Data processing method and system based on power operation and maintenance information

Abstract

This invention discloses a data processing method and system based on power operation and maintenance information, belonging to the field of power data processing technology. The method includes: collecting multi-source raw operation and maintenance data, preprocessing and windowing to form a feature vector window sequence; using the feature vector window sequence to construct a time-series deep network and generate equipment degradation coefficients; defining the actual age of the equipment based on the equipment degradation coefficient and health index, and modeling the instantaneous failure rate of the equipment to calculate the failure probability of the equipment within the future maintenance planning cycle; defining maintenance decision variables, calculating the load loss index of the maintenance decision combination based on the equipment failure probability to construct a comprehensive objective function with maintenance costs and constraints; defining reinforcement learning states, actions, and state transitions, defining immediate rewards based on the comprehensive objective function, and using a recurrent deep Q-network to learn maintenance strategies and generate optimal maintenance strategy suggestions for the power system. This forms a deployable intelligent maintenance platform.

Description

Technical Field

[0001] This invention relates to the field of power data processing technology, and in particular to data processing methods and systems based on power operation and maintenance information. Background Technology

[0002] As a critical national infrastructure, the reliability of power system equipment operation is directly related to the stability and security of the social economy. Traditionally, the maintenance strategy of power equipment mainly follows the preventive maintenance or post-maintenance model based on fixed cycles. These methods are difficult to accurately reflect the true health status of the equipment and are prone to "over-maintenance" or "under-maintenance". With the widespread application of sensor technology and information systems in the power grid, data-driven predictive maintenance has become a research hotspot. However, existing technologies often suffer from problems such as one-sided data utilization and disconnect between condition assessment and operation and maintenance decision-making, failing to form a closed-loop intelligent system from deep integration of multi-source information to global optimization decision-making.

[0003] Existing technologies still have several shortcomings that urgently need to be addressed. First, at the condition assessment level, most methods fail to effectively integrate and collaboratively analyze multimodal operation and maintenance information such as operating parameters, maintenance history, and environmental load, resulting in an insufficiently accurate and comprehensive characterization of equipment degradation processes. Second, at the risk modeling level, traditional reliability models often assume that the failure rate of equipment is a fixed value or only varies with time, ignoring the coupled influence of its actual degradation state and instantaneous operating stress (such as load and temperature), leading to large deviations in failure probability prediction and failing to provide a reliable basis for refined maintenance planning. Third, at the decision optimization level, existing maintenance scheduling methods mostly rely on static mathematical programming or heuristic rules, lacking sufficient consideration of power grid operation constraints (such as the N-1 safety criterion) and the long-term impact of maintenance actions, making it difficult to adaptively generate globally optimal strategies under complex conditions of multiple devices, multiple objectives, and multiple constraints. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a data processing method and system based on power operation and maintenance information, which solves the problems of one-sided data utilization, inaccurate risk assessment, and unreliable operation and maintenance decisions in the prior art.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a data processing method based on power operation and maintenance information, which includes, Collect raw operation and maintenance data from multiple sources, perform preprocessing and window partitioning to form a sequence of feature vector windows, use the feature vector window sequence to construct a time-series deep network and generate device degradation coefficients; Based on the equipment degradation coefficient and health index, the actual age of the equipment is defined, and the instantaneous failure rate of the equipment is modeled to calculate the failure probability of the equipment within the future maintenance planning cycle. Define maintenance decision variables, calculate the load loss index for maintenance decision combinations based on equipment failure probabilities, and construct a comprehensive objective function with maintenance costs and constraints; Define reinforcement learning states, actions, and state transitions, define immediate rewards based on the comprehensive objective function, and use a recurrent deep Q-network to learn maintenance strategies and generate optimal maintenance strategy suggestions for the power system.

[0007] As a preferred embodiment of the data processing method based on power operation and maintenance information described in this invention, the following steps are included: collecting multi-source raw operation and maintenance data, performing preprocessing and window partitioning to form a feature vector window sequence, using the feature vector window sequence to construct a time-series deep network and outputting the equipment degradation coefficient, including... For each device i in the power grid, operational status data, maintenance event data, and environmental and load data are collected at time t as raw operation and maintenance data; All original operation and maintenance data are resampled using a uniform time step, and then denoised, missing value imputed, and normalized. The preprocessed operation and maintenance data is then used to construct a feature vector window sequence using a sliding time window of length L. ; For each device i, the feature vector window sequence The input is fed into an LSTM network to obtain an intermediate representation vector. Based on the intermediate representation vector, the degradation coefficient is output through the output layer. ; Obtain pseudo-degeneracy labels from historical data As a supervisory signal, a supervisory loss function is constructed to jointly train the parameters of the deep neural network and the output layer. Using the trained deep neural network and output layer, the degradation coefficient is output. .

[0008] As a preferred embodiment of the data processing method based on power operation and maintenance information described in this invention, the step of defining the actual age of the equipment based on the equipment degradation coefficient and health index, performing instantaneous equipment failure rate modeling, and calculating the failure probability of the equipment within the future maintenance planning cycle includes, Obtain the calendar age of device i from commissioning to time t. and the design life given by the manufacturer Calculate the actual age of the device based on the degradation coefficient. ; Based on the actual age of the device In conjunction with reliable engineering principles, the failure rate of device i at time t is analyzed. Perform modeling; exist Based on this, calculate the next maintenance planning cycle. Failure probability of internal device i .

[0009] As a preferred embodiment of the data processing method based on power operation and maintenance information described in this invention, the step of defining maintenance decision variables and calculating load loss indicators for maintenance decision combinations based on equipment failure probabilities to construct a comprehensive objective function with maintenance costs and constraints includes: Discretize the maintenance planning cycle time axis into There are several maintenance periods, each with a length of [length missing]. Define maintenance decision variables When scheduling maintenance for equipment i within time period k. ,otherwise ; For each time period k, the load shedding amount that would result from a failure of equipment i under the current maintenance combination is obtained through power flow calculation. And calculate the load loss index EENS; For the entire maintenance planning cycle, taking into account EENS, direct maintenance costs, and potential failure losses, the objective function is defined. ; During the optimization process, maintenance resource constraints and maintenance time window constraints must be met.

[0010] As a preferred embodiment of the data processing method based on power operation and maintenance information described in this invention, the following steps are included: defining reinforcement learning states, actions, and state transitions; defining immediate rewards based on a comprehensive objective function; and using a recurrent deep Q-network for maintenance strategy learning. Starting at time period k, a state vector is constructed based on the degradation coefficient and failure rate of each device, combined with the current system load level and electricity consumption season. ; In each time period, the combination of maintenance decision variables is defined as an action vector. ; Execute action Subsequently, power flow calculations and N-1 checks are performed using a digital twin power grid model to obtain the updated power grid operating state, thus yielding the next state vector. ; Execute actions in each time period k Then, the instantaneous reward is calculated based on the objective function and the results of the digital twin power grid simulation. ; Construct a deep recurrent Q-network with RNN. ,in Indicates the main network parameters. This represents the hidden state of the previous time period, given a state. and the hidden state of the previous period Perform forward propagation, output the Q-values ​​of all actions, and update the hidden state; Select actions from the action space using an ε-greedy strategy. Receive an immediate reward after execution. and the next state Traverse all time periods to form an experience sequence. Stored in the experience replay pool; The target Q value is calculated using a small batch of empirical samples through the target network; Calculate the loss function based on the current Q value and the target Q value; After the loss is calculated, gradient backpropagation is performed. The Adam optimizer is used to update the parameters of the main network, and the parameters of the target network are updated through soft updates until the loss function no longer decreases significantly, resulting in a trained deep recurrent Q-network.

[0011] As a preferred embodiment of the data processing method based on power operation and maintenance information described in this invention, the step of performing power flow calculation and N-1 check using a digital twin power grid model includes: Based on the actual power grid's nodes, lines, transformers, generators, and protection configurations, an online simulation model is constructed. When the deep recurrent Q-network selects an action in time period k Then, in the digital twin, corresponding equipment maintenance and possible fault scenarios are set up, power flow calculation and N-1 check are run to obtain current, voltage and power flow distribution, calculate whether there is constraint violation, and count the load shedding amount under equipment failure conditions.

[0012] As a preferred embodiment of the data processing method based on power operation and maintenance information described in this invention, the step of generating optimal maintenance strategy suggestions for the power system includes: Using a trained deep recurrent Q-network, the optimal maintenance strategy for future planning cycles is calculated based on the current state vector. And it is distributed to each operation and maintenance unit through the scheduling system.

[0013] Secondly, this invention provides a data processing system based on power operation and maintenance information, comprising, The equipment assessment module collects real-time data on the operation, environment, and events of power equipment, performs preprocessing and sliding window segmentation, and uses a deep learning model to generate degradation coefficients that characterize the health status of the equipment. The risk prediction module calculates the equivalent actual age and instantaneous failure rate of the equipment based on its degradation coefficient, and predicts the probability of failure in a specific future maintenance cycle. The power grid simulation module constructs a digital twin model of the power grid, simulates the power flow and security status of the power grid under different maintenance strategies and fault scenarios, and quantifies the load loss indicators and the risk of exceeding safety limits. The modeling module is optimized by integrating failure probability, load loss, maintenance cost and resource constraints to construct a comprehensive maintenance optimization objective function that is centered on reliability and considers multiple constraints. The intelligent decision-making module models the maintenance and optimization problem as a reinforcement learning environment. Through interactive learning with the environment via a recurrent deep Q-network, it autonomously explores and outputs the globally optimal maintenance scheduling strategy.

[0014] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the data processing method based on power operation and maintenance information as described in the first aspect of the present invention.

[0015] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the data processing method based on power operation and maintenance information as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: by generating equipment degradation coefficients through multi-source data fusion, modeling instantaneous failure rates based on dynamic actual age, constructing maintenance optimization objectives integrating risk, cost, and constraints, and generating global adaptive maintenance strategies based on digital twins and reinforcement learning, it achieves deep learning of the actual degradation process of equipment. It combines degradation coefficients with reliability engineering to form a risk modeling system of "data-driven + mechanism model". By constructing a comprehensive objective function through power flow calculation, risk integration, and cost coupling, it transforms maintenance strategy optimization from experience-based planning to a quantitative decision-making model, forming a deployable intelligent maintenance platform that enables the scheduling system to automatically output maintenance suggestions. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of the data processing method based on power operation and maintenance information in Example 1.

[0019] Figure 2 This is a structural diagram of the data processing system based on power operation and maintenance information in Example 1.

[0020] Figure 3 This is a flowchart of the deep loop Q network maintenance strategy learning and output process in Example 1. Detailed Implementation

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0022] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0023] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation 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 single or selective embodiment that is mutually exclusive with other embodiments.

[0024] Example 1, referring to Figure 1 , Figure 2 and Figure 3 This is the first embodiment of the present invention, which provides a data processing method based on power operation and maintenance information, including the following steps: S1. Collect raw operation and maintenance data from multiple sources, perform preprocessing and window partitioning to form a feature vector window sequence, use the feature vector window sequence to construct a time-series deep network and generate device degradation coefficients; Specifically, for each device i in the power grid, operating status data, maintenance event data, and environmental and load data are collected at time t as raw operation and maintenance data; The maintenance event data includes the last maintenance time, the last maintenance type, the number of emergency repair records, and the number of planned maintenance operations. The environmental and load data include ambient temperature, humidity, load at the node, wind speed, and irradiance. All original operation and maintenance data are resampled using a uniform time step, and then denoised, missing value imputed, and normalized. The preprocessed operation and maintenance data is then used to construct a feature vector window sequence using a sliding time window of length L. ; For each device i, the feature vector window sequence The input is fed into an LSTM network to obtain an intermediate representation vector. Based on the intermediate representation vector, the degradation coefficient is output through the output layer. , represented as: ; ; in, This represents the intermediate representation vector of device i. Represents the Long Short-Term Memory network. This represents a feature vector window sequence of length L. This represents the degradation coefficient of device i. The closer the value is to 1, the more severe the degradation of device i is compared to similar devices. This represents the Sigmoid function. Indicates the weights of the output layer. Indicates bias; Obtain pseudo-degeneracy labels from historical data As a supervisory signal, a supervisory loss function is constructed to jointly train the parameters of the deep neural network and the output layer. Using the trained deep neural network and output layer, the degradation coefficient is output. .

[0025] By systematically collecting three types of heterogeneous data—operational status, maintenance events, and environmental and load data—and performing preprocessing such as time alignment, noise reduction, and normalization, a feature vector window sequence reflecting the short-term operating conditions and long-term history of equipment is constructed. Subsequently, an LSTM network is used to perform deep learning on this time series sequence to extract its high-dimensional intermediate representations. Finally, it is mapped to a quantified equipment degradation coefficient through a Sigmoid output layer. This achieves a precise mapping from multi-source, heterogeneous, and high-dimensional raw operation and maintenance data to a single, quantified, and interpretable core indicator of equipment health. Compared with traditional methods that only use a single type of data (such as current or temperature) or simple threshold comparisons, this approach, through data fusion and deep learning, can capture the nonlinear degradation patterns of equipment under complex operating stresses, thereby discovering potential defects earlier and more accurately. The output equipment degradation coefficient is not only a status indicator but also provides a unified and reliable input benchmark for subsequent risk quantification and optimization decisions, solving the problems of "state misjudgment" and "early warning delay" caused by insufficient data utilization in traditional methods.

[0026] S2. Define the actual age of the equipment based on the equipment degradation coefficient and health index, and perform instantaneous failure rate modeling to calculate the failure probability of the equipment within the future maintenance planning cycle. Specifically, obtain the calendar age of device i from commissioning to time t. and the design life given by the manufacturer Calculate the actual age of the device based on the degradation coefficient. , represented as: ; in, Indicates the actual age of device i; Based on the actual age of the device In conjunction with reliable engineering principles, the failure rate of device i at time t is analyzed. Modeling is performed, represented as: ; in, Indicates the reference failure rate of device i. , and Represents the fitted parameters, This indicates the average load of the node within the current window. This indicates the maximum design load of the node. Indicates the current device temperature. Indicates the maximum permissible temperature; exist Based on this, calculate the next maintenance planning cycle. Failure probability of internal device i , represented as: ; in, Indicates the length of the maintenance planning cycle. This represents the number of discrete-time steps.

[0027] By incorporating the equipment degradation coefficient into the definition of equipment age, a fixed "calendar age" is transformed into an "actual age" that varies with health status through a formula. Then, within the framework of a classic reliability model, this actual age is combined with instantaneous operating stresses (load rate, temperature rate) to construct a dynamic instantaneous failure rate model. Ultimately, the time-varying failure probability in future cycles is calculated. This breaks the rigid assumption in traditional reliability models that the failure rate only changes monotonically with time, establishing a dynamic risk prediction model strongly coupled with the real-time health status and operating conditions of the equipment. This makes the failure probability prediction no longer based on historical statistical averages, but on a precise profile of the equipment's "individuality" and "current state." This effectively solves the problem of "risk prediction distortion" caused by traditional models ignoring individual differences and real-time stress, laying a solid foundation for developing precise maintenance strategies that vary "from equipment to equipment and from time to time."

[0028] S3. Define maintenance decision variables, calculate the load loss index for the maintenance decision combination based on the equipment failure probability, and construct a comprehensive objective function with maintenance costs and constraints. Specifically, the maintenance planning cycle time axis is discretized into There are several maintenance periods, each with a length of [length missing]. Define maintenance decision variables When scheduling maintenance for equipment i within time period k. ,otherwise ; For each time period k, the load shedding amount that would result from a failure of equipment i under the current maintenance combination is obtained through power flow calculation. And calculate the load loss index EENS, expressed as: ; in, Indicators representing load loss This represents the load shedding amount that will result from a failure of device i during the k-th time period, obtained through power flow calculation. This represents the probability of device i failing within the k-th time period, expressed as... Calculations show that Indicates the length of each maintenance period; For the entire maintenance planning cycle, taking into account EENS, direct maintenance costs, and potential failure losses, the objective function is defined. , represented as: ; ; in, This represents all maintenance decision variables. This represents the unit cost of repairing device i at any given time. This represents the unit loss cost when device i fails. , and The weighting coefficients are determined through cross-validation. During the optimization process, maintenance resource constraints and maintenance time window constraints must be met; The maintenance resource constraints are expressed as follows: ; in, This represents the resources required for each maintenance of device i. This represents the total resources available for time period k; The maintenance time window constraint is expressed as follows: ; in, This represents the allowable maintenance window interval for device i, as specified by the scheduling plan.

[0029] By deeply integrating dynamic fault probability with the physical operating characteristics of the power grid, and using a digital twin model for power flow calculation and N-1 verification, the probability of equipment faults is transformed into a load loss index (EENS) that directly impacts the reliability of the system's power supply. Subsequently, EENS, maintenance costs, and fault losses are unified into a mathematical model, and practical constraints such as maintenance resources and time windows are embedded to construct a comprehensive objective function J(u). By seamlessly connecting the health risks at the equipment level to the operational risks at the system level, maintenance decisions are elevated from an isolated asset management problem to a system optimization problem concerning the safety, reliability, and economy of the power grid. Through the quantified EENS index, the macro-objective of "improving power supply reliability" becomes calculable and optimizable, thereby overcoming the shortcomings of traditional maintenance planning where "equipment maintenance is disconnected from power grid operation" and it is difficult to achieve a scientific trade-off between cost and reliability.

[0030] S4. Define reinforcement learning states, actions, and state transitions. Define immediate rewards based on the comprehensive objective function. Use a recurrent deep Q-network to learn maintenance strategies and generate optimal maintenance strategy suggestions for the power system.

[0031] Specifically, starting from time period k, a state vector is constructed based on the degradation coefficient and failure rate of each device, combined with the current system load level and electricity consumption season. , represented as: ; in, This represents the state vector for time period k, and N represents the total number of devices. and These represent the degradation coefficient and failure rate of the first device at the start of time period k, respectively. Indicates the total system load. Indicates seasonal category; In each time period, the combination of maintenance decision variables is defined as an action vector. , represented as: ; Execute action Subsequently, power flow calculations and N-1 checks are performed using a digital twin power grid model to obtain the updated power grid operating state, thus yielding the next state vector. ; Execute actions in each time period k Then, the instantaneous reward is calculated based on the objective function and the results of the digital twin power grid simulation. , represented as: ; in, Indicates an immediate reward. Indicates the penalty weight. This indicates the penalty amount when the power flow calculation and N-1 check fail to meet the safety constraints, and is determined according to the degree of violation and the penalty coefficient. Construct a deep recurrent Q-network with RNN. ,in Indicates the main network parameters. This represents the hidden state of the previous time period, given a state. and the hidden state of the previous period Perform forward propagation, output the Q-values ​​of all actions and update the hidden state, represented as: ; in, This represents the Q-value of all actions in time period k. This represents the hidden state of time period k; Select actions from the action space using an ε-greedy strategy. Receive an immediate reward after execution. and the next state Traverse all time periods to form an experience sequence. Stored in the experience replay pool; The target Q-value is calculated using a small batch of empirical samples through the target network, and is expressed as: ; in, Indicates the target Q value. Indicates the discount factor. Indicates the target network parameters. Indicates the state The following actions may be taken; The loss function is calculated based on the current Q-value and the target Q-value, and is expressed as: ; in, Represents the loss function. Indicates action Q value; After the loss calculation is completed, gradient backpropagation is performed. The Adam optimizer is used to update the parameters of the main network, and the parameters of the target network are updated through soft updates until the loss function no longer decreases significantly, resulting in a trained deep recurrent Q-network. Furthermore, based on the actual power grid's nodes, lines, transformers, generators, and protection configurations, an online simulation model is constructed. When the deep recurrent Q-network selects an action in time period k Then, in the digital twin, the corresponding equipment maintenance and possible fault scenarios are set up, power flow calculation and N-1 check are run to obtain current, voltage and power flow distribution, calculate whether there is constraint violation, and count the load shedding amount under equipment failure conditions. Furthermore, using a pre-trained deep recurrent Q-network, the optimal maintenance strategy for future planning cycles is calculated based on the current state vector. And it is distributed to each operation and maintenance unit through the scheduling system.

[0032] By defining the health status of equipment (degradation coefficient and failure rate) and system operating background as states, a full-dimensional decision input is achieved. Maintenance combinations are defined as actions, enabling the processing of high-dimensional combination decisions. After the actions are executed, power flow calculation and N-1 verification are performed through digital twins, realizing strategy learning under real physical constraints. The comprehensive objective function is transformed into an immediate reward, and the optimal strategy is autonomously learned through continuous interaction with the digital twin environment. Through the closed-loop structure of "state-action-physical simulation-reward" and the cyclic Q-network, the maintenance strategy has self-learning capabilities, realizing optimal maintenance planning for future cycles. Compared with traditional decision-making methods based on static rules or expert experience, it has significant advantages in intelligence and adaptability. It can automatically and efficiently generate optimal maintenance scheduling schemes that take into account both long-term benefits and short-term safety, greatly improving the safety, economy, and foresight of maintenance plans.

[0033] This embodiment also provides a data processing system based on power operation and maintenance information, including: The equipment assessment module collects real-time data on the operation, environment, and events of power equipment, performs preprocessing and sliding window segmentation, and uses a deep learning model to generate degradation coefficients that characterize the health status of the equipment. The risk prediction module calculates the equivalent actual age and instantaneous failure rate of the equipment based on its degradation coefficient, and predicts the probability of failure in a specific future maintenance cycle. The power grid simulation module constructs a digital twin model of the power grid, simulates the power flow and security status of the power grid under different maintenance strategies and fault scenarios, and quantifies the load loss indicators and the risk of exceeding safety limits. The modeling module is optimized by integrating failure probability, load loss, maintenance cost and resource constraints to construct a comprehensive maintenance optimization objective function that is centered on reliability and considers multiple constraints. The intelligent decision-making module models the maintenance and optimization problem as a reinforcement learning environment. Through interactive learning with the environment via a recurrent deep Q-network, it autonomously explores and outputs the globally optimal maintenance scheduling strategy.

[0034] This embodiment also provides a computer device applicable to the data processing method based on power operation and maintenance information, 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 data processing method based on power operation and maintenance information as proposed in the above embodiment. The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0035] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the data processing method based on power operation and maintenance information as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0036] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A data processing method based on power operation and maintenance information, characterized in that: include, Collect raw operation and maintenance data from multiple sources, perform preprocessing and window partitioning to form a sequence of feature vector windows, use the feature vector window sequence to construct a time-series deep network and generate device degradation coefficients; Based on the equipment degradation coefficient and health index, the actual age of the equipment is defined, and the instantaneous failure rate of the equipment is modeled to calculate the failure probability of the equipment within the future maintenance planning cycle. Define maintenance decision variables, calculate the load loss index for maintenance decision combinations based on equipment failure probabilities, and construct a comprehensive objective function with maintenance costs and constraints; Define reinforcement learning states, actions, and state transitions, define immediate rewards based on the comprehensive objective function, and use a recurrent deep Q-network to learn maintenance strategies and generate optimal maintenance strategy suggestions for the power system.

2. The data processing method based on power operation and maintenance information as described in claim 1, characterized in that: The process involves collecting raw operation and maintenance data from multiple sources, preprocessing and dividing it into windows to form a sequence of feature vector windows. This sequence of feature vector windows is then used to construct a time-series deep network and output the device degradation coefficient. For each device i in the power grid, operational status data, maintenance event data, and environmental and load data are collected at time t as raw operation and maintenance data; All original operation and maintenance data are resampled using a uniform time step, and then denoised, missing value imputed, and normalized. The preprocessed operation and maintenance data is then used to construct a feature vector window sequence using a sliding time window of length L. ; For each device i, the feature vector window sequence The input is fed into an LSTM network to obtain an intermediate representation vector. Based on the intermediate representation vector, the degradation coefficient is output through the output layer. ; Obtain pseudo-degeneracy labels from historical data As a supervisory signal, a supervisory loss function is constructed to jointly train the parameters of the deep neural network and the output layer. Using the trained deep neural network and output layer, the degradation coefficient is output. .

3. The data processing method based on power operation and maintenance information as described in claim 2, characterized in that: The method defines the actual age of the equipment based on the equipment degradation coefficient and health index, performs instantaneous equipment failure rate modeling, and calculates the failure probability of the equipment within the future maintenance planning cycle, including... Obtain the calendar age of device i from commissioning to time t. and the design life given by the manufacturer Calculate the actual age of the device based on the degradation coefficient. ; Based on the actual age of the device In conjunction with reliable engineering principles, the failure rate of device i at time t is analyzed. Perform modeling; exist Based on this, calculate the next maintenance planning cycle. Failure probability of internal device i .

4. The data processing method based on power operation and maintenance information as described in claim 3, characterized in that: The defined maintenance decision variables are used to calculate the load loss index for the maintenance decision combination based on the equipment failure probability, in order to construct a comprehensive objective function with maintenance costs and constraints, including: Discretize the maintenance planning cycle time axis into There are several maintenance periods, each with a length of [length missing]. Define maintenance decision variables When scheduling maintenance for equipment i within time period k. ,otherwise ; For each time period k, the load shedding amount that would result from a failure of equipment i under the current maintenance combination is obtained through power flow calculation. And calculate the load loss index EENS; For the entire maintenance planning cycle, taking into account EENS, direct maintenance costs, and potential failure losses, the objective function is defined. ; During the optimization process, maintenance resource constraints and maintenance time window constraints must be met.

5. The data processing method based on power operation and maintenance information as described in claim 4, characterized in that: The definition of reinforcement learning states, actions, and state transitions is described above. Immediate rewards are defined based on a comprehensive objective function, and a recurrent deep Q-network is used for repair policy learning. include, Starting at time period k, a state vector is constructed based on the degradation coefficient and failure rate of each device, combined with the current system load level and electricity consumption season. ; In each time period, the combination of maintenance decision variables is defined as an action vector. ; Execute action Subsequently, power flow calculations and N-1 checks are performed using a digital twin power grid model to obtain the updated power grid operating state, thus yielding the next state vector. ; Execute actions in each time period k Then, the instantaneous reward is calculated based on the objective function and the simulation results of the digital twin power grid. ; Construct a deep recurrent Q-network with RNN. ,in Indicates the main network parameters. This represents the hidden state of the previous time period, given a state. and the hidden state of the previous period Perform forward propagation, output the Q-values ​​of all actions, and update the hidden state; Select actions from the action space using an ε-greedy strategy. Receive an immediate reward after execution. and the next state Traverse all time periods to form an experience sequence. Stored in the experience replay pool; The target Q value is calculated using a small batch of empirical samples through the target network; Calculate the loss function based on the current Q value and the target Q value; After the loss is calculated, gradient backpropagation is performed. The Adam optimizer is used to update the parameters of the main network, and the parameters of the target network are updated through soft updates until the loss function no longer decreases significantly, resulting in a well-trained deep recurrent Q-network.

6. The data processing method based on power operation and maintenance information as described in claim 5, characterized in that: The power flow calculation and N-1 check using a digital twin power grid model include, Based on the actual power grid's nodes, lines, transformers, generators, and protection configurations, an online simulation model is constructed. When the deep recurrent Q-network selects an action in time period k Then, in the digital twin, corresponding equipment maintenance and possible fault scenarios are set up, power flow calculation and N-1 check are run to obtain current, voltage and power flow distribution, calculate whether there is constraint violation, and count the load shedding amount under equipment failure conditions.

7. The data processing method based on power operation and maintenance information as described in claim 6, characterized in that: The generated optimal maintenance strategy recommendations for the power system include, Using a trained deep recurrent Q-network, the optimal maintenance strategy for future planning cycles is calculated based on the current state vector. And it is distributed to each operation and maintenance unit through the scheduling system.

8. A data processing system based on power operation and maintenance information, based on the data processing method based on power operation and maintenance information according to any one of claims 1 to 7, characterized in that: include, The equipment assessment module collects real-time data on the operation, environment, and events of power equipment, performs preprocessing and sliding window segmentation, and uses a deep learning model to generate degradation coefficients that characterize the health status of the equipment. The risk prediction module calculates the equivalent actual age and instantaneous failure rate of the equipment based on its degradation coefficient, and predicts the probability of failure in a specific future maintenance cycle. The power grid simulation module constructs a digital twin model of the power grid, simulates the power flow and security status of the power grid under different maintenance strategies and fault scenarios, and quantifies the load loss indicators and the risk of exceeding safety limits. The modeling module is optimized by integrating failure probability, load loss, maintenance cost and resource constraints to construct a comprehensive maintenance optimization objective function that is centered on reliability and considers multiple constraints. The intelligent decision-making module models the maintenance and optimization problem as a reinforcement learning environment. Through interactive learning with the environment via a recurrent deep Q-network, it autonomously explores and outputs the globally optimal maintenance scheduling strategy.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the data processing method based on power operation and maintenance information as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the data processing method based on power operation and maintenance information as described in any one of claims 1 to 7.

Images