Life cycle evaluation and dynamic updating method for distributed resources

Abstract

The invention relates to the technical field of distributed resource management and optimization, in particular to a distributed resource life cycle evaluation and dynamic updating method, which comprises the following steps: data acquisition: collecting multi-modal data of distributed resources, including resource state data, use behavior data, environment data and historical maintenance data; feature extraction: carrying out feature engineering processing on the multi-modal data, and extracting key indexes such as a resource health state, a degradation rate and environmental sensitivity; building a life cycle evaluation model, building the model by using the time sequence data of the resources based on a multi-task learning framework, and predicting the health state and the residual life of the resources; and optimizing a dynamic updating strategy, and monitoring and dynamically adjusting in real time. The method provided by the invention overcomes the problems of insufficient life cycle dynamics and rigid updating strategy in a traditional method, has the characteristics of high efficiency, intelligence and accuracy, and is suitable for intelligent management and optimization of new energy equipment, industrial assets and other distributed resources.

Description

Technical Field

[0001] The present invention relates to the technical field of distributed resource management and optimization, in particular to a method for life cycle assessment and dynamic update of distributed resources, specifically a method for full life cycle simulation, health status assessment and dynamic update strategy optimization of distributed resources through multimodal data analysis, deep reinforcement learning and digital twin technology, which is applicable to the efficient management and optimization of new energy equipment, industrial assets and other distributed resources. Background Art

[0002] With the wide application of distributed resources, especially in the new energy field and industrial field, their management and optimization have become the core topics for improving resource utilization efficiency and reducing operating costs. However, existing distributed resource management technologies mostly focus on initial planning and static maintenance, and there are the following main problems and deficiencies in the full life cycle management of resources:

[0003] (1) Insufficient life cycle dynamics

[0004] Traditional resource management methods usually lack the ability to dynamically evaluate the full life cycle of resources, and it is difficult to monitor resource status in real time and adjust management strategies according to real-time changes. Since the health status and performance of resources are dynamically affected by multiple factors, management methods relying only on static data analysis appear lagging and inefficient.

[0005] (2) Insufficient consideration of multi-factor coupling

[0006] The aging and performance degradation of distributed resources are comprehensively affected by environmental conditions, usage behaviors and the characteristics of the resources themselves. In existing technologies, the interaction relationship of multimodal data is often ignored, and it is difficult to comprehensively and accurately evaluate the resource status.

[0007] (3) Lack of intelligent update mechanism

[0008] Current resource maintenance and update strategies are usually based on static rules, such as fixed maintenance cycles or simple state threshold judgments. Such rules often fail to adapt to the complex changes in the resource usage environment and status, resulting in insufficient or excessive maintenance, increasing operating costs and resource waste.

[0009] In view of the above problems, the deficiencies of existing technologies include:

[0010] The prediction accuracy of the health status of resources is relatively low;

[0011] The update strategy cannot dynamically adapt to the real-time status;

[0012] Lack of simulation and monitoring of the full life cycle of the resource life cycle.

[0013] Therefore, there is an urgent need for a new method to achieve efficient management and optimization of distributed resources, making the assessment of the resource life cycle more accurate, the dynamic update strategy more intelligent, and the management effect more efficient. Summary of the Invention

[0014] In view of the problems existing in the above or prior art, the present invention is proposed.

[0015] Therefore, the object of the present invention is to provide a method for life cycle assessment and dynamic update of distributed resources, which can solve the problems of inaccurate resource state assessment, lack of dynamic update strategy, and lack of full-cycle simulation in the prior art.

[0016] To solve the above technical problems, the present invention provides the following technical solutions: A method for life cycle assessment and dynamic update of distributed resources, which includes the following steps:

[0017] Collect multi-modal data of distributed resources to obtain the operating state parameters of the resources;

[0018] Extract features from the multi-modal data to establish a life cycle assessment model of the resources;

[0019] Based on the deep reinforcement learning algorithm, optimize the dynamic update strategy of the resources, and combine digital twin technology to monitor and optimize the operating state of the resources in real time.

[0020] As a preferred solution of the method for life cycle assessment and dynamic update of the distributed resources of the present invention, wherein: the data collection is carried out by collecting multi-modal data of distributed resources, including resource state data, usage behavior data, environmental data, and historical maintenance data:

[0021] Resource state data: including information such as power output, capacity attenuation, temperature, and voltage;

[0022] Usage behavior data: including operating duration, load pattern, and adjustment frequency;

[0023] Environmental data: including external environmental conditions such as temperature, humidity, and air quality;

[0024] Historical maintenance data: including the failure records of the resources, the number of repairs, and the corresponding repair effects.

[0025] As a preferred solution of the method for life cycle assessment and dynamic update of the distributed resources of the present invention, wherein: perform feature engineering processing on the multi-modal data to extract key indicators such as resource health index (HI), degradation rate (DR), and environmental sensitivity; the calculation formulas for the resource health index (HI) and degradation rate (DR) are respectively:

[0026] HI = Current performance × 100%

[0027]

[0028] Wherein, the HI represents the health degree of the resource, the DR represents the performance attenuation rate of the resource per unit time, ΔHI is the change amount of the health index, Δt is the change amount of time, and the current performance represents the actual working ability of the resource at the current moment.

[0029] As a preferred solution of the method for life cycle assessment and dynamic update of the distributed resource described in the present invention, wherein: the life cycle assessment model is based on a multi-task learning framework, uses time series modeling, and simultaneously predicts the health state and remaining useful life (RUL) of the resource, and is optimized by the following loss function:

[0030] L = α·L RUL +β·L HI

[0031] Wherein:

[0032] The L RUL represents the remaining useful life prediction error, L HI represents the health state prediction error, α and β are weight parameters of the loss function, and their calculation formula is:

[0033]

[0034] The L HI represents the health state prediction error, RUL i is the predicted remaining useful life, RUL i is the true remaining useful life, and its calculation formula is:

[0035]

[0036] The α and β are weight parameters of the loss function, is the predicted health state, HI t is the true health state, and T is the maximum length of the time step.

[0037] As a preferred solution of the method for life cycle assessment and dynamic update of the distributed resource described in the present invention, wherein: the dynamic update strategy optimization models the resource update problem as a Markov decision process (MDP), optimizes the dynamic update strategy of the resource through a deep reinforcement learning algorithm, generates decisions including maintenance, partial replacement, or complete replacement, and the dynamic update strategy optimization includes:

[0038] Model the dynamic update of resources as a Markov decision process (MDP), where the state s t includes the current health state, remaining life, and environmental conditions, and the action a t includes maintenance, partial replacement, and complete replacement;

[0039] Use a deep reinforcement learning algorithm to optimize the update policy and generate the update action a t = π(s t ; θ);

[0040] Define the reward function R t , taking into account the maintenance cost, resource performance, and replacement benefit comprehensively, with the goal of maximizing the long-term benefit:

[0041]

[0042] where T is the maximum length of the time step, and the γ is the discount factor, which is used to balance the current reward and future benefits, and R t represents the defined reward function.

[0043] As a preferred solution of the method for life cycle assessment and dynamic update of the distributed resources described in the present invention, wherein: the definition of the reward function of the dynamic update policy includes:

[0044] The reward R t includes three parts:

[0045] Maintenance cost: reducing the cost of frequent maintenance;

[0046] Resource performance: the benefit of improving resource performance through maintenance or replacement;

[0047] Replacement benefit: the evaluation of the performance improvement and cost optimization benefits brought by partial or complete replacement;

[0048] Dynamically optimize the reward function through a reinforcement learning algorithm to maximize the resource management benefit.

[0049] As a preferred solution of the method for life cycle assessment and dynamic update of the distributed resources described in the present invention, wherein: for the real-time monitoring and dynamic adjustment, a digital twin of the resource is constructed by combining digital twin technology, the running state of the resource is monitored in real time, and the model parameters and update policy are dynamically adjusted according to the feedback data of the twin; the application of digital twin technology includes the following steps:

[0050] Digital twin modeling: constructing a digital twin of the resource to simulate the running state, degradation behavior, and environmental interaction of the real resource;

[0051] Real-time data synchronization: updating the model parameters of the twin by collecting the running data of the resource in real time;

[0052] Future state prediction: Predict the future operating state of the twin prediction resources and evaluate the effectiveness of the update strategy;

[0053] Strategy adjustment: Dynamically adjust the life cycle assessment model and the reinforcement learning strategy according to the feedback data of the twin;

[0054] The training and verification of the model are carried out in the following ways:

[0055] Dataset construction: Use historical operation data and environmental condition data to construct a multi-modal dataset for training;

[0056] Model training: Optimize the life cycle assessment model through supervised learning and optimize the dynamic update strategy through reinforcement learning;

[0057] Verification and optimization: Verify the update strategy in the simulation environment and gradually optimize the model performance using an online learning mechanism;

[0058] The evaluation metrics of the model include:

[0059] Remaining useful life prediction error;

[0060] Mean squared error of health state prediction;

[0061] Cost-benefit ratio of the dynamic update strategy.

[0062] In a second aspect, an embodiment of the present invention provides a life cycle assessment and dynamic update system for distributed resources, which includes:

[0063] Data unit: Collect multi-modal data of distributed resources, including resource status data, usage behavior data, environmental data, and historical maintenance data;

[0064] Feature unit: Perform feature engineering processing on the multi-modal data, and extract key indicators such as resource health status (HealthIndex, HI), degradation rate (Degradation Rate, DR), and environmental sensitivity;

[0065] Simulation unit: Based on a multi-task learning framework, use the time series data of resources to build a model, and simultaneously predict the health state and remaining useful life (RemainingUseful Life, RUL) of the resources;

[0066] Optimization unit: Model the resource update problem as a Markov decision process (MDP), and optimize the dynamic update strategy of the resources through a deep reinforcement learning algorithm to generate decisions including maintenance, partial replacement, or full replacement;

[0067] Monitoring unit: Construct a digital twin of resources by combining digital twin technology, monitor the operating status of resources in real time, and dynamically adjust model parameters and update strategies according to the feedback data of the twin.

[0068] Thirdly, an embodiment of the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and: when the computer program instructions are executed by the processor, the steps of the method for life cycle assessment and dynamic update of distributed resources as described in the first aspect of the present invention are implemented.

[0069] Fourthly, an embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored, and: when the computer program instructions are executed by the processor, the steps of the method for life cycle assessment and dynamic update of distributed resources as described in the first aspect of the present invention are implemented.

[0070] Advantages of the present invention: Through the collection and analysis of multi-modal data, the complex impacts of resource status, environmental factors, and usage behaviors on resource performance and aging are comprehensively considered, significantly improving the accuracy of resource health status and remaining life prediction. In addition, by introducing deep reinforcement learning to optimize the dynamic update strategy, the maintenance and replacement decisions of resources can respond to changes in the environment and status in real time, with higher intelligence and flexibility. The application of digital twin technology further realizes the full-cycle simulation and monitoring of the resource life cycle, enabling the operating status and update strategy of resources to be evaluated and adjusted in real time, thereby ensuring the efficiency and stability of resource management. Compared with traditional methods, the present invention can significantly reduce the resource operation cost, reduce unnecessary maintenance and replacement, improve resource utilization efficiency, and extend the service life of resources, and is applicable to the intelligent management of new energy equipment, industrial assets, and other distributed resources. Description of the Drawings

[0071] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings. Among them:

[0072] Figure 1 It is a flow block diagram of the overall technical solution of the method for life cycle assessment and dynamic update of distributed resources according to an embodiment of the present invention.

[0073] Figure 2 It is a schematic diagram of the construction framework of the life cycle assessment model according to an embodiment of the present invention.

[0074] Figure 3The main flowchart of the online learning mechanism according to an embodiment of the present invention. Detailed implementation manners

[0075] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will describe the detailed implementation manners of the present invention with reference to the accompanying drawings of the specification.

[0076] In the following description, many specific details are set forth to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0077] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that can be included in at least one implementation manner of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that is mutually exclusive with other embodiments.

[0078] Embodiment 1

[0079] Refer to Figures 1 to 3 , which is the first embodiment of the present invention. This embodiment provides a method for life cycle assessment and dynamic update of distributed resources, which can achieve comprehensive monitoring and intelligent dynamic update of distributed resources through the integration of multi-modal data, the application of deep learning models, and the optimization strategy of reinforcement learning, including:

[0080] S1. Data collection and feature extraction

[0081] The first step of the life cycle assessment and dynamic update method of the present invention is data collection and feature extraction. This process provides the necessary data basis for subsequent model training and optimization. The goal of data collection is to comprehensively understand the operating conditions, usage behaviors, and environmental conditions of resources from multiple dimensions, so as to provide high-quality input for the life cycle assessment model.

[0082] (1) Data collection

[0083] 1) Resource status data: Collect real-time operating status data of resources, including power output, capacity attenuation, temperature, voltage, etc. Specifically, it includes the following items:

[0084] Power output P t : Represents the power output of the resource at the current moment, with the unit of watt (W) or kilowatt (kW).

[0085] Capacity attenuation C t: Represents the attenuation degree of the resource, which can be calculated by comparing the current capacity with the initial capacity. The formula is:

[0086]

[0087] Among them, C t : Represents the attenuation degree of the resource. The current performance represents the actual working ability of the resource at the current moment, and the initial performance is the standard performance value of the resource when it is first used.

[0088] If the current capacity of the resource is lower than the initial capacity, it means that the ability of the resource has declined.

[0089] Temperature T t : The temperature of the resource at the current moment, usually monitored in real time using a sensor. Temperature has an important impact on the operating efficiency and lifespan of the resource. Excessive temperature will accelerate the aging of the device.

[0090] Voltage V t : The voltage data of the resource. The stability of the voltage directly affects the working efficiency and safety of the resource.

[0091] 2) Usage behavior data: Record the usage behavior data of the resource, including running duration, load pattern, and adjustment frequency. Specifically include:

[0092] Running duration L t : Represents the running duration of the resource from the start of use to the current moment, in hours (h) or minutes (min). Long-term operation may lead to the aging of the resource, so monitoring the running duration is crucial.

[0093] Load pattern R t : The load pattern of the resource, usually represented by the load change rate. The load pattern can reveal the usage of the resource under different load conditions.

[0094]

[0095] Specifically, if the load pattern is often at a high load, the degradation rate of the resource is usually faster.

[0096] Adjustment frequency F t : The load adjustment frequency of the resource, indicating whether the workload of the resource changes frequently. Frequent load fluctuations may lead to increased wear of the device.

[0097] 3) Environmental data: Collect environmental factor data that affect resource degradation, such as temperature, humidity, air quality, etc. Specifically include:

[0098] Temperature T env : The environmental temperature. Changes in temperature will have a direct impact on the performance of the resource.

[0099] Humidity H env : Humidity in the air. Excessive humidity may cause rusting of metal parts or a decrease in insulation performance.

[0100] Air Quality AQI env : Air Quality Index. An environment with severe air pollution may accelerate the physical degradation of resources.

[0101] 4) Historical maintenance data: Obtain the historical maintenance records of resources, including information such as fault records, repair times, and repair effects. It mainly includes:

[0102] Fault record F record : Represents the number of times a resource fails and the type of fault. Fault records can help identify potential problems with the resource.

[0103] Number of repairs M t : Records the number of repairs to the resource, including regular maintenance and temporary repairs.

[0104] Furthermore, the criterion for determining a large number of repairs is as follows: If the cumulative number of repairs to the resource within a unit time exceeds 1.5 times the historical average number of repairs, it is determined that there are "a large number of repairs".

[0105] Repair effect E t : Calculate the repair effect through the capacity recovery rate (RecoveryRate, RR), and quantitatively evaluate the performance recovery of the resource after repair. The specific calculation formula is as follows:

[0106]

[0107] Among them, C repair Represents the capacity of the resource after repair, C pre-repair Represents the capacity of the resource before repair, C initial Represents the initial capacity of the resource.

[0108] Furthermore, when the capacity recovery rate after repair is greater than or equal to 80%, the repair effect is evaluated as "good"; when the recovery rate is between 50% - 80%, it is evaluated as "medium"; when it is less than 50%, it is evaluated as "poor".

[0109] It should be noted that the "number of repairs" adopts a dynamic threshold method based on the resource health status (Health Index, HI), and sets the determination criterion by combining historical data and resource categories. Specifically, if the cumulative number of repairs to the resource within a unit time (such as one operation cycle or one year) exceeds 1.5 times its historical average number of repairs, it is determined that there are "a large number of repairs"; this interval setting is based on the historical maintenance records of different types of resources, and a reasonable determination range is obtained through data analysis.

[0110] Specifically, the maintenance effect is mainly evaluated based on the degree of capacity attenuation recovery of the resource. The recovery rate (RR) is introduced as a key evaluation indicator. When the capacity recovery rate after maintenance is greater than or equal to 80%, the maintenance effect is evaluated as "good"; when the recovery rate is between 50% - 80%, it is evaluated as "medium"; and when it is less than 50%, it is evaluated as "poor".

[0111] (2) Feature extraction

[0112] After sufficient data is collected, key indicators are extracted through feature engineering, which will be used as inputs for life cycle assessment and dynamic update. The main ones include the following:

[0113] 1) Resource health index (HI): This indicator is used to quantify the current health status of the resource, and the calculation formula is:

[0114]

[0115] Among them, the current performance represents the actual working ability of the resource at the current moment, and the initial performance is the standard performance value when the resource is first used. The higher the HI value, the better the health status of the resource.

[0116] 2) Degradation rate (DR): This indicator is used to measure the speed of resource degradation, and the calculation formula is:

[0117]

[0118] Among them, ΔHI is the change in the health index, and Δt is the change in time. The degradation rate can reflect the acceleration degree of resource aging, which is helpful for predicting the remaining life of the resource and optimizing the maintenance plan.

[0119] 3) Environmental sensitivity: Through multiple regression analysis, the impact of environmental factors on resource degradation is evaluated. The specific calculation formula is:

[0120] Environmental sensitivity = β 0 + β 1 · T env + β 2 · H env + β 3 · AQI env

[0121] Among them, β 0 is the constant term, and β 1 , β 2 , β 3 are the regression coefficients of environmental factors. Environmental sensitivity can reflect the relative impact degree of environmental changes on resource degradation, AQIenv represents the air quality index, H env represents the humidity in the air, T env represents the ambient temperature.

[0122] It should be noted that by extracting and analyzing these features, the system can obtain important information about the health status and degradation rate of resources, providing data support for subsequent life cycle assessment and dynamic update strategies.

[0123] S2. Construction of life cycle assessment model:

[0124] After data collection and feature extraction are completed, the next step is to construct a life cycle assessment model, aiming to accurately predict the health status and remaining life of resources, so as to provide a scientific basis for resource maintenance and update. The present invention adopts a time series modeling method based on multi-task learning, which can consider multiple factors simultaneously and predict different indicators of resources.

[0125] (1) Model selection

[0126] To solve the multi-task prediction problem, the present invention selects a multi-task learning (Multi-task Learning, MTL) framework and combines it with a deep learning (Transformer) model to process time series data. The specific framework design is as Figure 2 shown. It should be noted that the deep learning (Transformer) model has strong advantages in processing long sequence data, can fully mine the long-term dependence relationship of historical data, and predict the future health status and remaining life. Multi-task learning can perform multiple prediction tasks simultaneously, improving the accuracy of the model.

[0127] Construction of the deep learning (Transformer) model:

[0128] ① Input Embedding: Convert multi-modal time series data into embedding vectors to meet the input requirements of the Transformer model.

[0129] ② Positional Encoding: Add positional encoding to retain the time series information of each data point in the sequence.

[0130] ③ Encoder: Composed of multiple layers of self-attention mechanisms and feed-forward neural networks, used to capture long-term dependence relationships and complex patterns in the input data.

[0131] ④ Feature Representation: The high-dimensional feature vectors extracted by the encoder are used as the input of multi-task learning.

[0132] In summary, the multi-task learning framework jointly optimizes multiple related tasks based on a shared feature representation, which can not only improve the overall prediction performance of the model, but also significantly enhance the generalization ability and data utilization efficiency of the model. First, the multi-task learning framework can utilize the correlation and shared information between different tasks to complement each other's feature representations, thereby reducing interference between tasks and achieving the effect of collaborative optimization. Second, through the shared feature extraction layer, the framework can more efficiently utilize limited data resources, especially showing more prominent advantages in scenarios with scarce data. In addition, the multi-task learning framework can also capture more comprehensive data features, enabling the model to have stronger adaptability in different application scenarios, which helps to improve its prediction accuracy and robustness in diverse real-world environments.

[0133] (2) Model Input and Output

[0134] 1) Input: Multimodal time series data of resources, including status data of resources, usage behavior data, environmental data, etc. These data will be converted into a format suitable for model training and fed into the multi-task learning model as input. Specifically, it includes:

[0135] ① Resource status data: Such as power, temperature, voltage, etc., which reflect the real-time operating performance of the resources.

[0136] ② Usage behavior data: Such as running duration, load pattern, etc., which characterize the usage situation and behavior pattern of the resources.

[0137] ③ Environmental data: Such as temperature, humidity, etc., which describe the external influence conditions of the environment where the resources are located.

[0138] It should be noted that multi-task learning (MTL) is a framework design concept used to achieve collaborative optimization of multiple related tasks through a shared feature representation; the multi-task learning model is a method or instantiated model for specifically implementing this framework; in the present invention, the multi-task learning framework is the overall guiding ideology for model design, and the specifically implemented multi-task learning model combines a deep learning (Transformer) structure to process time series data and achieve joint prediction of the health index (HI) and remaining useful life (RUL) of resources; therefore, it can be considered that the multi-task learning model of the present invention is a combination of the multi-task learning framework and a specific modeling technology (Transformer model).

[0139] Furthermore, the multi-task learning model incorporates a deep learning (Transformer) structure in its framework. The deep learning (Transformer) model extracts features from time series data through the self-attention mechanism and generates high-dimensional feature representations as the input for multi-task learning. Subsequently, collaborative predictions of the health state (HI) and remaining useful life (RUL) are achieved through a shared feature layer. The specific process is as follows:

[0140] ① Data input and embedding: Embed the multi-modal time series data into an input vector suitable for model calculation, and add position information to preserve the time dependence of the data.

[0141] ② Feature extraction and representation: Utilize the multi-layer self-attention mechanism of the deep learning (Transformer) model to extract long-term dependence features and form a shared high-dimensional feature representation.

[0142] ③ Task branch output: Generate prediction results for the health state (HI) and remaining useful life (RUL) through the task branch module of the multi-task learning framework.

[0143] 2) Output:

[0144] Remaining useful life prediction (RUL): The model predicts the remaining useful life (RUL) of the resource based on the input data. The prediction of the remaining useful life (RUL) estimates the time when the resource may fail in the future, helping maintenance personnel take measures in advance. The calculation formula is:

[0145] RUL = f(X t ; θ

[0146] where X t is the current input data, θ is the parameter of the model, and RUL is the remaining useful life.

[0147] Health state update (HI): The model also outputs the predicted health state of the resource, reflecting the current health condition of the resource. The calculation formula is:

[0148] HI t = g(X t ; θ

[0149] where g is the model function for predicting the health state, X t is the current input data, and θ is the parameter of the model.

[0150] Furthermore, the model predicts the remaining useful life of the resource based on the input data, that is, the time when it may fail in the future, thus helping maintenance personnel take measures in advance. The prediction results can be divided into different safety levels according to the following intervals:

[0151] Safe interval (RUL > 50%): The resource is operating well, and no special measures are required. Only routine monitoring is needed. For example, when the design life of a wind power generation device is 20 years and the predicted remaining life is more than 12 years, no additional maintenance operations are required.

[0152] Warning interval (30% ≤ RUL ≤ 50%): The operation status of the resource needs attention, and preventive maintenance is recommended. For example, when the design life of an industrial device is 10 years and the predicted remaining life is 3 - 5 years, a comprehensive inspection should be arranged and components should be stocked.

[0153] Risk interval (RUL < 30%): The operation status of the resource is at high risk, and maintenance or replacement measures need to be taken immediately. For example, when the design life of a photovoltaic cell is 25 years and the predicted remaining life is only 6 years, a replacement plan should be quickly formulated to avoid high costs caused by sudden failures.

[0154] Therefore, through the output of this model, the RUL values in different intervals can clearly correspond to actual resource management measures, providing scientific support for the health management and optimal allocation of resources.

[0155] (3) Loss function

[0156] To optimize the training of the multi - task learning model, the present invention defines a composite loss function that performs a weighted sum of the remaining useful life prediction error (L RUL ) and the health state prediction error (L HI ). The expression of the loss function is:

[0157] L = α·L RUL + β·L HI

[0158] where α and β are weighting coefficients that respectively adjust the importance of the remaining useful life prediction and the health state prediction in the loss function. L RUL represents the life prediction error, and L HI represents the health state prediction error.

[0159] The formula for the remaining useful life prediction error L RUL is:

[0160]

[0161] where RUL i is the predicted remaining useful life, and RUL i is the true remaining useful life.

[0162] The formula for the health state prediction error L HI is:

[0163]

[0164] Among them, is the predicted health state, HI t is the actual health state.

[0165] Specifically, by optimizing the loss function, the model can continuously improve the prediction accuracy, providing a more reliable basis for the life cycle assessment of resources.

[0166] S3. Optimization of Dynamic Update Strategy

[0167] One of the core innovations of the present invention is the optimization of the dynamic update strategy, specifically by introducing a reinforcement learning algorithm to achieve intelligent dynamic maintenance and replacement of resources. Different from the traditional static rule-based update method, the present invention models through the Markov decision process (MDP) and uses deep reinforcement learning (DRL) to automatically adjust the maintenance and update strategy to ensure that the resources are always in the best state during their life cycle.

[0168] (1) Problem Modeling

[0169] In the present invention, the dynamic update problem is modeled as a Markov decision process (MDP). The Markov decision process (MDP) provides a mathematical framework for describing how an agent makes decisions in an environment to maximize the long-term cumulative reward. In the context of the present invention, the Markov decision process (MDP) model is as follows:

[0170] 1) State set s t : Represents the current state of the resource, including the resource health state, remaining life, and environmental conditions. Specifically, the state can be represented by a vector:

[0171] s t = [HI t , RUL t , T env , H env , AQI env

[0172] Among them, HI t is the health state of the resource at time t, RUL t is the remaining life, T env , H env , AQI env are environmental factors.

[0173] 2) Action set a t : Represents the action taken by the agent at time t, mainly including three options:

[0174] Maintenance: Regular or emergency maintenance of the resource.

[0175] ​Partial replacement: When a part of the resource degrades, perform a local replacement.

[0176] Full replacement: When the resource cannot be repaired or continued to be used, perform a full replacement.

[0177] Furthermore, in the present invention, the partial replacement requirement of the resource is determined based on the dynamic monitoring of the resource health state (HI) and the prediction result of the remaining useful life (RUL);

[0178] Specifically, the determination rules are as follows:

[0179] Health state (HI) determination rule: When the resource health state (HI) drops to between 50% - 70%, it is determined that the resource has moderate degradation. At this time, it is recommended to perform a local replacement to avoid further degradation leading to a decline in overall performance; when the health state is lower than 50% but still higher than 30%, the resource enters a severe degradation state. Local replacement is still feasible, but the economy and operability of the replacement need to be evaluated.

[0180] Remaining useful life (RUL) auxiliary judgment: If the predicted remaining useful life (RUL) is still relatively long (for example, higher than 30% of the expected total life), it indicates that the resource as a whole still has a certain value of use. At this time, partial replacement is preferred to extend the service life; if the remaining useful life (RUL) is too short (lower than 30% of the total life), even if the HI is still within the range of partial replacement, a full replacement may be required to reduce the long-term maintenance cost.

[0181] 3) Reward R t : The reward function is used to quantify the benefit after the agent takes a certain action. In the present invention, the reward function comprehensively considers the maintenance cost, resource performance, and replacement benefit, and is specifically defined as:

[0182] R t = λ 1 · Performance improvement - λ 2 · Maintenance cost - λ 3 · Replacement cost

[0183] Among them, λ 1 ,λ 2 ,λ 3 is an adjustment coefficient used to balance the influence of various factors. The performance improvement refers to the improvement of the resource health state, and the maintenance cost and replacement cost respectively represent the costs required to perform maintenance and replacement operations.

[0184] (2) Reinforcement learning framework

[0185] To optimize the dynamic update strategy of resources, the present invention uses the Deep Reinforcement Learning (DRL) algorithm, specifically the Deep Deterministic Policy Gradient (DDPG) algorithm. The Deep Deterministic Policy Gradient (DDPG) is a reinforcement learning algorithm based on the actor-critic architecture that can effectively learn the optimal policy in a continuous action space.

[0186] Policy Network: The policy network generates an action a based on the current state information s t representing the best update strategy that the system should adopt. Its calculation formula is: t a

[0187] a t = π(s t ; θ)

[0188] where π is the policy function and θ is the parameter of the network. Through training, the policy network can output the optimal maintenance, partial replacement, or full replacement actions, and s t represents the current state information.

[0189] Value Network: The value network is used to evaluate the value of taking a certain action in the current state. It gives the expected cumulative return Q(s t , a t ). Its calculation formula is:

[0190]

[0191] where γ is the discount factor, represents the expected value operation, and s t represents the current state information. The goal of the value network is to minimize the prediction error of the Q value by optimizing the network parameter θ, so as to approximate the optimal value function.

[0192] (3) Optimization Objective

[0193] In reinforcement learning, the goal of the agent is to maximize the long-term cumulative reward. Therefore, the optimization objective is to maximize the following objective function through the training of the policy network and the value network:

[0194]

[0195] where T is the maximum length of the time step, and γ is the discount factor, representing the attenuation degree of future rewards. The reinforcement learning based on the Deep Deterministic Policy Gradient (DDPG) policy obtains the optimal agent by training the policy network and the value network. This agent can judge the lifespan of the resource based on the multi-modal temporal state of the distributed resource and make maintenance command decisions to achieve maximizing performance and minimizing cost.

[0196] S4. Digital Twin and Real-time Monitoring

[0197] To better support the dynamic assessment and update of resources, the present invention introduces digital twin technology. Digital twin can provide real-time data feedback by simulating the state changes of physical resources in real time and adjust the life cycle management strategy of resources according to the real-time feedback. The specific implementation steps are as follows:

[0198] (1) Digital Twin Modeling

[0199] Digital twin technology realizes the real-time simulation of physical resources by creating virtual copies of resources. Each physical resource has a corresponding digital twin, which can reflect the operating state, degradation behavior of the resource, and its interaction with the environment. The process of constructing a digital twin includes the following aspects:

[0200] Resource Modeling: According to the characteristics of physical resources, construct a mathematical model of the resources, including dynamic change models such as power output, capacity attenuation, and temperature.

[0201] Environment Modeling: Model the environmental factors where the resources are located, including temperature, humidity, air quality, etc., and associate them with the performance and degradation rate of the resources.

[0202] Data Synchronization: Synchronize the data of physical resources in real time, and transmit the collected resource status, environmental data, etc. to the digital twin system through sensors to ensure the accuracy of the digital twin.

[0203] Therefore, through the digital twin, the health status of the resources can be monitored in real time and its future status can be predicted. The digital twin can also help to verify the accuracy of the life cycle assessment model in real time and ensure the reliability of the model in practical applications.

[0204] (2) Real-time Monitoring and Feedback

[0205] Digital twin can not only monitor the resource status in real time, but also provide feedback and adjustment to the life cycle assessment model based on real-time data. Through the real-time monitoring of the digital twin, problems such as performance degradation and faults of resources can be detected immediately, and then the resource update strategy can be adjusted dynamically.

[0206] The specific implementation methods include:

[0207] Real-time Data Update: Through sensors and monitoring systems, obtain the real-time operation data of resources, such as power output, temperature, voltage, etc., and feedback these data to the digital twin in real time.

[0208] Twin Prediction: Digital twins utilize real-time data to predict the future health status and remaining lifespan of resources. These prediction results not only provide a basis for lifecycle assessment but also offer data support for the adjustment of dynamic update strategies.

[0209] Strategy Adjustment: Based on the feedback information from digital twins, the update strategy is adjusted in real time. If a resource is about to enter a faulty state, actions such as repair or replacement will be triggered; if the resource is still in good condition, the update may be delayed or the maintenance frequency may be optimized.

[0210] Therefore, through digital twin technology, the health management of resources becomes more intelligent and real-time, avoiding the lag and inaccuracy existing in traditional methods.

[0211] S5 Model Training and Validation

[0212] In the method of overall lifecycle assessment and dynamic update, model training and validation are crucial steps to ensure system performance. In this invention, historical operation data and simulation data are used for model training, multiple evaluation metrics are adopted to validate the model, and it is gradually optimized through an online learning mechanism.

[0213] (1) Training Objectives

[0214] The objective of the training phase is to optimize the lifecycle assessment model and dynamic update strategy through historical data and simulation data. During the training process, the model needs to continuously adjust parameters to improve prediction accuracy and strategy optimization effect. Specific objectives include:

[0215] Remaining Useful Life Prediction Error (RMSE): The accuracy of the model is measured by calculating the Root Mean Square Error (RMSE) between the predicted remaining useful life and the actual remaining useful life.

[0216]

[0217] Among them, RUL i is the predicted remaining useful life, and RUL i is the actual remaining useful life.

[0218] Cost-benefit Ratio of Update Strategy: Evaluate the ratio between the long-term benefits and costs generated by the model according to different strategies to ensure the economy and effectiveness of the update strategy.

[0219] (2) Validation and Optimization

[0220] In summary, in order to verify the effectiveness and robustness of the model, the present invention uses simulation data to verify the model multiple times. During the training process, the simulation data not only helps optimize the life cycle assessment model, but also is used to test the performance of different update strategies. By introducing an online learning mechanism, the model can be gradually optimized during the actual operation process, thereby improving the prediction accuracy and strategy effect.

[0221] Furthermore, this embodiment also provides a life cycle assessment and dynamic update system for distributed resources, which includes:

[0222] Data unit: Collect multi-modal data of distributed resources, including resource status data, usage behavior data, environmental data, and historical maintenance data.

[0223] Feature unit: Perform feature engineering processing on the multi-modal data, and extract key indicators such as resource health index (HI), degradation rate (DR), and environmental sensitivity.

[0224] Simulation unit: Based on the multi-task learning framework, use the time series data of the resource to build a model, and simultaneously predict the health status and remaining useful life (RUL) of the resource.

[0225] Optimization unit: Model the resource update problem as a Markov decision process (MDP), and optimize the dynamic update strategy of the resource through a deep reinforcement learning algorithm to generate decisions including maintenance, partial replacement, or complete replacement.

[0226] Monitoring unit: Combine digital twin technology to build a digital twin of the resource, monitor the operating status of the resource in real time, and dynamically adjust the model parameters and update strategy according to the feedback data of the twin.

[0227] This embodiment also provides a computer device applicable to the life cycle assessment and dynamic update method of distributed resources, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the life cycle assessment and dynamic update method of distributed resources proposed in the above embodiment.

[0228] This computer device can be a terminal, and this computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of this computer device is used to provide computing and control capabilities.

[0229] It should be noted that the memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a carrier network, NFC (Near Field Communication) or other technologies.

[0230] It should also be noted that the display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs or touchpads provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0231] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the method for realizing the life cycle assessment and dynamic update of distributed resources as proposed in the above embodiment.

[0232] It should be understood that each part of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiment, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following technologies well known in the art can be used: discrete logic circuits with logic gate circuits for implementing logical functions on data signals, application specific integrated circuits with appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0233] In summary, through the detailed description of the above specific embodiments, the present invention provides an efficient and intelligent method for life cycle assessment and dynamic update of distributed resources. This method comprehensively utilizes multi-modal data collection and feature extraction, a life cycle assessment model based on multi-task learning, a dynamic update strategy optimized by deep reinforcement learning, and real-time monitoring and feedback implemented by digital twin technology, effectively solving the problems of insufficient evaluation accuracy, lack of dynamics in the update strategy and insufficient full-cycle simulation ability in the prior art.

[0234] Embodiment 2

[0235] Refer to Figures 2 to 3 , which is the second embodiment of the present invention. This embodiment provides a method for life cycle assessment and dynamic update of distributed resources. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments; it includes,

[0236] 1) Simulation verification

[0237] Run the simulation environment under different scenarios, including normal operation, high temperature and high humidity, low temperature and low humidity, etc., to verify the prediction accuracy of the model for the health index (HI) and remaining useful life (RUL), as well as the benefit-cost optimization effect of the dynamic update strategy.

[0238] Table 1 in the reference shows the main evaluation indicators under different scenarios in the simulation results:

[0239] Table 1

[0240]

[0241]

[0242] 2) Online learning mechanism

[0243] The model can not only be initially trained with offline data, but also can perform online learning through real-time feedback in actual applications. The core of the online learning mechanism lies in the dynamic collection of real-time data and the adaptive optimization of the model. Through this mechanism, the model can continuously adjust its parameters over time to cope with new resource behavior patterns and environmental changes, ensuring the accuracy of prediction and the optimality of the update strategy.

[0244] The implementation steps of the online learning mechanism are as Figure 3 shown, mainly including the following modules:

[0245] Real-time data collection module: Dynamically collect the operation data of resources through the sensor network, including the health index (HI), remaining useful life (RUL), usage behavior (such as running duration, load pattern), and environmental data (such as temperature, humidity, air quality).

[0246] Digital twin feedback module: Input the real-time data into the digital twin, compare the model prediction value with the actual value, generate error feedback, and identify the deficiencies of the model.

[0247] Model optimization module: According to the error feedback results, online adjust the weight parameters in the multi-task learning framework and the reward function settings in the reinforcement learning strategy, so as to optimize the prediction ability of the model and the dynamic update strategy.

[0248] Online learning update module: Dynamically apply the optimized model parameters to the life cycle assessment and update strategy decision-making to achieve real-time adjustment.

[0249] Therefore, through the above steps, the online learning mechanism ensures the high adaptability and continuous optimization ability of the model, can respond to the complexity of environmental changes and resource behavior in real time, and improves the prediction accuracy and the economy of the update strategy.

[0250] Furthermore, the technical solution of the present invention can not only accurately predict the health status and remaining life of resources, but also adjust the maintenance and update strategies according to real-time environmental changes and resource status, significantly improving the utilization efficiency and economy of resources. By introducing an online learning mechanism, the model can be gradually optimized during the actual operation process, thereby improving the prediction accuracy and the effect of the strategy. In addition, the application of digital twin technology makes the full life cycle management of resources more real-time and intelligent, providing strong support for the optimization management of complex resource systems.

[0251] Table 2

[0252]

[0253]

[0254] As can be seen from Table 2, in the photovoltaic power generation system, the present invention uses real-time environmental data (such as high temperature and high humidity environment) to adjust the parameters of the health status prediction model, and optimizes the maintenance plan according to the RUL prediction result, effectively reducing the equipment failure rate; while in the traditional method, due to only relying on static maintenance cycles and empirical thresholds, it is difficult to respond to environmental changes in a timely manner, which may lead to resource waste or unexpected failures.

[0255] In summary, the present invention is applicable to the intelligent management and optimization of new energy equipment (such as wind power equipment, solar cells), industrial assets (such as manufacturing equipment, robot systems), and other distributed resources, and has broad application prospects and significant economic benefits.

[0256] It should be noted that the above embodiments are only used to illustrate the technical solution of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solution of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solution of the present invention, and it should be covered by the scope of the claims of the present invention.

Claims

1. A method for life cycle assessment and dynamic update of distributed resources, characterized in that: The following steps are involved: Collect multimodal data of distributed resources and obtain operating status parameters of resources; Extracting features from the multimodal data and establishing a life cycle assessment model for resources; Based on the deep reinforcement learning algorithm, the dynamic update strategy of resources is optimized, and the digital twin technology is combined to monitor and optimize the operating status of the resources in real time.

2. The method for life cycle assessment and dynamic update of distributed resources as claimed in claim 1, characterized in that: The data collection collects multimodal data of distributed resources, including resource status data, usage behavior data, environmental data and historical maintenance data: Resource status data: including information such as power output, capacity decay, temperature and voltage; Usage behavior data: including running time, load pattern and adjustment frequency; Environmental data: including external environmental conditions such as temperature, humidity and air quality; Historical maintenance data: including resource failure records, maintenance times, and corresponding maintenance results.

3. The method for life cycle assessment and dynamic update of distributed resources as claimed in claim 2, characterized in that: The multimodal data is subjected to feature engineering processing to extract key indicators such as resource health status, degradation rate and environmental sensitivity; the calculation formulas for the resource health status and degradation rate are respectively: HI = current performance × 100% Among them, the HI represents the health of the resource, the DR represents the performance decay rate of the resource in unit time, ΔHI is the change of the health index, Δt is the time change, and the current performance represents the actual working capacity of the resource at the current moment.

4. The method for life cycle assessment and dynamic update of distributed resources as claimed in claim 3, characterized in that: The life cycle assessment model is based on a multi-task learning framework and uses time series modeling to simultaneously predict the health status and remaining life of resources, and is optimized through the following loss function: L=α*L RUL +β*L HI in: The L RUL Remaining life prediction error, L HI represents the health status prediction error, α and β are the weight parameters of the loss function, and the calculation formula is: The L HI Represents the health status prediction error, RUL i is the predicted remaining useful life, RUL i is the actual remaining life, and its calculation formula is: The α and β are weight parameters of the loss function, is the predicted health status, HI t is the true health state, and T is the maximum length of the time step.

5. The method for life cycle assessment and dynamic update of distributed resources as claimed in claim 4, characterized in that: The dynamic update strategy optimization models the resource update problem as a Markov decision process, optimizes the dynamic update strategy of resources through a deep reinforcement learning algorithm, and generates decisions including maintenance, partial replacement or complete replacement. The dynamic update strategy optimization includes: The dynamic resource update is modeled as a Markov decision process, where the state s t : including current health status, remaining life and environmental conditions, action a t : Includes maintenance, partial replacement and complete replacement; Use deep reinforcement learning algorithm to optimize the update strategy and generate update action a t =π(s t ;θ); Define the reward function R t , taking into account maintenance costs, resource performance and replacement benefits, the goal is to maximize long-term benefits: Where T is the maximum length of the time step, γ is the discount factor used to balance the current reward and future benefits, and R t Represents the definition of reward function.

6. The method for life cycle assessment and dynamic update of distributed resources as claimed in claim 5, characterized in that: The reward function definition of the dynamic update strategy includes: Reward R t It consists of three parts: Maintenance cost: reduce the cost of frequent maintenance; Resource performance: the benefits of improving resource performance through maintenance or replacement; Replacement benefits: Evaluation of performance improvement and cost optimization benefits brought by partial or complete replacement; The reward function is dynamically optimized through reinforcement learning algorithm to maximize the benefits of resource management.

7. The method for life cycle assessment and dynamic update of distributed resources as claimed in claim 6, characterized in that: The real-time monitoring and dynamic adjustment are combined with digital twin technology to build a digital twin of resources, monitor the operating status of resources in real time, and dynamically adjust model parameters and update strategies based on twin feedback data; The application of digital twin technology includes the following steps: Digital twin modeling: building digital twins of resources to simulate the operating status, degradation behavior, and environmental interactions of real resources; Real-time data synchronization: updating the twin model parameters by collecting resource operation data in real time; Future state prediction: using the twin to predict the future operating state of resources and evaluate the effectiveness of update strategies; Strategy adjustment: dynamically adjusting the life cycle assessment model and reinforcement learning strategy based on the feedback data of the twin; The training and validation of the model are carried out in the following ways: Dataset construction: Use historical operation data and environmental condition data to build a multimodal dataset for training; Model training: Optimize the life cycle assessment model through supervised learning and optimize the dynamic update strategy through reinforcement learning; Verification and optimization: Verify the update strategy in a simulation environment and use online learning mechanisms to gradually optimize model performance; The evaluation indicators of the model include: Remaining life prediction error; Mean square error of health status prediction; Cost-effectiveness ratio of dynamic update strategy.

8. A distributed resource life cycle assessment and dynamic update system using the method according to any one of claims 1 to 7, characterized in that: Also includes, Data unit: collects multimodal data of distributed resources, including resource status data, usage behavior data, environmental data, and historical maintenance data; Feature unit: performing feature engineering processing on the multimodal data to extract key indicators such as resource health status, degradation rate and environmental sensitivity; Simulation unit: Based on a multi-task learning framework, it uses the time series data of resources to build models and simultaneously predict the health status and remaining life of resources; Optimization unit: Model the resource update problem as a Markov decision process, optimize the dynamic update strategy of resources through deep reinforcement learning algorithm, and generate decisions including maintenance, partial replacement or complete replacement; Monitoring unit: Combine digital twin technology to build digital twins of resources, monitor the operating status of resources in real time, and dynamically adjust model parameters and update strategies based on twin feedback data.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the distributed resource lifecycle assessment and dynamic update method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the distributed resource lifecycle assessment and dynamic update method described in any one of claims 1 to 7 are implemented.

Images