Electric power system optimization scheduling method based on artificial intelligence

Abstract

The invention discloses a power system optimization scheduling method based on artificial intelligence, and relates to the technical field of power system fault testing. The method comprises the steps of obtaining production monitoring data, operation data and management data under a target power system as a power data set, and performing preprocessing; constructing a load prediction model, inputting historical load data, user power consumption behaviors, meteorological data and calendar characteristics, and outputting a predicted power grid load curve; the technical key points are as follows: through combination with an equipment health state evaluation model, dynamic adjustment of health scores and subsequent real-time scheduling optimization, an optimal management scheme with extremely high linkage is formed, the effect of accurately identifying potential fault risks is realized, operation and maintenance personnel can take prevention measures in advance, and the risk of potential faults is prevented. The situation of misjudgment or missing report caused by neglecting external factors is avoided, and the pre-judgment and response capability of the whole scheduling scheme to the equipment health risk is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of power system fault testing, and specifically to an optimized scheduling method for power systems based on artificial intelligence. Background Art

[0002] At present, with the accelerating integration of digital and intelligent technologies and industries, data has leaped to become a new production factor, and artificial intelligence has become a new production tool. Artificial intelligence is used for assistance in power system fault testing and optimized scheduling; the data analysis of the parameters in the power system is to analyze the original data, search for specific types of patterns and trends, define the best parameters for creating a mining model with the analysis results, and apply these parameters to the entire data set to extract feasible patterns and detailed statistical information to obtain more effective information for understanding historical situations, monitoring the current situation, inferring and analyzing reasons, predicting future developments, or assisting in decision-making, etc.

[0003] Currently, traditional power system equipment health management mostly relies on static scoring models or simple threshold-based judgments, lacking an effective response mechanism to dynamic changes in the operating environment, resulting in lagging health scores and high misjudgment rates; for example, in scenarios of sudden load increases, traditional methods are difficult to timely identify the deteriorating trend of equipment health status, causing maintenance lags and even sudden failures, affecting power supply reliability; at the same time, existing solutions generally do not incorporate the load change rate as a key parameter into the scoring model, and even if partial consideration is given to it, it is impossible to achieve dynamic matching between the model and the operating scenario, resulting in poor scoring sensitivity and difficulty in providing effective support for scheduling decisions; in addition, most systems lack a closed-loop feedback mechanism, and model parameters are fixed for a long time or the update steps are inaccurate, unable to learn and optimize from actual operation results, resulting in the continuous existence of scoring deviations. This makes maintenance strategies often rely on empirical judgments, with unreasonable resource allocation, high maintenance costs, and insufficient system stability, seriously affecting the intelligent management level of power systems. Summary of the Invention

[0004] To achieve the above objectives, the present invention is realized through the following technical solutions: An optimized scheduling method for power systems based on artificial intelligence, including the steps of: obtaining production monitoring data, operation data, and management data under the target power system as a power data set, and performing preprocessing; Constructing a load prediction model, inputting historical load data, user electricity consumption behavior, meteorological data, and calendar features, and outputting a predicted grid load curve; synchronously extracting key performance indicators from the operation logs of the equipment and combining them with equipment ledger information as static features; using supervised learning to perform health scoring modeling to obtain an original health score H0; introducing a load factor L based on the load change rate to obtain a final health score H; where L = f(); Build a scheduling optimization objective function, which at least includes: minimizing power generation cost, balancing load, and N-1 security criterion. After constraint modeling, start the learning optimization algorithm and output the scheduling execution plan; according to the scheduling execution result, calculate the deviation of the equipment health score between the simulation and the actual scheduling execution; when the health score deviation exceeds the set index, perform parameter correction actions based on the health score deviation; otherwise, do not perform response actions.

[0005] Further, the production monitoring data at least includes: power generation, voltage and current, reserve capacity, equipment ledger information, and operation logs; the operation data at least includes: electricity price, user electricity consumption behavior, power sales situation, and load curve; the management data at least includes: personnel configuration, project progress, maintenance resources, and cost data.

[0006] Further, the content of preprocessing the power dataset is: using machine learning anomaly detection algorithms to identify and repair the abnormal data in the power dataset, converting the power dataset into a unified data structure, and synchronizing the time series of the power dataset.

[0007] Further, the meteorological data at least includes: temperature and humidity, wind speed, and light intensity, and the calendar features at least include: holidays and seasons; When building a load forecasting model, either LSTM or Transformer network is adopted; the sliding window method is used to construct training samples, and the data input into the load forecasting model is divided into a training set, a validation set, and a test set.

[0008] Further, the key performance indicators extracted from the equipment operation logs at least include: temperature, vibration, oil pressure, insulation resistance, and load rate; the combined equipment ledger information at least includes: model, manufacturing year, and maintenance times; When building a health score model, the content of feature extraction also includes: labeling whether a failure occurs to obtain a failure label; Among them, 0 indicates normal and 1 indicates failure.

[0009] Further, the process of constructing the equipment health score using supervised learning is as follows: Taking the failure label as the dependent variable and the key performance indicators and static features as the independent variables, train a classification model, and the classification model outputs: the probability of the equipment failing at a certain future moment, and map the predicted failure probability to a health score; The formula is: the original health score H0 = 100 * P fault ; where P fault represents the failure probability.

[0010] Further, the basis for obtaining the final health score H is as follows: The health score H is expressed as a combination of the original health score H0 and the load factor L: H = H0 + α * L; where α is the weight, with a value range from 0 to 1; and the function f is set as follows: According to different constraint conditions: Condition 1: The load change rate is within the standard threshold {Thd1, Thd2}; Condition 2: The load change rate is higher than Thd2; Condition 3: The load change rate is lower than Thd1; When Condition 1 is met, the function f adopts the basic model: ; In the formula, fl represents the future load, cl represents the current load, ld represents the load, and t represents the unit time; represents the load change rate; w1, w2 represent the weight coefficients, and their value ranges are both greater than 0; When Condition 2 is met, the function f adopts the non - linear enhancement model: ; When Condition 3 is met, the function f adopts the smoothing model: .

[0011] Furthermore, the content of the constraint modeling at least includes: Generator set start - stop constraint: Each generator is set with requirements for start - up time and stop time; Transmission line capacity constraint: Each transmission line is equipped with a maximum transmission power limit; Equipment health status warning constraint: When the health score of the equipment exceeds the set threshold, a warning is triggered.

[0012] Furthermore, the learning and optimization algorithms for start - up include: Deep Q - network and genetic + neural network hybrid optimization algorithms; The content of the scheduling execution plan at least includes: Generate the generator set start - stop plan: According to the current load demand and equipment health status, adjust the start - stop plan of the generator set; Load distribution plan: Distribute the load of each region to balance the load; Equipment scheduling strategy: The final health score H is inversely proportional to the load bearing capacity. When the health score H of the equipment exceeds the set threshold by a multiple Q, the equipment is suspended. When there are standby equipment, they are enabled; where the value range of Q is greater than 1.

[0013] Furthermore, the health score deviation: ΔH = H_1 - H_2; where H_1 represents the health score of the equipment actually scheduled and executed, and H_2 represents the health score of the equipment simulated and scheduled. The executed parameter correction actions are used to feedback and correct the weight coefficients in the target model adopted in the corresponding function f, and the content is as follows: ; In the formula, η represents the learning rate, and the value range of η is less than 0.01.

[0014] The present invention provides an optimized scheduling method for a power system based on artificial intelligence, which has the following beneficial effects: (1) By combining the equipment health status evaluation model, dynamically adjusting the health score, and subsequent real-time scheduling optimization, a highly coordinated optimization management scheme is formed, achieving the effect of accurately identifying potential fault risks, enabling maintenance personnel to take preventive measures in advance, avoiding misjudgment or missed reports caused by ignoring external factors, and solving the problem in traditional countermeasures that it is often difficult to accurately predict equipment failures due to the lack of an effective dynamic adjustment mechanism, resulting in increased maintenance costs and decreased system stability; (2) The load change rate is introduced as a key parameter in this scheme; On the one hand, it can reflect the stability and volatility of the operation state of the power system. According to the load change rate, the system operation state is divided into three types of constraint conditions. By introducing a constraint classification mechanism, an accurate matching between the function model and the actual operation scenario is achieved, avoiding the errors caused by using a single fixed function, and then improving the stability and credibility of the subsequent health score scheme during operation; on the other hand, as a function input parameter, the load change rate strengthens the sensitivity and accuracy of the load factor calculation, enabling the health score to reflect the direct impact of load changes on the equipment health state in real time, enhancing the sensitivity and prediction ability of the score, and solving the technical problem that the traditional static score model cannot timely warn of the deterioration of equipment health; (3) This scheme combines the load change rate with supervised learning and dynamic function modeling to realize a health score generation mechanism under multi-dimensional information fusion, which can achieve early warning, optimal resource allocation, and assist scheduling decisions through health scores, helping to reduce power outages caused by sudden failures; (4) Through the application of the feedback correction mechanism, on the one hand, the function model has the learning ability and can gradually approximate the real equipment response law, reflecting the self-adaptability of the AI system; on the other hand, a more accurate load factor L helps to correct the health score model, avoiding underestimating equipment risks due to ignoring external load pressure, and enhancing the prediction and response ability of the entire scheduling scheme to equipment health risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic diagram of the overall process of an optimized scheduling method for a power system based on artificial intelligence in the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0016] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0017] Embodiment 1: Please refer to Figure 1 , this embodiment provides an optimized scheduling method for a power system based on artificial intelligence. The solution given in this method realizes a closed-loop process around "perception - analysis - decision - feedback", integrates three types of big data in the power system (corresponding to production monitoring, operation, and management mentioned later), and realizes load forecasting, equipment health assessment, and scheduling optimization through AI modeling; the specific steps of this method are described as follows: S1. Data collection and integration (perception layer): S1.1. Power data set collection: Obtain production monitoring data, operation data, and management data under the target power system; Among them, the content of production monitoring data is: power generation, voltage and current, reserve capacity, equipment ledger information, and operation logs; sources: can be configured SCADA systems, PMU devices, and smart meters; the content of operation data is: electricity price, user electricity consumption behavior, power sales situation, and load curve; sources: can be configured marketing systems and market trading platforms; the content of management data is: personnel configuration, project progress, maintenance resources, and cost data; sources: can be configured ERP systems, collaborative office platforms, and material management systems; S1.2. Data cleaning and standardization processing: Use machine learning anomaly detection algorithms (such as Isolation Forest) to identify and repair abnormal data existing in the power data set collected in S1.1; and convert power data sets from different sources or systems into: a unified data structure for subsequent processing; among them, the unified data structure refers to: converting power data sets from different sources or systems into a standardized form with consistent formats, types, or units to ensure data compatibility; perform time series synchronization on the power data set to eliminate sampling frequency differences; S1.3. Establish a unified data set: Use distributed storage and computing platforms such as Hadoop, Spark, and Flink to build a power data set (that is, the power data set includes: production monitoring data, operation data, and management data); realize centralized storage and fast retrieval of multi-dimensional data, and support subsequent AI model training and real-time analysis.

[0018] S2. Load Forecasting and Equipment Health Status Modeling (Analysis Layer): S2.1. Build a load forecasting model based on LSTM / Transformer: S2.1.1. Model input features: Historical load data, user electricity consumption behavior, meteorological data, and calendar features; among them, in various types of data of the model input features, historical load data and user electricity consumption behavior (i.e., the clustered load curves) are from the power data set in S1; meteorological data includes temperature and humidity, wind speed, and light intensity, and calendar features include holidays and seasons; for meteorological data and calendar features, they can be obtained through the Internet of Things technology network; S2.1.2. Model selection and training: Adopt LSTM or Transformer network to capture long-term dependencies; use the sliding window method to construct training samples, which are the data input into the load forecasting model, and divide the data input into the load forecasting model into a training set, a validation set, and a test set; and use relevant indicators such as RMSE to evaluate the prediction accuracy; S2.1.3. Model result output: Predict the future grid load curve (i.e., the basis for future load); among them, the future time range can be selected according to needs. In this embodiment, it can be selected from 24 hours to 7 days (i.e., 168 hours) to achieve hourly prediction; in addition, it can also support the formulation of power generation plans, the optimization of peak shaving strategies, and demand response arrangements, which will not be elaborated here; S2.2. Equipment Health Status Assessment Model (PHM Model); S2.2.1. Feature extraction: Extract key performance indicators KPIs from the operation logs of the equipment (equipment used under the target power system): temperature, vibration, oil pressure, insulation resistance, and load rate, etc.; combine the equipment ledger information: model, manufacturing year, and maintenance times as static features, and mark whether a failure has occurred to obtain a failure label (0 means normal, 1 means failure); S2.2.2. Conduct health score modeling: Use supervised learning (such as any one of XGBoost or LightGBM) or unsupervised learning (such as Autoencoder) to conduct health score modeling to obtain the original health score H0; then, introduce the load factor L and dynamically adjust the score according to the results of different stages of grid load forecasting to obtain the final health score H; L = f(), where f is a function that can be designed according to specific circumstances; The process of constructing the equipment health score using supervised learning in this embodiment is as follows: Taking the fault label as the dependent variable and the KPIs and static features as the independent variables, a classification model is trained. What the classification model outputs is the probability of a fault occurring at a certain future moment of the device. The predicted fault probability is mapped to a health score, that is, the probability from 0 to 1 is converted to a score from 0 to 100. The formula is: Original health score H0 = 100 * P fault , where P fault represents the fault probability; for example: the input features of a certain device are predicted by the model to have a fault probability of 0.7 within the next week, then the health score is 70 points, indicating that key attention needs to be paid and maintenance needs to be arranged; this method uses historical data to train the model to achieve the mapping from multi-dimensional features to health scores, and has strong interpretability and practicability.

[0019] The adjusted health score is based on the following: The health score H is expressed as a combination of the original health score H0 and the load factor L: H = H0 + α * L, where α is the weight used to adjust the influence degree of the load factor, and the value range is from 0 to 1. In this embodiment, it is usually taken as 1; In this embodiment, the design method of the function f is as follows: According to different constraint conditions (involving load impact analysis): Condition 1: The load change rate is within the standard threshold {Thd1, Thd2}; Condition 2: The load change rate is higher than Thd2; Condition 3: The load change rate is lower than Thd1; When Condition 1 is met, the function f adopts the basic model: ; In the formula, fl represents the future load, cl represents the current load, ld represents the load, and t represents the unit time length; represents the load change rate (the change amount per unit time); w1 and w2 represent the weight coefficients, and the value range is greater than 0. Each weight coefficient can be adjusted according to specific situations, or the coefficient of variation method can be used to determine; it should be noted that the coefficient of variation method is a method of assigning weights to each index according to the variation degree of the current value and the target value of each evaluation index; if the value of a certain index has a large difference and can clearly distinguish each evaluated object, it means that the resolution information of this index is rich, so a larger weight should be given to this index; conversely, if the values of each evaluated object on a certain index have a small difference, then the ability of this index to distinguish each evaluation object is weak, so a smaller weight should be given to this index; this method directly uses the information contained in each index and calculates the weights of the indexes, so it has objectivity; The logic designed by the above basic model is explained as: When the load change rate is high, w2 should be large; if the load changes smoothly, it mainly depends on ; When condition two is met, the function f adopts a non-linear enhancement model: ; In the formula, the exponential function is used to amplify the influence of high load change rate, where e is the base, that is, the natural constant; is the exponent; The logical explanation is: the exponential function will significantly amplify its influence at high load change rate, causing the health score to rise rapidly. This method is particularly suitable for scenarios that need to strongly respond to high load change rate; When condition three is met, the function f adopts a smoothing model: ; In the formula, the logarithmic function log is used to smooth the influence of low load change rate; The logical explanation is: the logarithmic function will weaken its influence at low load change rate, making the change of health score more gentle; this method is applicable to the situation where the load changes relatively smoothly, avoiding over-sensitivity.

[0020] The reason for the above design of dynamically adjusting the score is as follows: Firstly, it improves accuracy: by considering real-time load conditions, the health score can be made more in line with the actual operating environment, avoiding misjudgment due to ignoring external factors. Secondly, it provides early warning: when the load increases, even if there is no obvious abnormality in the equipment currently, the health score can be adjusted to remind the operation and maintenance personnel to pay attention to potential risks, so as to take preventive measures. Finally, it optimizes resource allocation: based on the dynamically adjusted health score, the maintenance plan can be better arranged to ensure sufficient resources are invested at the most needed time, reducing unnecessary downtime and costs; this design not only improves the accuracy of equipment health management, but also enhances the flexibility and response speed of the entire power system, contributing to more intelligent operation and maintenance management; in summary, the higher the health score H, the worse the health state of the equipment, and they are inversely proportional; and the equipment is the relevant equipment used in the target power system.

[0021] The above solution forms a highly linked optimization management solution by combining the equipment health status assessment model, dynamically adjusting the health score, and subsequent real-time scheduling optimization, achieving the effect of accurately identifying potential fault risks. This method enables the operation and maintenance personnel to take preventive measures in advance, avoiding misjudgment or missed reports caused by ignoring external factors, and solving the problems in traditional technical solutions where it is often difficult to accurately predict equipment failures due to the lack of an effective dynamic adjustment mechanism, resulting in increased maintenance costs and decreased system stability; Among them, for the load change rate; On the one hand, it reflects the stability and volatility of the operation state of the power system and is a key parameter for judging whether the operating environment where the equipment is located is "severe". In this solution, the system operation state is divided into three types of constraint conditions according to the load change rate. By introducing this constraint classification mechanism, an accurate matching between the function model and the actual operation scenario is achieved, avoiding the errors caused by using a single fixed function, and then improving the stability and credibility during the operation of the subsequent health scoring scheme; On the other hand, as a function input parameter, the load change rate strengthens the sensitivity and accuracy of the load factor calculation. This design enables the health score to reflect the direct impact of load changes on the equipment health state in real time, enhancing the sensitivity and prediction ability of the score; especially in the case of sudden load increase or decrease, potential risks can be quickly identified, and by adjusting the health score, the operation and maintenance personnel are reminded to take preventive measures, solving the technical problem that the traditional static scoring model cannot timely warn of the deterioration of equipment health; In summary, in the power system, the health state of equipment is not only affected by its own aging degree but also deeply restricted by the external operation environment (such as load changes); therefore, relying solely on static features or historical fault data for health scoring often cannot comprehensively reflect the true state of the equipment; this solution introduces the load change rate as a key variable, combines supervised learning and dynamic function modeling, and realizes a health scoring generation mechanism under multi-dimensional information fusion, which can achieve early warning, optimize resource allocation, and assist scheduling decisions through health scoring, helping to reduce power outages caused by sudden failures.

[0022] S3. AI-Driven Scheduling Optimization Decision-Making (Decision-Making Layer): S3.1. Construct a scheduling optimization objective function (aiming to balance multiple key indicators): Minimize the generation cost, balance the load, and the N-1 security criterion; It should be noted that minimizing the generation cost: By optimizing the start-stop plan of the generating units, select the generating combination with the minimum cost; balancing the load: Allocate the load of each region to avoid local overload or underload; the N-1 security criterion: Ensure that even in the case of a failure of a certain key component, the entire power system can still operate normally, which is used to ensure power supply reliability; S3.2. Constraint Modeling (considering multiple constraint situations): Generating unit start-stop constraint: Each generator is set with requirements for start time and stop time to avoid frequent start and stop; Transmission line capacity constraint: Each transmission line has a maximum transmission power limit to avoid exceeding its design capacity; Equipment health state warning constraint: When the health score of the equipment exceeds the set threshold, a warning is triggered; S3.3. Start the learning optimization algorithm: The learning optimization algorithms in operation include: Deep Q Network and Genetic + Neural Network Hybrid Optimization Algorithm; Among them, the Deep Q Network: gradually learns the optimal strategy by simulating the decision-making process of dispatchers in complex environments; the Genetic + Neural Network Hybrid Optimization Algorithm: searches for the global optimal solution by integrating or combining the global search ability of genetic algorithms and the powerful fitting ability of neural networks; S3.4. Dispatch Output: The dispatch execution output is obtained based on the learning optimization algorithm, and it includes: Generate unit start-stop plan: Adjust the start-stop plan of the generating units according to the current load demand and equipment health status; Load distribution plan: Distribute the loads of each region to balance the load; Equipment dispatch strategy: The lower the equipment health score, the more load it is arranged to bear, and the health score is inversely proportional to the load-bearing amount; When the health score of the equipment exceeds Q times the set threshold, the equipment is suspended, and when there are standby equipment, they are enabled; Among them, Q > 1. For example: 2 times the set threshold. If the health score of the equipment exceeds 2 times the set threshold, it means that the equipment is faulty or approaching a fault and needs to be shut down for maintenance.

[0023] The above solution adopts the technical solution of combining a load forecasting model with reinforcement learning or a hybrid optimization algorithm, achieving the ability to efficiently respond to emergencies. This enables the power system to quickly make adjustments when facing extreme weather or other unforeseen events, solves the related problems that traditional dispatch optimization methods only rely on fixed rules, have poor flexibility, are difficult to quickly make effective responses in case of emergencies, and are prone to large-scale power outages, and realizes and ensures power supply reliability.

[0024] S3.5. Feedback Correction Mechanism (Feedback Layer): According to the dispatch execution result, calculate the deviation of the equipment health score between the simulation and the actual dispatch execution: ΔH = H_1 - H_2, where ΔH represents the deviation of the equipment health score, H_1 represents the equipment health score of the actual dispatch execution, and H_2 represents the equipment health score of the simulated dispatch execution; When ΔH exceeds the set index (indicating that the deviation is outside the acceptable range), parameter correction actions are taken based on the health score deviation to feedback and correct the weight coefficients in the target model adopted by the function f in S2.2.2. The content is as follows: ; In the formula, η represents the learning rate, which controls the amplitude of each correction (usually set to 0.01 or smaller); The above two calculated results, namely w1' and w2' represent the corrected weight coefficients; Logical description: The deviation of the health score directly reflects the actual impact of the scheduling behavior on the health status of the equipment. If there is a significant deviation between the simulated health score and the actual result, it indicates that the load factor function model fails to accurately reflect the response law of the equipment in the real environment. By using this deviation for reverse correction, the weights of the function model can be dynamically adjusted to better adapt to different equipment types, operating environments, and load change patterns. Conversely, no response action is taken. In summary, the entire content of the S3 solution can provide more efficient guidance for the subsequent triggering of early warnings through training in a simulation environment and optimization using the scheduling execution results. Effect description: Through the application of the feedback correction mechanism, on the one hand, the function model is enabled to have learning ability and can gradually approach the real equipment response law, reflecting the self-adaptability of the AI system. On the other hand, a more accurate load factor L helps to correct the health score model, avoiding underestimating the equipment risk due to ignoring the external load pressure and enhancing the prediction and response ability of the entire scheduling solution to equipment health risks.

[0025] Embodiment 2: Based on Embodiment 1, this embodiment also provides an optimized scheduling system for a power system based on artificial intelligence. The system includes various function modules or units that operate in sequence, specifically as follows: Perception and acquisition unit: Obtain production monitoring data, operation data, and management data under the target power system as a power data set and perform preprocessing. Analysis and management unit: Build a load forecasting model, input historical load data, user electricity consumption behavior, meteorological data, and calendar features, and output the predicted grid load curve. Synchronously extract key performance indicators from the operation logs of the equipment and combine them with the equipment inventory information as static features. Use supervised learning to build a health score model to obtain the original health score H0. Introduce the load factor L based on the load change rate to obtain the final health score H. Among them, L = f(current load, future load). Decision and feedback unit: Build a scheduling optimization objective function, including at least: minimizing the generation cost, balancing the load, and the N - 1 security criterion. After constraint modeling, start the learning optimization algorithm to output the scheduling execution plan. Calculate the deviation of the equipment health score between the simulation and the actual scheduling execution according to the scheduling execution result. When the health score deviation exceeds the set index, perform parameter correction actions based on the health score deviation; conversely, no response action is taken.

[0026] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those of ordinary skill in the art will appreciate that the units and algorithm steps of the examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, or in a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution.

[0027] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units. They may be located in one place or distributed over multiple network units. Some or all of the units can be selected according to actual needs to achieve the objectives of the solution of this embodiment.

[0028] As described above, the specific implementation manners of the present application are only provided, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered by the protection scope of the present application.

Claims

1. An artificial intelligence-based optimal scheduling method for power systems, characterized in that, The steps include: obtaining production monitoring data, operation data, and management data under the target power system as a power data set and performing preprocessing; Constructing a load forecasting model, inputting historical load data, user electricity consumption behavior, meteorological data, and calendar features, and outputting the predicted grid load curve; Synchronously extracting key performance indicators from the operation logs of the devices and combining them with the device ledger information as static features; Using supervised learning to build a health score model to obtain the original health score H0; introducing a load factor L based on the load change rate to obtain the final health score H; where L = f(); f is a function; Constructing a scheduling optimization objective function, which at least includes: minimizing the power generation cost, balancing the load, and the N-1 security criterion. After constraint modeling, start the learning optimization algorithm to output the scheduling execution plan; according to the scheduling execution result, calculate the deviation of the device health score between the simulation and the actual scheduling execution; when the health score deviation exceeds the set index, then perform parameter correction actions based on the health score deviation; otherwise, do not perform response actions.

2. The method for optimizing the power system scheduling based on artificial intelligence according to claim 1, wherein: The production monitoring data at least includes: power generation, voltage and current, reserve capacity, device ledger information, and operation logs; the operation data at least includes: electricity price, user electricity consumption behavior, power sales situation, and load curve; the management data at least includes: personnel configuration, project progress, maintenance resources, and cost data.

3. A method for optimizing the scheduling of a power system based on artificial intelligence according to claim 1, characterized in that: The content of preprocessing the power data set is: using a machine learning anomaly detection algorithm to identify and repair the abnormal data existing in the power data set, converting the power data set into a unified data structure, and performing time series synchronization on the power data set.

4. A power system optimal scheduling method based on artificial intelligence according to claim 1, characterized in that: The meteorological data at least includes: temperature and humidity, wind speed, and light intensity, and the calendar features at least include: holidays and seasons; When constructing the load forecasting model, either the LSTM or the Transformer network is adopted; the sliding window method is used to construct training samples, and the data input into the load forecasting model is divided into a training set, a validation set, and a test set.

5. The optimization scheduling method of a power system based on artificial intelligence according to claim 1, characterized in that: Extracting key performance indicators from the operation logs of the devices at least includes: temperature, vibration, oil pressure, insulation resistance, and load rate; the combined device ledger information at least includes: model, manufacturing year, and maintenance times; When building the health score model, the content of feature extraction also includes: marking whether a failure occurs to obtain a failure label; Among them, 0 represents normal and 1 represents failure.

6. The optimization scheduling method of a power system based on artificial intelligence according to claim 5, characterized in that: The process of constructing the device health score using supervised learning is as follows: Taking the failure label as the dependent variable and the key performance indicators and static features as the independent variables to train a classification model, and the classification model outputs: the probability of failure that the device will occur at a certain future moment, and maps the predicted failure probability to a health score; The formula is: original health score H0 = 100 * P fault ; where P fault represents the failure probability.

7. A method for optimizing the scheduling of a power system based on artificial intelligence according to claim 1, characterized in that: The basis for obtaining the final health score H is as follows: The health score H is expressed as a combination of the original health score H0 and the load factor L: H = H0 + α * L; where α is the weight, and the value range is from 0 to 1; and the setting process of the function f is as follows: According to different constraint conditions: Condition 1: The load change rate is within the standard threshold {Thd1, Thd2}; Condition 2: The load change rate is higher than Thd2; Condition 3: The load change rate is lower than Thd1; When condition one is met, the function f adopts the basic model: ； Where, fl represents the future load, cl represents the current load, ld represents the load, and t represents the unit time duration; represents the load change rate; w1 and w2 represent the weight coefficients, and their value ranges are both greater than 0; When condition two is met, the function f adopts the non-linear enhancement model: ； where e represents the natural constant; When condition three is met, the function f adopts the smoothing model: 。 8. The optimization scheduling method of a power system based on artificial intelligence according to claim 1, characterized in that: The content of the constraint modeling includes at least: start-stop constraints of generating units: each generator is set with requirements for start time and stop time; transmission line capacity constraints: each transmission line is equipped with a maximum transmission power limit; equipment health status warning constraints: when the health score of the equipment exceeds the set threshold, a warning is triggered.

9. The method for optimizing the scheduling of a power system based on artificial intelligence according to claim 1, characterized in that: The learning optimization algorithms for startup include: deep Q network and genetic + neural network hybrid optimization algorithm; The content of the scheduling execution plan includes at least: Generate the start-stop plan of generating units: adjust the start-stop plan of generating units according to the current load demand and equipment health status; load distribution plan: distribute the load of each region to balance the load; equipment scheduling strategy: the final health score H is inversely proportional to the load-bearing capacity. When the health score H of the equipment exceeds the set threshold by a multiple Q, the equipment is suspended. When there are standby equipment, they are enabled; where the value range of Q is greater than 1.

10. A method for optimizing the scheduling of a power system based on artificial intelligence according to claim 7, characterized in that: Health score deviation: ΔH = H_1 - H_2; where H_1 represents the health score of the equipment actually scheduled and executed, and H_2 represents the health score of the equipment simulated and scheduled. The executed parameter correction action is used to feedback and correct the weight coefficient in the target model adopted in the corresponding function f, and the content is as follows: ； where η represents the learning rate, and the value range of η is less than 0.01; w1' and w2' represent the corrected weight coefficients.

Images