, and revaluation decisions. Traditional depreciation calculation methods primarily rely on fixed formulas and rules within financial systems. Common methods include the straight-line method and the declining balance method. These methods typically use a set depreciation period, depreciation rate, and original asset value as parameters, and perform linear or nonlinear deductions on the depreciation value at fixed intervals. These methods are simple to calculate and straightforward to operate, making them suitable for accounting needs. However, they fail to reflect the actual usage status of equipment and the impact of policy changes on the evolution of asset value.

[0003] In existing technology, most depreciation simulation systems are integrated into ERP platforms or asset management systems. A common practice is to manually adjust depreciation parameters based on asset master data, trigger a batch depreciation recalculation module, and output simulation results. Some systems support parameter versions with different depreciation codes, years, or start dates. Through branching simulation, multiple comparable depreciation curves are generated to assist in policy selection. However, these methods rely on static rule-based logic processing and lack the ability to learn from historical depreciation behavior or predict future trends. Furthermore, they are unable to characterize the dynamic impact of policy perturbations on a timeline.

[0004] In recent years, machine learning-based forecasting methods have been gradually applied in asset management. Some methods attempt to fit historical book net value sequences through multi-layer perceptrons, recurrent neural networks, or time series models to output forecast values ​​for future periods. These methods have improved forecast flexibility to a certain extent, but most rely on discrete-time modeling structures, making it difficult to handle non-uniformly spaced depreciation records and inject structured policy change behaviors into the model. In addition, different assets have different depreciation starting points, making it difficult to align time series and resulting in structural inconsistencies in model training samples. For behaviors such as depreciation code switching and age modification that are common in actual operations and maintenance, such models can often only be processed as input features and are unable to express their real-time intervention in state evolution during the forecasting process.

[0005] Currently, there is a lack of a technical solution that can model the evolution of net asset value in the continuous time domain, support the injection of perturbations from multiple depreciation strategies, and provide a clearly structured prediction and solution mechanism. Existing methods are unable to accurately model the impact of policy adjustments at any point in the asset lifecycle on future depreciation trends, nor can they simultaneously address the modeling requirements of temporal continuity, perturbation response, and predictive interpretability. Therefore, a continuous prediction method that establishes a dynamic functional mapping between the initial state of an asset, historical depreciation behavior, and policy changes is urgently needed to accurately model time-varying perturbations in asset depreciation trend forecasting.

[0006] Therefore, how to provide a method for predicting equipment asset depreciation based on machine learning is an urgent problem that needs to be solved by those skilled in the art. Summary of the Invention

[0007] One objective of the present invention is to propose a machine learning-based method for predicting equipment asset depreciation. This method combines modeling of asset historical state sequences with a policy perturbation injection mechanism to construct a neural ordinary differential system that continuously solves on a unified timeline. This method dynamically predicts equipment depreciation trajectories through asset state normalization, identification of policy change events, and the construction of a time-gated function coupled with a perturbation function. This method boasts high temporal modeling accuracy, a clear policy response structure, and continuous and smooth prediction results, making it suitable for a variety of business scenarios, including asset valuation, depreciation simulation, and financial calculations.

[0008] A method for predicting equipment asset depreciation based on machine learning according to an embodiment of the present invention includes the following steps: S1. Collect data on equipment assets, construct an asset status sequence dataset, perform time normalization on asset status sequences with different depreciation start dates, and generate a normalized asset status input sequence; S2. Input the normalized asset status input sequence into the neural ordinary differential equation prediction model, construct a differential equation system with the change of asset status over time as the modeling target, parameterize the right-hand side function through the neural network function, and use a numerical solver to generate the asset depreciation evolution path; S3. Identify depreciation policy changes during the asset lifecycle, extract the change time, change type, and change magnitude of each policy change operation, and construct a depreciation policy disturbance event sequence; S4. Construct a time gating function to dynamically control whether the disturbance term is injected into the differential system according to the activation time of the depreciation strategy disturbance event sequence, and establish a disturbance coupling function; S5. Inject the perturbation coupling function into the Neural Ordinary Differential Equation prediction model to form a perturbation coupled differential system. Perform continuous-time numerical solution on the system and output the predicted trajectory of the net asset value after considering the influence of the strategy perturbation. S6. Map the net asset value prediction trajectory to the actual time interval to form a depreciation trend prediction result of the corresponding asset.

[0009] Optionally, the time normalization process in S1 includes: taking the depreciation start date of each asset as the normalization starting point, mapping the corresponding state time series to the standard time interval [0,1] according to the life cycle, and normalizing the time point by the formula Calculate, where is the original time point, The starting time for asset depreciation, The end time of the current observation period is normalized timestamp The combination of asset state variables corresponding to the time point forms the input pair , construct a normalized asset state input sequence.

[0010] Optionally, the S2 specifically includes: S21. Input the normalized asset state into each pair of data in the sequence Input to the state encoder network, where represents the normalized timestamp, Represents the asset state variables at the corresponding time point, including the asset original value, depreciation code, depreciation period and book value, and outputs the initial hidden state vector ; S22, based on the initial hidden state , construct a system of neural ordinary differential equations in the form: ; in, Normalized time The hidden state under are the trainable neural network parameters, is a neural network function used to represent the derivative of the asset state over time; S23, in the normalized time interval Set the numerical solution time point set ,by 、 As the boundary, the ODE solver is used to perform continuous-time numerical solution on the neural ODE system, and the ODE at each time point is calculated. The corresponding hidden state ; S24, each time point The corresponding hidden state Input to the decoder network, decode and generate the corresponding asset book net value prediction value , forming a predicted value sequence .

[0011] Optionally, the S3 specifically includes: S31. Obtain the strategic parameter change records of the equipment assets within the depreciation cycle, filter out the historical change operations including depreciation code change, depreciation period modification, or depreciation start date adjustment, and extract the time point of each change record. , parameter value before change and the changed parameter values ; S32. For each change operation, calculate the disturbance value according to the change type , where when the change is a depreciation code change, the parameter values ​​before and after the change are encoded into discrete vectors and the encoding difference is calculated as , when the change is to modify the depreciation period, calculate When the depreciation start date is changed, the normalized time offset corresponding to the time difference before and after the adjustment is calculated as ; S33, changing the time point and the corresponding disturbance value The disturbance event , and arrange them in chronological order to generate a disturbance event sequence ,in, Indicates the number of valid policy changes that exist during the asset lifecycle; S34, the disturbance event sequence Stored in the disturbance management structure, complete the unified encapsulation of asset disturbance events.

[0012] Optionally, the disturbance value for the depreciation code change type in S32 The calculation operation also includes: the depreciation key parameter value before the change and the changed depreciation key parameter value The generated encoding vectors are normalized separately using the vector normalization function , the normalized vector difference is taken as the final perturbation value, recorded as , for all perturbation values The dimensions are uniformly adjusted to a fixed length and zero padding is used to complete dimension alignment.

[0013] Optionally, the S4 specifically includes: S41, in the normalized time interval Set a fixed step size , generating a time point sequence ,in, , , is the total number of discrete time steps; S42. At every point in time Initialize the perturbation value , and traverse the disturbance event sequence , for each disturbance event , to determine whether the conditions are met , if satisfied, then the disturbance value Add the disturbance value corresponding to the current time point , and take the accumulated result as the new disturbance value at the time point; S43, the disturbance values ​​corresponding to all time points are combined into a vector in chronological order , and in accordance with the time point sequence Store in corresponding order to form a time series of disturbance values; S44. At every point in time , call the asset hidden state , calculate the output of the right-hand function of the neural network , the disturbance value corresponding to the time point Add to generate update terms ,Will Store update vector sequence .

[0014] Optionally, the disturbance value time series generated in S43 The following processing steps are included before storage: for each disturbance value in the sequence Perform a numeric type conversion operation to convert all Mapped to a floating-point vector of fixed dimension, if the original perturbation value is a scalar, it is expanded to a length of The sparse vector is filled with zero values, and the precision standardization is performed on all disturbance value vectors, and all elements in each vector are normalized to the interval according to the maximum absolute value. , the processed vectors maintain the original sequence order and form a unified format vector structure.

[0015] Optionally, the S5 specifically includes: S51. Read normalized time point sequence and the corresponding update item sequence , read the initial hidden state , set the time step of the differential solution , construct the hidden state sequence array , and assign the initial state to ; S52, initialization loop index , perform the following iterative operations: at each time step , from the array Read the previous state , from the array Read the update item corresponding to the previous time step , perform the Euler integration operation and calculate the hidden state of the current time step ; Write the results to an array No. item; update the index to , repeat the process until ; S53. Create a book net value forecast value array , initialize the loop index , in each loop, from the array Read hidden state ,Will Input to the decoder network and perform forward reasoning to calculate the book value prediction value , write the calculation results into the array No. Item, update index , repeat until ; S54. Establish a dual-channel data structure in the memory structure to store the normalized time point sequence separately Array of predicted book value values , pair them one by one by index, and each pair Write the structure fields to complete the timing matching and cache operations of the prediction sequence.

[0016] Optionally, the S6 specifically includes: S61. Read the normalized time point sequence, the book net value forecast value sequence, the depreciation start time, and the depreciation end time to determine the actual time interval corresponding to the depreciation; S62. For each normalized time point in the normalized time point sequence, calculate its corresponding time point in the actual time interval, and write all calculated actual time points into the actual time point sequence in order; S63. Combine each actual time point in the actual time point sequence with the predicted value at the same index position in the book net value predicted value sequence into a set of data pairs, and write all the data pairs into the net asset value prediction mapping structure in index order; S64. Perform format standardization processing on each set of data in the net asset value forecast mapping structure, convert each actual time point into a timestamp representation in a unified format, adjust each forecast value to a set floating point precision value, and write them into the depreciation trend forecast result table in chronological order.

[0017] The beneficial effects of the present invention are: Continuous-time modeling of the equipment asset depreciation process is realized: by introducing the neural ordinary differential equation structure, the present invention can gradually integrate the asset status on the normalized time axis, avoiding the shortcomings of traditional discrete models in time accuracy and interpolation fitting, and can more finely characterize the dynamic change process of net asset value, adapting to the unified modeling requirements of non-uniformly spaced data input and assets with multiple starting points.

[0018] A disturbance injection mechanism that can respond to strategy changes has been constructed: in response to behaviors such as depreciation code switching, age adjustment, and start time correction in the asset depreciation process, the present invention embeds strategy changes in the form of coupling terms into the differential system through disturbance event sequence extraction and time gating function construction, supporting strategy disturbances to trigger intervention in the state evolution process at any time point, effectively improving the model's ability to fit real asset depreciation behavior.

[0019] A prediction result process with a clear structure and standardized output has been formed: by mapping the model solution results with the actual time interval and completing the formatting of the prediction values, the present invention facilitates docking with the existing asset management system while maintaining the stability of the calculation process, thereby improving the system compatibility of the prediction results and the efficiency of business implementation. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0021] Figure 1 This is an overall flow chart of a method for predicting equipment asset depreciation based on machine learning proposed by the present invention; Figure 2 This is a diagram of asset status normalization and time axis mapping for a machine learning-based equipment asset depreciation prediction method proposed by the present invention; Figure 3 This is the net value prediction trajectory mapping and output diagram of the equipment asset depreciation prediction method based on machine learning proposed by the present invention. DETAILED DESCRIPTION

[0022] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0023] refer to Figure 1-3 , a method for predicting equipment asset depreciation based on machine learning, comprising the following steps: S1. Collect data on equipment assets, construct an asset status sequence dataset, perform time normalization on asset status sequences with different depreciation start dates, and generate a normalized asset status input sequence; S2. Input the normalized asset status input sequence into the neural ordinary differential equation prediction model, construct a differential equation system with the change of asset status over time as the modeling target, parameterize the right-hand side function through the neural network function, and use a numerical solver to generate the asset depreciation evolution path; S3. Identify depreciation policy changes during the asset lifecycle, extract the change time, change type, and change magnitude of each policy change operation, and construct a depreciation policy disturbance event sequence; S4. Construct a time gating function to dynamically control whether the disturbance term is injected into the differential system according to the activation time of the depreciation strategy disturbance event sequence, and establish a disturbance coupling function; S5. Inject the perturbation coupling function into the Neural Ordinary Differential Equation prediction model to form a perturbation coupled differential system. Perform continuous-time numerical solution on the system and output the predicted trajectory of the net asset value after considering the influence of the strategy perturbation. S6. Map the net asset value prediction trajectory to the actual time interval to form a depreciation trend prediction result of the corresponding asset.

[0024] The present invention introduces a perturbation-coupled neural ordinary differential equation structure for equipment asset depreciation prediction, realizing the fusion modeling of three types of information: asset status, time series, and strategy parameters. It overcomes the structural defect of traditional discrete models that cannot express dynamic strategy changes, and effectively improves the continuity of depreciation trajectory and the accuracy of prediction response.

[0025] In this embodiment, the time normalization process in S1 includes: taking the depreciation start date of each asset as the normalization starting point, mapping the corresponding state time series to the standard time interval [0,1] according to the life cycle, and normalizing the time point by the formula Calculate, where is the original time point, The starting time for asset depreciation, The end time of the current observation period is normalized timestamp The combination of asset state variables corresponding to the time point forms the input pair , construct a normalized asset state input sequence.

[0026] The present invention introduces a unified time normalization mechanism in the asset status construction process, standardizes and aligns the status sequence based on the asset start time, solves the problem of training sample time misalignment caused by inconsistent depreciation starting points of multiple assets, and improves the consistency and expression ability of the model input structure.

[0027] In this embodiment, S2 specifically includes: S21. Input the normalized asset state into each pair of data in the sequence Input to the state encoder network, where represents the normalized timestamp, Represents the asset state variables at the corresponding time point, including the asset original value, depreciation code, depreciation period and book value, and outputs the initial hidden state vector ; S22, based on the initial hidden state , construct a system of neural ordinary differential equations in the form: ; in, Normalized time The hidden state under are the trainable neural network parameters, is a neural network function used to represent the derivative of the asset state over time; S23, in the normalized time interval Set the numerical solution time point set ,by 、 As the boundary, the ODE solver is used to perform continuous-time numerical solution on the neural ODE system, and the ODE at each time point is calculated. The corresponding hidden state ; S24, each time point The corresponding hidden state Input to the decoder network, decode and generate the corresponding asset book net value prediction value , forming a predicted value sequence .

[0028] The present invention adopts the neural ordinary differential equation to model the depreciation process, replacing the traditional discrete structure with the time-continuous modeling method, allowing the net asset value to evolve in a differential form over time, with higher time resolution and numerical stability, and adapting to the non-uniformly spaced bookkeeping scenarios.

[0029] In this embodiment, S3 specifically includes: S31. Obtain the strategic parameter change records of the equipment assets within the depreciation cycle, filter out the historical change operations including depreciation code change, depreciation period modification, or depreciation start date adjustment, and extract the time point of each change record. , parameter value before change and the changed parameter values ; S32. For each change operation, calculate the disturbance value according to the change type , where when the change is a depreciation code change, the parameter values ​​before and after the change are encoded into discrete vectors and the encoding difference is calculated as , when the change is to modify the depreciation period, calculate When the depreciation start date is changed, the normalized time offset corresponding to the time difference before and after the adjustment is calculated as ; S33, changing the time point and the corresponding disturbance value The disturbance event , and arrange them in chronological order to generate a disturbance event sequence ,in, Indicates the number of valid policy changes that exist during the asset lifecycle; S34, the disturbance event sequence Stored in the disturbance management structure, complete the unified encapsulation of asset disturbance events.

[0030] By constructing a disturbance event sequence, the present invention represents the change operations such as depreciation code, life, and start date in a structured manner, forming a clear disturbance injection path, providing a logical entry for subsequent time dynamic control and policy response mechanism, and improving the accuracy of policy change modeling.

[0031] In this embodiment, the disturbance value for the depreciation code change type in S32 is The calculation operation also includes: the depreciation key parameter value before the change and the changed depreciation key parameter value The generated encoding vectors are normalized separately using the vector normalization function , the normalized vector difference is taken as the final perturbation value, recorded as , for all perturbation values The dimensions are uniformly adjusted to a fixed length and zero padding is used to complete dimension alignment.

[0032] The present invention introduces perturbation vector normalization and dimension unification operations in the depreciation code perturbation processing process, so that the coding difference has numerical stability and consistency with the input vector, which improves the learnability of the perturbation model and the training stability of the neural network.

[0033] In this embodiment, the S4 specifically includes: S41, in the normalized time interval Set a fixed step size , generating a time point sequence ,in, , , is the total number of discrete time steps; S42. At every point in time Initialize the perturbation value , and traverse the disturbance event sequence , for each disturbance event determining whether the condition is met , if the condition is met, adding the disturbance value to the disturbance value corresponding to the current time point , and taking the accumulation result as the new disturbance value of the time point S43, grouping the disturbance values corresponding to all time points into a vector in time sequence , and storing them in the order corresponding to the time point sequence to form a disturbance value time sequence S44, at each time point , calling the asset hidden state , calculating the output of the neural network right end function , adding the disturbance value corresponding to the time point to generate an update item , and storing in the update vector sequence .

[0034] The present application designs a time-gated control structure and a disturbance coupling logic, which realizes the coupling and fusion of policy change and predicted state path by activating the disturbance value at a specific time point and writing it into the right end function of the differential equation, thereby effectively improving the dynamic response capability of the model to policy switching behavior.

[0035] In the present embodiment, the disturbance value time sequence generated in S43 includes the following processing steps before storage: performing a numerical type conversion operation on each disturbance value in the sequence , mapping all into a fixed-dimension floating-point vector, if the original disturbance value is a scalar, expanding it into a sparse vector with a length of and filling it with zero values according to a preset rule, performing precision standardization processing on all disturbance value vectors, normalizing all elements in each vector to the interval according to the maximum absolute value, and the processed vectors maintain the original sequence order to form a uniform format vector structure.

[0036] The present application performs dimension standardization and precision normalization processing on the disturbance vector, unifies the vector structure of the disturbance input, reduces the modeling error caused by inconsistent input dimensions, and enhances the compatibility and training convergence performance of the model to multiple types of disturbance values.

[0037] In the present embodiment, S5 specifically includes: S51, reading the normalized time point sequence and the corresponding update item sequence , reading the initial hidden state , setting the differential solving time step , and constructing a hidden state sequence array , and assign the initial state to ; S52, initialization loop index , perform the following iterative operations: at each time step , from the array Read the previous state , from the array Read the update item corresponding to the previous time step , perform the Euler integration operation and calculate the hidden state of the current time step ; Write the results to an array No. item; update the index to , repeat the process until ; S53. Create a book net value forecast value array , initialize the loop index , in each loop, from the array Read hidden state ,Will Input to the decoder network and perform forward reasoning to calculate the book value prediction value , write the calculation results into the array No. Item, update index , repeat until ; S54. Establish a dual-channel data structure in the memory structure to store the normalized time point sequence separately Array of predicted book value values , pair them one by one by index, and each pair Write the structure fields to complete the timing matching and cache operations of the prediction sequence.

[0038] The present invention constructs a hidden state update process based on the numerical integration method, and performs Euler iteration calculation in combination with the disturbance coupling update term, thereby avoiding the state jump problem in the discrete modeling process and ensuring the continuity of state evolution and the accuracy and stability of differential system solution.

[0039] In this embodiment, S6 specifically includes: S61. Read the normalized time point sequence, the book net value forecast value sequence, the depreciation start time, and the depreciation end time to determine the actual time interval corresponding to the depreciation; S62. For each normalized time point in the normalized time point sequence, calculate its corresponding time point in the actual time interval, and write all calculated actual time points into the actual time point sequence in order; S63, combine each actual time point in the actual time point sequence with the predicted value at the same index position in the net worth prediction value sequence into a set of data pairs, and write all the data pairs into the net worth prediction mapping structure in index order; S64, perform format standardization processing on each data pair in the net worth prediction mapping structure, convert each actual time point into a timestamp representation in a unified format, adjust each predicted value to a set floating point precision value, and write in time order into the depreciation trend prediction result table.

[0040] The present application maps the normalized time points in the prediction results to the actual time interval, binds the actual time points and the net worth prediction values through structuring, completes the time alignment and format standardization of the result output, and facilitates the integration and visualization of the prediction trajectory in the asset management system.

[0041] Embodiment 1: In order to verify the feasibility of the present application in implementation, the present application is applied to the depreciation prediction scene of the equipment assets of a large manufacturing enterprise. The enterprise has more than 12000 fixed assets, covering heavy processing equipment, transmission lines, power equipment, intelligent warehouse equipment and other asset types, with many asset quantities, various types, high value and frequent changes. The management department is faced with many realistic problems such as frequent adjustment of depreciation strategy, lagging prediction work and distorted simulation results.

[0042] In actual operation, the depreciation simulation system of the enterprise originally uses the standard depreciation engine integrated with the ERP system, only supports the static simulation function after the depreciation code and the service life are changed, cannot reflect the real-time impact of strategy adjustment on future net worth, and all predictions are completed based on fixed time steps and rule calculation methods, lacking unified processing ability for the depreciation start time of different equipment. In the budget preparation period of multiple projects, the average deviation between the equipment financial prediction results and the final financial clearance results is more than 14%, and the deviation of individual high-value equipment is even more than 28%, causing budget misjudgment and financial lag.

[0043] In this embodiment, the method of the present application is deployed in the asset management platform of the enterprise, the neural ordinary differential equation is used to build the depreciation evolution path, and the original value, depreciation code, service life, start depreciation time and historical net worth record of all equipment assets are converted into a standard format and subjected to time normalization processing. For equipment assets with depreciation strategy adjustment behavior, the disturbance time points and parameter differences are extracted by calling the historical adjustment records in the background to automatically generate a disturbance event sequence.

[0044] Taking a certain group of typical equipment as an example, 90 heavy numerical control equipment put into use after 2017 are selected, the original values are all in the interval of 1.6 million to 3 million, the depreciation life is uniformly 10 years, the balance decreasing method is adopted, and the initial depreciation time span is between January, 2017 and December, 2019. The historical data show that 37 equipment modified the depreciation code or the depreciation life in the running process due to policy adjustment. For the equipment with disturbance of the depreciation strategy, the application constructs a time gating function and a disturbance coupling function to dynamically inject the disturbance factor into the prediction model.

[0045] The application is used for predicting the future 36-month net book value of the equipment cluster, and compared with the traditional static rule method and the LSTM model without disturbance learning. In the prediction accuracy evaluation, the actual depreciation net book value is taken as the reference, and the mean absolute error (MAE) and the relative error (MAPE) of the three methods in different time periods are calculated respectively. The results show that: the overall average absolute error of the application on all equipment for 36 months is 18,300 yuan, and the relative error is 3.1%; while the relative error of the static rule simulation method is 10.7%, and the relative error of the LSTM model is 6.2%. The error improvement of the application is the most obvious on the equipment with high frequency of strategy adjustment.

[0046] At the same time, the consistency of the prediction results and the actual depreciation trend is visualized and analyzed, and the net value trajectory curve output by the application shows a smooth response and a continuous transition before and after the change of the depreciation code, without inflection point mutation or depreciation fault, with high explainability and simulation fidelity. In the project financial simulation, the monthly budget deviation of the prediction value of the application is controlled within ±3%, which is about 70% less than the prediction fluctuation of the traditional method.

[0047] Table 1 shows the error comparison data of the original asset parameters, strategy change history, actual depreciation net value and the prediction results of part of the heavy equipment, all of which are derived from the business records in the real running environment: Table 1: Comparison table of heavy equipment depreciation prediction effect ; The above table 1 shows the depreciation prediction effect of the application on typical heavy equipment assets, covering the comparison data of the original asset parameters, strategy change situation, actual net value and predicted net value of seven groups of equipment, and reflecting the precision level of the prediction results of the application through the mean absolute error (MAE) and the relative error index. From the asset original value distribution, the sample equipment original value range is between 1.604 million and 3 million, which has a certain representativeness. In terms of depreciation strategy, 5 equipment have adjustment behavior of depreciation code or depreciation life, and 2 equipment remain the original strategy unchanged, which can be used for comparative analysis of the response ability of the prediction model to the disturbance event.

[0048] Comparing the actual net value with the predicted net value of the present invention, it can be seen that the predicted results of all samples are highly close to the actual net value in the 36th month, with the maximum absolute error being 34,900 yuan and the minimum being only 3,600 yuan. The relative error control is balanced, with most samples below 2% and the lowest being only 0.23%. In devices with policy changes, such as HDX-014 and HDX-050, the present invention still accurately tracks changes in net value after adjusting the depreciation period, demonstrating a strong disturbance response capability. In devices HDX-035 and HDX-078, where the policy remains unchanged, the predicted values ​​are highly consistent with the actual values, indicating that the model has a stable non-disturbance depreciation trajectory learning capability.

[0049] Furthermore, for equipment with high plateau values ​​and high depreciation intensity (such as HDX-066), the proposed method maintains a 1.96% error rate despite significant fluctuations, demonstrating its effective control capabilities for high-value assets. Overall, the data shown in Table 1 demonstrates that the proposed method outperforms traditional depreciation estimation methods in multiple aspects, including unified time modeling, strategic disturbance handling, and prediction accuracy control. This demonstrates the feasibility of its application in complex industrial asset environments.

[0050] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for predicting equipment asset depreciation based on machine learning, characterized in that: The steps include: S1. Collect data on equipment assets, construct an asset status sequence dataset, perform time normalization on asset status sequences with different depreciation start dates, and generate a normalized asset status input sequence; S2. Input the normalized asset status input sequence into the neural ordinary differential equation prediction model, construct a differential equation system with the change of asset status over time as the modeling target, parameterize the right-hand side function through the neural network function, and use a numerical solver to generate the asset depreciation evolution path; S3. Identify depreciation policy changes during the asset lifecycle, extract the change time, change type, and change magnitude of each policy change operation, and construct a depreciation policy disturbance event sequence; S4. Construct a time gating function to dynamically control whether the disturbance term is injected into the differential system according to the activation time of the depreciation strategy disturbance event sequence, and establish a disturbance coupling function; S5. Inject the perturbation coupling function into the Neural Ordinary Differential Equation prediction model to form a perturbation coupled differential system. Perform continuous-time numerical solution on the system and output the predicted trajectory of the net asset value after considering the influence of the strategy perturbation. S6. Map the net asset value prediction trajectory to the actual time interval to form a depreciation trend prediction result of the corresponding asset.

2. The equipment asset depreciation prediction method based on machine learning according to claim 1 is characterized in that: The time normalization process in S1 includes: taking the depreciation start date of each asset as the normalization starting point, mapping the corresponding state time series to the standard time interval [0,1] according to the life cycle, and normalizing the time point by the formula Calculate, where is the original time point, The starting time for asset depreciation, The end time of the current observation period is normalized timestamp The combination of asset state variables corresponding to the time point forms the input pair , construct a normalized asset state input sequence.

3. The equipment asset depreciation prediction method based on machine learning according to claim 2 is characterized in that: The S2 specifically includes: S21. Input the normalized asset state into each pair of data in the sequence Input to the state encoder network, where represents the normalized timestamp, Represents the asset state variables at the corresponding time point, including the asset original value, depreciation code, depreciation period and book value, and outputs the initial hidden state vector ; S22, based on the initial hidden state , construct a system of neural ordinary differential equations in the form: ； in, Normalized time The hidden state under are the trainable neural network parameters, is a neural network function used to represent the derivative of the asset state over time; S23, in the normalized time interval Set the numerical solution time point set ,by 、 As the boundary, the ODE solver is used to perform continuous-time numerical solution on the neural ODE system, and the ODE at each time point is calculated. The corresponding hidden state ; S24, each time point The corresponding hidden state Input to the decoder network, decode and generate the corresponding asset book net value prediction value , forming a predicted value sequence .

4. The equipment asset depreciation prediction method based on machine learning according to claim 3 is characterized in that: The S3 specifically includes: S31. Obtain the strategic parameter change records of the equipment assets within the depreciation cycle, filter out the historical change operations including depreciation code change, depreciation period modification, or depreciation start date adjustment, and extract the time point of each change record. , parameter value before change and the changed parameter values ; S32. For each change operation, calculate the disturbance value according to the change type , where when the change is a depreciation code change, the parameter values ​​before and after the change are encoded into discrete vectors and the encoding difference is calculated as , when the change is to modify the depreciation period, calculate When the depreciation start date is changed, the normalized time offset corresponding to the time difference before and after the adjustment is calculated as ; S33, changing the time point and the corresponding disturbance value The disturbance event , and arrange them in chronological order to generate a disturbance event sequence ,in, Indicates the number of valid policy changes that exist during the asset lifecycle; S34, the disturbance event sequence Stored in the disturbance management structure, complete the unified encapsulation of asset disturbance events.

5. The equipment asset depreciation prediction method based on machine learning according to claim 4 is characterized in that: The disturbance value for the depreciation code change type in S32 The calculation operation also includes: the depreciation key parameter value before the change and the changed depreciation key parameter value The generated encoding vectors are normalized separately using the vector normalization function , the normalized vector difference is taken as the final perturbation value, recorded as , for all perturbation values The dimensions are uniformly adjusted to a fixed length and zero padding is used to complete dimension alignment.

6. The equipment asset depreciation prediction method based on machine learning according to claim 5 is characterized in that: The S4 specifically includes: S41, in the normalized time interval Set a fixed step size , generating a time point sequence ,in, , , is the total number of discrete time steps; S42. At every point in time Initialize the perturbation value , and traverse the disturbance event sequence , for each disturbance event , to determine whether the conditions are met , if satisfied, then the disturbance value The disturbance value added to the current time point , and take the accumulated result as the new disturbance value at the time point; S43, the disturbance values ​​corresponding to all time points are combined into a vector in chronological order , and in accordance with the time point sequence Store in corresponding order to form a time series of disturbance values; S44. At every point in time , call the asset hidden state , calculate the output of the right-hand function of the neural network , the disturbance value corresponding to the time point Add to generate update terms ,Will Store update vector sequence .

7. The equipment asset depreciation prediction method based on machine learning according to claim 6 is characterized in that: The disturbance value time series generated in S43 The following processing steps are included before storage: for each disturbance value in the sequence Perform a numeric type conversion operation to convert all Mapped to a floating-point vector of fixed dimension, if the original perturbation value is a scalar, it is expanded to a length of The sparse vector is filled with zero values, and the precision standardization is performed on all disturbance value vectors, and all elements in each vector are normalized to the interval according to the maximum absolute value. , the processed vectors maintain the original sequence order and form a unified format vector structure.

8. The equipment asset depreciation prediction method based on machine learning according to claim 7 is characterized in that: The S5 specifically includes: S51. Read normalized time point sequence and the corresponding update item sequence , read the initial hidden state , set the time step of the differential solution , construct the hidden state sequence array , and assign the initial state to ; S52, initialization loop index , perform the following iterative operations: at each time step , from the array Read the previous state , from the array Read the update item corresponding to the previous time step , perform the Euler integration operation and calculate the hidden state of the current time step: ； Write the results to an array No. item; update the index to , repeat the process until ; S53. Create a book net value forecast value array , initialize the loop index , in each loop, from the array Read hidden state ,Will Input to the decoder network and perform forward reasoning to calculate the book value prediction value , write the calculation results into the array No. Item, update index , repeat until ; S54. Establish a dual-channel data structure in the memory structure to store the normalized time point sequence separately Array of predicted book value values , pair them one by one by index, and each pair Write the structure fields to complete the timing matching and cache operations of the prediction sequence.

9. The equipment asset depreciation prediction method based on machine learning according to claim 8 is characterized in that: The S6 specifically includes: S61. Read the normalized time point sequence, the book net value forecast value sequence, the depreciation start time, and the depreciation end time to determine the actual time interval corresponding to the depreciation; S62. For each normalized time point in the normalized time point sequence, calculate the corresponding time point in the actual time interval, and write all calculated actual time points into the actual time point sequence in order; S63: Combine each actual time point in the actual time point sequence with the predicted value at the same index position in the book net value predicted value sequence into a set of data pairs, and write all the data pairs into the net asset value prediction mapping structure in index order; S64. Perform format standardization processing on each set of data in the net asset value forecast mapping structure, convert each actual time point into a timestamp representation in a unified format, adjust each forecast value to a set floating point precision value, and write them into the depreciation trend forecast result table in chronological order.

Images