A high-pressure unit integrated control method and system

Abstract

The present application relates to the technical field of high-pressure unit control system, and particularly relates to a high-pressure unit integrated control method and system, wherein the high-pressure unit integrated control method comprises: obtaining start-stop costs and maximum loads of each generator in the high-pressure unit; predicting expected load demand of the high-pressure unit in a future period; based on the expected load demand and the maximum loads of each generator, selecting a target generator combination satisfying the expected load demand from each generator with the minimum overall start-stop cost of the high-pressure unit as an optimization target; and performing integrated control on each generator in the target generator combination. The present application realizes the mode change of the high-pressure unit from the current optimal to the period optimal by fusing future load prediction and start-stop cost optimization, reduces the overall operation cost and improves the energy efficiency level under the premise of guaranteeing power supply reliability.

Description

Technical Field

[0001] This invention relates to the field of high-voltage generator control system technology, specifically to an integrated control method and system for high-voltage generators. Background Technology

[0002] High-voltage generating units refer to parallel generator systems that output high voltage levels. They typically consist of multiple high-voltage generators that operate in parallel to achieve power aggregation and mutual backup, directly connecting to the high-voltage power grid to meet the demands for large-capacity, high-reliability power supply. Among these, hydroelectric turbine generator units in hydropower stations, with their rapid start-up and shutdown and flexible operation, play a crucial role in peak shaving, frequency regulation, and emergency backup within the power grid, significantly improving the safety, stability, and economic efficiency of the power system.

[0003] The control of existing high-voltage generator units in hydropower stations generally adopts a discrete architecture, with three independent modules: the unit's LCU control panel, the excitation control panel, and the high-voltage switchgear. Integrating these three modules into a single "high-voltage generator integrated control panel" can significantly reduce signal transmission loss and interference, while also significantly improving grid connection speed and accuracy. However, in practical engineering applications, existing high-voltage generator control methods typically base their decisions solely on the current real-time load status. Under this control method, when there are significant fluctuations in the total load, the existing discrete architecture suffers from signal communication delays between modules. These delays mask the deficiencies of real-time state-based decision-making, preventing the generator from reacting immediately to drastic load fluctuations and thus avoiding severe system oscillations. If a control method integrating the three modules is adopted, the communication delay between modules is significantly reduced. The control system's perception delay of load changes is minimal. If the traditional "current instantaneous load-based" control logic is still used, it will lead to severe electrical oscillations in the entire high-voltage generator system, while also significantly exacerbating the mechanical wear of critical components such as the turbine generator's main shaft and key pins, resulting in high maintenance costs. Summary of the Invention

[0004] This invention provides an integrated control method and system for high-voltage generator units to solve existing problems.

[0005] The integrated control method and system for high-voltage generator units of the present invention adopts the following technical solution: In a first aspect, one embodiment of the present invention provides an integrated control method for a high-voltage generator unit, the method comprising: obtaining the start-up and shutdown costs and maximum loads of each generator in the high-voltage generator unit; predicting the expected load demand of the high-voltage generator unit in a future period; based on the expected load demand and the maximum loads of each generator, selecting a target generator combination from the generators that meets the expected load demand with the goal of minimizing the overall start-up and shutdown cost of the high-voltage generator unit; and performing integrated control on each generator in the target generator combination to meet the expected load demand of the high-voltage generator unit in the future period.

[0006] Further, the step of selecting a target generator combination that meets the expected load demand from among the generators, based on the expected load demand and the maximum load of each generator, with the goal of minimizing the overall start-up and shutdown cost of the high-voltage unit, includes: dividing each generator in the high-voltage unit into a set of originally operating generators and a set of inactive generators; determining the basic cost; wherein the basic cost is the start-up and shutdown cost assuming all generators in the originally operating generator set are shut down; determining the selection cost of each generator; wherein, for the generators in the originally operating generator set, the selection cost is the negative value of the corresponding generator start-up and shutdown cost; for the generators in the inactive generator set, the selection cost is the corresponding generator start-up and shutdown cost; and selecting the target generator combination that meets the expected load demand based on the expected load demand, the maximum load of each generator, and the selection cost of each generator, with the goal of minimizing the overall start-up and shutdown cost.

[0007] Furthermore, the step of selecting the target generator combination that meets the expected load demand based on the expected load demand, the maximum load of each generator, and the selection cost of each generator, with the goal of minimizing the overall start-up and shutdown cost, includes: determining the unit cost coefficient of each generator; sorting each generator based on the unit cost coefficient; and determining the target generator combination according to the sorting result and the expected load demand.

[0008] Further, determining the target generator combination based on the sorting results and the expected load demand includes: selecting generators sequentially according to the sorting results until the cumulative load exceeds the expected load demand to obtain candidate generator combinations; determining a pruning threshold based on the total selection cost of the candidate generator combinations; and optimizing the selection process by pruning based on the pruning threshold to determine the target generator combination.

[0009] Furthermore, determining the selection cost of each generator also includes: for each generator in the original set of operating generators, including the potential start-up and shutdown costs of restarting each generator after it has been shut down in the selection cost of the corresponding generator.

[0010] Furthermore, predicting the expected load demand of the high-voltage generator unit in the future includes: acquiring historical production data and future production plans of the high-voltage generator unit; wherein the historical production data includes historical production plans and historical actual loads for corresponding periods; constructing a load prediction model based on the historical production plans and the historical actual loads; inputting the future production plans into the load prediction model to predict the load change trend of the high-voltage generator unit in the future period; and determining the expected load demand for the future period based on the load change trend.

[0011] Further, determining the expected load demand for the future period based on the load change trend includes: identifying high-load operating segments within the future period based on the load change trend; determining time nodes based on load change characteristics between adjacent high-load operating segments; wherein the time nodes are determined based on the maximum load values ​​between adjacent high-load periods; dividing the future period into at least one sub-period based on the time nodes; and determining the expected load demand for each sub-period based on the maximum load values ​​and a preset tolerance within each sub-period.

[0012] Furthermore, obtaining the start-up and shutdown costs of each generator in the high-voltage unit includes: obtaining equipment loss parameters and maintenance cost parameters of each generator during the start-up process; determining the equivalent start-up loss cost of each generator based on the equipment loss parameters and the maintenance cost parameters, and using the equivalent start-up loss cost as the start-up and shutdown cost.

[0013] Furthermore, the integrated control of each generator in the target generator group includes: performing electrical parameter matching and grid connection operations on each generator to synchronize the output voltage, frequency, and phase of each generator with the bus; and after successful grid connection, distributing the load to each generator based on the optimal efficiency point and real-time load of each generator.

[0014] Secondly, another embodiment of the present invention provides an integrated control system for a high-voltage generator unit, including a host computer and a high-voltage generator unit connected to the host computer, wherein: The high-voltage unit includes multiple generators; The host computer is used to obtain the start-up and shutdown costs and maximum loads of each generator in the high-voltage generator set; predict the expected load demand of the high-voltage generator set in the future; based on the expected load demand and the maximum loads of each generator, with the goal of minimizing the overall start-up and shutdown cost of the high-voltage generator set, select a target generator combination from the generators that meets the expected load demand; and perform integrated control on each generator in the target generator combination to meet the expected load demand of the high-voltage generator set in the future period. The system also includes an integrated control panel, which integrates a unit LCU control module, an excitation control module, and a high-voltage switchgear module to perform the integrated control.

[0015] The beneficial effects of the technical solution of the present invention are: In this embodiment of the invention, the start-up and shutdown costs and maximum loads of each generator in the high-voltage generator set are obtained; the expected load demand of the high-voltage generator set in the future is predicted; based on the expected load demand and the maximum load of each generator, with the minimum overall start-up and shutdown cost of the high-voltage generator set as the optimization objective, a target generator combination that meets the expected load demand is selected from each generator; and the generators in the target generator combination are controlled in an integrated manner to meet the expected load demand of the high-voltage generator set in the future.

[0016] This invention, by integrating future load forecasting and start-up / shutdown cost optimization, enables high-voltage generator units to shift from current optimal to time-period optimal operation, reducing overall operating costs and improving energy efficiency while ensuring power supply reliability. Furthermore, the production plan-driven load forecasting mechanism allows high-voltage generator units to anticipate future load changes and pre-arrange generator start-up and shutdown, effectively avoiding unit oscillations and response lags caused by real-time load fluctuations. Additionally, by quantifying the start-up / shutdown costs of each generator and selecting the target generator combination with the goal of minimizing overall start-up / shutdown costs, it avoids additional energy losses and equipment lifespan reduction caused by frequent generator start-up / shutdown switching. Moreover, by accurately configuring generator combinations based on expected future load demands, it prevents multiple generators from operating under load during off-peak periods, ensuring that each generator operates within its high-efficiency range and reducing mechanical wear caused by underload.

[0017] Furthermore, this invention integrates the three major modules of the traditional discrete architecture—the unit LCU control panel, the excitation control panel, and the high-voltage switchgear—into a single integrated control panel. This allows a single panel to perform the functions of the previous three panels, reducing the number of interconnecting cables between panels, significantly reducing signal transmission loss and interference, and substantially improving grid connection speed and accuracy. Simultaneously, by combining a control method based on future load prediction, it eliminates system electrical oscillations that may be caused by reduced communication latency, achieving synergistic effects between hardware integration and control optimization. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart illustrating the integrated control method for high-voltage generator units provided in this application embodiment; Figure 2 A flowchart illustrating an integrated control method for high-voltage generator units in a specific application scenario provided in this application embodiment; Figure 3 This is a schematic diagram of the architecture of the integrated control system for high-voltage generator units provided in the embodiments of this application. Detailed Implementation

[0020] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following detailed description, in conjunction with the accompanying drawings and preferred embodiments, provides a detailed account of the specific implementation methods, structures, features, and effects of the method proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0022] It should be noted that the integrated control method for high-voltage generator units provided by this invention is based on an integrated control panel at the hardware level. In traditional hydropower station high-voltage generator units, control functions are independently performed by three separate modules: the unit LCU control panel, the excitation control panel, and the high-voltage switchgear. Signals are transmitted between these modules via numerous interconnecting cables, resulting in electromagnetic interference and signal attenuation. Furthermore, the communication delays during grid connection coordination are significant, limiting grid connection speed and accuracy. Therefore, this invention integrates these three modules into a single integrated control panel for high-voltage generator units in its hardware architecture. This integrated control panel specifically includes: The generator set LCU control module is responsible for collecting operating status parameters of the generator set, such as voltage, current, power, frequency, and temperature, executing sequential control (such as start-up and shutdown procedures), protection logic (such as overvoltage, overcurrent, and frequency anomaly protection), and data communication with the host computer. Excitation control module: used to regulate the excitation current of the generator, dynamically control the generator terminal voltage and reactive power, realize the reactive power balance between the generator and the power grid, and cooperate in completing voltage matching during grid connection; High-voltage switchgear module: integrates grid-connected circuit breaker, disconnecting switch, instrument transformer and protection device, used to realize electrical connection and safe disconnection between generator and high-voltage bus, and has the ability to quickly clear faults.

[0023] Through the high integration of the aforementioned modules, the integrated control panel centrally performs the signal acquisition, logic operation, and control command issuance functions that were originally handled by three separate cabinets. This eliminates the need for numerous connecting cables and intermediate interfaces between cabinets, thereby significantly reducing electromagnetic interference and attenuation during signal transmission. Simultaneously, because the communication links between modules have changed from external cables to internal buses or shared memory, the information exchange latency has been reduced from milliseconds to microseconds. This allows for faster voltage, frequency, and phase matching during grid connection, significantly improving the speed and accuracy of grid connection operations.

[0024] However, the low-latency environment brought about by the integrated control panel also places higher demands on the control strategy. In the traditional discrete architecture, the inherent communication delay between modules objectively masks the shortcomings of the "current instantaneous load-based" control logic, causing the generator to not immediately respond to minor load fluctuations, and the system instead gains a certain degree of inertial buffer. When using an integrated control panel, the system is extremely sensitive to load changes. If the traditional instantaneous decision-making logic is still used, it is very easy to cause electrical oscillations in the unit system and exacerbate the mechanical wear of key components such as the turbine generator main shaft and key pins. To this end, this method combines the rapid response capability of the integrated control panel, predicts the load change trend in future periods, and optimizes the scheduling by combining the start-up and shutdown costs of each generator, transforming from "current instantaneous optimal" to "future period global optimal," thereby fundamentally solving the system oscillation problem that may be caused by reduced communication latency and achieving synergistic effects between hardware integration and control strategy.

[0025] The following description, in conjunction with the accompanying drawings, details a specific scheme for an integrated control method for high-voltage generator units provided by the present invention.

[0026] like Figure 1 As shown in the figure, this application provides an integrated control method for high-voltage generator units, including: Step S110: Obtain the start-up and shutdown costs and maximum load of each generator in the high-voltage unit.

[0027] Before step S110 above, historical production data of the high-voltage unit and equipment parameters of each generator can be obtained. The historical production data includes historical production planning and historical actual load data for the corresponding period. The equipment parameters of each generator include historical maintenance records and maintenance costs of each generator, as well as rated load parameters of each generator.

[0028] The aforementioned start-up and shutdown cost is the comprehensive cost incurred by a generator from a shutdown state to grid-connected operation, including the equivalent life loss cost due to mechanical wear and tear during startup. Given that the shutdown cost from a generator shutdown state to restart is much lower than the comprehensive cost incurred during startup, this cost is usually directly quantified as the startup cost, serving as a key indicator for evaluating the economic efficiency of generator state switching. The embodiments of this application can determine the generator start-up and shutdown cost using at least one of the following methods: The first method of determination: based on historical statistical data; In this approach, the actual cost data generated by each generator in the high-voltage unit during each start-up and shutdown operation within its historical operating cycle can be collected and statistically analyzed to establish a correlation between start-up and shutdown costs and equipment operating status. For example, the implementation method involves recording the start-up time, manual assistance costs, and additional maintenance costs incurred in the short term after start-up for each start-up and shutdown process. Regression analysis or arithmetic averaging of the historical data is then performed to determine the unit start-up and shutdown cost of each generator under different operating conditions.

[0029] The second method of determination: based on the equipment depreciation model; In this approach, a quantitative relationship can be established between the generator's total life-cycle depreciation and the frequency of start-stop operations, converting the impact of each start-stop operation on the remaining lifespan of the equipment into economic costs. For example, based on the generator's design lifespan, the statistical relationship between historical start-stop frequency and total operating time, the equivalent operating time loss or lifespan reduction rate caused by a single start-stop operation is calculated. Combined with the equipment's purchase cost and residual value assessment, the equipment depreciation cost allocated to each start-stop operation is determined and added to the direct operating costs during the startup process to form the complete start-stop cost.

[0030] The third method of determination: based on real-time monitoring data; In this approach, a real-time monitoring unit can be deployed during generator startup to dynamically collect and calculate the actual resource consumption of the current start-up and shutdown operation. For example, the implementation method involves real-time monitoring of auxiliary energy consumption, startup time, temperature rise rate of key components, and vibration parameters during startup. Based on the real-time collected energy consumption costs and equipment status parameters, the energy consumption cost and equipment loss assessment value of the current startup process are dynamically calculated, thereby determining the actual cost of this start-up and shutdown operation.

[0031] The fourth method of determination: based on equipment wear and tear and maintenance costs; Optionally, the start-up and shutdown costs of each generator in the high-voltage unit are obtained, including: obtaining equipment loss parameters and maintenance cost parameters of each generator during the start-up process; and determining the equivalent start-up loss cost of each generator based on the equipment loss parameters and the maintenance cost parameters, as the start-up and shutdown cost.

[0032] Equipment loss parameters characterize the equivalent loss to the mechanical life of the generator during a single start-up process. They are quantified using the average equivalent operating time of the single start-up loss, denoted as [missing information]. Its dimension is time. Equipment loss parameters can be calculated and determined using statistical regression methods based on the operating time, start-stop frequency, and total maintenance cost matrix of multiple generators of the same model. Maintenance cost parameters include the annual equivalent total operating time. Total maintenance cost for the corresponding generator in one year ,in The dimension of is time. The unit of measurement is currency. The equivalent cost of startup losses is quantified by converting mechanical lifespan depletion into equivalent maintenance costs; the calculation formula is... This formula transforms mechanical wear into economic costs, providing a quantitative basis for subsequent optimization. This implementation method can convert physical mechanical wear into economic cost quantification, enabling subsequent optimization algorithms to compare the start-up and shutdown economics of different generators under a unified monetary unit. By distinguishing the start-up and shutdown cost differences between the originally operating generator and the inactive generator, it provides a quantitative basis for the optimization model to differentiate the benefits of state switching.

[0033] Of course, the shutdown cost of the generator can also be considered to construct a more complete quantitative model of start-up and shutdown costs. For example, in some application scenarios with higher requirements for cost accounting accuracy, in addition to the equipment life loss during the start-up process, the energy loss, auxiliary operation costs, and additional life loss caused by mechanical stress during the generator shutdown process can be included in the start-up and shutdown cost calculation. In this case, the start-up and shutdown cost is defined as the sum of the start-up cost and the shutdown cost, with both units unified in monetary units, and the total economic indicator of state switching is formed by algebraic superposition.

[0034] The maximum load mentioned above refers to the maximum electrical power that each generator in the high-voltage unit can continuously output under safe and stable operating conditions. It is a key parameter characterizing the generator capacity boundary and is used as a load capacity constraint when selecting generator combinations for subsequent optimization, ensuring that the total load of the selected generator combinations can meet the expected load demand for the corresponding period. The methods for obtaining the maximum load include: (1) directly obtaining the rated load capacity specified in the generator equipment nameplate or technical specifications and using it as the maximum load; (2) determining it based on the actual load peak data recorded by the generator during historical operation and after correction with the safety margin coefficient; (3) deriving its maximum load capacity for sustainable operation through thermal balance calculation based on the generator's design parameters (such as winding temperature rise limit, insulation class and cooling system rated capacity).

[0035] Step S120: Predict the expected load demand of the high-voltage unit in the future period.

[0036] It is understandable that step S120 above, which predicts the expected load demand of the high-voltage generator unit in the future, aims to transform the generator start-up and shutdown control decision from a passive response mode based on the current instantaneous load to an active forward-looking scheduling mode based on the load change trend in the future. By knowing the power demand of the high-voltage generator unit in advance before the actual load change occurs, the start-up and shutdown sequence and combination configuration of each generator can be determined in advance, reserving the necessary response time for the generator start-up process. This avoids control lag and power supply tension caused by temporary generator start-up when the load suddenly increases. At the same time, it expands the unit scheduling strategy from instantaneous local optimum to global optimum in the future, reduces frequent start-up and shutdown operations caused by load fluctuations, and ensures that each generator operates in a load range close to the optimal efficiency point in the corresponding period.

[0037] The embodiments of this application can predict the expected load demand of high-voltage units in future periods using at least one of the following methods: The first method: prediction based on time series statistical models; This method analyzes the time-series dependencies of historical load data of high-voltage generator units to establish a statistical prediction model to infer future load trends. The implementation method is as follows: collect actual load time-series data of high-voltage generator units in the past continuous period, identify trend terms, periodic terms and random fluctuation terms in the data sequence, use an autoregressive integral moving average model to fit and estimate the parameters of the historical load time series, use the estimated model parameters for extrapolation prediction, thereby obtaining the load prediction value at each time point in the future period, and determine the expected load demand based on the extreme values ​​or mean values ​​in the predicted time series.

[0038] The second approach: prediction based on deep learning methods; This approach constructs a deep neural network to learn the nonlinear dynamic features in historical load data, enabling deep fitting and prediction of complex load patterns. For example, it uses multidimensional state parameters (including historical load, ambient temperature, and operating time) from the historical operation of the high-voltage unit as input feature sequences. The network is trained using a network structure with temporal memory capabilities, such as a long short-term memory network or a gated recurrent unit. The network weight parameters are optimized using a backpropagation algorithm. After training, known conditions for future periods are input into the trained network model, and the corresponding load prediction results are output, thereby determining the expected load demand for future periods.

[0039] Long Short-Term Memory (LSTM) networks and gated recurrent units (GRUs) and other temporal neural network structures are well-known to those skilled in the art. The specific implementation details and underlying mathematical principles of their network topology construction, forward propagation computation, backpropagation training, and parameter optimization have been thoroughly described in existing deep learning literature and open-source framework documentation, and are considered mature existing technologies. Therefore, this application will not elaborate on the specific internal structure, training process, and convergence mechanism of the aforementioned network models; relevant technical details can be found in existing technical materials.

[0040] The third method: prediction based on multi-factor regression; This method establishes a mathematical regression relationship between load influencing factors and future electricity demand to achieve load estimation based on planned information. The implementation method includes, for example, collecting production and operation parameters related to the electricity demand of high-voltage generating units (including production schedules, expected operating rates, seasonal electricity consumption patterns, equipment maintenance plans, etc.), performing correlation analysis and feature screening on historical production parameters and actual load data for corresponding periods, constructing a multiple regression equation and estimating the regression coefficients of each influencing factor, and then substituting the production and operation parameters for future periods into the regression equation for calculation to obtain the expected load demand for future periods.

[0041] The fourth approach: a prediction method based on historical production data and machine learning models; Optionally, step S120 may include: acquiring historical production data and future production plans of the high-voltage unit; wherein, the historical production data includes historical production plans and historical actual loads for corresponding periods; constructing a load prediction model based on historical production plans and historical actual loads; inputting the future production plan into the load prediction model to predict the load change trend of the high-voltage unit in the future period; and determining the expected load demand for the future period based on the load change trend.

[0042] The aforementioned historical production data includes the production plan indicators and corresponding actual electricity consumption records of the production system served by the high-voltage generator units within historical time periods. Historical production planning refers to planned parameters such as the planned production tasks, equipment deployment quantities, and process scheduling for each historical period. Its dimensions are determined based on the specific indicator type; for example, planned output can be expressed in units or tons, and equipment deployment quantities can be expressed in units. Historical actual load refers to the time-series data of the actual power output of the high-voltage generator units within the corresponding period, with its dimension being power units. Future production planning refers to the production task schedules and operation plans already established for the period to be predicted, and its data structure is consistent with historical production planning. The load prediction model is a mathematical mapping model trained based on historical production data, capable of converting the input production planning feature vector into the output load prediction value.

[0043] The aforementioned load prediction model can employ an ensemble learning approach based on gradient boosting decision trees. Its main architecture consists of multiple classification and regression trees arranged sequentially to form an additive model. In each iteration, the newly trained decision tree fits the residual gradient of the preceding model on the training data. Overfitting risk is controlled by introducing a regularization term that includes leaf node weights and tree complexity. The construction process of the gradient boosting decision tree is as follows: Historical production plans are feature-encoded to obtain the model input feature matrix. A loss function is defined using the historical actual load as the regression target value. After initializing constant prediction values, it enters the iterative training phase. In each iteration, the negative gradient between the current model output and the actual load is calculated as a pseudo-residual. A new decision tree is fitted based on this pseudo-residual, and the optimal leaf weights are solved. The contribution of the new tree is scaled using a shrinkage coefficient and then accumulated into the existing model until a preset number of iterations is reached or the loss function converges. Finally, a prediction model composed of the weighted sum of multiple trees is output.

[0044] The aforementioned load prediction model can also employ an ensemble learning approach based on random forests. This can utilize a Bagging framework to construct multiple decorrelational decision trees in parallel. Bootstrapping and random feature subspace methods enhance the diversity among base learners, and the arithmetic mean of the prediction results from multiple trees is used as the final output to reduce prediction variance. The random forest construction process involves, for example, bootstrapping with replacement from historical production data to generate several training subsets. Decision trees are independently constructed on each subset, and feature subsets are randomly selected when splitting nodes to determine the optimal split point. The growth depth of a single tree or the minimum number of samples per leaf node is controlled to prevent overfitting. After all decision trees are trained, future production plans are input into each decision tree to obtain their respective load prediction values. The average of the prediction results from all trees is then used as the final load prediction output for the future period.

[0045] The aforementioned load prediction model can also be implemented using a feedforward neural network based on a multilayer perceptron. Its main architecture includes a sequentially connected input layer, several hidden layers, and an output layer. Information is transferred between layers via a fully connected weight matrix, and a nonlinear activation function is used to transform the input features layer by layer to learn the complex nonlinear mapping relationship between production planning and load. The construction process of this feedforward neural network is as follows: determine the number of input layer nodes based on the feature dimensions of historical production planning; design a single-hidden-layer or multi-hidden-layer network topology with an appropriate number of neurons and select an activation function; normalize the input features and target load to stabilize the training process; initialize the network weights and calculate the model output through forward propagation; use mean squared error as the loss function to measure prediction bias; use the backpropagation algorithm to calculate gradients layer by layer and update the weight parameters; introduce learning rate decay and early stopping mechanisms during iterative optimization until the validation set loss no longer decreases, thus obtaining the trained load prediction model.

[0046] It is understood that the aforementioned gradient boosting decision trees, random forests, and multilayer perceptrons are all existing technologies known to those skilled in the art. Their network topology, loss function design, optimization algorithms, hyperparameter tuning strategies, and convergence analysis, along with their underlying mathematical principles, have been systematically described in existing technical literature, open-source algorithm frameworks, and engineering practice manuals. Therefore, this application will not elaborate on the internal structure, training process, inference mechanism, and parameter configuration methods of the aforementioned load prediction models; relevant technical content can be found in existing publicly available materials.

[0047] Optionally, the above-mentioned determination of expected load demand for future periods based on load change trends includes: identifying high-load operating segments within the future period based on load change trends; determining time nodes based on load change characteristics between adjacent high-load operating segments; wherein the time nodes are determined based on the maximum load values ​​between adjacent high-load periods; dividing the future period into at least one sub-period based on the time nodes; and determining the expected load demand for each sub-period based on the maximum load values ​​and preset tolerances within each sub-period.

[0048] It is understandable that the above scheme divides the future time period into several sub-time periods, aiming to match the granularity of high-voltage unit start-up and shutdown scheduling with the actual rhythm of future load changes, thereby avoiding the coarseness of control strategies caused by using the entire future time period as a single decision unit. Since the load in the future time period often exhibits non-uniform fluctuation characteristics, if the generator combination is uniformly configured based solely on the global maximum load within that time period, multiple generators will be in an underloaded state for a long time in sub-time periods with relatively low loads, resulting in decreased energy efficiency and mechanical damage. Conversely, by identifying high-load operating periods and determining time nodes based on the load change characteristics between adjacent high-load operating periods, the continuous future time period can be divided into sub-time periods with relatively independent load characteristics. This allows each sub-time period to independently select the optimal generator combination based on its specific expected load demand, preventing frequent start-ups and shutdowns and system oscillations caused by load fluctuations, ensuring that each generator operates in a load range close to its optimal efficiency point within the corresponding sub-time period, and reserving necessary response time for generator start-up and shutdown based on time nodes, thus achieving a unity of forward-looking control and fine-grained scheduling.

[0049] The aforementioned high-load operating periods refer to continuous time intervals where power consumption remains at a high level within the future load change trend. This is typically identified by comparing predicted load values ​​with preset power thresholds and performing cluster analysis on extreme points that meet the threshold conditions. These periods represent time segments where power demand is highly concentrated during production. The purpose of identifying high-load operating periods is to accurately locate the peak load intervals and their boundaries within the future period, providing a reference benchmark for analyzing transition intervals between adjacent high-load periods. This allows for the rational determination of time nodes based on load change characteristics between adjacent high-load operating periods, thus scientifically dividing the future period into at least one relatively independent sub-period. This ensures that the generator configuration for each sub-period matches the actual load demand within that period, avoiding issues such as generator underload during off-peak periods or insufficient capacity during peak periods caused by using the entire future period as a unified scheduling unit. Simultaneously, it allows for reasonable response time for generator start-up and shutdown control, achieving dynamic adaptation between high-voltage unit operation strategies and load fluctuation rhythms.

[0050] The embodiments of this application can identify high load running periods through at least one of the following identification methods: (1) First identification method: identification method based on fixed power threshold and density clustering; In this embodiment, high load candidate points are screened by setting a power ratio threshold, and the candidate points are clustered based on density reachability to extract continuous high load intervals; First, the load change trend of future periods is smoothed to eliminate random fluctuation interference, and the time point when the smoothed load value is greater than 50% of the preset limit power is determined as a high load candidate point. The time coordinates of the load extreme points corresponding to these candidate points are extracted, and the DBSCAN (density-based spatial clustering) algorithm is used to perform spatial clustering of the extreme points. The density-connected extreme points are assigned to the same cluster. Each effective cluster represents a high load running period, and the start and end times of this period are defined by the earliest and latest time points in the cluster. (2) The second identification method: the adaptive judgment method based on statistical quantiles; This implementation method dynamically determines the high load judgment boundary by analyzing the load probability distribution characteristics to adapt to baseline fluctuations; calculates the upper quantile statistics of the load change trend in the future period, initially marks the continuous interval with the load value greater than the set upper quantile as the high load candidate interval, introduces the minimum duration constraint to screen the candidate interval, filters out the instantaneous peak interval with the duration less than the preset minimum time window, and determines the continuous load segment that meets the duration condition as the high load running segment, so that the identification standard can be adaptively adjusted with the change of the overall load level. (3) The third identification method: the identification method based on slope analysis and extreme value detection; this embodiment identifies the starting point and peak plateau period of load rise by analyzing the local change characteristics of the load curve; calculates the first difference of the load change trend to obtain the load change rate, identifies the inflection point of the load change rate from positive to negative as the local load maximum, sets the load amplitude threshold to eliminate local maxima below the threshold, and determines the high load running period between adjacent local maxima where the load is continuously higher than the threshold and the change rate tends to be flat. This method focuses on capturing the structural characteristics of the load entering the high power plateau stage. (4) Fourth identification method: identification method based on energy accumulation of sliding window; This embodiment distinguishes between continuous high load intervals and instantaneous power peaks by calculating the load accumulation effect within a unit time window; a fixed length time sliding window is set, and the cumulative amount of load over time within the window is calculated as the energy index of the window. Windows with energy indices greater than a preset energy threshold are marked as high energy windows. Adjacent high energy windows are merged and their continuity is checked. The time range covered by consecutively appearing high energy windows is determined as the high load running period, thereby avoiding misjudging isolated instantaneous load impacts as continuous high load running states.

[0051] An optional implementation method for determining time nodes based on load change characteristics between adjacent high-load operating periods includes: analyzing the predicted load change trend. Smoothing is performed to eliminate random fluctuations and noise interference, resulting in a smoothed load change trend. On the smoothed load curve, for the transition interval between two adjacent high-load operating periods, a traversal analysis is performed along the direction of load decrease to identify the load maxima during the relatively low-load period within this interval. When three load maxima that meet the matching conditions are identified between adjacent high-load operating periods, the time coordinate corresponding to the leftmost load maxima is marked as a time node; simultaneously, the time coordinate corresponding to the rightmost load maxima is shifted forward by 1.2 times the maximum generator start-up time and marked as another time node. This forward shifting operation aims to reserve necessary response time for generator start-up control. The time interval between the current time node and the next time node is denoted as... This time interval serves as the basic time unit for subsequently dividing the generator operating periods and determining the expected load demand for each sub-period.

[0052] After obtaining the time node, a lower bound can be set for the generator's operating duration, denoted as . The lower bound is the maximum of the minimum continuous operating time and the minimum downtime among all generators, that is: This is to prevent the generator from frequently switching between start-stop states due to excessively short operating time. When a certain time node is reached, the time interval between the current time node and the next time node is recorded as _____. Based on this, the operating time of the generator can be determined. The working time is taken from the aforementioned lower bound. With time interval The larger of the two, i.e. Subsequently, based on the predicted future load change trends... The duration of the interception is The load trend segment is denoted as This serves as the basis for load analysis within the sub-period, thereby dividing the continuous future period into at least one sub-period with a clear duration boundary and independent load characteristics, enabling each sub-period to independently configure generator combinations based on its specific load trend segment.

[0053] For each sub-period, the corresponding load trend segment Identifying maximum load values This maximum load value represents the peak level of electricity demand during this sub-period. Considering that the actual load in production often deviates from the predicted value, to prevent insufficient generator capacity due to sudden load increases, in... Based on this, a preset tolerance ratio is added as a safety margin, and the result of the addition is recorded as the expected maximum load. The expected maximum load This serves as the expected load demand for that sub-period, guiding the selection of subsequent generator combinations and ensuring that the selected generator combinations can still meet power supply requirements under load fluctuations.

[0054] Step S130: Based on the expected load demand and the maximum load of each generator, with the goal of minimizing the overall start-up and shutdown cost of the high-voltage unit, select the target generator combination that meets the expected load demand from among the generators.

[0055] The target generator combination can be obtained through at least one of the following solution methods in step S130 above: The first solution method: a step-by-step selection method based on a greedy strategy; This implementation method uses iterative selection of locally optimal devices to quickly construct a feasible solution. All generators are sorted in ascending order of start-up and shutdown costs. An empty set is initialized as a candidate combination. Each generator is traversed sequentially, and the device with the lowest start-up and shutdown cost that has not yet been selected is selected each time. This device is included in the candidate combination, and its maximum load is accumulated until the accumulated load is greater than or equal to the expected load requirement for the first time. The device set obtained at this time is the target generator combination. This method has low computational complexity and is easy to implement. It is suitable for scenarios with high requirements for solution timeliness and relatively low requirements for accuracy.

[0056] The second solution method: adopting an exact search method based on the branch and bound method; This implementation method searches for the globally optimal combination in the solution space through systematic enumeration and boundary pruning. The generator selection problem is modeled as a combinatorial optimization problem in a 0-1 decision variable space. The optimization objective is to minimize the overall start-up and shutdown cost, and the constraint is that the cumulative load meets the expected load demand. A search tree is constructed and the selection state of each generator is determined layer by layer. During the search process, the lower bound of the current partial solution is calculated. If the lower bound exceeds the overall start-up and shutdown cost of the currently known optimal solution, the branch is pruned and not expanded again until all valid branches are traversed and the target generator combination with the minimum start-up and shutdown cost is output. This method can guarantee the optimality of the solution, but the computational cost increases exponentially with the number of devices.

[0057] The third solution method: a heuristic search method based on genetic algorithms; This implementation approximates the global optimum through multiple generations of iteration by simulating the biological evolution process. The selection state of each generator is encoded as a binary chromosome string, where each bit corresponds to the selection decision of a generator. After randomly generating an initial population, individuals are evaluated using the overall start-up and shutdown cost as the fitness function. The population evolves generation by generation through selection, crossover, and mutation operations. Load satisfaction rate is introduced as a constraint to penalize infeasible individuals. After reaching a preset number of generations or population convergence, the target generator combination corresponding to the chromosome with the best fitness is output. This method is suitable for scenarios with a large number of devices and a complex solution space.

[0058] The fourth solution method: based on heuristic initial solutions and dynamic programming optimization; Optionally, the above-mentioned selection of a target generator combination that meets the expected load demand, based on the expected load demand and the maximum load of each generator, with the goal of minimizing the overall start-up and shutdown cost of the high-voltage generator unit, includes: dividing the generators in the high-voltage generator unit into a set of originally operating generators and a set of inactive generators; determining the base cost; wherein the base cost is the start-up and shutdown cost assuming all generators in the originally operating generator set are shut down; determining the selection cost of each generator; wherein, for generators in the originally operating generator set, the selection cost is the negative of the corresponding generator start-up and shutdown cost; for generators in the inactive generator set, the selection cost is the corresponding generator start-up and shutdown cost; and selecting a target generator combination that meets the expected load demand based on the expected load demand, the maximum load of each generator, and the selection cost of each generator, with the goal of minimizing the overall start-up and shutdown cost.

[0059] The core of the above solution lies in distinguishing the cost structure of the generator current operating state reconfiguration optimization problem, quantifying the continuous operating benefits of the original operating generator as a negative selection cost, thereby establishing a combinatorial optimization model with the goal of minimizing the overall start-up and shutdown cost, and realizing the global economic optimization of the generator combination configuration for future periods.

[0060] The aforementioned "original operating generator set" refers to the group of generators already in grid-connected operation at the time of the optimization decision. These generators are currently bearing load and outputting power to the grid. The "non-activated generator set," on the other hand, refers to the group of generators currently in standby mode and not yet connected to the high-voltage grid. These generators are in either cold or hot standby mode and require a startup process before they can be put into operation. The division between these two sets is based on the real-time operational status monitoring results of the high-voltage units at the time of the decision and evolves dynamically with the execution of generator start-up and shutdown control.

[0061] The base cost refers to the total start-up and shutdown costs assumed in the optimization model if all generators in the original set of operating generators were shut down. This cost serves as a benchmark reference value for subsequent optimization calculations. Let the base cost be denoted as... Its mathematical expression is ,in This represents the original set of operating generators. Indicates the first The start-up and shutdown costs of a generator are measured in monetary units. The introduction of a base cost aims to establish a unified cost measurement benchmark, allowing any subsequent selection strategy to be economically evaluated based on increments or decrements relative to this benchmark.

[0062] Selection cost refers to the economic cost incurred or saved when selecting a specific generator to include in the target combination during the optimization process. This cost is defined with opposite signs depending on the set to which the generator belongs, to characterize the economic differences in state switching. For the original set of operating generators... The generator in the selection cost Defined as the corresponding generator start-up and shutdown cost The negative value, that is The negative sign indicates that continuing to use the generator avoids the start-up and restart costs required after it has been shut down, reflecting a saving in relative base costs; for a collection of generators that are not in use... The generator in the selection cost Defined as the start-up and shutdown cost of the corresponding generator itself, that is This represents the additional startup cost required to activate the generator. The unit of measurement for this cost is the same as that for start-up and shutdown costs, both in monetary units.

[0063] The above solution method is based on a quantification mechanism of state switching benefits. By setting the selection cost of the originally running generator to a negative value, the optimization model tends to retain the originally running generator to obtain negative cost contribution (i.e., cost savings) when minimizing the overall start-up and shutdown costs, thereby avoiding unnecessary downtime and the energy loss and equipment wear caused by subsequent restarts. At the same time, for the unactivated generators, the positive selection cost ensures that the optimization model only activates them when additional capacity is truly needed to meet the expected load demand, preventing redundant configuration. The sum of the base cost and the selection cost constitutes the total cost term in the optimization objective function, enabling the final selected target generator combination to achieve the minimum net cost relative to the baseline case of completely shutting down the originally running generators, while meeting the expected load demand constraints.

[0064] Optionally, the above-mentioned selection of a target generator combination that meets the expected load demand, the maximum load of each generator, and the selection cost of each generator, with the goal of minimizing the overall start-up and shutdown cost, includes: determining the unit cost coefficient of each generator; ranking each generator based on the unit cost coefficient of each generator; and determining the target generator combination based on the ranking results and the expected load demand.

[0065] The aforementioned unit cost coefficient is a normalized economic indicator characterizing the selection cost corresponding to each unit of maximum load capacity of a generator. It is used to quantitatively evaluate the economic efficiency of different generators when undertaking a unit load task, providing a quantitative basis for subsequent cost-benefit-based ranking and pruning optimization. The unit cost coefficient can be obtained by calculating the ratio of the generator selection cost to the maximum load; its mathematical expression is as follows: ,in Indicates the first The unit cost coefficient of a generator is a ratio of monetary units to power units (e.g., yuan / kilowatt or yuan / megawatt). This represents the cost of selecting the generator, expressed in monetary units. This indicates the generator's maximum load capacity, measured in power units. A negative unit cost coefficient indicates that the corresponding generator is a previously operating generator, and each unit of load capacity brings net cost savings; a positive unit cost coefficient indicates that the corresponding generator is not in use, and each unit of load capacity incurs additional start-up and shutdown costs.

[0066] One possible implementation for determining the unit cost coefficient includes: directly performing a division operation using the already determined selection cost and maximum load parameters; for each generator in the high-voltage unit, obtaining its selection cost and maximum load within the current optimization cycle, performing a real number division operation with the selection cost as the numerator and the maximum load as the denominator, and obtaining the quotient as the unit cost coefficient for that generator. This operation can be efficiently executed in embedded control units with limited computing resources, making it suitable for online optimization scenarios with high real-time requirements. Alternatively, a normalization-based approach can be used to enhance the comparability between generators of different capacity levels by unifying the dimensions and numerical scale; firstly, the maximum load of all generators is normalized, and the maximum load of each generator is divided by the maximum single-unit load in the unit to obtain a relative capacity coefficient; simultaneously, the selection cost is divided by the maximum start-up and shutdown cost in the unit to obtain a relative cost coefficient; then, the ratio of the relative cost coefficient to the relative capacity coefficient is used as the normalized unit cost coefficient. This approach eliminates the interference of absolute numerical differences on the ranking results, enabling economic comparisons of generators of different capacity levels on a unified scale.

[0067] Ranking generators based on unit cost coefficients can be achieved using a direct ranking method based on numerical comparison. This method determines the ranking priority by comparing the values ​​of the unit cost coefficients of each generator. All generators are then arranged in ascending order according to their unit cost coefficient values. Generators with smaller unit cost coefficients rank higher in the ranking sequence, indicating lower net cost per unit load capacity (or greater savings compared to the original operating generator). These generators should be given priority consideration in the subsequent selection process to achieve the maximum load fulfillment rate or maximum cost savings per unit investment.

[0068] The above scheme achieves a fair comparison of the economic efficiency of generators of different capacity levels by transforming the absolute economic indicator of selection cost into the relative efficiency indicator of unit cost coefficient. This avoids the problem of large-capacity generators being systematically excluded due to their high absolute cost when simply comparing start-up and shutdown costs. The unit cost coefficient establishes a quantitative bridge between load capacity and economic efficiency, enabling the optimization algorithm to identify the optimal generator combination that meets the expected load demand with the minimum net cost. Furthermore, the ranking result of this coefficient provides a priority benchmark for subsequent pruning optimization, improving the solution efficiency of large-scale combinatorial optimization problems.

[0069] Optionally, determining the target generator combination based on the ranking results and expected load demand includes: selecting generators sequentially according to the ranking results until the cumulative load exceeds the expected load demand to obtain candidate generator combinations; determining a pruning threshold based on the total selection cost of the candidate generator combinations; and optimizing the selection process based on the pruning threshold to determine the target generator combination.

[0070] A candidate generator set refers to the initial set of equipment formed during a greedy selection process based on unit cost coefficients, sequentially including generators until the cumulative load first meets the expected load demand. This set can be obtained, for example, by arranging all generators in ascending order of unit cost coefficients to obtain a cost-efficiency sequence. Starting from the head of this sequence, generators are selected sequentially. For each selected generator, its maximum load is added to the current total load, and its selection cost is also added to the current total cost. This selection process continues until the cumulative load first exceeds the expected load demand. The selected generators at this point constitute the candidate generator set. This candidate set ensures that the load demand is met, but there may be equipment redundancy; that is, removing some generators will not cause the cumulative load to fall below the expected load demand.

[0071] Furthermore, it is understandable that after selecting generators sequentially according to the sorting results until the cumulative load exceeds the expected load demand, redundant equipment in the generator combination can be removed. Specifically, let the cumulative load of the selected generators when selecting generators sequentially according to the sorting results until the cumulative load exceeds the expected load demand be denoted as . Among the selected generators, those with a maximum load less than [a certain value] are removed. The generator with the highest cost is selected, and the removal operation is repeated until no generator that meets the above conditions is found. The set of the remaining generators is then used as the candidate generator combination.

[0072] The pruning threshold is the sum of the optimal start-stop costs determined based on the candidate generator combinations mentioned above. This serves as an upper bound for the subsequent optimization search process, used to exclude inferior solution branches. The pruning threshold is denoted as... Its mathematical expression is ,in, Basic cost, measured in monetary units. For candidate generator combinations The algebraic sum of the costs of selecting all generators, with the dimension in monetary units, is therefore... The unit of measurement is currency. The pruning threshold represents the total cost of the better feasible solutions obtained through the greedy strategy. Any solution branch with a total cost exceeding this threshold cannot be better than the current candidate combination and can therefore be excluded in subsequent searches.

[0073] Pruning optimization based on a pruning threshold can be implemented using a lower bound estimation approach. Specifically, during the recursive or search phase of the selection process, the minimum possible cost required to fill all loads from the current partial solution is estimated, and this lower bound is compared with the pruning threshold to determine whether to terminate the expansion of the branch. Specifically, in the currently selected generator set... Estimate the minimum total cost under the given conditions. Its calculation formula is In the formula, This represents the sum of the selection costs of the currently selected generators, expressed in monetary units. This represents the sum of the maximum loads of the currently selected generators, expressed in units of power. The unit of measurement is power, representing the expected load demand. The smallest unit cost coefficient among the generators was not selected. The dimension is the ratio of monetary units to power units; therefore, the dimension of the difference in parentheses is power units. The product has units of monetary value after multiplication. Its dimension is monetary units. When Greater than When the condition is met, it indicates that even if the remaining load is filled with the least cost unselected generator, the total cost of this branch will still exceed the pruning threshold. At this point, the branch is pruned and its subsequent selected branches are no longer expanded.

[0074] The above scheme uses a greedy strategy to quickly obtain a reasonably good feasible solution with acceptable quality and uses its cost as a pruning threshold, effectively compressing the solution space for subsequent optimization searches. Furthermore, during the recursive selection process, an optimistic estimate (lower bound estimate) of the remaining filling cost is made using a unit cost coefficient, identifying and pruning branches that are unlikely to produce a better solution in advance. This reduces the computational complexity of the combinatorial optimization problem, enabling the efficient determination of the target generator combination that minimizes the overall start-up and shutdown cost even when there are a large number of generators.

[0075] Optionally, the above determination of the selection cost of each generator also includes: for each generator in the original set of operating generators, including the potential start-up and shutdown costs of restarting each generator after it has been shut down in the selection cost of the corresponding generator.

[0076] The potential start-stop cost after shutdown refers to the predicted economic expenditure incurred when a generator is shut down at the time of optimization decision-making, and then needs to be restarted in subsequent operating cycles due to increased load demand. Potential start-stop cost quantifies the hidden costs of short-sighted decisions, aiming to avoid system oscillations and additional energy losses caused by frequent generator start-stop switching through forward-looking cost accounting. One method for determining potential start-stop cost is to assume that the resource consumption and equipment losses required for restarting the generator in the future are consistent with the current start-stop cost, directly using the calculated current start-stop cost as the potential start-stop cost value. This simplifies the prediction model and ensures consistency in cost accounting without requiring additional collection or calculation of independent parameters.

[0077] In the above scheme, the final selected generator set The load constraints that need to be satisfied exhibit dynamic characteristics, and their mathematical expression is as follows: ,in Represents a set The sum of the maximum loads of all generators in the system, with the upper bound of this constraint being: Its special feature is that the upper bound is not a fixed constant, but depends on the minimum load capacity contained in the set E to be solved. That is, the value of the upper bound is dynamically adjusted as the solution set changes, forming a self-coupling structure in which the constraint boundary depends on the decision result.

[0078] The aforementioned dynamic constraints make the optimization problem difficult to solve directly using standard algorithms. Traditional combinatorial optimization problems typically have fixed resource capacity constraints or fixed objective boundaries, while here, due to the upper bound parameter... With decision variables The coupling and correlation mean that the exact load and upper limit cannot be known in advance before the generators are selected, making it impossible to decouple the verification of constraints from the optimization of the objective function. If an exhaustive method is used to traverse all possible subsets, the computational complexity increases exponentially with the number of generators; if algorithms such as dynamic programming are used, the state transition depends on the undetermined upper bound parameters, making it impossible to directly establish recursive relationships, leading to a circular dependency problem in the solution.

[0079] To overcome the aforementioned problem-solving difficulties, a problem decomposition strategy is adopted to break down the original dynamically constrained optimization problem into several static subproblems with fixed boundaries. Specifically, all subproblems in the high-voltage unit are first decomposed into subproblems. Each generator is based on its maximum load. Sort in ascending order to obtain an ordered sequence. Then enumerate the minimum load devices and construct... One feasible candidate set : No. The candidate set is Each candidate set Force the inclusion of the generator with the smallest load in the set. This determines the minimum load parameter for the subproblem. This transforms the original dynamic upper bound into a fixed value. .

[0080] In each candidate set's corresponding subproblem, since the minimum load capacity has been predetermined... The original dynamic constraints are transformed into static boundary conditions. At this point, the problem is simplified to a 0-1 knapsack problem with a fixed initial value: the maximum capacity of the knapsack is... The lower limit is The weight of the items is equal to the maximum load of each generator. The value of an item is the cost of selection. The dynamic programming algorithm is used to optimize the solution of each subproblem, and only the candidate set is considered during the state transition process. The generators in the dataset are recursively calculated to minimize the total start-up and shutdown costs under load constraints.

[0081] Complete all After solving the subproblems, we obtain at most A locally optimal strategy is proposed. For each subproblem, the total start-up and shutdown cost is calculated, which serves as the base cost. The sum of the costs of selecting the chosen generators. Compare the optimal solutions to each subproblem, and select the set of generators corresponding to the scheme with the minimum total start-up and shutdown cost as the final target generator combination. The above-described decomposition solution strategy transforms dynamic constraints into static constraints by enumerating the minimum load, thereby reducing computational complexity while maintaining global optimality and achieving a balance between computational efficiency and solution accuracy.

[0082] Step S140: Perform integrated control of each generator in the target generator set to meet the expected load demand of the high-voltage unit in the future period.

[0083] The aforementioned integrated control refers to incorporating the start-up and shutdown operations, electrical state coordination, and power distribution of each generator in a high-voltage generating unit into a unified centralized control framework. Based on the optimized target generator combination, synchronized grid connection and operation scheduling are implemented, forming a closed-loop control mechanism from strategic decision-making to physical execution. The purpose of integrated control is to ensure that the target generator combination selected by the optimization algorithm can be coordinated and consistently put into operation on the high-voltage grid, avoiding electrical parameter mismatch, grid connection impact, or load distribution imbalance caused by independent operation of individual generators. This ensures that the high-voltage generating unit reliably meets and efficiently responds to expected load demands in the future, guaranteeing a balance between power supply continuity and operational economy.

[0084] Optionally, the above-mentioned integrated control of each generator in the target generator combination includes: performing electrical parameter matching and grid connection operations on each generator to synchronize the output voltage, frequency and phase of each generator with the bus; and after successful grid connection, distributing the load to each generator based on the optimal efficiency point and real-time load of each generator.

[0085] Electrical parameter matching refers to the technical process of adjusting and correcting key electrical parameters such as voltage, frequency, and phase at the generator's output end before it is connected to the high-voltage power grid, ensuring synchronization with the corresponding parameters on the high-voltage bus (grid side). This process is a necessary prerequisite for ensuring the generator's safe and shock-free grid connection. If grid connection is initiated without electrical parameter matching, it will generate huge circulating currents and mechanical torque impacts, leading to serious consequences such as generator winding insulation breakdown, shaft torsional vibration damage, or a sudden drop in grid voltage. Grid connection operation refers to the process of physically connecting the generator's output end to the high-voltage bus by closing the grid connection switch or synchronous circuit breaker after electrical parameter matching is completed, enabling it to formally connect to the power grid and participate in power transmission and voltage support.

[0086] Electrical parameter matching and grid connection operations can be performed based on an automatic synchronization detection unit. Specifically, the automatic synchronization detection unit continuously collects the effective value, frequency, and phase angle of the bus voltage, and simultaneously collects the corresponding parameters of the generator terminal voltage to be connected. It calculates the amplitude difference, frequency difference, and phase angle difference between the two. When the amplitude difference is less than a preset voltage deviation threshold, the frequency difference is less than a preset slip threshold, and the phase angle difference is close to zero, a closing command is issued to the grid connection switch to complete the grid connection operation. Alternatively, electrical parameter matching and grid connection operations can be performed based on manual adjustment and quasi-synchronous grid connection. Specifically, the operator observes the voltage, frequency, and phase relationship displayed on the synchronization meter or oscilloscope, manually adjusts the generator's excitation regulator to change the output voltage amplitude, and adjusts the prime mover's speed control handle to change the speed and thus adjust the frequency. When the synchronization meter pointer slowly rotates to near zero (indicating a phase difference close to zero), a closing command is issued, and the grid connection switch is manually closed to complete the grid connection. This method is suitable for low levels of automation or as a backup measure when the automatic synchronization device fails.

[0087] The aforementioned optimal efficiency point refers to the operating condition where a generator achieves the highest energy conversion efficiency under a specific load rate (typically 70% to 85% of its rated capacity). This point is determined by the generator manufacturer through type testing and is indicated on the technical nameplate or characteristic curve. Load sharing refers to the process of allocating total power proportionally or according to an optimization strategy among multiple generators operating in parallel, based on the total load demand and the characteristics of each individual generator. The aim is to ensure that each generator operates within a load range close to its optimal efficiency point.

[0088] After successful grid connection, load allocation based on optimal efficiency points and real-time load can be implemented using an equal-capacity proportional allocation method. This method distributes the total load according to the same proportion of each generator's rated capacity. For example, the ratio of the high-voltage unit's real-time total load to each generator's rated capacity is calculated. The total load is then multiplied by each generator's rated capacity to obtain the target power each generator should handle. The prime mover output power of each generator is adjusted to track the target power value. This method is simple to implement but may cause some generators to deviate from their optimal efficiency points. Alternatively, an optimal efficiency point tracking method can be used. This method prioritizes operating each generator at its optimal efficiency point load rate. For example, the power value corresponding to each generator's optimal efficiency point is determined. The real-time total load is preferentially allocated to generators that have not yet reached their optimal efficiency point until they do. If there is still a surplus in the total load, it is allocated to the next generator. Real-time monitoring and feedback adjustments maintain each generator operating within its optimal efficiency range. This method ensures optimal overall unit energy efficiency.

[0089] The above scheme eliminates voltage, frequency, and phase differences at the moment of grid connection by matching electrical parameters, suppresses the grid connection inrush current within the equipment's tolerance range, and ensures synchronous interconnection between the generator and the grid. By load allocation based on the optimal efficiency point, it avoids bidirectional efficiency loss due to overloading of one generator and underloading of another, allowing each generator to operate under the condition of highest energy conversion efficiency. This reduces mechanical wear per unit of power generation while ensuring power supply reliability, and enables the high-voltage unit to respond economically to expected load demand in the future.

[0090] To facilitate understanding of the working principle of the above-mentioned integrated control method for high-voltage units, this application also provides an application example of this method in a certain application scenario. For example... Figure 2 As shown, in this application scenario, the above-mentioned integrated control method for high-voltage units mainly includes: Step 1: Obtain historical production data and parameters of each generator; Historical production data is collected through production logs and operation records. This data includes historical production plans and corresponding historical actual loads for the corresponding periods. Equipment parameters for each generator are collected synchronously, covering the maximum load, average equivalent operating time of a single start-up loss, total equivalent operating time per year, and annual maintenance costs. Based on these equipment parameters, the start-up and shutdown costs of each generator are determined. These start-up and shutdown costs are equivalent to the start-up loss costs, and their calculation formula is as follows: .

[0091] Step 2: Predict future load changes and arrange generator starting strategies accordingly; An XGBoost (Extreme Gradient Boosting) load prediction model is constructed based on historical production plans and actual loads from historical production data. Future production plans are input into this model to predict load change trends. High-load operating periods are identified based on these trends: time points where the predicted power exceeds the limit power by 50% are identified as candidate high-load points. DBSCAN clustering is performed on the extreme points among these candidate points, with each valid cluster representing a high-load operating period. Time nodes are determined based on load change characteristics between adjacent high-load operating periods: the load maxima between adjacent high-load operating periods are traversed on the smoothed load curve. When three matching points are identified, the leftmost point is marked as a time node, and the rightmost point is marked as another time node by pushing forward 1.2 times the maximum generator start-up time. Based on these time nodes, the future period is divided into at least one sub-period: the minimum operating duration is set as the maximum value of the minimum continuous operating time and minimum downtime among all generators. The larger of this minimum operating duration and the time interval between adjacent time nodes is taken as the operating duration, and the future period is divided accordingly. For each sub-period, the maximum load value is taken from the load trend segment of that sub-period and a 20% tolerance is added as a preset tolerance to determine the expected load demand for each sub-period, which is denoted as the expected maximum load. .

[0092] All generators in the high-voltage generator unit are divided into a set of originally operating generators and a set of inactive generators. For each generator in the high-voltage generator unit, its maximum load is determined. Start-up and shutdown costs Two attributes. To ensure production efficiency while reducing costs, the load and... It must be greater than the expected maximum load. At the same time, there should be no redundant equipment; removing any equipment will lead to increased load and... Less than the expected maximum load That is, the final set of generators selected. To meet At the same time, it is necessary to ensure that the energy consumed during the state transition is minimized, that is, to minimize the total start-up and shutdown costs.

[0093] To satisfy the above dynamic constraints, the original problem is split into several subproblems. The splitting method is as follows: first, all... Each generator operates according to its maximum load. Sort in ascending order to obtain the sequence Enumerating the minimum load device as a mandatory inclusion item, a total of One feasible candidate set : , ,…, ,…, In each sequence, based on selecting the generator with the minimum load, the following steps are taken respectively: Optimization is performed using the Dynamic Programming (DP) algorithm. The value of is determined, and the problem is simplified to a 0-1 knapsack problem with a fixed initial value.

[0094] The base cost is the sum of the start-up and shutdown costs assuming all generators in the original set of operating generators are shut down, denoted as . ,in This represents the original set of operating generators. The selection cost for each generator is determined, where for each generator in the original set, the selection cost is the negative of the corresponding generator's start-up and shutdown cost. For generators in the set of generators that are not in use, the cost is selected as the start-up and shutdown cost of the corresponding generator, i.e. For each generator in the original set of operating generators, the potential start-up and shutdown costs of restarting each generator after it has been shut down are included in the selection cost of the corresponding generator, in order to avoid frequent start-ups and shutdowns caused by short-sighted decisions.

[0095] Calculate the unit cost coefficient for each generator. This unit cost coefficient is used to quantify the selection cost corresponding to each generator per unit of maximum load. The calculation formula is as follows: .

[0096] The cost efficiency sequence is obtained by sorting the generators in ascending order based on their unit cost coefficients. Devices are selected sequentially from front to back in the sequence until the sum of the loads on these devices exceeds the expected maximum load. Let the total load of the selected equipment at this time be . Remove devices with a maximum load less than [a certain value] from the selected devices. Simultaneously, select the generator with the highest cost and repeat this step until no generator meets the conditions. At this point, the set of remaining selected generators constitutes a relatively optimal feasible solution. The algebraic sum of the basic cost and the selection costs of each generator in the above set is denoted as the optimal start-up and shutdown cost sum. The optimal start-stop cost and pruning threshold are used as the basis for this.

[0097] Using the dynamic programming algorithm to evaluate each feasible candidate set Before solving, the feasible candidate set is... The generators in the list are arranged in ascending order according to their unit cost coefficient, and then... Pruning is performed using the upper bound. The minimum total cost is estimated during the DP recursion process. : In the formula, This refers to the set of generators selected in the current state. The minimum total cost is... Higher than optimal start-stop costs and If this happens, the expansion of that state is terminated early, and no further expansion of that state is permitted.

[0098] final One feasible candidate set Get the most Each result is used to calculate the total start-up and shutdown cost. ,in The set of generators selected for the corresponding results. The sum of the base cost and the cost of selecting the chosen generator is expressed in monetary units. This represents the scheme with the minimum total start-stop cost. This is the optimal generator starting strategy.

[0099] Step 3: Perform integrated control based on the calculated generator starting strategy; After determining the target generator combination, the generators within the combination are controlled as a whole to meet the expected load demand of the high-voltage unit during the specified time period. Electrical parameter matching and grid connection operations are performed on each generator in the target generator combination: the voltage regulators of each generator are adjusted to make the effective output voltage equal to the bus voltage; the prime mover speed governors of each generator are adjusted to change the generator speed so that the output frequency matches the bus frequency; the phase of each generator is aligned with the bus voltage, and grid connection is performed to synchronize the output voltage, frequency, and phase of each generator with the bus. After successful grid connection, load distribution is performed on each generator based on its optimal efficiency point and real-time load to ensure that each generator operates within its high-efficiency range, thereby meeting the power supply demand during the specified time period and optimizing overall energy efficiency.

[0100] like Figure 3 As shown, based on the same inventive concept, this application also provides an integrated control system 200 for high-voltage generator units, including a host computer 210 and a high-voltage generator unit 220 connected to the host computer 210, wherein: High-voltage unit 220, including multiple generators; The host computer 210 is used to obtain the start-up and shutdown costs and maximum loads of each generator in the high-voltage generator set 220; predict the expected load demand of the high-voltage generator set 220 in the future; based on the expected load demand and the maximum load of each generator, with the goal of minimizing the overall start-up and shutdown cost of the high-voltage generator set 220, select a target generator combination that meets the expected load demand from the generators; and perform integrated control on each generator in the target generator combination to meet the expected load demand of the high-voltage generator set 220 in the future.

[0101] System 200 also includes an integrated control panel 230, which integrates a unit LCU control module, an excitation control module, and a high-voltage switchgear module for performing integrated control.

[0102] It is understood that the above-mentioned integrated control system 200 for high-voltage generator units can realize any one of the functions of the integrated control method for high-voltage generator units provided in the embodiments of this application. The method embodiment section describes the way each function is realized and the working principle. The system embodiment section will not repeat the description.

[0103] In this embodiment, the three major modules of the traditional discrete architecture—the unit LCU control panel, the excitation control panel, and the high-voltage switchgear—are integrated into a single high-voltage unit integrated control panel. This integrated control panel centrally handles signal acquisition, logic operations, and control command issuance, significantly reducing signal transmission loss and interference from inter-module connecting cables, and substantially improving grid connection speed and accuracy. Because this invention employs a control method based on future load change trends, even with the significant reduction in communication delays between modules due to hardware integration, electrical oscillations in the high-voltage unit system will not occur during periods of severe load fluctuations. Simultaneously, it avoids mechanical wear on critical components such as the turbine generator main shaft and key pins, achieving synergistic effects between hardware integration and control optimization.

[0104] This invention is now complete.

[0105] In summary, in this embodiment of the invention, the start-up and shutdown costs and maximum loads of each generator in the high-voltage generator unit are obtained; the expected load demand of the high-voltage generator unit in the future period is predicted; based on the expected load demand and the maximum load of each generator, with the minimum overall start-up and shutdown cost of the high-voltage generator unit as the optimization objective, a target generator combination that meets the expected load demand is selected from each generator; and the generators in the target generator combination are controlled in an integrated manner to meet the expected load demand of the high-voltage generator unit in the future period. This invention integrates future load forecasting and start-up / shutdown cost optimization to transform high-voltage generator units from current optimal to time-period optimal operation modes, reducing overall operating costs and improving energy efficiency while ensuring power supply reliability. Furthermore, a production plan-driven load forecasting mechanism enables high-voltage generator units to anticipate future load changes and pre-arrange generator start-up and shutdown, effectively avoiding unit oscillations and response lags caused by real-time load fluctuations. Additionally, by quantifying the start-up / shutdown costs of each generator and selecting the target generator combination with the minimum overall start-up / shutdown cost as the optimization objective, it avoids additional energy losses and equipment lifespan reduction caused by frequent generator start-up / shutdown switching. Moreover, by accurately configuring generator combinations based on expected future load demands, it prevents multiple generators from operating under load during off-peak periods, ensuring that each generator operates within its high-efficiency range and reducing mechanical wear caused by underload.

[0106] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An integrated control method for high-voltage generator units, characterized in that, The method includes: Obtain the start-up and shutdown costs and maximum load of each generator in the high-voltage unit; Predict the expected load demand of the high-voltage unit in the future; Based on the expected load demand and the maximum load of each generator, with the goal of minimizing the overall start-up and shutdown cost of the high-voltage unit, a target generator combination that meets the expected load demand is selected from among the generators. The generators in the target generator set are controlled in an integrated manner to meet the expected load demand of the high-voltage unit in the future period. The step of selecting a target generator combination that meets the expected load demand based on the expected load demand and the maximum load of each generator, with the goal of minimizing the overall start-up and shutdown cost of the high-voltage generator unit, includes: dividing each generator in the high-voltage generator unit into a set of originally operating generators and a set of inactive generators; determining the base cost; wherein the base cost is the start-up and shutdown cost assuming all generators in the originally operating generator set are shut down; determining the selection cost of each generator; wherein, for each generator in the originally operating generator set, the selection cost is the negative value of the corresponding generator start-up and shutdown cost; for each generator in the inactive generator set, the selection cost is the corresponding generator start-up and shutdown cost; and selecting the target generator combination that meets the expected load demand based on the expected load demand, the maximum load of each generator, and the selection cost of each generator, with the goal of minimizing the overall start-up and shutdown cost; wherein the final selected generator set... satisfy: ； in, Represents a set The sum of the maximum loads of all generators in the system; Represents a set Middle The maximum load of the generator; This indicates the expected load demand; Selecting the target generator combination includes: arranging the generators in ascending order of maximum load, enumerating the generators with the minimum load and constructing multiple candidate sets, each candidate set containing the corresponding generator with the minimum load; for each candidate set, transforming the dynamic load constraint into a 0-1 knapsack problem with fixed upper and lower bounds, and solving it using dynamic programming with the maximum load of each generator as the weight and the selection cost of each generator as the value; comparing the total start-up and shutdown costs corresponding to each candidate set, and selecting the generator set corresponding to the candidate set with the minimum total start-up and shutdown cost as the target generator combination.

2. The integrated control method for high-voltage generator units according to claim 1, characterized in that, The process of selecting the target generator combination that meets the expected load demand, based on the expected load demand, the maximum load of each generator, and the selection cost of each generator, with the goal of minimizing the overall start-up and shutdown cost, includes: Determine the unit cost coefficient for each of the aforementioned generators; The generators are ranked based on their unit cost coefficients. The target generator combination is determined based on the sorting results and the expected load demand.

3. The integrated control method for high-voltage generator units according to claim 2, characterized in that, Determining the target generator combination based on the sorting results and the expected load demand includes: Based on the sorting results, generators are selected sequentially until the cumulative load exceeds the expected load requirement, thus obtaining a candidate generator combination. The pruning threshold is determined based on the total selection cost of the candidate generator combinations; Based on the pruning threshold, the selection process is optimized by pruning to determine the target generator combination.

4. The integrated control method for high-voltage generator units according to any one of claims 1-3, characterized in that, The determination of the selection cost for each of the generators also includes: For each generator in the original set of operating generators, the potential start-up and shutdown costs of restarting each generator after it has been shut down are included in the selection cost of the corresponding generator.

5. The integrated control method for high-voltage generator units according to any one of claims 1-3, characterized in that, The prediction of the expected load demand of the high-voltage unit in the future includes: Obtain historical production data and future production plans for high-voltage generator units; wherein, the historical production data includes historical production plans and historical actual loads for corresponding periods; Based on the historical production plan and the historical actual load, a load prediction model is constructed; The future production plan is input into the load forecasting model to predict the load change trend of the high-voltage unit in the future period. Based on the load change trend, the expected load demand for the future period is determined.

6. The integrated control method for high-voltage generator units according to claim 5, characterized in that, Determining the expected load demand for the future period based on the load change trend includes: Based on the load change trend, identify the high-load running segments in the future time period; Time nodes are determined based on load variation characteristics between adjacent high-load operating periods; wherein, the time nodes are determined based on the maximum load values ​​between adjacent high-load periods. Based on the aforementioned time points, the future time period is divided into at least one sub-time period; For each sub-time period, the expected load demand for each sub-time period is determined based on the maximum load value and preset tolerance within the sub-time period.

7. The integrated control method for high-voltage generator units according to any one of claims 1-3, characterized in that, The acquisition of the start-up and shutdown costs of each generator in the high-voltage unit includes: Obtain the equipment loss parameters and maintenance cost parameters of each generator during the startup process; Based on the equipment loss parameters and the maintenance cost parameters, the equivalent cost of starting loss for each generator is determined, and the equivalent cost of starting loss is used as the start-stop cost.

8. The integrated control method for high-voltage generator units according to any one of claims 1-3, characterized in that, The integrated control of each generator in the target generator assembly includes: Electrical parameter matching and grid connection operations are performed on each of the generators to synchronize the output voltage, frequency, and phase of each generator with the bus. After successful grid connection, load distribution is performed on each generator based on its optimal efficiency point and real-time load.

9. An integrated control system for high-voltage generator units, characterized in that, It includes a host computer and a high-voltage generator unit connected to the host computer, wherein: The high-voltage unit includes multiple generators; The host computer is used to obtain the start-up and shutdown costs and maximum loads of each generator in the high-voltage generator set; predict the expected load demand of the high-voltage generator set in the future; based on the expected load demand and the maximum loads of each generator, with the goal of minimizing the overall start-up and shutdown cost of the high-voltage generator set, select a target generator combination from the generators that meets the expected load demand; and perform integrated control on each generator in the target generator combination to meet the expected load demand of the high-voltage generator set in the future period. The system also includes an integrated control panel, which integrates a unit LCU control module, an excitation control module, and a high-voltage switchgear module to perform the integrated control. The step of selecting a target generator combination that meets the expected load demand based on the expected load demand and the maximum load of each generator, with the goal of minimizing the overall start-up and shutdown cost of the high-voltage generator unit, includes: dividing each generator in the high-voltage generator unit into a set of originally operating generators and a set of inactive generators; determining the base cost; wherein the base cost is the start-up and shutdown cost assuming all generators in the originally operating generator set are shut down; determining the selection cost of each generator; wherein, for each generator in the originally operating generator set, the selection cost is the negative value of the corresponding generator start-up and shutdown cost; for each generator in the inactive generator set, the selection cost is the corresponding generator start-up and shutdown cost; and selecting the target generator combination that meets the expected load demand based on the expected load demand, the maximum load of each generator, and the selection cost of each generator, with the goal of minimizing the overall start-up and shutdown cost; wherein the final selected generator set... satisfy: ； in, Represents a set The sum of the maximum loads of all generators in the system; Represents a set Middle The maximum load of the generator; This indicates the expected load demand; Selecting the target generator combination includes: arranging the generators in ascending order of maximum load, enumerating the generators with the minimum load and constructing multiple candidate sets, each candidate set containing the corresponding generator with the minimum load; for each candidate set, transforming the dynamic load constraint into a 0-1 knapsack problem with fixed upper and lower bounds, and solving it using dynamic programming with the maximum load of each generator as the weight and the selection cost of each generator as the value; comparing the total start-up and shutdown costs corresponding to each candidate set, and selecting the generator set corresponding to the candidate set with the minimum total start-up and shutdown cost as the target generator combination.

Images