An intelligent operation and maintenance method and system for an energy management platform

Abstract

The application discloses an intelligent operation and maintenance method and system for an energy management platform, relates to the technical field of intelligent operation and maintenance, and comprises the following steps: analyzing data and generating a multi-objective optimization function by using an NLP technology, constructing a user behavior portrait, adjusting weight parameters of the multi-objective function by using a reinforcement learning algorithm combined with an Actor strategy network, obtaining an energy scheduling scheme by using an improved bald eagle search algorithm and prey escape energy mechanism, constructing a digital twin model of physical space and virtual space mapping, and optimizing the energy scheduling scheme and implementing the same by using the digital twin model combined with a soft constraint mechanism. Through the fusion of NLP technology semantic analysis and the reinforcement learning algorithm, the multi-objective scheduling strategy is dynamically optimized, and the improved bald eagle search algorithm and the digital twin model are combined to realize simulation verification and closed-loop feedback, so that the intelligentization and optimization level of the energy system operation are improved, and the stability, economy and low-carbon operation capability of the system are also improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance technology, and in particular to an intelligent operation and maintenance method and system for an energy management platform. Background Technology

[0002] With the continuous integration of information technology and industrial control technology, energy management systems are gradually evolving towards intelligence, integration, and platformization. Especially under the guidance of carbon neutrality goals, the requirements for energy efficiency optimization and energy dispatch in buildings, industrial parks, and other scenarios are becoming increasingly stringent, prompting energy management platforms to become the core infrastructure supporting the collaborative operation of multi-source heterogeneous energy systems. Currently, energy management platforms widely integrate functions such as energy consumption data acquisition, load forecasting, dispatch decision-making, and visual monitoring, and connect various sensing devices to a unified system through IoT technology. However, traditional platforms still face significant technical bottlenecks in areas such as multi-objective optimization control, dynamic user behavior response, dispatch scheme flexibility, and system closed-loop capabilities. For example, existing technologies mostly adopt static or rule-driven dispatch strategies, lacking the ability to perform deep semantic analysis of high-dimensional energy data. Furthermore, insufficient modeling of user behavior characteristics in optimization strategies leads to untimely energy dispatch response, low control accuracy, and poor system convergence.

[0003] Furthermore, although some studies have attempted to introduce machine learning models to optimize scheduling logic, significant limitations remain in multi-objective trade-offs and model generalization. For example, current optimization algorithms mostly employ fixed objective functions for scheduling, lacking the ability to dynamically adjust weights and struggling to balance multiple objectives such as energy consumption costs, comfort, and carbon emissions. Simultaneously, the coupling between reinforcement learning and evolutionary algorithms is not yet tight, and scheduling strategies cannot fully reflect the optimal response path under dynamic environmental changes. Digital twin technology, as an advanced tool integrating physical systems and virtual simulation models, has been initially used for energy consumption visualization and status monitoring, but its depth of participation and feedback accuracy in the energy scheduling process remain low, failing to achieve the goal of simultaneously prioritizing real-time prediction and flexible adjustment. Therefore, how to construct an intelligent operation and maintenance system with semantic understanding capabilities, dynamic behavior modeling, adaptive scheduling optimization, and closed-loop feedback mechanisms has become a key issue urgently needing breakthroughs in the field of energy management. Summary of the Invention

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

[0005] Therefore, this invention provides an intelligent operation and maintenance method and system for energy management platforms, which solves the problem of insufficient multi-objective scheduling optimization and dynamic behavior modeling capabilities in energy management platforms.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides an intelligent operation and maintenance method for an energy management platform, which includes collecting energy data and uploading it to a database, using NLP technology to analyze the data and generate a multi-objective optimization function;

[0008] Construct user behavior profiles, use reinforcement learning algorithms combined with Actor policy networks to adjust the weight parameters of the multi-objective function, and use an improved vulture search algorithm and prey escape energy mechanism based on the updated multi-objective optimization function to obtain an energy scheduling scheme. Construct a digital twin model that maps physical space to virtual space, and optimize the energy scheduling scheme and implement it by combining the digital twin model with a soft constraint mechanism.

[0009] The experience data generated during the implementation of the plan is uploaded to the database to achieve closed-loop feedback.

[0010] As a preferred embodiment of the intelligent operation and maintenance method for an energy management platform described in this invention, the energy scheduling scheme obtained based on the updated multi-objective optimization function using the improved vulture search algorithm and prey escape energy mechanism refers to defining the energy scheduling scheme matrix. ;

[0011] Generate Z vulture individuals, each representing an energy scheduling matrix. Normalize and initialize the energy scheduling matrix. ;

[0012] The objective function value of each individual is calculated based on the multi-objective function F. ;

[0013] Select the individual with the smallest objective function value to obtain the current optimal scheduling matrix. ;

[0014] Calculate the escape energy of the prey for each individual. ;

[0015] Each individual selects an update iteration method based on the prey escape energy value and sets a prey escape energy threshold. and ,and ;

[0016] like Choose to explore new update methods;

[0017] like Choose the soft capture update method;

[0018] like Choose the hard-encirclement update method;

[0019] After performing the hard-enclosure update, calculate the objective function value of the individual that performed the hard-enclosure. ;

[0020] like If the current hunting method is successful, continue iterative updates; otherwise, immediately replace the current hunting method with dive hunting and use the dive hunting update method to update the individual again.

[0021] When the maximum number of iterations G is reached, the individual with the smallest objective function value is selected. The scheduling matrix of the scheme represented by this individual is then used as the current optimal scheme scheduling matrix. ;

[0022] For the scheme matrix Each device and time period in the matrix is ​​denormalized to obtain the scheduling scheme matrix. .

[0023] As a preferred embodiment of the intelligent operation and maintenance method for an energy management platform described in this invention, the step of using NLP technology to analyze data and generate a multi-objective optimization function refers to first using a seasonal-trend decomposition method to decompose time-series energy consumption data into trend components. Seasonal ingredients and random noise ;

[0024] Extract trend and seasonal components as time series feature vectors. ;

[0025] Extract the device operating status from the logs at the time points corresponding to the time series features as log text. The BERT model is used to perform bidirectional semantic encoding on log text to generate log embedding vectors. The time series feature vector and the log embedding vector are combined to form the final semantic feature vector. ;

[0026] Based on feature vectors ,from Extract energy efficiency and load distribution indices to construct objective functions that minimize total energy consumption. With the objective function of balancing power consumption optimization ,from Extract log relationships to construct an optimization objective function that maximizes the proportion of renewable energy use. ;

[0027] The objective functions are combined according to their weights to generate the overall objective function F.

[0028] As a preferred embodiment of the intelligent operation and maintenance method for an energy management platform described in this invention, the following steps are included: Constructing a user behavior profile involves using a reinforcement learning algorithm combined with an Actor policy network to adjust the weight parameters of a multi-objective function. This involves defining a user feature matrix based on data collected from a database. The matrix includes user usage time features, user power usage features, user load fluctuation features, and user equipment startup frequency features. The feature matrix is ​​then normalized using the z-score normalization method to obtain a user normalized matrix. Behavioral features are extracted from the user normalized matrix to construct the behavior profile of the i-th user. ;

[0029] Generate a user profile set based on each user's behavioral profile. ;

[0030] Define state vectors based on user behavior profiles Based on the Actor policy network, according to the state vector With parameters Output action vector ;

[0031] Based on action vectors, the result of the multi-objective function is used as the reward function. ;

[0032] Calculate expected reward The gradient is calculated using the policy gradient method. And perform gradient ascent, based on Update parameters ;

[0033] Each time parameters are updated Then, based on the Actor policy network, a set of weights is output using the updated parameters, and the expected reward for each weight is calculated. The difference between the two expected rewards is then calculated. Set a difference threshold; if If the value exceeds the threshold, continue updating the parameters. Otherwise, it indicates that the gradient update has converged and will not continue;

[0034] Use the parameters from the last update. Generate action vectors The optimal weights for the objective function.

[0035] As a preferred embodiment of the intelligent operation and maintenance method for an energy management platform described in this invention, the step of collecting energy data and uploading it to the database refers to collecting time-series energy consumption data, user energy consumption characteristic data, and log data, and uploading the collected data to the database.

[0036] As a preferred embodiment of the intelligent operation and maintenance method for energy management platform described in this invention, wherein: the construction of a digital twin model mapping physical space and virtual space, the optimization of energy scheduling scheme and implementation of instructions through the digital twin model combined with soft constraint mechanism, and the construction of a digital twin model B of the entity object through the location information, type and power capacity information of the equipment;

[0037] Will As a control vector injected into the digital twin model, a soft constraint mechanism is used to determine whether the scheduling violates the energy consumption upper limit constraint, scheduling smoothness constraint, and system load balance constraint.

[0038] Computing devices at time steps Internal operating energy consumption Set the upper limit for device energy consumption. ,like This indicates that the constraint conditions are met; otherwise, it indicates that the equipment energy consumption exceeds the upper limit and violates the energy consumption upper limit constraint.

[0039] The difference in scheduling values ​​between any two adjacent moments of the computing device Set the maximum scheduling jump value If we say If the constraints are met, it means that the scheduling amplitude is too large and the scheduling smoothness constraint is not met.

[0040] Calculate the difference between the total scheduled power of all devices at time t and the expected load, and set the tolerance range. ,like This indicates that the constraints are met; otherwise, it indicates that the system supply and demand are unbalanced, violating the system load balance constraints.

[0041] If the scheduling scheme satisfies all constraints, based on Simulate and predict the state of each device ;

[0042] The actual status of the data acquisition equipment after it is in operation Calculate the error between the predicted state and the actual state. The optimized scheduling scheme is calculated based on the error value. *;

[0043] The final output is the optimized solution. *Scheduling equipment.

[0044] As a preferred embodiment of the intelligent operation and maintenance method for an energy management platform described in this invention, the step of uploading the experience data generated during the execution of the plan to the database to achieve closed-loop feedback refers to uploading the adjustment experience generated during the generation and optimization of the scheduling plan to the database to establish closed-loop feedback.

[0045] Secondly, the present invention provides an intelligent operation and maintenance system for an energy management platform, comprising:

[0046] The energy data acquisition module is used to collect raw data from the energy system in real time and upload it to the database.

[0047] The Natural Language Processing module is used to analyze user needs using NLP technology and extract optimization goals from them.

[0048] The user behavior profiling module is used to build user behavior profiles based on the database.

[0049] The Actor policy network module is used to dynamically adjust the weight coefficients of each objective in the multi-objective optimization function by utilizing user behavior profiles and historical scheduling feedback.

[0050] The scheduling scheme generation module performs a global optimization search based on the vulture optimization algorithm combined with the prey escape energy mechanism to generate a scheduling scheme.

[0051] The soft constraint verification module is used to verify the scheduling scheme and determine its feasibility.

[0052] The digital twin simulation module is used to build a digital twin model that maps physical space to virtual space, and to optimize the scheduling scheme based on the simulation results.

[0053] 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 intelligent operation and maintenance method for an energy management platform as described in the first aspect of the present invention.

[0054] 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 intelligent operation and maintenance method for an energy management platform as described in the first aspect of the present invention.

[0055] The beneficial effects of this invention are as follows: By constructing an intelligent operation and maintenance method for energy management that integrates data acquisition, semantic parsing, intelligent optimization, digital twinning, and feedback learning, the accuracy, flexibility, and intelligence level of energy scheduling are significantly improved. Natural language processing technology is used to perform semantic analysis on energy data and related operation and maintenance information, constructing a multi-objective optimization function, which effectively improves the model's expressive ability and adaptability to complex objectives. User behavior profiles are constructed and combined with reinforcement learning and an Actor policy network to dynamically adjust the weights of the optimization function, achieving real-time adaptation to user responses and behavioral changes, thereby improving the personalization and robustness of the scheduling strategy. An improved vulture search algorithm combined with a prey escape energy mechanism is used to generate scheduling schemes, balancing global optimization capabilities with dynamic adjustment capabilities, significantly improving optimization efficiency and accuracy. By constructing a digital twin model that maps physical and virtual spaces and combining it with a soft constraint mechanism for scheduling implementation, the scheme has higher executability and system flexible control capabilities. Finally, the experience data during the execution process is uploaded to the database, forming a complete closed-loop feedback system, enhancing the model's self-learning ability and long-term evolution effect. Attached Figure Description

[0056] 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.

[0057] Figure 1 This is a flowchart of the intelligent operation and maintenance method for the energy management platform in Example 1.

[0058] Figure 2 This is a structural diagram of the intelligent operation and maintenance system used in the energy management platform in Example 1. Detailed Implementation

[0059] 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.

[0060] 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.

[0061] 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.

[0062] Example 1, referring to Figure 1 and Figure 2 This is the first embodiment of the present invention, which provides an intelligent operation and maintenance method for an energy management platform, including the following steps:

[0063] S1: Collect energy data and upload it to the database. Use NLP technology to analyze the data and generate a multi-objective optimization function.

[0064] Specifically, collecting energy data and uploading it to the database refers to collecting time-series energy consumption data, user energy consumption characteristic data, and log data, and then uploading the collected data to the database.

[0065] By collecting energy data and uploading it to the database, centralized acquisition and standardized storage of multi-source heterogeneous energy consumption information are achieved, providing basic support for subsequent data-driven optimization. This effectively improves the platform's transparency and accuracy in energy usage under different scenarios, thereby enhancing the response efficiency of energy analysis and strategy formulation, and strengthening the integrity and availability of energy system data.

[0066] Further analysis of the data using NLP techniques and generation of multi-objective optimization functions first involves using a seasonal-trend decomposition method to decompose the time-series energy consumption data into trend components. Seasonal ingredients and random noise The expression is:

[0067] ,

[0068] in, Here is the original data, and t is the time.

[0069] Extract trend and seasonal components as time series feature vectors. ;

[0070] Extract the device operating status from the logs at the time points corresponding to the time series features as log text. The BERT model is used to perform bidirectional semantic encoding on log text to generate log embedding vectors. The time series feature vector and the log embedding vector are combined to form the final semantic feature vector. ;

[0071] Based on feature vectors ,from Extract energy efficiency and load distribution indices to construct objective functions that minimize total energy consumption. With the objective function of balancing power consumption optimization ,from Extract log relationships to construct an optimization objective function that maximizes the proportion of renewable energy use. ;

[0072] The total energy consumption objective function The expression is:

[0073] ,

[0074] Where T is the total time period. Let be the power at time t;

[0075] The objective function for balancing power consumption The expression is:

[0076] ,

[0077] in, This represents the current actual load of the system. This represents the theoretical ideal load average.

[0078] The objective function for maximizing the proportion of renewable energy use The expression is:

[0079] ,

[0080] in, The power output of renewable energy, This represents the total power requirement of the system.

[0081] The objective functions are combined according to their weights to generate the overall objective function F:

[0082] ,

[0083] in, , , These are the weights for the first, second, and third optimization objectives, respectively.

[0084] By using NLP technology to analyze data and generate multi-objective optimization functions, the implicit semantic information in energy data and related operation and maintenance records is mined and structured, thereby constructing a scheduling function that takes into account multiple objectives such as energy consumption, economy, environmental impact and user needs. This breaks through the traditional model construction method based on hard-coded rules, has stronger scene adaptability and objective expression ability, and improves the intelligence level of optimization function generation and the ability to adapt to dynamic trade-offs of multiple objectives.

[0085] S2: Construct user behavior profiles, use reinforcement learning algorithms combined with Actor policy networks to adjust the weight parameters of the multi-objective function, and use an improved vulture search algorithm and prey escape energy mechanism based on the updated multi-objective optimization function to obtain an energy scheduling scheme. Construct a digital twin model that maps physical space to virtual space, and optimize the energy scheduling scheme and implement it by combining the digital twin model with soft constraint mechanisms.

[0086] Specifically, the construction of user behavior profiles involves using reinforcement learning algorithms combined with an Actor policy network to adjust the weight parameters of a multi-objective function. This involves defining a user feature matrix based on data collected from a database. The matrix includes user usage time features, user power consumption features, user load fluctuation features, and user device startup frequency features. The z-score normalization method is used to normalize the feature matrix, resulting in a user normalized matrix. Behavioral features are then extracted from this normalized matrix to construct the behavior profile of the i-th user. :

[0087] ,

[0088] in, Let i be the usage time characteristic of the i-th user. Let i be the power consumption characteristics of the i-th user. For the load fluctuation characteristics of the i-th user, The startup frequency characteristic of the i-th user equipment;

[0089] Generate a user profile set based on each user's behavioral profile. :

[0090] ,

[0091] Where h represents the total number of users;

[0092] Define state vector :

[0093] ,

[0094] Based on the Actor policy network, according to the state vector Output action vector :

[0095] ,

[0096] in, For the neural network policy function, the input state vector is... Output a set of weights , For learnable parameters, and 1;

[0097] Based on action vectors, the result of the multi-objective function is used as the reward function. :

[0098] ,

[0099] Calculate expected reward :

[0100] ,

[0101] in, To find the state vector The mathematical expectation;

[0102] Calculate the gradient using the policy gradient method. And perform gradient ascent:

[0103] ,

[0104] in, For parameters Find the gradient. is the logarithm of the action density of the policy network;

[0105] based on Update parameters :

[0106] ,

[0107] Among them, the set The learning rate;

[0108] Each time parameters are updated Then, based on the Actor policy network, a set of weights is output using the updated parameters, and the expected reward for each weight is calculated. The difference between the two expected rewards is then calculated. Set a difference threshold; if If the value exceeds the threshold, continue updating the parameters. Otherwise, it indicates that the gradient update has converged and will not continue;

[0109] Use the parameters from the last update. Generate action vectors The optimal weights for the objective function.

[0110] By constructing user behavior profiles and using reinforcement learning algorithms combined with Actor policy networks to dynamically adjust the weight parameters of multi-objective functions, the system achieves the ability to optimize scheduling based on user demand fluctuations and behavioral patterns. The reinforcement learning mechanism enables the system to continuously iterate itself in an interactive environment, while the Actor policy network further improves the efficiency and accuracy of policy updates, thereby enhancing the personalization and adaptability of energy scheduling schemes and achieving a synergistic improvement in user response optimization and system flexible control.

[0111] Further, based on the updated multi-objective optimization function, an improved vulture search algorithm and prey escape energy mechanism are used to obtain energy scheduling schemes, defining the energy scheduling scheme matrix. :

[0112] ,

[0113] Where m represents the m-th device, and M represents the total number of devices. It is a real matrix with M rows and T columns;

[0114] Generate Z vulture individuals, each representing an energy scheduling matrix. Normalize and initialize the energy scheduling matrix. :

[0115] ,

[0116] in, Let r be the scheduling value of the m-th device in the n-th individual of the 0th iteration at time t, where r is a random factor and r (0,1), , Let m be the minimum and maximum power that device m can operate at time t;

[0117] Calculate the objective function value for each individual based on a multi-objective function. :

[0118] ,

[0119] in, Let be the fitness value of the nth individual in the g-th iteration. Let be the matrix representing the schemes of the nth individual in the g-th iteration;

[0120] Select the individual with the smallest objective function value to obtain the current optimal scheduling matrix. :

[0121] ,

[0122] in, This represents the optimal scheduling scheme among the individuals in the g-th iteration.

[0123] Calculate the escape energy of the prey for each individual. :

[0124] , ,

[0125] in, The initial energy is randomly assigned to prevent each individual from making the same judgment. The maximum number of iterations is set.

[0126] Each individual selects an update iteration method based on the prey escape energy value and sets a prey escape energy threshold. and ,and ;

[0127] like Choose the exploratory update method; the specific expression is:

[0128]

[0129] in, Let be the matrix of schemes represented by the nth individual in the (g+1)th iteration. Let be the matrix representing a scheme of a randomly selected vulture individual in the g-th iteration;

[0130] like Choose the soft capture update method. The specific expression is:

[0131] ,

[0132] in, To control the step size, The average value of an individual in the g-th iteration;

[0133] like Choose the hard capture update method, the specific expression is:

[0134] ,

[0135] in, , The disturbance factor is randomly generated;

[0136] After performing the hard-enclosure update, calculate the objective function value of the individual that performed the hard-enclosure:

[0137] ,

[0138] like If the current hunting method is successful, continue the iterative update; otherwise, immediately replace the current hunting method with dive hunting and update the individual again using the dive hunting update method.

[0139]

[0140] in, , The coefficients are constants.

[0141] When the maximum number of iterations G is reached, the individual with the smallest objective function value is selected. The scheduling matrix of the scheme represented by this individual is then used as the current optimal scheme scheduling matrix. ;

[0142] For the scheme matrix Each device and time period in the data is denormalized to obtain... For subsequent simulation and execution:

[0143] ,

[0144] in, It is a matrix containing the actual scheduling scheme for each device.

[0145] By combining the updated multi-objective optimization function with the improved vulture search algorithm and the prey escape energy mechanism to generate an energy scheduling scheme, the global optimization capability and local fine-tuning capability in the search process are synergistically enhanced. The improved vulture search algorithm introduces the dynamic change mechanism of prey escape energy, which can simulate the fluctuations and disturbances of real energy systems, improve the robustness of the scheduling scheme in nonlinear and dynamic environments, and improve the scheduling accuracy and algorithm convergence speed.

[0146] Furthermore, a digital twin model mapping physical and virtual spaces is constructed. This digital twin model, combined with soft constraint mechanisms, optimizes energy dispatching schemes and implements them. By utilizing the location, type, and power capacity information of equipment, a digital twin model B of the entity objects is constructed.

[0147] ,

[0148] in, For the m-th device in the digital twin model, , , Let m be the spatial location of the device. For the type of device m, The maximum power of device m;

[0149] Will As a control vector injected into the digital twin model, a soft constraint mechanism is used to determine whether the scheduling violates the energy consumption upper limit constraint, scheduling smoothness constraint, and system load balance constraint.

[0150] Computing devices at time steps Internal operating energy consumption :

[0151] ,

[0152] Set the upper limit for device energy consumption. ,like This indicates that the constraint conditions are met; otherwise, it indicates that the equipment energy consumption exceeds the upper limit and violates the energy consumption upper limit constraint.

[0153] The difference in scheduling values ​​between any two adjacent moments of the computing device :

[0154] ,

[0155] Set the maximum scheduling jump value If we say If the constraints are met, it means that the scheduling amplitude is too large and the scheduling smoothness constraint is not met.

[0156] Calculate the difference between the total scheduled power of all devices and the expected load at time t:

[0157] ,

[0158] in, The expected load of the system after scheduling;

[0159] Set tolerance range ,like This indicates that the constraints are met; otherwise, it indicates that the system supply and demand are unbalanced, violating the system load balance constraints.

[0160] If the scheduling scheme satisfies all constraints, based on Simulate and predict the state of each device :

[0161] ,

[0162] in, For the disturbance term, , These are the coefficients obtained by fitting historical equipment operation data with disturbance data;

[0163] The actual status of the data acquisition equipment after it is in operation Calculate the error between the predicted state and the actual state. :

[0164] ,

[0165] The optimized scheduling scheme based on error value calculation *:

[0166] ,

[0167] in, For adjustment coefficients;

[0168] The final output is the optimized solution. *Scheduling equipment.

[0169] By constructing a digital twin model that maps physical and virtual spaces, and combining it with a soft constraint mechanism to optimize and implement scheduling schemes, real-time simulation and flexible control of the energy system's operating status are achieved. The digital twin model enhances the platform's ability to visualize and predict the actual operating status, while the soft constraint mechanism balances system safety boundaries and scheduling flexibility, achieving optimal resource allocation while maintaining operational stability, thus improving the system's controllability, precision, and feasibility of scheduling implementation.

[0170] S3: Upload the experience data generated during the execution of the plan to the database to achieve closed-loop feedback.

[0171] Specifically, uploading the experience data generated during the execution of the plan to the database to achieve closed-loop feedback means uploading the adjustment experience generated during the generation and optimization of the scheduling plan to the database to establish closed-loop feedback.

[0172] By uploading the experience data generated during the execution of the plan to the database, a complete loop system of data-model-scheduling-implementation-data re-evolution was established, which significantly improved the system's self-evolution capability, enabling the optimization algorithm and model to gradually adapt to different scenarios and objectives through continuous learning, thereby improving the level of intelligent operation and maintenance and long-term operating performance.

[0173] This embodiment also provides an intelligent operation and maintenance system for an energy management platform, including:

[0174] The energy data acquisition module is used to collect raw data from the energy system in real time and upload it to the database.

[0175] The Natural Language Processing module is used to analyze user needs using NLP technology and extract optimization goals from them.

[0176] The user behavior profiling module is used to build user behavior profiles based on the database.

[0177] The Actor policy network module is used to dynamically adjust the weight coefficients of each objective in the multi-objective optimization function by utilizing user behavior profiles and historical scheduling feedback.

[0178] The scheduling scheme generation module performs a global optimization search based on the vulture optimization algorithm combined with the prey escape energy mechanism to generate a scheduling scheme.

[0179] The soft constraint verification module is used to verify the scheduling scheme and determine its feasibility.

[0180] The digital twin simulation module is used to build a digital twin model that maps physical space to virtual space, and to optimize the scheduling scheme based on the simulation results.

[0181] This embodiment also provides a computer device applicable to an intelligent operation and maintenance method for an energy management platform, comprising: 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 intelligent operation and maintenance method for an energy management platform as proposed in the above embodiment.

[0182] 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.

[0183] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the intelligent operation and maintenance method for an energy management platform 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.

[0184] In summary, this invention significantly improves the accuracy, flexibility, and intelligence of energy scheduling by constructing an intelligent operation and maintenance method for energy management that integrates data acquisition, semantic parsing, intelligent optimization, digital twinning, and feedback learning. Natural language processing (NLP) technology is used to perform semantic analysis on energy data and related operation and maintenance information, constructing a multi-objective optimization function that effectively enhances the model's expressive ability and adaptability to complex objectives. User behavior profiles are constructed and combined with reinforcement learning and an Actor policy network to dynamically adjust the weights of the optimization function, achieving real-time adaptation to user responses and behavioral changes, thereby improving the personalization and robustness of the scheduling strategy. An improved vulture search algorithm combined with a prey escape energy mechanism is used to generate scheduling schemes, balancing global optimization capabilities with dynamic adjustment capabilities, significantly improving optimization efficiency and accuracy. By constructing a digital twin model that maps physical and virtual spaces and combining it with a soft constraint mechanism for scheduling implementation, the scheme possesses higher executability and system flexibility. Finally, the experience data during execution is uploaded to a database, forming a complete closed-loop feedback system, enhancing the model's self-learning ability and long-term evolution performance.

[0185] 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. An intelligent operation and maintenance method for an energy management platform, characterized in that: include, Collect energy data and upload it to a database; use NLP technology to analyze the data and generate multi-objective optimization functions. Construct user behavior profiles, use reinforcement learning algorithms combined with Actor policy networks to adjust the weight parameters of the multi-objective function, and use an improved vulture search algorithm and prey escape energy mechanism based on the updated multi-objective optimization function to obtain an energy scheduling scheme. Construct a digital twin model that maps physical space to virtual space, and optimize the energy scheduling scheme and implement it by combining the digital twin model with a soft constraint mechanism. The experience data generated during the implementation of the plan is uploaded to the database to achieve closed-loop feedback; The energy scheduling scheme obtained based on the updated multi-objective optimization function using the improved vulture search algorithm and prey escape energy mechanism refers to the definition of the energy scheduling scheme matrix. , where m is the m-th device and t is the time; Generate Z vulture individuals, each representing an energy scheduling matrix. Normalize and initialize the energy scheduling matrix. ,in Refers to the 0th iteration; The objective function value of each individual is calculated based on the multi-objective function F. ,in Refers to the g-th iteration; Select the individual with the smallest objective function value to obtain the current optimal scheduling matrix. ; Calculate the escape energy of the prey for each individual. ; Each individual selects an update iteration method based on the prey escape energy value and sets a prey escape energy threshold. and ,and ; like Choose to explore new update methods; like Choose the soft-encirclement update method; like Choose the hard-encirclement update method; After performing the hard-enclosure update, calculate the objective function value of the individual that performed the hard-enclosure. ; like If the current hunting method is successful, continue iterative updates; otherwise, immediately replace the current hunting method with dive hunting and use the dive hunting update method to update the individual again. When the maximum number of iterations G is reached, the individual with the smallest objective function value is selected. The scheduling matrix of the scheme represented by this individual is then used as the current optimal scheme scheduling matrix. ; For the scheme matrix Each device and time period in the matrix is ​​denormalized to obtain the scheduling scheme matrix. ; The construction of user behavior profiles involves using a reinforcement learning algorithm combined with an Actor policy network to adjust the weight parameters of a multi-objective function. This involves defining a user feature matrix based on data collected from a database. The matrix includes user usage time features, user power consumption features, user load fluctuation features, and user device startup frequency features. The feature matrix is ​​then normalized using the z-score normalization method to obtain a user normalized matrix. Behavioral features are extracted from the user normalized matrix to construct the behavior profile of the i-th user. ; Generate a user profile set based on each user's behavioral profile. ; Define state vectors based on user behavior profiles Based on the Actor policy network, according to the state vector With parameters Output action vector ; Based on action vectors, the result of the multi-objective function is used as the reward function. ; Calculate expected reward The gradient is calculated using the policy gradient method. And perform gradient ascent, based on Update parameters ; Each time parameters are updated Then, based on the Actor policy network, a set of weights is output using the updated parameters, and the expected reward for each weight is calculated. The difference between the two expected rewards is then calculated. Set a difference threshold; if If the value exceeds the threshold, continue updating the parameters. Otherwise, it indicates that the gradient update has converged and will not continue; Use the parameters from the last update. Generate action vectors The optimal weights for the objective function.

2. The intelligent operation and maintenance method for an energy management platform as described in claim 1, characterized in that: The process of using NLP technology to analyze data and generate multi-objective optimization functions first involves using a seasonal-trend decomposition method to decompose time-series energy consumption data into trend components. Seasonal ingredients and random noise ; Extract trend and seasonal components as time series feature vectors. ; Extract the device operating status from the logs at the time points corresponding to the time series features as log text. The BERT model is used to perform bidirectional semantic encoding on log text to generate log embedding vectors. The time series feature vector and the log embedding vector are combined to form the final semantic feature vector. ; Based on feature vectors ,from Extract energy efficiency and load distribution indices to construct objective functions that minimize total energy consumption. With the objective function of balancing power consumption optimization ,from Extract log relationships to construct an optimization objective function that maximizes the proportion of renewable energy use. ; The objective functions are combined according to their weights to generate the overall objective function F.

3. The intelligent operation and maintenance method for an energy management platform as described in claim 2, characterized in that: The collection of energy data and its upload to the database refers to the collection of time-series energy consumption data, user energy consumption characteristic data, and log data, and the upload of the collected data to the database.

4. The intelligent operation and maintenance method for an energy management platform as described in claim 3, characterized in that: The construction of a digital twin model that maps physical space to virtual space, combined with a soft constraint mechanism, optimizes energy dispatching schemes and implements instructions. By using the location, type, and power capacity information of equipment, a digital twin model B of the entity object is constructed. Will As a control vector injected into the digital twin model, a soft constraint mechanism is used to determine whether the scheduling violates the energy consumption upper limit constraint, scheduling smoothness constraint, and system load balance constraint. Computing devices at time steps Internal operating energy consumption Set the upper limit for device energy consumption. ,like This indicates that the constraint conditions are met; otherwise, it indicates that the equipment energy consumption exceeds the upper limit and violates the energy consumption upper limit constraint. The difference in scheduling values ​​between any two adjacent moments of the computing device Set the maximum scheduling jump value If we say If the constraints are met, it means that the scheduling amplitude is too large and the scheduling smoothness constraint is not met. Calculate the difference between the total scheduled power of all devices at time t and the expected load, and set the tolerance range. ,like This indicates that the constraints are met; otherwise, it indicates that the system supply and demand are unbalanced, violating the system load balance constraints. If the scheduling scheme satisfies all constraints, based on Simulate and predict the state of each device ; The actual status of the data acquisition equipment after it is in operation Calculate the error between the predicted state and the actual state. The optimized scheduling scheme is calculated based on the error value. *; The final output is the optimized solution. *Scheduling equipment.

5. The intelligent operation and maintenance method for an energy management platform as described in claim 4, characterized in that: The process of uploading the experience data generated during the execution of the plan to the database to achieve closed-loop feedback refers to uploading the adjustment experience generated during the generation and optimization of the scheduling plan to the database to establish closed-loop feedback.

6. An intelligent operation and maintenance system for an energy management platform, based on the intelligent operation and maintenance method for an energy management platform according to any one of claims 1 to 5, characterized in that: include, The energy data acquisition module is used to collect raw data from the energy system in real time and upload it to the database. The Natural Language Processing module is used to analyze user needs using NLP technology and extract optimization goals from them. The user behavior profiling module is used to build user behavior profiles based on the database. The Actor policy network module is used to dynamically adjust the weight coefficients of each objective in the multi-objective optimization function by utilizing user behavior profiles and historical scheduling feedback. The scheduling scheme generation module performs a global optimization search based on the vulture optimization algorithm combined with the prey escape energy mechanism to generate a scheduling scheme. The soft constraint verification module is used to verify the scheduling scheme and determine its feasibility. The digital twin simulation module is used to build a digital twin model that maps physical space to virtual space, and to optimize the scheduling scheme based on the simulation results.

7. 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 intelligent operation and maintenance method for an energy management platform as described in any one of claims 1 to 5.

8. 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 intelligent operation and maintenance method for an energy management platform as described in any one of claims 1 to 5.

Images