Specific model construction method and system based on asset digital intelligence management platform

Abstract

The invention relates to the technical field of asset management, discloses a specific model construction method and system based on an asset digital intelligent management platform, and aims at solving the problems that a traditional asset management mode is low in efficiency, inaccurate in prediction, incomplete in evaluation, not intelligent in decision making and the like. The method comprises the step of constructing an asset performance prediction model, an asset health degree evaluation model and an asset management decision model. The performance prediction model predicts future performance indexes based on asset configuration and environmental condition data; the health degree evaluation model comprehensively considers multiple factors to evaluate the current health state of the assets; the management decision model models a management process into a Markov decision process, and optimizes decision actions through a deep reinforcement learning algorithm. According to the system, effective integration and cooperative work of the models are realized, the intelligent level and decision-making efficiency of asset management are improved, and a brand-new and intelligent solution is provided for asset management.

Description

Technical Field

[0001] The present invention relates to the field of asset management technology, and specifically to a method and system for constructing a specific model based on an asset digital management platform. Background Art

[0002] With the rapid development of information technology and the deepening of digital transformation, the field of asset management is facing unprecedented challenges and opportunities. Traditional asset management methods often rely on regular inspections and post-maintenance, which is not only inefficient, but also difficult to cope with the complexity of asset performance changes over time and environmental conditions. Especially in key areas such as large-scale industrial equipment, infrastructure, and intelligent transportation systems, the efficient and reliable operation of assets is directly related to production safety, service quality, and economic benefits.

[0003] In recent years, with the rise of big data, artificial intelligence and the Internet of Things, digital asset management platforms have gradually become an important means to improve asset management. By integrating sensors, data acquisition systems, cloud computing and advanced analysis algorithms, digital asset management platforms can monitor asset status in real time, predict performance changes, evaluate health, and provide a scientific basis for asset management decisions. However, how to effectively build and use these models to achieve intelligent and automated asset management is still a difficult problem in current research and practice. Existing asset management models and methods often have the following problems: Inaccurate performance prediction: Traditional performance prediction models are often based on simple statistical methods or empirical formulas, which make it difficult to accurately capture the complex relationship between asset performance and environmental conditions and configuration information, resulting in large deviations in prediction results.

[0004] Incomplete health assessment: The health status of assets is affected by many factors, including historical maintenance records, current performance indicators, real-time environmental conditions, etc. Existing health assessment models often only consider some factors and lack a comprehensive assessment system, resulting in inaccurate assessment results.

[0005] Unintelligent decision-making process: Asset management decisions involve multiple dimensions and complex factors. Traditional decision-making methods often rely on manual experience and rules, which are difficult to adapt to dynamically changing environments and asset conditions, resulting in low decision-making efficiency and low decision-making quality.

[0006] Low model integration: Performance prediction, health assessment and asset management decision-making are three key links in asset management, but existing models and methods often exist in isolation and lack effective integration and coordination, resulting in poor overall management results. Summary of the invention

[0007] The purpose of the present invention is to provide a specific model construction method and system based on an asset digital management platform to solve the problems raised in the above background technology.

[0008] To achieve the above object, the present invention provides the following technical solution: a specific model construction method based on an asset digital management platform, the method comprising: Step 1: Construct an asset performance prediction model to predict the performance indicators of assets in the future. The input data includes: asset configuration information and environmental condition data. The output data is: asset predicted performance indicator data. The model training data includes: historical asset configuration information, historical environmental condition data and corresponding actual performance indicator data. Step 2: Construct an asset health assessment model to assess the current health status of the asset; its input data includes: asset configuration information, historical maintenance records, current performance indicator data and real-time environmental condition data; its output data is: asset current health status indicators; model training data includes: historical asset configuration information, maintenance records, performance indicator data, historical environmental condition data and corresponding asset health status assessment results; Step 3: Construct an asset management decision model and model the asset management process as a Markov decision process, which is used to make asset management decisions based on the output of the asset performance prediction model and the output of the asset health assessment model. The specific definitions include: State space: defined as the combination of an asset’s real-time performance indicator data, configuration information, and environmental condition data; Action space: defined as the set of maintenance, adjustment, and optimization operations that can be performed; Reward function: Calculate rewards or penalties based on performance deviation and health deviation, give positive rewards to operations that reduce deviations, and give penalties to operations that increase deviations; Output strategy: defined as the method of selecting the best action in the current state, which is continuously optimized by deep reinforcement learning algorithm; Step 4: Perform real-time management based on the trained asset performance prediction model, asset health assessment model, and asset management decision model, and automatically execute the decision results or transmit them to the management personnel for confirmation and execution through the management system.

[0009] Preferably, the performance indicators include the operating efficiency, energy consumption, temperature, vibration, pressure, flow and failure rate of the assets; the environmental parameters include ambient temperature, humidity, air pressure and electromagnetic interference; the configuration information includes the model, specification, manufacturer and service life of the assets; the maintenance records include maintenance time, maintenance type, maintenance cost and maintenance effect.

[0010] Preferably, the executable maintenance, adjustment and optimization operation set includes: adjusting operating parameters, replacing parts, cleaning equipment, lubricating equipment, calibrating equipment, upgrading software, adjusting data collection frequency and accuracy, and scheduling preventive maintenance.

[0011] Preferably, the implementation method of the reward function includes: Obtain real-time performance indicator data of the asset; compare the real-time performance indicator data with the predicted performance indicator data and calculate the performance deviation; Obtain real-time health assessment data of assets; compare the current health status indicators output by the asset health assessment model with the preset health standards or historical health data, and calculate the health deviation; Rewards or penalties are calculated based on performance deviation and health deviation. The specific formula is: in, A reward term representing a performance deviation, Rewards representing health deviations, and are the weight coefficients of the performance deviation and health deviation reward items respectively, and .

[0012] Preferably, the asset performance prediction model is constructed using a long short-term memory network LSTM algorithm, and its network structure specifically includes: Input layer: used to receive training data of the model; Hidden layer: contains at least one LSTM hidden layer. Each LSTM hidden layer contains several LSTM units. Each LSTM unit is equipped with a forget gate, input gate and output gate to control the flow of information and state update. Output layer: Receives the output of the hidden layer and maps the output of the LSTM network to the predicted performance indicator value through a fully connected layer.

[0013] Preferably, the step of training the asset performance prediction model includes: Step 101: Initialize the weights and bias items of the LSTM network using a random initialization method; Step 102: Select a loss function to measure the difference between the model prediction value and the true value, and use the mean square error MSE as the loss function; Step 103: Use the back propagation algorithm and the Adam optimization algorithm to calculate the gradient according to the loss function, update the weights and bias items of the LSTM network, and iteratively train the network until the loss function converges; Step 104: Save the trained LSTM network model as the asset performance prediction model in the asset digital management platform.

[0014] Preferably, the asset health assessment model is constructed using a support vector machine (SVM) algorithm, and its training steps include: Step 201: Use historical asset configuration information, maintenance records, performance indicator data, and historical environmental condition data in the training data as input features, use corresponding asset health status assessment results as labels, and construct a training sample set for the support vector machine algorithm; Step 202: using the support vector machine algorithm to train the training sample set, by finding the optimal hyperplane to maximize the interval between different health status categories, while considering the soft interval to accommodate some misclassification situations; Step 203: During the training process, the penalty parameter C and kernel function parameters of the support vector machine algorithm are optimized by cross-validation method.

[0015] Preferably, the asset management decision model is constructed using a deep deterministic policy gradient (DDPG) algorithm, and its training steps include: S1. Initialize the Actor network and the Critic network: The Actor network is used to generate actions, and the Critic network is used to evaluate the value of actions; S2. Initialize the experience playback buffer D: Set the buffer size N to store the state transition experience generated during the asset management process, including state s, action a, reward r, next state s' and whether the termination flag is done; S3. For each training round: S301: Initialize the state s at the beginning of the round; S302: Before the round is terminated, the following steps are executed in a loop: i) The Actor network selects action a according to the current state s and adds exploration noise; ii) Execute action a, observe the reward r, the next state s' and the termination flag done; iii) Store the experience (s, a, r, s', done) in the experience replay buffer D; iv) Randomly sample a batch of experience from the experience replay buffer D to update the network parameters; v) Use the sampled experience to update the parameters of the Critic network to minimize the prediction error of the value function; vi) Use the sampled experience to update the parameters of the Actor network to maximize the action value evaluated by the Critic network; vii) Update the current state s to the next state s'; S303: When the round ends, reset the state s to the new round start state; S4: During the training process, the performance of the asset management decision model is regularly evaluated, and based on the evaluation results, the weight coefficient of the reward function, the learning rate of the model, the size of the exploration noise, and the size of the experience replay buffer are adjusted; S5: Repeat steps S3 and S4 until a satisfactory performance level is reached.

[0016] Preferably, in step S4, the average reward value is used as the performance evaluation indicator of the current round of training, and the specific implementation steps include: S401: After each training round, the average reward value of the round is calculated as the performance evaluation indicator of the current round iteration; S402: If the average reward value of the current iteration is higher than the average reward value of the previous iteration, the current model parameters are kept unchanged and the next iteration is continued; S403: If the average reward value of the current iteration is lower than the average reward value of the previous iteration, adjust the weight coefficient of the reward function, the learning rate of the model, the size of the exploration noise and the size of the experience replay buffer, reinitialize the parameters of the Actor network and the Critic network, and start a new round of training.

[0017] Preferably, a specific model building system based on an asset digital management platform comprises: The asset performance prediction model building module is used to build an asset performance prediction model. The module receives the asset configuration information and environmental condition data as input and outputs the asset prediction performance index data. When training the model, the module uses the historical asset configuration information, historical environmental condition data and the corresponding actual performance index data for training. The asset health assessment model building module is used to build an asset health assessment model. The module receives asset configuration information, historical maintenance records, current performance indicator data, and real-time environmental condition data as input, and outputs the asset's current health status indicator. When training the model, the module uses historical asset configuration information, maintenance records, performance indicator data, historical environmental condition data, and corresponding asset health status assessment results for training; The asset management decision model construction module is used to construct the asset management decision model. This module models the asset management process as a Markov decision process and makes decisions based on the output of the asset performance prediction model and the output of the asset health assessment model. The state space is defined as a combination of the real-time performance indicator data, configuration information and environmental condition data of the asset, and the action space is a set of executable maintenance, adjustment and optimization operations. The reward function calculates rewards or penalties based on performance deviations and health deviations. The output strategy is a method for selecting the optimal action in the current state, and the generated decision actions are continuously optimized through a deep reinforcement learning algorithm. The management execution module is used to perform real-time management based on the trained asset performance prediction model, asset health assessment model and asset management decision model, automatically execute the decision results or transmit them to the management personnel for confirmation and execution through the management system.

[0018] Compared with the prior art, the present invention has the following beneficial effects: The present invention constructs an asset performance prediction model, comprehensively considers the configuration information and environmental condition data of the asset, and uses historical data to train the model, which can accurately capture the complex relationship between asset performance and environmental conditions and configuration information. This significantly improves the accuracy of performance prediction, allowing asset managers to understand the future performance indicators of assets in advance and provide a scientific basis for decision-making.

[0019] The asset health assessment model integrates asset configuration information, historical maintenance records, current performance indicator data, and real-time environmental condition data to build a comprehensive assessment system. This comprehensive assessment method can more accurately reflect the current health status of assets, provide more reliable health indicators for asset management, and help to promptly discover and deal with potential problems.

[0020] The present invention models the asset management process as a Markov decision process, and realizes the intelligentization of asset management decision-making by defining the state space, action space, reward function and output strategy. This decision-making method can adapt to the dynamically changing environment and asset status, automatically adjust the decision-making strategy according to real-time data, and significantly improve the decision-making efficiency and quality.

[0021] The present invention effectively integrates the performance prediction model, health assessment model and asset management decision model, and realizes the collaborative work between the models. This integrated management method can make full use of the advantages of each model, improve the overall management effect, and realize the intelligence and automation of asset management. Through real-time management and automatic execution or transmission of decision results, the present invention can significantly improve the intelligence level and decision-making efficiency of asset management. This helps to reduce manual intervention and management costs, while improving the operating efficiency and reliability of assets, providing strong guarantees for the production safety, service quality and economic benefits of enterprises. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a working principle diagram of the specific model building method based on the asset digital management platform described in the present invention; Figure 2 A flowchart of an asset performance prediction model built using the Long Short-Term Memory Network (LSTM) algorithm; Figure 3 Flowchart for building an asset management decision model based on the Deep Deterministic Policy Gradient (DDPG) algorithm. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] See also Figure 1-3 The present invention provides a technical solution: a specific model construction method based on an asset digital management platform. The following are the specific implementation steps of the method: Step 1: Build an asset performance prediction model Collect and organize historical asset configuration information, historical environmental condition data, and corresponding actual performance indicator data as model training data. Asset configuration information includes but is not limited to asset specifications, models, production dates, etc.; environmental condition data includes environmental factors such as temperature, humidity, and pressure; actual performance indicator data is the actual operating performance data of assets within a specific time period.

[0025] Use machine learning algorithms (such as random forests, support vector machines, or neural networks) to build asset performance prediction models. Use asset configuration information and environmental condition data as input features of the model, and asset prediction performance indicator data as output. Train the model with training data so that it can accurately predict the performance indicators of the asset in the future.

[0026] Step 2: Build an asset health assessment model Collect historical asset configuration information, maintenance records, performance indicator data, historical environmental condition data, and corresponding asset health status assessment results as model training data. Maintenance records include asset repair history, replacement parts records, etc.; health status assessment results are expert assessments of asset health status or health levels based on certain standards.

[0027] The asset health assessment model is constructed by using multi-source information fusion technology and combining machine learning algorithms (such as Bayesian networks, fuzzy comprehensive evaluation or deep learning). The asset configuration information, historical maintenance records, current performance indicator data and real-time environmental condition data are used as the input of the model, and the asset's current health status indicator is used as the output of the model. The model is trained with training data so that it can comprehensively and accurately assess the current health status of the asset.

[0028] Step 3: Build an asset management decision model Model the asset management process as a Markov decision process. Define the state space as a combination of the asset's real-time performance indicator data, configuration information, and environmental condition data to fully reflect the asset's current state. Define the action space as a set of executable maintenance, adjustment, and optimization operations, such as regular maintenance, component replacement, parameter adjustment, etc. Design a reward function to calculate rewards or penalties based on performance deviations and health deviations. Give positive rewards to operations that reduce deviations and penalties to operations that increase deviations to guide the decision model to make decisions that are beneficial to asset performance and health.

[0029] The asset management decision model is trained using deep reinforcement learning algorithms (such as deep Q network, policy gradient or A3C, etc.). Through continuous trial and error and learning, the generated decision actions are optimized to form a method to select the best action in the current state, that is, the output strategy.

[0030] Step 4: Real-time management Real-time management is performed based on trained asset performance prediction models, asset health assessment models, and asset management decision models. The configuration information, performance indicator data, and environmental condition data of assets are acquired in real time through sensors and data acquisition systems. Real-time data is input into the asset performance prediction model and asset health assessment model to obtain the predicted performance indicators and health status indicators of the assets. The predicted performance indicators and health status indicators are input into the asset management decision model to obtain the optimal asset management decision action. The decision results are automatically executed or transmitted to the management personnel for confirmation and execution through the management system. If the decision result is to perform maintenance operations, the maintenance process is automatically or manually triggered; if the decision result is to adjust parameters, the asset parameters are automatically or manually adjusted to optimize performance. Through real-time management, efficient and reliable operation of assets is ensured.

[0031] The present invention will be further described below in conjunction with Examples 1 to 5: Embodiment 1:

[0032] The input data of the asset performance prediction model include: asset configuration information (model, specification, manufacturer, service life) and environmental parameters (ambient temperature, humidity, air pressure, electromagnetic interference). The output data include: asset prediction performance indicators (operating efficiency, energy consumption, temperature, vibration, pressure, flow, failure rate).

[0033] Specific implementation method: Use machine learning algorithms to train historical data, establish a mapping relationship between performance indicators and environmental parameters and configuration information, and thus predict performance indicators in the future.

[0034] The input data of the asset health assessment model includes: asset configuration information, historical maintenance records, current performance indicators and environmental parameters. The output data includes: the asset's current health status indicator, which can be a comprehensive score or health level.

[0035] The implementation method is: combine expert experience, rule reasoning and machine learning algorithms to integrate and analyze multi-source information to evaluate the health status of assets. Consider the maintenance type, cost and effect in the maintenance record, as well as the abnormal conditions of current performance indicators and environmental parameters, and comprehensively judge the health of assets.

[0036] The asset management decision model models the asset management process as a Markov decision process. The specific definitions include: State space: Defined as the combination of an asset’s real-time performance metrics, configuration information, and environmental parameters.

[0037] Action space: includes executable maintenance, adjustment and optimization operations such as adjusting operating parameters, replacing parts, cleaning equipment, lubricating equipment, calibrating equipment, upgrading software, adjusting data collection frequency and accuracy, and arranging preventive maintenance.

[0038] Reward function: Design a reward or penalty mechanism based on factors such as performance deviation (the difference between the actual performance index and the predicted performance index), health deviation (the difference between the current health status index and the ideal health status), and maintenance cost. Positive rewards are given to operations that reduce deviations and reduce costs; penalties are given to operations that increase deviations and increase costs.

[0039] Output strategy: Use deep reinforcement learning algorithm to learn and optimize the decision-making process and generate a strategy for selecting the optimal action in the current state.

[0040] The performance indicators, environmental parameters and configuration information of assets are acquired in real time through sensors and data acquisition systems. The raw data is preprocessed, such as denoising and normalization, to improve the accuracy and stability of the model.

[0041] Input real-time data into the asset performance prediction model to obtain the predicted performance indicators for a period of time in the future. Input real-time data and historical maintenance records into the asset health assessment model to obtain the current health status indicators of the asset.

[0042] The predicted performance indicators and health status indicators are used as inputs to the asset management decision model to obtain the optimal asset management decision action. According to the decision action, the corresponding maintenance, adjustment and optimization operations are performed. For example, if the predicted performance indicators show that the asset is about to fail and the health status indicator is low, the decision model may recommend to immediately schedule preventive maintenance or replace key components.

[0043] In an intelligent manufacturing factory, the method of the present invention is applied to intelligently manage key equipment on the production line. By real-time monitoring of equipment operating efficiency, energy consumption, temperature and other indicators, as well as environmental parameters and configuration information, the future performance changes and health status of the equipment are predicted. According to the prediction results and the recommendations of the decision model, the equipment operating parameters are adjusted in time, preventive maintenance is arranged or parts are replaced to ensure stable operation and efficient production of the production line. Embodiment 2:

[0044] The asset performance prediction model is constructed using the long short-term memory network (LSTM) algorithm. The following is the specific implementation of the model: 1. LSTM network structure design: Input layer: used to receive training data for the model, including asset configuration information (such as model, specification, manufacturer, age), environmental parameters (such as ambient temperature, humidity, air pressure, electromagnetic interference), and historical performance indicators (such as operating efficiency, energy consumption, temperature, vibration, pressure, flow). Input data is preprocessed at the input layer, such as standardization or normalization, to ensure data consistency and comparability.

[0045] Hidden layer: contains at least one LSTM hidden layer, each LSTM hidden layer consists of several LSTM units. Each LSTM unit has a forget gate, input gate and output gate. These gating mechanisms are used to control the flow of information and state updates, allowing the LSTM network to capture and remember long-term dependencies in time series data. The forget gate determines which information should be discarded from the memory, the input gate determines which new information should be added to the memory, and the output gate generates output based on the current memory state.

[0046] Output layer: Receives the output of the hidden layer and maps the output of the LSTM network to the predicted performance indicator value through a fully connected layer. The output layer can be designed according to specific needs, such as outputting a single performance indicator or outputting multiple performance indicators at the same time.

[0047] Second, the steps for training the asset performance prediction model include: Step 101: Initialize the weights and biases of the LSTM network The random initialization method is used to give initial values ​​to the weights and bias items of the LSTM network. Random initialization helps to break the symmetry, allowing the network to learn different features during training.

[0048] Step 102: Choose a loss function The mean square error (MSE) is selected as the loss function to measure the difference between the model prediction value and the true value. MSE is a commonly used regression loss function that can reflect the average deviation between the predicted value and the true value.

[0049] Step 103: Train the LSTM network The back propagation algorithm is used to calculate the gradient of the loss function with respect to the network weights and bias terms. The Adam optimization algorithm is used to update the network weights and bias terms. The Adam algorithm combines the advantages of the momentum method and the RMSprop method to accelerate convergence and improve training results. The network is iterated until the loss function converges or the preset training rounds are reached.

[0050] Step 104: Save the trained LSTM network model The trained LSTM network model is saved as a file or database record so that it can be called and used in the asset digital management platform. The saved model includes parameters such as network structure, weights, and bias items, which can be used to predict the performance of new input data.

[0051] Taking the key equipment in a smart manufacturing factory as an example, the application of LSTM network in asset performance prediction is illustrated.

[0052] Data preparation: Collect historical operating data of key equipment, including configuration information, environmental parameters, and performance indicators. Clean and preprocess the data, such as removing outliers, filling missing values, and standardizing.

[0053] Model training: Use the preprocessed data to train the LSTM network model, following steps 101 to 104 above. During the training process, hyperparameters such as the number of layers, number of units, and learning rate of the LSTM network can be adjusted to optimize model performance.

[0054] Performance prediction: Deploy the trained LSTM network model to the asset digital management platform. Collect the operating data of key equipment in real time and input it into the LSTM network model for performance prediction. The model outputs the predicted performance indicator values, such as operating efficiency and energy consumption, which are used to monitor the operating status of the equipment and predict future performance changes.

[0055] Decision support: Based on the prediction results, combined with the equipment health assessment and maintenance records, a reasonable maintenance plan and management strategy can be formulated. For example, when the predicted performance indicators show that the equipment is about to fail or performance degradation, preventive maintenance or replacement of parts can be arranged in advance to avoid production interruptions and losses. Embodiment 3:

[0056] The asset health assessment model is constructed using the support vector machine (SVM) algorithm. The following is a specific implementation of the model: The SVM model training steps include: Step 201: Construct a training sample set Input feature selection: Extract historical asset configuration information from the training data, including asset model, specification, manufacturer, service life, etc. Extract maintenance records, including maintenance time, maintenance type, maintenance cost, and maintenance effect, etc. Extract performance indicator data, such as operating efficiency, energy consumption, temperature, vibration, pressure, flow, etc. Extract historical environmental condition data, such as ambient temperature, humidity, air pressure, electromagnetic interference, etc. Use the above data as input features of the SVM algorithm.

[0057] Tag definition: Based on the actual health status of the asset, the asset health status assessment results are used as tags. The health status can be divided into multiple categories, such as "excellent", "good", "average", "poor", etc. The specific number of categories can be determined according to actual needs.

[0058] Construct a training sample set: Combine the input features and corresponding labels into training samples to construct a training sample set for the SVM algorithm.

[0059] Step 202: Training the SVM model Finding the optimal hyperplane: The training sample set is trained using the SVM algorithm, with the goal of finding an optimal hyperplane that maximizes the interval between different health status categories. The optimal hyperplane can best distinguish assets in different health states.

[0060] Consider soft interval: In actual training, there may be some misclassification, that is, some samples are mistakenly classified into other health status categories. In order to accommodate these misclassifications, the concept of soft interval is introduced to allow a certain degree of classification error. By adjusting the penalty parameter C, the tolerance for misclassification can be controlled.

[0061] Step 203: Optimize SVM parameters Cross-validation method: Use cross-validation method to optimize the penalty parameter C and kernel function parameters of the SVM algorithm. Divide the training sample set into several subsets, take one of the subsets as the validation set in turn, and train the remaining subsets as the training set. Through multiple cross-validations, evaluate the model performance under different parameter combinations and select the optimal parameter combination.

[0062] Kernel function selection: In the SVM algorithm, the choice of kernel function has an important impact on the model performance. Commonly used kernel functions include linear kernel, polynomial kernel, radial basis function (RBF) kernel, etc. During the cross-validation process, you can try different kernel functions and select the kernel function that best suits the current data.

[0063] Taking the health assessment of power equipment as an example, the application of SVM algorithm in asset health assessment is explained.

[0064] Data preparation: Collect historical operation data of power equipment, including configuration information, maintenance records, performance indicators and environmental condition data. Clean and preprocess the data, such as removing outliers, filling missing values, and standardizing.

[0065] Model training: According to the above steps 201 to 203, the preprocessed data is used to train the SVM model. During the training process, the penalty parameter C and the kernel function parameters are optimized by cross-validation method to select the optimal parameter combination.

[0066] Health assessment: Deploy the trained SVM model to the digital asset management platform. Collect the operating data of the power equipment in real time and input it into the SVM model for health assessment. The model outputs the health status assessment results of the power equipment, such as "excellent", "good", "average", "poor", etc.

[0067] Decision support: According to the health assessment results, formulate reasonable maintenance plans and management strategies. For example, when the health status assessment result of power equipment is "poor", preventive maintenance or equipment replacement can be arranged to avoid power outages caused by equipment failure. Embodiment 4:

[0068] This embodiment constructs an asset management decision model based on the deep deterministic policy gradient (DDPG) algorithm, and optimizes the asset management strategy in combination with the reward function. The following is a detailed implementation of this method: 1. Implementation method of reward function: Obtain real-time performance indicator data: Obtain asset performance indicator data in real time from the asset digital management platform, such as operating efficiency, energy consumption, temperature, vibration, etc.

[0069] Calculate performance deviation: Compare the real-time performance indicator data with the predicted performance indicator data and calculate the deviation between the two. The predicted performance indicator data can be obtained through the asset performance prediction model.

[0070] Obtain real-time health assessment data: Obtain asset health assessment data in real time from the asset digital management platform, that is, the current health status indicators output by the asset health assessment model.

[0071] Calculate health deviation: Compare the current health status indicators with the preset health standards or historical health data to calculate the health deviation.

[0072] Calculate rewards or penalties: Calculate rewards or penalties based on performance deviation and health deviation. The specific formula is:

[0073] in, A reward term representing a performance deviation, Rewards representing health deviations, and are the weight coefficients of the performance deviation and health deviation reward items respectively, and .

[0074] The specific value of the reward or penalty can be set according to actual needs and business logic. For example, the smaller the performance deviation, the greater the reward; the smaller the health deviation, the greater the reward.

[0075] 2. Construction method of asset management decision model: Initialize the Actor network and the Critic network: The Actor network is used to generate actions, that is, to select appropriate maintenance or management strategies based on the current asset status. The Critic network is used to evaluate the value of actions, that is, to determine whether the actions generated by the Actor network are good or not.

[0076] Initialize the experience playback buffer D: Set the buffer size N to store the state transition experience generated during the asset management process, including state s, action a, reward r, next state s' and whether the termination flag is done.

[0077] For each training round: S301: Initialize the state s at the beginning of the round: obtain the status information of the current asset from the digital asset management platform as the state s at the beginning of the round.

[0078] S302: Before the round ends, the following steps are executed in a loop: i) Actor network selects action a: The Actor network selects action a based on the current state s and adds exploration noise to increase exploration.

[0079] ii) Execute action a: Execute action a in the digital asset management platform and observe the reward r, the next state s', and the termination flag done.

[0080] iii) Store experience: Store the experience (s, a, r, s', done) in the experience replay buffer D.

[0081] iv) Randomly sample experience: Randomly sample a batch of experience from the experience replay buffer D to update the network parameters.

[0082] v) Update Critic network parameters: Use the sampled experience to update the parameters of the Critic network to minimize the prediction error of the value function.

[0083] vi) Update the Actor network parameters: Use the sampled experience to update the parameters of the Actor network to maximize the action value evaluated by the Critic network.

[0084] vii) Update current state s: Update the current state s to the next state s'.

[0085] S303: When the round ends: reset state s to the new round start state and continue training for the next round.

[0086] Regularly evaluate model performance: S401: Calculate the average reward value: After each training round, calculate the average reward value of the round as the performance evaluation indicator of the current round iteration.

[0087] S402: Determine whether the performance has improved: If the average reward value of the current iteration is higher than the average reward value of the previous iteration, keep the current model parameters unchanged and continue to the next iteration.

[0088] S403: Adjust model parameters: If the average reward value of the current iteration is lower than the average reward value of the previous iteration, adjust the weight coefficient of the reward function, the learning rate of the model, the size of the exploration noise and the size of the experience replay buffer, reinitialize the parameters of the Actor network and the Critic network, and start a new round of training.

[0089] Repeat training: Repeat steps S3 and S4 until a satisfactory performance level is achieved, that is, the average reward value of the model stabilizes at a high level and no longer increases significantly. Embodiment 5:

[0090] The present invention further expands its application in building asset management, land asset management and office asset management, and comprehensively connects the system functions of asset digital management through intelligent and AI-facilitated operations; the specific methods include: 1. Building asset management: In terms of building asset management, the system integrates the building's configuration information (such as construction age, structure type, area, etc.), environmental condition data (such as temperature and humidity, light intensity, air quality, etc.) and historical maintenance records.

[0091] Performance prediction model: The system uses LSTM network to build a performance prediction model for key performance indicators such as building energy consumption and equipment failure rate. The model input includes building configuration information, historical environmental conditions data and energy consumption data, and the output is energy consumption prediction and failure rate prediction for a period of time in the future, providing a scientific basis for building energy efficiency management and maintenance plans.

[0092] Health assessment model: Combining the building's real-time performance indicators, historical maintenance records, and environmental condition data, the system uses the SVM algorithm to build a health assessment model. This model can comprehensively assess the overall health of the building and promptly identify potential problems such as equipment aging and energy efficiency degradation.

[0093] Decision management model: The building management process is modeled as a Markov decision process, and the system generates the optimal maintenance, adjustment and optimization strategy through the DDPG algorithm. For example, based on the predicted energy consumption and failure rate, the system can automatically adjust the building's lighting, air conditioning and other system parameters to reduce energy consumption and reduce the occurrence of failures.

[0094] In addition, the system is integrated with the contract management module, information management module and capital management module to achieve all-round management of building assets. The contract management module is responsible for the management of building lease contracts and maintenance contracts; the information management module is responsible for the entry, query and update of building information; and the capital management module is responsible for the income and expenditure management of building-related expenses.

[0095] 2. Land Asset Management: In terms of land asset management, the system focuses on land use efficiency, value changes and environmental impacts.

[0096] Performance prediction model: Use LSTM network to predict the future value change of land. The input data includes the location, use, historical value data of the land, and surrounding environmental factors (such as transportation, supporting facilities, etc.).

[0097] Health assessment model: Combining land use records, environmental conditions, and changes in laws and regulations, the system uses the SVM algorithm to assess the health of the land, that is, its ability to continue to generate economic benefits and maintain environmental sustainability.

[0098] Decision-making management model: Through the DDPG algorithm, the system generates the optimal land use strategy, such as adjusting land use and improving the environment, to maximize the long-term value of the land.

[0099] At the same time, the system is integrated with the contract management module to manage land lease contracts, land use right transfer contracts, etc.; integrated with the information management module to record basic land information, historical usage records, etc.; and integrated with the fund management module to process land-related income and expenditure.

[0100] 3. Office Asset Management: In terms of office asset management, the system focuses on the performance, usage efficiency and maintenance costs of office equipment.

[0101] Performance prediction model: Use LSTM network to predict the future efficiency and failure rate of office equipment. Input data includes equipment type, age, historical usage data, and office environment conditions.

[0102] Health assessment model: Combining the equipment's maintenance records, real-time performance indicators, and office environment conditions, the system uses the SVM algorithm to assess the health of the equipment and promptly identify and address potential problems.

[0103] Decision management model: Through the DDPG algorithm, the system generates the optimal maintenance, replacement and upgrade strategies for office equipment to reduce maintenance costs and improve usage efficiency.

[0104] In addition, the system is integrated with the contract management module to manage the purchase contracts and maintenance contracts of office equipment; integrated with the information management module to record the basic information and usage records of the equipment; and integrated with the capital management module to handle the purchase, maintenance and scrapping costs related to the equipment.

[0105] The present invention also includes a specific model building system based on the asset digitalization management platform, the system comprising: The asset performance prediction model building module is used to build an asset performance prediction model. The module receives the asset configuration information and environmental condition data as input and outputs the asset prediction performance index data. When training the model, the module uses the historical asset configuration information, historical environmental condition data and the corresponding actual performance index data for training. The asset health assessment model building module is used to build an asset health assessment model. The module receives asset configuration information, historical maintenance records, current performance indicator data, and real-time environmental condition data as input, and outputs the asset's current health status indicator. When training the model, the module uses historical asset configuration information, maintenance records, performance indicator data, historical environmental condition data, and corresponding asset health status assessment results for training; The asset management decision model construction module is used to construct the asset management decision model. This module models the asset management process as a Markov decision process and makes decisions based on the output of the asset performance prediction model and the output of the asset health assessment model. The state space is defined as a combination of the real-time performance indicator data, configuration information and environmental condition data of the asset, and the action space is a set of executable maintenance, adjustment and optimization operations. The reward function calculates rewards or penalties based on performance deviations and health deviations. The output strategy is a method for selecting the optimal action in the current state, and the generated decision actions are continuously optimized through a deep reinforcement learning algorithm. The management execution module is used to perform real-time management based on the trained asset performance prediction model, asset health assessment model and asset management decision model, automatically execute the decision results or transmit them to the management personnel for confirmation and execution through the management system.

[0106] The implementation of the system refers to the above embodiment and will not be described in detail in the specification.

[0107] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0108] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A specific model construction method based on the asset digital management platform, characterized in that: The method comprises: Step 1: Construct an asset performance prediction model to predict the performance indicators of assets in the future. The input data includes: asset configuration information and environmental condition data. The output data is: asset predicted performance indicator data. The model training data includes: historical asset configuration information, historical environmental condition data and corresponding actual performance indicator data. Step 2: Construct an asset health assessment model to assess the current health status of the asset; its input data includes: asset configuration information, historical maintenance records, current performance indicator data and real-time environmental condition data; its output data is: asset current health status indicators; model training data includes: historical asset configuration information, maintenance records, performance indicator data, historical environmental condition data and corresponding asset health status assessment results; Step 3: Construct an asset management decision model and model the asset management process as a Markov decision process, which is used to make asset management decisions based on the output of the asset performance prediction model and the output of the asset health assessment model. The specific definitions include: State space: defined as the combination of an asset’s real-time performance indicator data, configuration information, and environmental condition data; Action space: defined as the set of maintenance, adjustment, and optimization operations that can be performed; Reward function: Calculate rewards or penalties based on performance deviation and health deviation, give positive rewards to operations that reduce deviations, and give penalties to operations that increase deviations; Output strategy: defined as the method of selecting the best action in the current state, which is continuously optimized by deep reinforcement learning algorithm; Step 4: Perform real-time management based on the trained asset performance prediction model, asset health assessment model, and asset management decision model, and automatically execute the decision results or transmit them to the management personnel for confirmation and execution through the management system.

2. The specific model construction method based on the asset digitalization management platform according to claim 1 is characterized by: The performance indicators include the asset's operating efficiency, energy consumption, temperature, vibration, pressure, flow and failure rate; the environmental parameters include ambient temperature, humidity, air pressure and electromagnetic interference; the configuration information includes the asset's model, specifications, manufacturer and service life; the maintenance records include maintenance time, maintenance type, maintenance cost and maintenance effect.

3. The specific model construction method based on the asset digitalization management platform according to claim 1 is characterized in that: The executable maintenance, adjustment and optimization operation set includes: adjusting operating parameters, replacing parts, cleaning equipment, lubricating equipment, calibrating equipment, upgrading software, adjusting data collection frequency and accuracy, and scheduling preventive maintenance.

4. The specific model construction method based on the asset digitalization management platform according to claim 1 is characterized in that: The implementation method of the reward function includes: Obtain real-time performance indicator data of the asset; compare the real-time performance indicator data with the predicted performance indicator data and calculate the performance deviation; Obtain real-time health assessment data of assets; compare the current health status indicators output by the asset health assessment model with the preset health standards or historical health data, and calculate the health deviation; Rewards or penalties are calculated based on performance deviation and health deviation. The specific formula is: in, A reward term representing a performance deviation, Rewards representing health deviations, and are the weight coefficients of the performance deviation and health deviation reward items, respectively, and .

5. The specific model construction method based on the asset digitalization management platform according to claim 1 is characterized in that: The asset performance prediction model is constructed using the long short-term memory network LSTM algorithm, and its network structure specifically includes: Input layer: used to receive training data of the model; Hidden layer: contains at least one LSTM hidden layer. Each LSTM hidden layer contains several LSTM units. Each LSTM unit is equipped with a forget gate, input gate and output gate to control the flow of information and state update. Output layer: Receives the output of the hidden layer and maps the output of the LSTM network to the predicted performance indicator value through a fully connected layer.

6. The specific model construction method based on the asset digitalization management platform according to claim 5 is characterized in that: The steps to train an asset performance prediction model include: Step 101: Initialize the weights and bias items of the LSTM network using a random initialization method; Step 102: Select a loss function to measure the difference between the model prediction value and the true value, and use the mean square error MSE as the loss function; Step 103: Use the back propagation algorithm and the Adam optimization algorithm to calculate the gradient according to the loss function, update the weights and bias items of the LSTM network, and iteratively train the network until the loss function converges; Step 104: Save the trained LSTM network model as the asset performance prediction model in the asset digital management platform.

7. The specific model construction method based on the asset digitalization management platform according to claim 1 is characterized in that: The asset health assessment model is constructed using the support vector machine (SVM) algorithm, and its training steps include: Step 201: Use historical asset configuration information, maintenance records, performance indicator data, and historical environmental condition data in the training data as input features, use corresponding asset health status assessment results as labels, and construct a training sample set for the support vector machine algorithm; Step 202: using the support vector machine algorithm to train the training sample set, by finding the optimal hyperplane to maximize the interval between different health status categories, while considering the soft interval to accommodate some misclassification situations; Step 203: During the training process, the penalty parameter C and kernel function parameters of the support vector machine algorithm are optimized by cross-validation method.

8. The specific model construction method based on the asset digitalization management platform according to claim 1 is characterized by: The asset management decision model is constructed using the deep deterministic policy gradient DDPG algorithm, and its training steps include: S1. Initialize the Actor network and the Critic network: The Actor network is used to generate actions, and the Critic network is used to evaluate the value of actions; S2. Initialize the experience playback buffer D: Set the buffer size N to store the state transition experience generated during the asset management process, including state s, action a, reward r, next state s' and whether the termination flag is done; S3. For each training round: S301: Initialize the state s at the beginning of the round; S302: Before the round is terminated, the following steps are executed in a loop: i) The Actor network selects action a according to the current state s and adds exploration noise; ii) Execute action a, observe the reward r, the next state s' and the termination flag done; iii) Store the experience (s, a, r, s', done) in the experience replay buffer D; iv) Randomly sample a batch of experience from the experience replay buffer D to update the network parameters; v) Use the sampled experience to update the parameters of the Critic network to minimize the prediction error of the value function; vi) Use the sampled experience to update the parameters of the Actor network to maximize the action value evaluated by the Critic network; vii) Update the current state s to the next state s'; S303: When the round ends, reset the state s to the new round start state; S4: During the training process, the performance of the asset management decision model is regularly evaluated, and based on the evaluation results, the weight coefficient of the reward function, the learning rate of the model, the size of the exploration noise, and the size of the experience replay buffer are adjusted; S5: Repeat steps S3 and S4 until a satisfactory performance level is reached.

9. The specific model construction method based on the asset digitalization management platform according to claim 8 is characterized in that: In step S4, the average reward value is used as the performance evaluation indicator of the current round of training, and the specific implementation steps include: S401: After each training round, the average reward value of the round is calculated as the performance evaluation indicator of the current round iteration; S402: If the average reward value of the current iteration is higher than the average reward value of the previous iteration, the current model parameters are kept unchanged and the next iteration is continued; S403: If the average reward value of the current iteration is lower than the average reward value of the previous iteration, adjust the weight coefficient of the reward function, the learning rate of the model, the size of the exploration noise and the size of the experience replay buffer, reinitialize the parameters of the Actor network and the Critic network, and start a new round of training.

10. A specific model building system based on an asset digital management platform, characterized in that: The system includes: The asset performance prediction model building module is used to build an asset performance prediction model. The module receives the asset configuration information and environmental condition data as input and outputs the asset prediction performance index data. When training the model, the module uses the historical asset configuration information, historical environmental condition data and the corresponding actual performance index data for training. The asset health assessment model building module is used to build an asset health assessment model. The module receives asset configuration information, historical maintenance records, current performance indicator data, and real-time environmental condition data as input, and outputs the asset's current health status indicator. When training the model, the module uses historical asset configuration information, maintenance records, performance indicator data, historical environmental condition data, and corresponding asset health status assessment results for training; The asset management decision model construction module is used to construct the asset management decision model. This module models the asset management process as a Markov decision process and makes decisions based on the output of the asset performance prediction model and the output of the asset health assessment model. The state space is defined as a combination of the real-time performance indicator data, configuration information and environmental condition data of the asset, and the action space is a set of executable maintenance, adjustment and optimization operations. The reward function calculates rewards or penalties based on performance deviations and health deviations. The output strategy is a method for selecting the optimal action in the current state, and the generated decision actions are continuously optimized through a deep reinforcement learning algorithm. The management execution module is used to perform real-time management based on the trained asset performance prediction model, asset health assessment model and asset management decision model, automatically execute the decision results or transmit them to the management personnel for confirmation and execution through the management system.

Images