Intelligent management platform suitable for power plant

Abstract

The invention discloses an intelligent management platform suitable for a power plant, and relates to the technical field of power plant management, and the platform employs an intelligent scheduling system and a DDPG algorithm, and dynamically adjusts a power generation strategy according to the real-time equipment health score, residual life prediction, and the change of energy price and load demand. The power plant can quickly make adjustments when the load demand changes sharply, preferentially dispatch efficient and low-emission equipment, improve the overall energy efficiency and reduce energy waste, and through integration of a sensor network and edge computing, key operation parameters of the equipment are collected in real time, and a dynamic maintenance plan is generated based on health scores and residual life prediction of the equipment. The equipment about to break down is preferentially processed, the random forest model is used for health state evaluation and fault probability prediction, the health score of the equipment can accurately reflect the running state of the equipment, the maintenance time is dynamically adjusted in combination with the load condition of the equipment, and excessive maintenance or maintenance lag is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of power plant management, and in particular to an intelligent management platform suitable for power plants. Background Art

[0002] As an important infrastructure in modern society, the operating efficiency, energy efficiency management and equipment maintenance of power plants play a key role in the stability and economy of power supply. With the increasing fluctuations in electricity demand, frequent fluctuations in energy prices and increasingly stringent environmental regulations, the management needs of power plants are also increasing.

[0003] Traditional power generation dispatch management platforms rely on preset dispatching schemes, which determine the operation of generator sets based on preset rules and historical data. However, since such schemes are based on fixed rules, they cannot dynamically respond to real-time load fluctuations and changes in energy prices. As a result, when power demand fluctuates sharply, power plants cannot adjust load distribution in time, affecting overall energy efficiency. At the same time, traditional dispatching schemes ignore the real-time operating status of power generation equipment and cannot effectively balance the operating efficiency of each unit with equipment maintenance requirements, resulting in the failure to fully utilize efficient units and excessive operation of inefficient units, increasing equipment losses and maintenance costs.

[0004] In order to respond to real-time load fluctuations and changes in energy prices, power plants mostly use preset operation plans and load forecasting models for optimization. However, the accuracy of load forecasting models is limited and real-time scheduling is not possible under high-frequency fluctuation scenarios. Therefore, there is an urgent need for an intelligent management platform suitable for power plants to solve such problems. Summary of the invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] The present invention provides an intelligent management platform suitable for power plants to solve the problems that static scheduling rules are difficult to cope with the complexity and variability in actual production, and at the same time overly rely on historical data and cannot fully utilize real-time data.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] On the one hand, the present invention provides a smart management platform suitable for power plants, which includes:

[0009] The sensor network consists of sensors deployed on key equipment in the power plant, which is responsible for collecting the operating parameters of the equipment in real time.

[0010] Edge computing devices, the operating parameters of the equipment are transmitted to the edge computing devices through the sensor network for preliminary processing, including data filtering, aggregation and anomaly detection, and output of equipment health status information;

[0011] The intelligent maintenance system predicts the health status of the equipment, evaluates the remaining useful life (RUL) of the equipment, and generates a health score for the equipment based on the equipment health status information.

[0012] Digital twin models are used to dynamically simulate equipment status, load demand, energy consumption and emissions. Digital twin models use sensors and intelligent maintenance systems to generate data and simulate different power generation strategies.

[0013] The intelligent dispatching system integrates the operating parameters collected by sensors, the health scores and remaining life predictions generated by the intelligent maintenance system, and optimizes load distribution based on the DDPG reinforcement learning algorithm in combination with external conditions to perform intelligent dispatching optimization of power generation equipment.

[0014] Furthermore, the key equipment in the sensor network refers to equipment including boilers, turbines, and generators.

[0015] The operating parameters of the equipment include temperature, vibration, pressure, current, voltage and oil quality.

[0016] Furthermore, in the intelligent dispatching system, the external conditions incorporated include energy prices and demand fluctuations.

[0017] In another aspect, the present invention provides a smart management method applicable to a power plant, comprising:

[0018] Step S1, equipment data collection and preliminary processing,

[0019] Based on the sensor network, the operating parameters of the power plant equipment are collected in real time, including temperature, vibration, pressure, current, voltage and oil quality detection parameters. The edge computing device performs preliminary processing on the operating parameter data, including data filtering, aggregation and anomaly detection, and outputs the health status information of the basic equipment.

[0020] Step S2: Equipment health status assessment and prediction.

[0021] The intelligent maintenance system uses the random forest machine learning algorithm to evaluate the health status of equipment and identify early signs of failure based on the real-time and historical operating parameter data of the equipment.

[0022] Generate a health score and remaining useful life RUL for each device,

[0023] Step S3: Dynamic update of the digital twin model.

[0024] Based on the real-time operating parameter data and health scores of the equipment, a digital twin model of the power plant is established and dynamically updated, which covers equipment status, load demand, energy consumption and emissions.

[0025] The digital twin model simulates different scheduling strategies and predicts future equipment status and energy usage.

[0026] Step S4, intelligent power generation dispatch optimization,

[0027] The intelligent dispatching system uses the deep deterministic policy gradient algorithm DDPG to optimize load dispatching based on equipment health scores, remaining life prediction and digital twin models.

[0028] The dispatch strategy dynamically adjusts the load distribution of equipment, giving priority to the use of high-efficiency, low-emission units, and responds according to external conditions.

[0029] Step S5, predictive maintenance execution,

[0030] Based on the equipment health status and remaining life prediction, the intelligent maintenance system generates a dynamic predictive maintenance plan, prioritizing equipment that may be about to fail.

[0031] Furthermore, in step S1, the basic equipment health status information is generated in the following manner:

[0032] Carry out data collection, including: temperature T(t), vibration V(t), pressure P(t), current I(t), voltage U(t) and oil quality Q(t), where t is the time variable, i.e. the time series during the collection process;

[0033] The Kalman filter is used to process the data. Assuming that the original signal collected by the sensor is x i (t), after filtering, the obtained signal is The state update equation of the Kalman filter is:

[0034] in, represents the signal value after filtering, K(t) represents the Kalman gain, x i (t) represents the original signal value, H represents the observation matrix,

[0035] The update formula of the prediction covariance is: P(t) = (iK(t)H)P(t-1), where P(t) represents the prediction covariance matrix, I represents the identity matrix,

[0036] The weighted average method is used for aggregation:

[0037] S(t)=w T ·T(t)+w V ·V(t)+w P ·P(t)+w I I(t)+w U ·U(t)+w QQ(t), where S(t) represents the comprehensive status value of the device, and w T ,w V ,w P ,w I ,w U ,w Q represents the weight of each parameter, T(t), V(t), P(t), I(t), U(t), Q(t) represent the equipment operation parameters collected,

[0038] Use MahalanobisDistance to detect abnormal values ​​of equipment operating parameters:

[0039] Among them, D M (t) represents the Mahalanobis distance at time t, and X(t) represents the operating parameter vector of the equipment at time t, in the form of X(t) = [T(t), V(t), P(t), I(t), U(t), Q(t)] T , μ represents the historical mean vector of each operating parameter, Σ represents the covariance matrix of the operating parameters,

[0040] If D M (t) exceeds a certain threshold τ, the current device state is considered abnormal.

[0041] According to the equipment's operating parameters, comprehensive status value S(t) and abnormal detection result D M (t), calculate the health score H(t) of the device, and the health score is defined as:

[0042] Among them, H(t) represents the health score of the device at time t, ranging from 0 to 1, α, β represent adjustment parameters, S(t) represents the comprehensive status value of the device, and D M (t) represents the Mahalanobis distance of the device, and outputs the basic device health status information.

[0043] Furthermore, in step S1, the output basic equipment health status information includes:

[0044] The real-time health score H(t) of the device,

[0045] The comprehensive status parameter S(t) reflects the overall operating status of the equipment.

[0046] And the anomaly detection result D M (t) is used to determine whether the equipment has a potential failure risk.

[0047] Furthermore, in step S2, the equipment health status assessment and prediction method is:

[0048] The real-time operating parameters and historical operating parameters of the equipment are organized into a feature matrix X(t), and the fault label Y(t) is set (to indicate whether the equipment is operating normally), X(t) = [T(t), V(t), P(t), I(t), U(t), Q(t)], where T(t) represents the temperature data of the equipment at time t, V(t) represents the vibration data of the equipment at time t, P(t) represents the pressure data of the equipment at time t, I(t) represents the current data of the equipment at time t, U(t) represents the voltage data of the equipment at time t, and Q(t) represents the oil quality data of the equipment at time t. The historical data will be used to train the random forest model, and the real-time data will be used to evaluate the current health status of the equipment.

[0049] Equipment historical operating parameter data X h As feature input, the device failure label Y h As output, a random forest model is constructed. The model construction process is:

[0050] Among them, f RF (X h ) represents the output of the random forest model, i.e., the classification result of the predicted device status, X h represents the historical operating parameter feature matrix, T i (X h ) represents the output result of the i-th decision tree, N represents the number of decision trees,

[0051] Each decision tree T i Make decisions based on different operating parameters of the equipment and use an ensemble learning method (voting mechanism) to classify the health status of the equipment.

[0052] Based on the trained random forest model f RF , using real-time operating parameter data X r Evaluate the health status of the equipment. After real-time data is input into the model, the health status prediction of the equipment is obtained:

[0053] in, Indicates the health status of the device predicted by the model, with a value of 1 for normal or 0 for fault. r represents the real-time operating parameter matrix of the device at time t,

[0054] Except classification results The random forest model also outputs the probability of device failure p fault (t):

[0055] Among them, p fault (t) represents the probability of equipment failure at time t, p i (X r) represents the failure probability predicted by the i-th decision tree,

[0056] When p fault When (t) exceeds a certain preset threshold value τ, the device is judged to be in an early failure state and an early warning is triggered.

[0057] Based on the failure probability p fault (t), generates the device health score H(t), which reflects the current health status of the device:

[0058] H(t)=1-p fault (t), where H(t) represents the health score of the device at time t, ranging from 0 to 1, and p fault (t) represents the probability of equipment failure,

[0059] The real-time health score H(t) and historical operation data of the equipment are used to predict the remaining service life RUL of the equipment in combination with the model. Assuming t fail The expected failure time of the equipment:

[0060] RUL(t)=t fail -t, where RUL(t) represents the remaining useful life of the equipment at time t, and t fail It indicates the time when the equipment is expected to fail, which is predicted by the model, and t indicates the current time point;

[0061] The final output includes:

[0062] Health score H(t), used to evaluate the current health status of the device.

[0063] Remaining useful life RUL(t) is used to predict the remaining useful life of the equipment at the current time point.

[0064] Failure probability p fault (t), used to identify early signs of failure.

[0065] Furthermore, in step S3, a digital twin model is established and dynamically updated based on the real-time operating parameter data and health score of the equipment:

[0066] The state of the power plant equipment is modeled based on the real-time operating parameters X(t) and health score H(t) of the equipment. Assuming that the state vector of the equipment is S(t), the state update equation is:

[0067] S(t+1)=A·S(t)+B·X(t)+C·H(t), where S(t) represents the state vector of the device at time t, A represents the transfer matrix of the device state, X(t) represents the real-time operating parameter vector of the device at time t, B represents the control matrix, which describes the impact of the device operating parameters on the state change, H(t) represents the health score of the device at time t, and C represents the impact matrix of the health score, which describes the impact of the health score on the device state.

[0068] The state update equation is used to describe the dynamic update of the device state, combining the real-time operating parameters X(t) and the health score H(t) to simulate the evolution of the device state over time.

[0069] Construct a load demand model to describe the load demand D(t) of the power plant at different times:

[0070] D(t)=α1D(t-1)+α2W(t)+α3P e (t)+∈(t), where D(t) represents the load demand of the power plant at time t, D(t-1) represents the load demand at the previous time step, W(t) represents the weather conditions at time t, and P e (t) represents the energy price at time t, α1, α2, α3 represent the sensitivity coefficients of load demand to historical load, weather and energy price, ∈(t) represents the noise or error term of load demand,

[0071] Build energy consumption and emissions models:

[0072]

[0073] Where E(t) represents the energy consumption of the power plant at time t, R(t) represents the emissions of the power plant at time t, and S i (t) represents the state of device i at time t, D(t) represents the load demand of the power plant at time t, γ i represents the energy efficiency coefficient of equipment i, δ i represents the emission coefficient of equipment i,

[0074] Based on the equipment status S(t), load demand D(t) and health score H(t), multi-objective optimization is performed to maximize equipment efficiency and minimize energy consumption and emissions. The optimization problem is expressed as:

[0075]

[0076] Constraints:

[0077] S(t+1)=A·S(t)+B·X(t)+C·H(t),

[0078] D(t)=α1D(t-1)+α2W(t)+α3P e (t),

[0079] Objective function: Minimize the total energy consumption within the time period T and total emissions

[0080] The deep deterministic policy gradient algorithm DDPG is used to optimize the problem and finally output the optimal scheduling strategy. This strategy dynamically adjusts the load distribution of the equipment according to the equipment health score H(t), real-time operating parameters X(t) and load demand D(t).

[0081] According to the current state S(t) of the equipment, load demand D(t) and the simulation results of the scheduling strategy, the future equipment state and energy usage are predicted. The future state S(t+k) is calculated recursively by the state equation:

[0082] Among them, S(t+k) represents the equipment status prediction at time t+k, A k Represents the k-th power of the state transfer matrix, describing the state of the device after k time steps.

[0083] Furthermore, in step S4, load scheduling optimization is performed based on DDPG:

[0084] Define the state s(t), action a(t) and reward function r(t),

[0085] The state s(t) represents the system state of the power plant at time t, including:

[0086] s(t)=[H(t),RUL(t),D(t),P e (t)], where H(t) represents the health score of the equipment at time t, RUL(t) represents the remaining useful life prediction of the equipment at time t, D(t) represents the load demand at time t, and P e (t) represents the energy price at time t,

[0087] Action a(t) represents the load distribution strategy for the device at time t, including the load distribution ratio of each device:

[0088] a(t)=[α1(t),α2(t),…,α n (t)], where α i (t) represents the load distribution ratio of equipment i at time t, satisfying

[0089] The reward function r(t) is calculated from equipment efficiency, energy consumption, and emissions. The goal is to maximize equipment efficiency and minimize energy consumption and emissions:

[0090] Among them, E i (t) represents the energy consumption of device i at time t, R i (t) represents the emission of equipment i at time t, γ i , δ i represents the energy efficiency and emission weight coefficient of equipment i,

[0091] The Actor-Critic architecture is adopted, in which the Actor network is used to generate the action a(t), and the Critic network is used to evaluate the value function Q(s(t), a(t)) of the action:

[0092] Actor network: generates load distribution action a(t) based on the current state s(t),

[0093] a(t)=π θ (s(t))+N(t), where π θ (s(t)) represents the policy network with parameter θ, generating the load distribution action a(t), N(t) represents the noise term, and the exploration noise strategy is used for exploration.

[0094] Critic network: evaluates the value of state s(t) and action a(t), and updates it using the Bellman equation.

[0095] Q φ (s(t),a(t))=r(t)+γQ φ′ (s(t+1),π θ′ (s(t+1))), where Q φ (s(t), a(t)) represents the Q value function of the current Critic network, γ represents the discount factor, which measures the contribution of future rewards, φ, φ′ represents the Critic network parameters of the current network and the target network, π θ′ (s(t+1)) represents the next action generated by the target network,

[0096] The Actor network updates the policy network parameters θ based on the Q value evaluated by the Critic network to maximize the Q value:

[0097] in, Represents the gradient of the policy network, that is, updating the policy network based on the Critic evaluation, represents the gradient of Q value to action a(t), represents the gradient of the action generated by the policy network,

[0098] Critic network update, based on minimizing the Bellman error, updates the Critic network parameters φ:

[0099]

[0100] Among them, L(φ) represents the loss function of the Critic network, and the Q value function is updated based on the Bellman error.

[0101] Dynamic load distribution strategy:

[0102] After several training sessions, the DDPG model learns the optimal load distribution strategy. The load distribution strategy a(t) at each time step dynamically adjusts the load distribution of the equipment, giving priority to high-efficiency, low-emission units. According to the health score H(t) and remaining service life RUL(t) of the equipment, the scheduling strategy will reduce the load distribution of high-risk equipment and optimize the utilization efficiency of the equipment.

[0103] Among them, α i (t) represents the load distribution ratio of equipment i at time t, H i (t) represents the health score of device i, that is, the health status of the device, RUL i (t) represents the remaining useful life of device i, that is, the length of time the device can operate,

[0104] The DDPG model is based on external conditions, energy price P e (t) and load demand D(t) for dynamic response. The dispatch strategy gives priority to the use of high-efficiency equipment when energy prices are high, and gives priority to the equipment with high health scores when load demand increases. The dynamic response rule is:

[0105] a(t)=π θ (s(t))if P e (t)<τ e and D(t)>τ d , where τ e represents the threshold of energy price, when the price is lower than τ e Scheduling is performed when d Represents the load demand threshold. When the load demand is higher than τ d Priority scheduling.

[0106] Furthermore, in step S5, dynamic predictive maintenance is performed based on the health status and remaining life of the equipment:

[0107] According to the health score H of the device i (t) and remaining useful life RUL i (t), calculate the maintenance priority P i (t), the higher the priority, the closer the equipment is to failure, and the maintenance will be prioritized.

[0108] Among them, Pi (t) represents the maintenance priority of equipment i at time t, H i (t) represents the health score of device i at time t, and its value range is [0,1]. The lower the value, the worse the health status. i (t) represents the remaining service life of equipment i at time t, which indicates the length of time before the equipment fails in the future, and α represents the control health score H i (t) The weight of the impact on the priority, β represents the remaining useful life RUL i (t) weight of impact on priority,

[0109] According to the priority of the device P i (t), select the equipment that needs to be maintained first. Assuming that the power plant can handle k equipment at each time step t, the k equipment with the highest maintenance priority is selected for maintenance. The selected equipment set is M(t):

[0110] Where M(t) represents the set of equipment that needs to be maintained at time t, arg · Indicates that maintenance priority P is selected i (t) is the largest k devices, n is the total number of devices,

[0111] Calculate the maintenance interval for each device:

[0112] Among them, Δt i Indicates the next maintenance interval of device i, RUL i (t) represents the remaining useful life of equipment i at time t, H i (t) represents the health score of device i at time t, λ represents the adjustment parameter to control the impact of the health score on the maintenance interval,

[0113] When the health score of the device is H i (t) Lower or remaining life RUL i When (t) is short, the maintenance interval Δt i will be shorter, indicating that more frequent maintenance is required.

[0114] In the dynamic maintenance plan, the impact of load on the maintenance plan is considered, and the remaining life of the equipment is dynamically updated:

[0115] RUL i (y+1)=RUL i (t)-μ·L i (y), where RUL i (t+1) represents the remaining useful life of device i at time t+1, RUL i (t) represents the remaining useful life of equipment i at time t, Li (t) represents the load level of equipment i at time t, μ represents the load influence coefficient, which represents the influence of load level on equipment life consumption,

[0116] Higher load L i (t) will accelerate the aging of the equipment and shorten its remaining life (RUL) i (t), which requires maintenance in advance.

[0117] Carry out comprehensive maintenance cost optimization,

[0118] C(t)=∑ i∈m(t) (C maint,i +C downtime,i (t)), where C(t) represents the total cost of the maintenance plan at time t, and C maint,i represents the direct maintenance cost of equipment i, C downtime,i (t) represents the downtime cost of equipment i due to maintenance,

[0119] Intelligent maintenance system combined with equipment health score H i (t), remaining life RUL i (t) and load level L i (t), calculate the maintenance priority P i (t), thereby dynamically generating equipment maintenance plans and optimizing maintenance time intervals Δt i , by optimizing the maintenance plan, minimize maintenance costs and ensure the normal operation of equipment.

[0120] The beneficial effects of the present invention are:

[0121] The present invention, through the intelligent scheduling system and DDPG algorithm, dynamically adjusts the power generation strategy according to the real-time equipment health score, remaining life prediction, energy price and load demand changes. The power plant can make rapid adjustments when the load demand changes sharply, and give priority to scheduling high-efficiency and low-emission equipment, thereby improving overall energy efficiency and reducing energy waste.

[0122] The present invention, through the integration of sensor networks and edge computing, can collect the key operating parameters of the equipment in real time, use Kalman filters and Mahalanobis distances to achieve data filtering, anomaly detection and real-time evaluation of the equipment health status, and can promptly identify early signs of equipment failure and prevent inefficient equipment from over-operating.

[0123] The present invention generates a dynamic maintenance plan based on the health score and remaining life prediction of the equipment through a predictive maintenance system, gives priority to equipment that is about to fail, and uses a random forest model to perform health status assessment and failure probability prediction. The health score of the equipment can accurately reflect the operating status of the equipment, and dynamically adjusts the maintenance time in combination with the load condition of the equipment to avoid excessive maintenance or delayed maintenance.

[0124] The present invention uses a digital twin model to dynamically simulate the actual operating status of the power plant, covering equipment status, load demand, energy consumption and emissions. It can simulate the effects of different scheduling strategies in real time, predict future equipment status and energy usage through state update equations, and continuously adjust strategies according to actual operating conditions.

[0125] The present invention, through the optimized intelligent scheduling system, prioritizes high-efficiency, low-emission equipment in load distribution and dynamically adjusts the operating status according to the health score and emission level of the equipment. The load demand and energy consumption model helps optimize energy utilization efficiency, minimizes the use of high-emission equipment, and achieves the goal of reducing carbon emissions. BRIEF DESCRIPTION OF THE DRAWINGS

[0126] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0127] Figure 1 It is a schematic diagram of the structure of the intelligent management platform of the present invention;

[0128] Figure 2 It is a schematic diagram of the flow chart of the intelligent management method of the present invention. DETAILED DESCRIPTION

[0129] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0130] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0131] 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 term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0132] Example 1, reference Figure 1 , this embodiment provides a smart management platform suitable for a power plant, including:

[0133] The sensor network consists of sensors deployed on key equipment in the power plant, which is responsible for collecting the operating parameters of the equipment in real time.

[0134] Key equipment in the sensor network refers to equipment including boilers, turbines, and generators.

[0135] The equipment’s operating parameters include temperature, vibration, pressure, current, voltage, and oil quality;

[0136] Edge computing devices, the operating parameters of the equipment are transmitted to the edge computing devices through the sensor network for preliminary processing, including data filtering, aggregation and anomaly detection, and output of equipment health status information;

[0137] The intelligent maintenance system predicts the health status of the equipment, evaluates the remaining useful life (RUL) of the equipment, and generates a health score for the equipment based on the equipment health status information.

[0138] Digital twin models are used to dynamically simulate equipment status, load demand, energy consumption and emissions. Digital twin models use sensors and intelligent maintenance systems to generate data and simulate different power generation strategies.

[0139] The intelligent dispatching system integrates the operating parameters collected by sensors, the health scores and remaining life predictions generated by the intelligent maintenance system, and optimizes the load distribution based on the DDPG reinforcement learning algorithm in combination with external conditions, and performs intelligent dispatching optimization for power generation equipment.

[0140] In the intelligent dispatching system, the external conditions incorporated include energy prices and demand fluctuations.

[0141] Example 2, reference Figure 2 This embodiment provides a smart management method applicable to a power plant, comprising the following steps:

[0142] Step S1, equipment data collection and preliminary processing,

[0143] Based on the sensor network, the operating parameters of the power plant equipment are collected in real time, including temperature, vibration, pressure, current, voltage and oil quality detection parameters. The edge computing device performs preliminary processing on the operating parameter data, including data filtering, aggregation and anomaly detection, and outputs the health status information of the basic equipment.

[0144] In step S1, the basic equipment health status information is generated in the following manner:

[0145] Carry out data collection, including: temperature T(t), vibration V(t), pressure P(t), current I(t), voltage U(t) and oil quality Q(t), where t is the time variable, i.e. the time series during the collection process;

[0146] The Kalman filter is used to process the data. Assuming that the original signal collected by the sensor is x i (t), after filtering, the obtained signal is The state update equation of Kalman filter is:

[0147] in, represents the signal value after filtering, K(t) represents the Kalman gain, x i (t) represents the original signal value, G represents the observation matrix,

[0148] The update formula of the prediction covariance is: P(t) = (Ik(t)H)P(t-1), where P(t) represents the prediction covariance matrix, I represents the identity matrix,

[0149] The weighted average method is used for aggregation:

[0150] S(t)=w T ·T(t)+w V ·V(t)+w P ·P(t)+w I I(t)+w U ·U(t)+w Q Q(t), where S(t) represents the comprehensive status value of the device, and w T ,w V ,w P ,w I ,w U ,w Q represents the weight of each parameter, T(t), V(t), P(t), I(t), U(t), Q(t) represent the equipment operation parameters collected,

[0151] Use MahalanobisDistance to detect abnormal values ​​of equipment operating parameters:

[0152] Among them, D M (t) represents the Mahalanobis distance at time t, and X(t) represents the operating parameter vector of the equipment at time t, in the form of X(t) = [T(t), V(t), P(t), I(t), U(t), Q(t)] T , μ represents the historical mean vector of each operating parameter, Σ represents the covariance matrix of the operating parameters,

[0153] If D M (t) exceeds a certain threshold τ, the current device state is considered abnormal.

[0154] According to the equipment's operating parameters, comprehensive status value S(t) and abnormal detection result D M(t), calculate the health score H(t) of the device, and the health score is defined as:

[0155] Among them, H(t) represents the health score of the device at time t, ranging from 0 to 1, α, β represent adjustment parameters, S(t) represents the comprehensive status value of the device, and D M (t) represents the Mahalanobis distance of the device and outputs basic device health status information;

[0156] In step S1, the output basic equipment health status information includes:

[0157] The real-time health score H(t) of the device,

[0158] The comprehensive status parameter S(t) reflects the overall operating status of the equipment.

[0159] And the anomaly detection result D M (t), used to determine whether the equipment has a potential failure risk,

[0160] Specifically, the temperature, vibration, and pressure parameters of the equipment are collected in real time through the sensor network, and combined with edge computing devices for real-time processing. With the help of the Kalman filter, noise can be effectively filtered out, and various parameters can be aggregated to generate the comprehensive health status of the equipment. When the equipment status is abnormal, potential problems can be discovered in time and countermeasures can be taken, significantly reducing the probability of equipment failure. By detecting anomalies through Mahalanobis distance, deviations in equipment operation can be identified at an early stage.

[0161] Step S2: Equipment health status assessment and prediction.

[0162] The intelligent maintenance system uses the random forest machine learning algorithm to evaluate the health status of equipment and identify early signs of failure based on the real-time and historical operating parameter data of the equipment.

[0163] Generate a health score and remaining useful life RUL for each device,

[0164] In step S2, the equipment health status assessment and prediction method is:

[0165] The real-time operating parameters and historical operating parameters of the equipment are organized into a feature matrix X(t), and the fault label Y(t) is set (to indicate whether the equipment is operating normally), X(t) = [T(t), V(t), P(t), I(t), U(t), Q(t)], where T(t) represents the temperature data of the equipment at time t, V(t) represents the vibration data of the equipment at time t, P(t) represents the pressure data of the equipment at time t, I(t) represents the current data of the equipment at time t, U(t) represents the voltage data of the equipment at time t, and Q(t) represents the oil quality data of the equipment at time t. The historical data will be used to train the random forest model, and the real-time data will be used to evaluate the current health status of the equipment.

[0166] Equipment historical operating parameter data X h As feature input, the device failure label Y h As output, a random forest model is constructed. The model construction process is:

[0167] Among them, f RF (X h ) represents the output of the random forest model, i.e., the classification result of the predicted device status, X h represents the historical operating parameter feature matrix, T i (X h ) represents the output result of the oth decision tree, N represents the number of decision trees,

[0168] Each decision tree T i Make decisions based on different operating parameters of the equipment and use an ensemble learning method (voting mechanism) to classify the health status of the equipment.

[0169] Based on the trained random forest model f RF , using real-time operating parameter data X r Evaluate the health status of the equipment. After real-time data is input into the model, the health status prediction of the equipment is obtained:

[0170] in, Indicates the health status of the device predicted by the model, with a value of 1 for normal or 0 for fault. r represents the real-time operating parameter matrix of the device at time t,

[0171] Except classification results The random forest model also outputs the probability of device failure p fault (t):

[0172] Among them, p fault (t) represents the probability of equipment failure at time t, p i (X r) represents the failure probability predicted by the i-th decision tree,

[0173] When p fault When (t) exceeds a certain preset threshold value τ, the device is judged to be in an early failure state and an early warning is triggered.

[0174] Based on the failure probability p fault (t), generates the device health score H(t), which reflects the current health status of the device:

[0175] H(t)=1-p fault (t), where H(t) represents the health score of the device at time t, ranging from 0 to 1, and p fault (t) represents the probability of equipment failure,

[0176] The real-time health score H(t) and historical operation data of the equipment are used to predict the remaining service life RUL of the equipment in combination with the model. Assuming t fail The expected failure time of the equipment:

[0177] RUL(t)=t fail -t, where RUL(t) represents the remaining useful life of the equipment at time t, and t fail It indicates the time when the equipment is expected to fail, which is predicted by the model, and t indicates the current time point;

[0178] The final output includes:

[0179] Health score H(t), used to evaluate the current health status of the device.

[0180] Remaining useful life RUL(t) is used to predict the remaining useful life of the equipment at the current time point.

[0181] Failure probability p fault (t), used to identify early signs of failure;

[0182] Specifically, based on the deep deterministic policy gradient algorithm DDPG, it can dynamically adjust the load distribution of equipment according to the equipment's health score, remaining service life prediction and real-time operating status. It can give priority to scheduling high-efficiency and low-emission equipment, reduce the use of high-energy-consuming equipment, and respond to changes in external conditions in real time to maximize energy utilization efficiency and minimize carbon emissions.

[0183] In traditional power plant management, the operating status is mostly determined by monitoring systems and historical data analysis, but there is a lack of specific and refined data. When the load demand changes drastically or the equipment status is unclear, it is difficult to make effective and timely operating status judgments. In addition, the monitoring system is scattered in various subsystems, making it difficult to achieve unified data integration of the equipment and systems of the entire plant, and it is impossible to perform real-time status perception and optimization at the overall level. To address such problems, most power plants currently rely on monitoring data and regular equipment inspections to troubleshoot problems, and perform optimization through manual analysis of historical data. However, this processing method lags behind in the face of real-time changes, and cannot achieve dynamic optimization under complex working conditions and coordinated operation of multiple devices. For this reason, the present invention proposes the following scheme to perform targeted optimization:

[0184] Step S3: Dynamic update of the digital twin model.

[0185] Based on the real-time operating parameter data and health scores of the equipment, a digital twin model of the power plant is established and dynamically updated, which covers equipment status, load demand, energy consumption and emissions.

[0186] The digital twin model simulates different scheduling strategies and predicts future equipment status and energy usage.

[0187] In step S3, a digital twin model is established and dynamically updated based on the real-time operating parameter data and health score of the equipment:

[0188] The state of the power plant equipment is modeled based on the real-time operating parameters X(t) and health score H(t) of the equipment. Assuming that the state vector of the equipment is S(t), the state update equation is:

[0189] S(t+1)=A·S(t)+B·X(t)+C·H(t), where S(t) represents the state vector of the device at time t, A represents the transfer matrix of the device state, X(t) represents the real-time operating parameter vector of the device at time t, B represents the control matrix, which describes the impact of the device operating parameters on the state change, H(t) represents the health score of the device at time t, and C represents the impact matrix of the health score, which describes the impact of the health score on the device state.

[0190] The state update equation is used to describe the dynamic update of the device state, combining the real-time operating parameters X(t) and the health score H(t) to simulate the evolution of the device state over time.

[0191] Construct a load demand model to describe the load demand D(t) of the power plant at different times:

[0192] D(t)=α1D(t-1)+α2W(t)+α3Pe(t)+∈(t), where D(t) represents the load demand of the power plant at time t, D(t-1) represents the load demand at the previous time step, W(t) represents the weather conditions at time t, and P e (t) represents the energy price at time t, α1, α2, α3 represent the sensitivity coefficients of load demand to historical load, weather and energy price, ∈(t) represents the noise or error term of load demand,

[0193] Build energy consumption and emissions models:

[0194]

[0195] Where E(t) represents the energy consumption of the power plant at time t, R(t) represents the emissions of the power plant at time t, and S i (t) represents the state of device i at time t, D(t) represents the load demand of the power plant at time t, γ i represents the energy efficiency coefficient of equipment i, δ i represents the emission coefficient of equipment i,

[0196] Based on the equipment status S(t), load demand D(t) and health score H(t), multi-objective optimization is performed to maximize equipment efficiency and minimize energy consumption and emissions. The optimization problem is expressed as:

[0197]

[0198] Constraints:

[0199] S(t+1)=A·S(t)+B·X(t)+C·H(t),

[0200] D(t)=α1D(t-1)+α2W(t)+α3P e (t),

[0201] Objective function: Minimize the total energy consumption within the time period T and total emissions

[0202] The deep deterministic policy gradient algorithm DDPG is used to optimize the problem and finally output the optimal scheduling strategy. This strategy dynamically adjusts the load distribution of the equipment according to the equipment health score H(t), real-time operating parameters X(t) and load demand D(t).

[0203] According to the current state S(t) of the equipment, load demand D(t) and the simulation results of the scheduling strategy, the future equipment state and energy usage are predicted. The future state S(t+k) is calculated recursively by the state equation:

[0204] Among them, S(t+k) represents the equipment status prediction at time t+k, A k represents the kth power of the state transfer matrix, describing the state of the device after k time steps,

[0205] Specifically, the predictive maintenance system uses the random forest algorithm to evaluate the health status of the equipment, and combines it with the remaining useful life (RUL) prediction to generate a dynamic maintenance plan for each device. Through the comprehensive evaluation of the health score and remaining life, it can identify equipment that may be about to fail in advance and generate maintenance priorities to avoid waste of resources and equipment failure caused by excessive or delayed maintenance.

[0206] Step S4, intelligent power generation dispatch optimization,

[0207] The intelligent dispatching system uses the deep deterministic policy gradient algorithm DDPG to optimize load dispatching based on equipment health scores, remaining life prediction and digital twin models.

[0208] The dispatch strategy dynamically adjusts the load distribution of equipment, giving priority to the use of high-efficiency, low-emission units, and responds according to external conditions.

[0209] In step S4, load scheduling optimization is performed based on DDPG:

[0210] Define the state s(t), action a(t) and reward function r(t),

[0211] The state s(t) represents the system state of the power plant at time t, including:

[0212] s(t)=[H(t),RUL(t),D(t),P e (t)], where H(t) represents the health score of the equipment at time t, RUL(t) represents the remaining useful life prediction of the equipment at time t, D(t) represents the load demand at time t, and P e (t) represents the energy price at time t,

[0213] Action a(t) represents the load distribution strategy for the device at time t, including the load distribution ratio of each device:

[0214] a(t)=[α1(t),α2(t),…,α n (t)], where α i (t) represents the load distribution ratio of equipment i at time t, satisfying

[0215] The reward function r(t) is calculated from equipment efficiency, energy consumption, and emissions. The goal is to maximize equipment efficiency and minimize energy consumption and emissions:

[0216] Among them, E i (t) represents the energy consumption of device i at time t, R i (t) represents the emission of equipment i at time t, γ i , δ i represents the energy efficiency and emission weight coefficient of equipment i,

[0217] The Actor-Critic architecture is adopted, in which the Actor network is used to generate the action a(t), and the Critic network is used to evaluate the value function Q(s(t), a(t)) of the action:

[0218] Actor network: generates load distribution action a(t) based on the current state s(t),

[0219] Among them, π θ (s(t)) represents the policy network with parameter θ, generating the load distribution action a(t), represents the noise term, explored using the exploration noise strategy,

[0220] Critic network: evaluates the value of state s(t) and action a(t), and updates it using the Bellman equation.

[0221] Q φ (s(t),a(t))=r(t)+γQ φ′ (s(t+1),π θ′ (s(t+1))), where Q φ (s(t), a(t)) represents the Q value function of the current Critic network, γ represents the discount factor, which measures the contribution of future rewards, φ, φ′ represents the Critic network parameters of the current network and the target network, π θ′ (s(t+1)) represents the next action generated by the target network,

[0222] The Actor network updates the policy network parameters θ based on the Q value evaluated by the Critic network to maximize the Q value:

[0223] in, Represents the gradient of the policy network, that is, updating the policy network based on the Critic evaluation, represents the gradient of Q value to action a(t), represents the gradient of the action generated by the policy network,

[0224] Critic network update, based on minimizing the Bellman error, updates the Critic network parameters φ:

[0225] Among them, L(φ) represents the loss function of the Critic network, and the Q value function is updated based on the Bellman error.

[0226] Dynamic load distribution strategy:

[0227] After several training sessions, the DDPG model learns the optimal load distribution strategy. The load distribution strategy a(t) at each time step dynamically adjusts the load distribution of the equipment, giving priority to high-efficiency, low-emission units. According to the health score H(t) and remaining service life RUL(t) of the equipment, the scheduling strategy will reduce the load distribution of high-risk equipment and optimize the utilization efficiency of the equipment.

[0228] Among them, α i (t) represents the load distribution ratio of equipment i at time t, H i (t) represents the health score of device i, that is, the health status of the device, RUL i (t) represents the remaining useful life of device i, that is, the length of time the device can operate,

[0229] The DDPG model is based on external conditions, energy price P e (t) and load demand D(t) for dynamic response. The dispatch strategy gives priority to the use of high-efficiency equipment when energy prices are high, and gives priority to the equipment with high health scores when load demand increases. The dynamic response rule is:

[0230] a(t)=π θ (s(t))if P e (t)<τ e and D(t)>τ d , where τ e represents the threshold of energy price, when the price is lower than τ e Scheduling is performed when d Represents the load demand threshold. When the load demand is higher than τ d Time priority scheduling;

[0231] Specifically, through the real-time operating parameters and health scores of the equipment, the digital twin model can effectively simulate the entire operating status of the power plant and dynamically update the system. The model can not only simulate different power generation scheduling strategies, but also predict future equipment status, load demand, energy consumption and emissions, avoiding problems such as excessive energy consumption or excessive emissions in actual operations.

[0232] Step S5, predictive maintenance execution,

[0233] Based on the equipment health status and remaining life prediction, the intelligent maintenance system generates a dynamic predictive maintenance plan to prioritize equipment that may be about to fail.

[0234] In step S5, dynamic predictive maintenance is performed based on the equipment health status and remaining life:

[0235] According to the health score H of the device i (t) and remaining useful life RUL i (t), calculate the maintenance priority P i (t), the higher the priority, the closer the equipment is to failure, and the maintenance will be prioritized.

[0236] Among them, P i (t) represents the maintenance priority of equipment i at time t, H i (t) represents the health score of device i at time t, and its value range is [0,1]. The lower the value, the worse the health status. i (t) represents the remaining useful life of equipment i at time t, which indicates the length of time before the equipment fails in the future, and α represents the control health score H i (t) The weight of the impact on the priority, β represents the remaining useful life RUL i (t) weight of impact on priority,

[0237] According to the priority of the device P i (t), select the equipment that needs to be maintained first. Assuming that the power plant can handle k equipment at each time step t, the k equipment with the highest maintenance priority is selected for maintenance. The selected equipment set is M(t):

[0238] Where M(t) represents the set of equipment that needs to be maintained at time t, arg · Indicates that maintenance priority P is selected i (t) is the largest k devices, n is the total number of devices,

[0239] Calculate the maintenance interval for each device:

[0240] Among them, Δt i Indicates the next maintenance interval of device i, RUL i (t) represents the remaining useful life of equipment i at time t, H i (t) represents the health score of device i at time t, λ represents the adjustment parameter to control the impact of the health score on the maintenance interval,

[0241] When the health score of the device is H i (t) Lower or remaining life RUL iWhen (t) is short, the maintenance interval Δt i will be shorter, indicating that more frequent maintenance is required.

[0242] In the dynamic maintenance plan, the impact of load on the maintenance plan is considered, and the remaining life of the equipment is dynamically updated:

[0243] RUL i (t+1) = RUL i (t)-μ·L i (t), where RUL i (t+1) represents the remaining useful life of device i at time t+1, RUL i (t) represents the remaining useful life of equipment i at time t, L i (t) represents the load level of equipment i at time t, μ represents the load influence coefficient, which represents the influence of load level on equipment life consumption,

[0244] Higher load L i (t) will accelerate the aging of the equipment and shorten its remaining life (RUL) i (t), which requires maintenance in advance.

[0245] Carry out comprehensive maintenance cost optimization,

[0246] C(t)=∑ i∈M(t) (C maint,i +C downtime,i (t)), where C(t) represents the total cost of the maintenance plan at time t, and C maint,i represents the direct maintenance cost of equipment i, C downtime,i (t) represents the downtime cost of equipment i due to maintenance,

[0247] Intelligent maintenance system combined with equipment health score H i (t), remaining life RUL i (t) and load level L i (t), calculate the maintenance priority P i (t), thereby dynamically generating equipment maintenance plans and optimizing maintenance time intervals Δt i , by optimizing the maintenance plan, minimize maintenance costs and ensure the normal operation of equipment.

[0248] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A smart management platform for power plants, characterized by: include, The sensor network consists of sensors deployed on key equipment in the power plant, which is responsible for collecting the operating parameters of the equipment in real time. Edge computing devices, the operating parameters of the equipment are transmitted to the edge computing devices through the sensor network for preliminary processing, including data filtering, aggregation and anomaly detection, and output of equipment health status information; The intelligent maintenance system predicts the health status of the equipment, evaluates the remaining useful life (RUL) of the equipment, and generates a health score for the equipment based on the equipment health status information. Digital twin models are used to dynamically simulate equipment status, load demand, energy consumption and emissions. Digital twin models use sensors and intelligent maintenance systems to generate data and simulate different power generation strategies. The intelligent dispatching system integrates the operating parameters collected by sensors, the health scores and remaining life predictions generated by the intelligent maintenance system, and optimizes load distribution based on the DDPG reinforcement learning algorithm in combination with external conditions to perform intelligent dispatching optimization of power generation equipment.

2. The intelligent management platform for power plants according to claim 1, characterized in that: Key equipment in the sensor network refers to equipment including boilers, turbines, and generators. The operating parameters of the equipment include temperature, vibration, pressure, current, voltage and oil quality.

3. The intelligent management platform for power plants according to claim 1, characterized in that: In the intelligent dispatching system, the external conditions incorporated include energy prices and demand fluctuations.

4. A smart management method applicable to a power plant, characterized in that: A smart management platform suitable for a power plant according to any one of claims 1 to 3, comprising the following steps: Step S1, equipment data collection and preliminary processing, based on the sensor network, real-time collection of power plant equipment operating parameters, including temperature, vibration, pressure, current, voltage and oil quality detection parameters, edge computing equipment performs preliminary processing on the operating parameter data, including data filtering, aggregation and anomaly detection, and outputs basic equipment health status information, Step S2, equipment health status assessment and prediction, the intelligent maintenance system uses the random forest machine learning algorithm to assess the equipment health status and identify early signs of failure based on the real-time operating parameter data and historical operating parameter data of the equipment. Generate a health score and remaining useful life RUL for each device, Step S3: Dynamic update of the digital twin model. Based on the real-time operating parameter data and health score of the equipment, a digital twin model of the power plant is established and dynamically updated. The model covers equipment status, load demand, energy consumption, and emissions. The digital twin model simulates different scheduling strategies and predicts future equipment status and energy usage. Step S4, intelligent power generation dispatch optimization, the intelligent dispatch system uses the deep deterministic policy gradient algorithm DDPG to optimize load dispatch based on equipment health score, remaining life prediction and digital twin model, The dispatch strategy dynamically adjusts the load distribution of equipment, giving priority to the use of high-efficiency, low-emission units, and responds according to external conditions. Step S5, predictive maintenance execution, based on the equipment health status and remaining life prediction, the intelligent maintenance system generates a dynamic predictive maintenance plan, giving priority to equipment that may soon fail.

5. The intelligent management method applicable to a power plant according to claim 4, characterized in that: In step S1, the basic equipment health status information is generated in the following manner: Carry out data collection, including: temperature T(t), vibration V(t), pressure P(t), current i(t), voltage U(t) and oil quality Q(t), where t is the time variable, i.e. the time series during the collection process; The Kalman filter is used to process the data. Assuming that the original signal collected by the sensor is x i (t), after filtering, the obtained signal is The state update equation of the Kalman filter is: in, represents the signal value after filtering, K(t) represents the Kalman gain, x i (t) represents the original signal value, H represents the observation matrix, The update formula of the prediction covariance is: P(t) = (IK(t)H)P(t-1), where P(t) represents the prediction covariance matrix, I represents the identity matrix, The weighted average method is used for aggregation: S(t)=w T ·T(t)+w V ·V(t)+w P ·P(t)+w I ·I(t)+w U ·U(t)+w Q ·Q(t), Among them, S(t) represents the comprehensive status value of the equipment, w T ,w V ,w P ,w I ,w U ,w Q represents the weight of each parameter, T(t), V(t), P(t), I(t), U(t), Q(t) represent the equipment operation parameters collected, Use MahalanobisDistance to detect abnormal values ​​of equipment operating parameters: Among them, D M (t) represents the Mahalanobis distance at time t, and X(t) represents the operating parameter vector of the equipment at time t, in the form of X(t) = [T(t), V(t), P(t), I(t), U(t), Q(t)] T , μ represents the historical mean vector of each operating parameter, Σ represents the covariance matrix of the operating parameters, If D M (t) exceeds a certain threshold τ, the current device state is considered abnormal. According to the equipment's operating parameters, comprehensive status value S(t) and abnormal detection result D M (t), calculate the health score H(t) of the device, and the health score is defined as: Among them, H(t) represents the health score of the device at time t, ranging from 0 to 1, α, β represent adjustment parameters, S(t) represents the comprehensive status value of the device, and D M (t) represents the Mahalanobis distance of the device, and outputs the basic device health status information.

6. The intelligent management method applicable to a power plant according to claim 5, characterized in that: In step S1, the output basic equipment health status information includes: The real-time health score H(t) of the device, The comprehensive status parameter S(t) reflects the overall operating status of the equipment. And the anomaly detection result D M (t) is used to determine whether the equipment has a potential failure risk.

7. The intelligent management method applicable to a power plant according to claim 6, characterized in that: In step S2, the equipment health status assessment and prediction method is: The real-time operating parameters and historical operating parameters of the equipment are organized into a feature matrix X(t), and the fault label Y(t) is set, X(t) = [T(t), V(t), P(t), I(t), U(t), Q(t)], where T(t) represents the temperature data of the equipment at time t, V(t) represents the vibration data of the equipment at time t, P(t) represents the pressure data of the equipment at time t, I(t) represents the current data of the equipment at time t, U(t) represents the voltage data of the equipment at time t, and Q(t) represents the oil quality data of the equipment at time t. Equipment historical operating parameter data X h As feature input, the device failure label Y h As output, a random forest model is constructed. The model construction process is: Among them, f RF (X h ) represents the output of the random forest model, i.e., the classification result of the predicted device status, X h represents the historical operating parameter feature matrix, T i (X h ) represents the output result of the i-th decision tree, N represents the number of decision trees, Each decision tree T i Make decisions based on different operating parameters of the equipment and use ensemble learning methods to classify the health status of the equipment. Based on the trained random forest model f RF , using real-time operating parameter data X r Evaluate the health status of the equipment. After real-time data is input into the model, the health status prediction of the equipment is obtained: in, Indicates the health status of the device predicted by the model, with a value of 1 for normal or 0 for fault. r represents the real-time operating parameter matrix of the device at time t, The random forest model also outputs the probability of device failure p fault (t): Among them, p fault (t) represents the probability of equipment failure at time t, p i (X r ) represents the failure probability predicted by the i-th decision tree, When p fault When (t) exceeds a certain preset threshold value τ, the device is judged to be in an early failure state and an early warning is triggered. Based on the failure probability p fault (t), generates the device health score H(t), which reflects the current health status of the device: H(t)=1-p fault (t), where H(t) represents the health score of the device at time t, ranging from 0 to 1, and p fault (t) represents the probability of equipment failure, The real-time health score H(t) and historical operation data of the equipment are used to predict the remaining service life RUL of the equipment in combination with the model. Assuming t fail The expected failure time of the equipment: RUL(t)=t fail -t, where RUL(t) represents the remaining useful life of the equipment at time t, and t fail It indicates the time when the equipment is expected to fail, which is predicted by the model, and t indicates the current time.

8. The intelligent management method applicable to a power plant according to claim 7, characterized in that: In step S3, a digital twin model is established and dynamically updated based on the real-time operating parameter data and health score of the equipment: The state of the power plant equipment is modeled based on the real-time operating parameters X(t) and health score H(t) of the equipment. Assuming that the state vector of the equipment is S(t), the state update equation is: S(t+1)=A·S(t)+B·X(t)+C·H(t), where S(t) represents the state vector of the device at time t, A represents the transfer matrix of the device state, X(t) represents the real-time operating parameter vector of the device at time t, B represents the control matrix, which describes the impact of the device operating parameters on the state change, H(t) represents the health score of the device at time t, and C represents the impact matrix of the health score, which describes the impact of the health score on the device state. Construct a load demand model to describe the load demand D(t) of the power plant at different times: D(t)=α1D(t-1)+α2W(t)+α3P e (t)+∈(t), where D(t) represents the load demand of the power plant at time t, D(t-1) represents the load demand at the previous time step, W(t) represents the weather conditions at time t, and P e (t) represents the energy price at time t, α1, α2, α3 represent the sensitivity coefficients of load demand to historical load, weather and energy price, ∈(t) represents the noise or error term of load demand, Build energy consumption and emissions models: Where E(t) represents the energy consumption of the power plant at time t, R(t) represents the emissions of the power plant at time t, and S i (t) represents the state of device i at time t, D(t) represents the load demand of the power plant at time t, γ i represents the energy efficiency coefficient of equipment i, δ i represents the emission coefficient of equipment i, Based on the equipment status S(t), load demand D(t) and health score H(t), multi-objective optimization is performed to maximize equipment efficiency and minimize energy consumption and emissions. The optimization problem is expressed as: Constraints: S(t+1)=A·S(t)+B·X(t)+C·H(t), D(t)=α1D(t-1)+α2W(t)+α3P e (t), Objective function: Minimize the total energy consumption within the time period T and total emissions According to the current state S(t) of the equipment, load demand D(t) and the simulation results of the scheduling strategy, the future equipment state and energy usage are predicted. The future state S(t+k) is calculated recursively by the state equation: Among them, S(t+k) represents the equipment status prediction at time t+k, A k Represents the k-th power of the state transfer matrix, describing the state of the device after k time steps.

9. The intelligent management method applicable to a power plant according to claim 8, characterized in that: In step S4, load scheduling optimization is performed based on DDPG: Define the state s(t), action a(t) and reward function r(t), The state s(t) represents the system state of the power plant at time t, including: s(t)=[H(t),RUL(t),D(t),P e (t)], where H(t) represents the health score of the equipment at time t, RUL(t) represents the remaining useful life prediction of the equipment at time t, D(t) represents the load demand at time t, and P e (t) represents the energy price at time t, Action a(t) represents the load distribution strategy for the device at time t, including the load distribution ratio of each device: a(t)=[α1(t),α2(t),…,α n (t)], where α i (t) represents the load distribution ratio of equipment i at time t, satisfying The reward function r(t) is calculated from equipment efficiency, energy consumption, and emissions. The goal is to maximize equipment efficiency and minimize energy consumption and emissions: Among them, E i (t) represents the energy consumption of device i at time t, R i (t) represents the emission of equipment i at time t, γ i ,δ i represents the energy efficiency and emission weight coefficient of equipment i, The Actor-Critic architecture is adopted, in which the Actor network is used to generate the action a(t), and the Critic network is used to evaluate the value function Q(s(t), a(t)) of the action: Actor network: generates load distribution action a(t) based on the current state s(t), Among them, π θ (s(t)) represents the policy network with parameter θ, generating the load distribution action a(t), represents the noise term, explored using the exploration noise strategy, Critic network: evaluates the value of state s(t) and action a(t), and updates it using the Bellman equation. Q φ (s(t),a(t))=e(r)+γQ φ′ (s(r+1),π θ′ (s(t+1))), where Q φ (s(t), a(t)) represents the Q-value function of the current Critic network, γ represents the discount factor, which measures the contribution of future rewards, φ,φ ′ Represents the Critic network parameters of the current network and the target network, π θ′ (s(t+1)) represents the next action generated by the target network, The Actor network updates the policy network parameters θ based on the Q value evaluated by the Critic network to maximize the Q value: in, Represents the gradient of the policy network, that is, updating the policy network based on the Critic evaluation, represents the gradient of Q value to action a(t), represents the gradient of the action generated by the policy network, Critic network update, based on minimizing the Bellman error, updates the Critic network parameters φ: Among them, L(φ) represents the loss function of the Critic network, and the Q value function is updated based on the Bellman error. Dynamic load distribution strategy: Among them, α i (t) represents the load distribution ratio of equipment i at time t, H i (t) represents the health score of device i, that is, the health status of the device, RUL i (t) represents the remaining useful life of device i, that is, the length of time the device can operate, The DDPG model is based on external conditions, energy price P e (t) and load demand D(t) for dynamic response. The dispatch strategy gives priority to the use of high-efficiency equipment when energy prices are high, and gives priority to the equipment with high health scores when load demand increases. The dynamic response rule is: a(t)=π θ (s(t))if P e (t)<τ e and D(t)>τ d , where τ e represents the threshold of energy price, when the price is lower than τ e When scheduling, τ d Represents the load demand threshold. When the load demand is higher than τ d Time priority scheduling.

10. The intelligent management method applicable to a power plant according to claim 9, characterized in that: In step S5, dynamic predictive maintenance is performed based on the equipment health status and remaining life: According to the health score H of the device i (t) and remaining useful life RUL i (t), calculate the maintenance priority P i (t), the higher the priority, the closer the equipment is to failure, and the maintenance will be prioritized. Among them, P i (t) represents the maintenance priority of equipment i at time t, H i (t) represents the health score of device i at time t, and its value range is [0,1]. The lower the value, the worse the health status. i (t) represents the remaining service life of equipment i at time t, which indicates the length of time before the equipment fails in the future, and α represents the control health score H i (t) The weight of the impact on the priority, β represents the remaining useful life RUL i (t) weight of impact on priority, According to the priority of the device P i (t), select the equipment that needs to be maintained first. Assuming that the power plant can handle k equipment at each time step t, the k equipment with the highest maintenance priority is selected for maintenance. The selected equipment set is M(t): Where M(t) represents the set of equipment that needs to be maintained at time t, and arg represents the maintenance priority P. i (t) is the largest k devices, n is the total number of devices, Calculate the maintenance interval for each device: Among them, Δt i Indicates the next maintenance interval of device i, RUL i (t) represents the remaining useful life of equipment i at time t, H i (t) represents the health score of device i at time t, λ represents the adjustment parameter to control the impact of the health score on the maintenance interval, In the dynamic maintenance plan, the impact of load on the maintenance plan is considered, and the remaining life of the equipment is dynamically updated: RUL i (t+1) = RUL i (t)-μ·L i (t), where RUL i (t+1) represents the remaining useful life of device i at time t+1, RUL i (t) represents the remaining useful life of equipment i at time t, L i (t) represents the load level of equipment i at time t, μ represents the load influence coefficient, which represents the influence of load level on equipment life consumption, Carry out comprehensive maintenance cost optimization, C(t)=∑ i∈M(t) (C maint,i +C downtime,i (t)), where C(t) represents the total cost of the maintenance plan at time t, and C maint,i represents the direct maintenance cost of equipment i, C downtime,i (t) represents the downtime cost of equipment i due to maintenance.

Images