Multi-energy-storage thermal power generating unit optimization scheduling method

Abstract

The invention discloses a multi-energy-storage thermal power generating unit optimal scheduling method, which belongs to the technical field of power system scheduling, and specifically comprises the following steps: acquiring operation data and power grid load data of a multi-energy-storage thermal power generating unit; constructing a load prediction model based on the collected data to obtain a load prediction result; taking the minimum comprehensive cost as a target, and combining a thermal power generating unit output constraint, an energy storage equipment charging and discharging constraint and a power grid security constraint to construct an optimal scheduling model; solving the optimal scheduling model by adopting an improved intelligent optimization algorithm to obtain an optimal scheduling scheme; performing real-time scheduling on the multi-energy-storage thermal power generating unit according to the optimal scheduling scheme; the method can improve the scheduling flexibility of the multi-energy-storage thermal power generating unit, effectively deals with the load fluctuation of the power grid, and is suitable for the optimal scheduling task of the multi-energy-storage thermal power generating unit in a complex power grid environment.

Description

Technical Field

[0001] The present invention belongs to the technical field of power system dispatching, and in particular relates to an optimization dispatching method for multiple energy storage thermal power generating units. Background Art

[0002] With the large-scale integration of renewable energy generation, the volatility and uncertainty of the power system have increased significantly. As a key regulating power source in the power system, the scheduling and operation of thermal power units faces even greater challenges. At the same time, to improve energy efficiency and reduce carbon emissions, an increasing number of energy storage devices, such as batteries, flywheel storage, and pumped hydro, are being integrated with thermal power units to form multi-storage thermal power systems. However, current scheduling methods for multi-storage thermal power units face numerous challenges. For example, traditional scheduling methods often focus solely on a single objective, such as minimizing operating costs while ignoring factors such as energy utilization and carbon emissions, making it difficult to achieve optimal overall efficiency. Furthermore, the characteristics of different energy storage devices vary significantly, including charge and discharge efficiency, storage capacity, and response speed. Therefore, achieving coordinated and optimized scheduling of thermal power units and multiple energy storage devices to cope with complex load fluctuations and the uncertainties of renewable energy generation is a pressing technical challenge. Summary of the Invention

[0003] In response to the shortcomings of the existing technology, the present invention proposes a method for optimizing the scheduling of multiple energy storage thermal power units, which collects the operating data and grid load data of the multiple energy storage thermal power units; constructs a load forecasting model based on the collected data to obtain load forecast results; takes minimizing the comprehensive cost as the goal, combines the output constraints of the thermal power units, the charging and discharging constraints of the energy storage equipment, and the safety constraints of the grid to construct an optimization scheduling model; adopts an improved intelligent optimization algorithm to solve the optimization scheduling model to obtain the optimal scheduling plan; and performs real-time scheduling of the multiple energy storage thermal power units according to the optimal scheduling plan. This method can improve the scheduling flexibility of the multiple energy storage thermal power units, effectively cope with grid load fluctuations, and is suitable for the optimization scheduling tasks of multiple energy storage thermal power units in complex grid environments.

[0004] To achieve the above object, the present invention provides the following technical solutions: A method for optimizing the scheduling of multiple energy storage thermal power units, comprising: Collect operating data of multiple energy storage thermal power units and grid load data; Building a load forecasting model based on the operation data and the grid load data, and obtaining a load forecasting result through the load forecasting model; With the goal of minimizing comprehensive costs, an optimal dispatch model is constructed by combining the thermal power unit output constraint sub-model, energy storage device charging and discharging constraints, and grid security constraints. An improved intelligent optimization algorithm is used to solve the optimal scheduling model to obtain an optimal scheduling scheme, and the multi-energy storage thermal power unit is scheduled in real time according to the optimal scheduling scheme.

[0005] Specifically, the optimal scheduling scheme includes target output values of each thermal power unit in each time interval, start-stop state instructions, charge-discharge power curves and SOC target trajectories of each energy storage device, distribution results of rotating reserve capacity among the thermal power units and the energy storage devices, and an accompanying scheme economic analysis report and a network loss control strategy based on load prediction results and power grid structure.

[0006] Specifically, the operation data and the grid load data of the multi-energy storage thermal power unit are collected, including: The real-time output, coal consumption rate, start-stop state data of the thermal power unit, and the state of charge, charge-discharge power, and charge-discharge efficiency data of the energy storage device are collected by sensors and a data transmission network as the operation data of the multi-energy storage thermal power unit; The historical load curve, real-time load value, load fluctuation amplitude, peak-valley period division, and load prediction error interval are collected as the grid load data; The operation data and the grid load data of the multi-energy storage thermal power unit are spatiotemporally aligned by an edge computing node and are stored by blockchain technology.

[0007] Specifically, the load prediction model is constructed based on the operation data and the grid load data, and the load prediction results are obtained by the load prediction model, including: The time series features are extracted from the preprocessed grid load data; the time series features include hourly load trend, daily periodic fluctuation features, weekly regularity features, and meteorological factor correlation features; Based on the time series features, a long short-term memory network with an attention mechanism is constructed as a load prediction model; the attention mechanism is used to dynamically allocate the weights of different historical period data; The dataset containing historical time series features and corresponding historical load values is used as a training set, and the Adam optimizer is used to minimize the mean square error of the predicted value and the actual value to train the constructed load prediction model to obtain a trained load prediction model; The real-time time series features obtained from the real-time collected grid load data are input into the trained load prediction model to obtain the load prediction results of the preset time period.

[0008] Specifically, the process of constructing the optimal scheduling model includes: The decision variables of the optimization scheduling model are determined and a time scale is determined; the decision variables include an output variable of a thermal power unit and a charging and discharging variable of an energy storage device; the output variable of the thermal power unit refers to an output of each thermal power unit in each scheduling period, and the charging and discharging variable of the energy storage device refers to a charging power or discharging power of each energy storage device in each scheduling period; A target function of minimum comprehensive cost is constructed based on the decision variables; the comprehensive cost includes a fuel cost of the thermal power unit, a start-stop cost of the thermal power unit, a charging and discharging loss cost of the energy storage device, a cycle life loss cost of the energy storage device, and a grid safety and load shortage penalty cost; The constructed thermal power unit output constraint sub-model, the energy storage device charging and discharging constraint, and the grid safety constraint are incorporated into the target function; The target function and all the above constraints are integrated to form an optimization scheduling model; the optimization scheduling model takes the output variable of the thermal power unit and the charging and discharging variable of the energy storage device as the solving object, and outputs a scheduling scheme parameter that satisfies the minimum comprehensive cost and all the constraints.

[0009] Specifically, the process of constructing the thermal power unit output constraint sub-model includes: Historical operation data of the thermal power unit are collected, and a coal consumption characteristic curve and an output response rate parameter of the thermal power unit are extracted; the coal consumption characteristic curve is obtained by nonlinearly fitting historical output values and corresponding coal consumption amounts by using a cubic polynomial regression method, and an expression thereof is a function of coal consumption amount with respect to output; the output response rate parameter is determined by counting a climbing rate and a descending rate in the historical operation data, and the minimum start-up time of the unit from shutdown to start-up completion and the minimum shutdown time of the unit from operation to shutdown are recorded to form a start-stop constraint; the climbing rate is a maximum output increment of the unit per unit time; and the descending rate is a maximum output decrement of the unit per unit time; A nonlinear mapping relationship between the thermal power unit output and the coal consumption amount is established based on the coal consumption characteristic curve to obtain a calculated value of the coal consumption amount at a corresponding output level; In combination with the output response rate parameter, upper and lower limits of output adjustment of the thermal power unit per unit time are set to form a climbing constraint of the thermal power unit; A thermal power unit output constraint sub-model is constructed based on the output response rate parameter and the upper and lower limits of output adjustment; the thermal power unit output constraint sub-model includes an upper and lower limit constraint, a climbing constraint, and a start-stop constraint.

[0010] Specifically, the energy storage device charging and discharging constraint includes a charging and discharging power constraint and a state of charge constraint; The charging and discharging power constraint refers to that the charging power does not exceed the rated charging power, the discharging power does not exceed the rated discharging power, and the charging and discharging are not performed at the same time; The state of charge constraint refers to keeping the state of charge in a preset safe interval and returning to an initial value at the end of a scheduling period, and the state of charge is a ratio of current stored energy to rated capacity, which is calculated based on a state of charge of a previous period, a charging and discharging amount of the current period and charging and discharging efficiency.

[0011] Specifically, the power grid safety constraint includes a power balance constraint, a line flow constraint and a node voltage constraint. The power balance constraint refers to a sum of thermal power unit output and effective discharging amount of energy storage being equal to a sum of load, network loss and charging consumption of energy storage. The line flow constraint refers to line transmission power not exceeding a thermal stability limit. The node voltage constraint refers to a node voltage being maintained between 95% and 105% of a rated voltage.

[0012] Specifically, the improved intelligent optimization algorithm is used to solve the optimization scheduling model to obtain an optimal scheduling scheme, including: Initializing a particle swarm, each particle representing a scheduling scheme, a dimension of the particle corresponding to a total number of scheduling periods, and a dimension value of each dimension respectively representing output of all thermal power units and charging and discharging power of all energy storage devices in the period; Performing particle updating operation based on the initialized particle swarm to generate a new particle swarm, including: randomly selecting two particles as parent particles, randomly selecting 2-3 continuous or non-continuous scheduling periods in all dimensions of the particles, exchanging dimension values of the two particles in the selected periods to form two child particles; for each child particle, randomly selecting a dimension corresponding to a scheduling period, randomly generating a new value within a constraint range of the dimension corresponding to the scheduling period to replace an original value, to obtain an initially updated particle swarm; performing constraint checking on the initially updated particle swarm, if there is a value not satisfying the thermal power unit output constraint, the energy storage device charging and discharging constraint or the power grid safety constraint, then modifying through a constraint repair mechanism to finally form a new particle swarm; the constraint range of the dimension corresponding to the scheduling period refers to an allowed range of the thermal power unit output and an allowed range of the energy storage device charging and discharging power in the period; the constraint repair mechanism refers to adjusting an out-of-limit output value to a nearest constraint boundary; Calculating an adaptability value of each particle in the new particle swarm; the adaptability value is obtained based on a target function of the optimization scheduling model with weighting; The fitness value of each particle in the new particle group is compared with the historical optimal fitness value, if the fitness value of each particle in the current new particle group is less than the historical optimal fitness value, the historical optimal position of the particle is updated as the current particle position, and the position with the minimum fitness value is selected from the historical optimal positions of all particles as a candidate of the global optimal position, if the fitness value of the candidate position is less than the fitness value of the current global optimal position, the global optimal position is updated; The iteration is performed to a preset number of times, and a particle corresponding to the global optimal position is selected as an optimized scheduling scheme.

[0013] Specifically, the real-time scheduling of the multiple energy storage thermal power units according to the optimal scheduling scheme comprises: The optimized scheduling scheme is disassembled into thermal power unit output instructions and energy storage device charging and discharging instructions for each period according to the scheduling period; the thermal power unit output instruction is a target output value of each thermal power unit in the corresponding period; and the energy storage device charging and discharging instruction is a target charging power or discharging power of each energy storage device in the corresponding period; The disassembled thermal power unit output instructions are sent to a thermal power unit control system, and the energy storage device charging and discharging instructions are sent to an energy storage management system, so that the thermal power units are controlled to adjust the real-time output according to the target output value by the control system, and the energy storage devices are controlled to charge or discharge according to the target power by the energy storage management system; During the operation of the thermal power units and the energy storage devices according to the instructions, the operating state of the multiple energy storage thermal power units and the actual value of the power grid load are monitored in real time; The corresponding period load prediction result output by the load prediction model is called, and the monitored actual value of the power grid load is compared with the corresponding load prediction result to calculate a deviation value; When the deviation value does not exceed a preset threshold, the instructions corresponding to the current optimal scheduling scheme are maintained, and the devices are controlled to continue operating through the control system; When the deviation value exceeds the preset threshold, the optimal scheduling scheme is dynamically adjusted based on the real-time operation data to generate an adjusted scheduling scheme; The adjusted scheduling scheme is converted into thermal power unit output adjustment instructions and energy storage device charging and discharging adjustment instructions for the corresponding period, and is sent to the corresponding control system to control the thermal power units and the energy storage devices to operate according to the adjusted instructions.

[0014] The present application provides a kind of multi-energy storage thermal power unit optimization scheduling method, through the data-driven and intelligent optimization of whole process, realize the precision and high efficiency of scheduling decision, improve the economy and security of system operation, this method ensures the comprehensiveness and reliability of operating data and grid load data from data acquisition link, the load prediction model based on this can accurately capture the load variation law, and the optimization scheduling model and improved intelligent optimization algorithm of fusion multi-constraint condition, can realize the minimization of comprehensive cost under the premise of meeting the output limit of thermal power unit, energy storage charge and discharge safety and grid stability requirement, effectively reduce fuel consumption, equipment loss and load shortage penalty and other costs.

[0015] The present application provides a kind of multi-energy storage thermal power unit optimization scheduling method, through dynamic real-time scheduling mechanism realizes the flexible response to actual operating condition, breaks the limitation of traditional static scheduling.Optimal scheduling scheme can be decomposed into specific period equipment instruction, ensure that thermal power unit and energy storage device are efficiently operated according to plan, when the actual value and prediction result of grid load appear deviation, system can be based on real-time data quickly adjust scheduling scheme, generate adaptive adjustment instruction, guarantee power supply and demand balance and grid safety and stability.This prediction-optimization-execution-adjustment closed loop mechanism not only improves the adaptability of multi-energy storage thermal power unit to load fluctuation and equipment state change, but also prolongs the service life of equipment and reduces the incidence of safety accidents. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 It is a kind of multi-energy storage thermal power unit optimization scheduling method schematic diagram of the present application; Figure 2 It is a kind of multi-energy storage thermal power unit optimization scheduling method principle flow chart of the present application; Figure 3 It is a kind of multi-energy storage thermal power unit optimization scheduling method optimal scheduling scheme implementation flow chart of the present application. DETAILED DESCRIPTION

[0017] Example 1: Please refer to Figure 1 And Figure 2 The present application provides an embodiment: a kind of multi-energy storage thermal power unit optimization scheduling method, the method includes S1~S4, including the following steps: S1: the operating data and grid load data of multi-energy storage thermal power unit are collected; S2: based on the operating data and grid load data, a load prediction model is constructed, and a load prediction result is obtained by the load prediction model; S3: with the minimization of comprehensive cost as target, combine thermal power unit output constraint submodel and energy storage device charge and discharge constraint and grid safety constraint, construct optimization scheduling model; Further, the charge-discharge constraints of the energy storage device include a charge power constraint, a discharge power constraint, a state of charge (SOC) constraint, and a charge-discharge frequency constraint. The charge power constraint and the discharge power constraint limit the power of the energy storage device during charging and discharging, respectively. The SOC constraint ensures that the power of the energy storage device is within a certain range, avoiding overcharging or over-discharging. The charge-discharge frequency constraint limits the number of charge-discharge cycles within a certain time period, extending the service life of the device.

[0018] The grid safety constraints include line flow constraints, node voltage constraints, and frequency constraints. The line flow constraints ensure that the transmission power of each line in the grid does not exceed its rated capacity. The node voltage constraints ensure that the voltage of each node in the grid is within the allowed deviation range. The frequency constraints maintain the frequency of the grid near the rated value, ensuring stable operation of the grid.

[0019] S4: An improved intelligent optimization algorithm is used to solve the optimization scheduling model to obtain an optimal scheduling scheme, and the multi-energy storage thermal power unit is scheduled in real time according to the optimal scheduling scheme.

[0020] Embodiment 2: Please refer to Figure 3 In this embodiment, the optimal scheduling scheme includes the target output value of each thermal power unit in each time interval, the start-stop state instruction, the charge-discharge power curve of each energy storage device, the SOC target trajectory, the allocation result of the rotating reserve capacity among the thermal power units and the energy storage devices, and the accompanying scheme economic analysis report and the grid loss control strategy based on the load prediction results and the grid structure.

[0021] It needs to be explained that the target output value of each thermal power unit in each time interval refers to the specific power generation value set for each thermal power unit in each scheduling time interval, such as 15 minutes or 1 hour, in the optimized scheduling scheme. It is a core parameter determined based on load prediction results, coal consumption characteristics of thermal power units, grid safety constraints and other factors, and directly determines the real-time operation state of the thermal power unit. Functionally, the setting of the target output value needs to meet three requirements: first, to match the load demand of the power grid, to ensure that the output of the thermal power unit and the charging and discharging of the energy storage device can cover the load consumption and network loss in the current period after coordination; second, to meet the technical constraints of the thermal power unit, such as not exceeding the maximum output upper limit, not being lower than the minimum technical output, and the output change between adjacent periods not exceeding the ramp-up or ramp-down rate, to avoid damage to the unit due to frequent and severe adjustments; third, to achieve economic optimization, by referring to the coal consumption characteristic curve, the output level that minimizes fuel costs is selected under the premise of meeting the first two requirements. Unlike the empirical output allocation in traditional scheduling, the target output value of this method is accurately calculated through an optimized scheduling model and intelligent algorithm, and can dynamically adapt to real-time conditions such as load fluctuations and changes in energy storage state. For example, during the load peak period, the target output value will be moderately increased to ensure power supply; during the load valley period, it will be lowered to the economic output interval, while leaving space for the energy storage device to charge, this precise output setting can reduce fuel consumption of thermal power units and improve operating efficiency.

[0022] The start-stop state instruction of each thermal power unit is a control instruction for the running or shutdown state of the thermal power unit, including the time point of unit start-up, the output ramp-up curve during start-up, and the time point of shutdown, the output drop curve during shutdown, which is the core basis for ensuring safe and economic start-stop of the unit. In traditional thermal power unit scheduling, start-stop decisions often rely on manual judgment, which can easily lead to problems such as "premature start-up resulting in waste" or "late start-up failing to meet load". The start-stop state instruction of this method is generated based on the "start-stop constraints" in the optimized scheduling model, and needs to strictly follow three rules: first, the minimum start-up time constraint, which means the time from shutdown to complete start-up of the unit should not be shorter than the technical limit, such as 4 hours for coal-fired units and 1 hour for gas-fired units, to avoid overheating or mechanical damage due to rapid start-up; second, the minimum shutdown time constraint, which means the time from operation to complete shutdown of the unit should not be shorter than the technical limit, such as 2 hours for coal-fired units and 0.5 hours for gas-fired units, to prevent frequent start-stop from exacerbating equipment wear and tear; third, economic trade-off, the start-up cost and standby cost during shutdown need to be included in the comprehensive cost calculation, and the instruction will choose the start-stop time that minimizes the total cost.

[0023] The charging and discharging power curve is a continuous curve that describes the change of charging power or discharging power of the energy storage device over time in each scheduling period. The charging and discharging power curve of this method is generated based on the charging and discharging constraints of the energy storage device and the optimization objective.

[0024] The state of charge (SOC) target trajectory of each energy storage device refers to the path of the target state of charge of the energy storage device changing with time in each scheduling period, and is a key indicator to ensure the safe and sustainable operation of the energy storage device, wherein the target state of charge is the ratio of the current stored energy to the rated capacity; the design of the SOC target trajectory needs to balance three core requirements: first, the safety interval constraint, all points on the trajectory need to fall within the preset safety interval to avoid overcharging leading to battery bulging and short circuit, or over-discharging leading to permanent capacity attenuation of the battery; second, the periodic recovery constraint, the end point of the trajectory at the end of the scheduling period, such as 24 hours later, needs to be consistent with the SOC value of the starting point to ensure that the state of the energy storage device is reusable and avoid state imbalance caused by long-term one-way charging and discharging; third, the cooperativity constraint, the trajectory needs to be linked with the charging and discharging power curve, and the SOC target value of any period is determined by the SOC value of the previous period, the charging and discharging power and efficiency of the current period.

[0025] The spinning reserve capacity refers to the standby power generation capacity reserved by the system to cope with uncertain factors such as sudden load growth and equipment failure, and the allocation result between thermal power units and energy storage devices is a buffer mechanism to ensure the safe and stable operation of the power grid, and directly determines the ability of the system to respond to disturbances.

[0026] The operation data and grid load data of the multi-energy storage thermal power unit are collected, including: A1: collecting real-time output, coal consumption rate, start-stop state data of the thermal power unit, and state of charge, charging and discharging power, charging and discharging efficiency data of the energy storage device as the operation data of the multi-energy storage thermal power unit through sensors and a data transmission network; Further, the data transmission network adopts a hybrid architecture of industrial Ethernet and 5G private network, key data such as real-time output of the thermal power unit is transmitted through the 5G private network, and non-real-time data such as historical load curve is transmitted through the industrial Ethernet, and the transmission data is encapsulated using the Modbus TCP protocol; the encapsulated transmission data is used as the received data of the edge computing node.

[0027] A2: collecting historical load curve, real-time load value, load fluctuation amplitude, peak-valley period division and load prediction error interval as the grid load data; A3: aligning the operation data of the multi-energy storage thermal power unit and the grid load data in time and space through the edge computing node, and storing the data through the blockchain technology.

[0028] Further, the time and space alignment through the edge computing node includes: matching the collected operation data of the multi-energy storage thermal power unit and the grid load data according to the time stamp, interpolating and correcting the data with a time stamp deviation exceeding 0.5 seconds, and obtaining a standardized data set after time and space alignment; the standardized data set is used as input data for extracting time sequence features.

[0029] Further, the data with a timestamp deviation exceeding 0.5 seconds is corrected by interpolation, and the missing data in the deviation period is calculated by a linear interpolation method; the data after the interpolation correction is included in the standardized data set; the linear interpolation method is a prior art content in the field and is not the creative scheme of the present application, and is not described here.

[0030] Further, the blockchain technology is used for storage when the alliance chain architecture is adopted, each edge computing node is used as an accounting node, the storage data includes original collection data, space-time alignment results and data check codes, and the storage interval is 1 hour; the blockchain data of the storage is used for subsequent tracing and auditing of the scheduling scheme.

[0031] Further, the specific process of the blockchain technology storage is that the edge computing node calculates the SHA-256 hash value of the standardized data set, packs the hash value, data timestamp and node ID into transaction information, writes into a block after verification by the consensus mechanism of the alliance chain, and each block contains the hash value of the previous block to form a chain structure; the generated block data is used for integrity check during subsequent data tracing, wherein the SHA-256 hash calculation method is a prior art content in the field and is not the creative scheme of the present application, and is not described here.

[0032] In summary, in the whole process of the optimal scheduling of the multi-energy storage thermal power unit, the data collection link is the basis for the efficient operation of the whole system, and its functionality is reflected in the multiple guarantees for the comprehensiveness, real-time, reliability and security of the data. The innovative design of this link makes it impossible to be replaced by traditional data collection methods. In traditional thermal power unit scheduling, data collection is often limited to basic operating parameters of the thermal power unit, and there are problems such as low collection frequency, fragmented data, transmission delay, etc., which leads to a lack of accurate data support for scheduling decisions. The present method constructs a comprehensive data collection system through sensors and data transmission networks, and realizes deep perception of the operating state of the multi-energy storage thermal power unit and the load characteristics of the power grid. Specifically, for the thermal power unit, the sensor captures real-time changes in its output, coal consumption rate fluctuations, start-stop state switching and other key data, which accurately reflect the real-time operating efficiency and energy consumption level of the unit; for energy storage equipment, it focuses on collecting parameters such as state of charge, charging and discharging power and efficiency, to ensure a clear understanding of the energy state and conversion efficiency of the energy storage equipment. At the same time, the collection of power grid load data not only includes historical curves and real-time values, but also covers load fluctuation amplitude, peak-valley period division and prediction error interval, which provides multi-dimensional feature support for subsequent load prediction.

[0033] More importantly, the method introduces an edge computing node for spatio-temporal alignment processing, which completely solves the problem of time asynchronization and space different sources in traditional data processing. The edge computing node can complete real-time processing near the data source, accurately match various data of thermal power units, energy storage devices and power grids according to time stamp, correct small time deviation, and ensure consistency of all data in time and space dimensions. This processing method avoids the delay of data transmission to the cloud for processing, and saves valuable time for subsequent model training and scheduling decision. In addition, the application of blockchain technology provides irreplaceable guarantee for data security. In the power system, the authenticity and integrity of data are directly related to the reliability of scheduling decision. Traditional data storage methods are vulnerable to tampering, loss and other risks, while the alliance chain architecture of blockchain makes each edge computing node become an accounting node, generates a storage record every hour, includes original data, alignment results and check code, forms an unalterable chain structure. This storage method not only traces the whole life cycle of data, but also provides authoritative verification basis when disputes arise, ensuring the credibility of data from collection to application. It can be said that the data collection link of the method builds a highly reliable data foundation through the triple functions of comprehensive perception, real-time processing and secure storage, providing irreplaceable data support for load forecasting, model building and scheduling execution, which is difficult to achieve by traditional data collection methods.

[0034] The load prediction model is constructed based on the operation data and the power grid load data, and a load prediction result is obtained through the load prediction model, including: B1: extracting time sequence features from preprocessed power grid load data; the time sequence features include hourly load trend, daily cycle fluctuation features, weekly regularity features and meteorological factor correlation features; Further, the hourly time granularity correlation refers to binding the hourly average meteorological data with the corresponding hourly average load value to generate an hourly correlation array containing meteorological parameters and load values; the hourly correlation array is used to extract daily cycle fluctuation features.

[0035] Further, the daily cycle fluctuation features are extracted by Fourier transform, and the load data of continuous 7 days is subjected to Fourier decomposition to obtain the amplitude and phase of the frequency component with a period of 24 hours to form a daily cycle feature vector; the daily cycle feature vector and the hourly load trend are jointly input into the load prediction model.

[0036] Further, the meteorological factor correlation features include environmental temperature, humidity, wind speed and precipitation probability, which are collected in real time by a meteorological monitoring terminal and correlated with the power grid load data at an hourly time granularity to form a meteorological-load correlation feature set; the meteorological-load correlation feature set is input into the load prediction model as a component of the time sequence features.

[0037] B2: constructing a long short-term memory network with fusion attention mechanism as the load prediction model based on the time sequence features; the attention mechanism is used to dynamically allocate the weights of different historical period data, wherein the long short-term memory network is prior art content in the art and is not the inventive scheme of the present application, and will not be described here; B3: taking a data set containing historical time sequence features and corresponding historical load values as a training set, training the constructed load prediction model by minimizing the mean square error of the predicted value and the actual value through an Adam optimizer, and obtaining a trained load prediction model; Further, the minimizing the mean square error of the predicted value and the actual value through the Adam optimizer comprises: setting an initial learning rate to 0.001, attenuating the learning rate to 0.9 times the current value every 100 iterations, and stopping training when the change amount of the mean square error of 50 consecutive iterations is less than 0.0001; The model parameters at the time of training stop are used as the final parameters of the load prediction model.

[0038] B4: inputting real-time time sequence features obtained according to real-time collected power grid load data into the trained load prediction model to obtain load prediction results for a preset time period.

[0039] It should be emphasized that load forecasting is the navigation system of multi-energy storage thermal power unit scheduling. The load forecasting model of the long short-term memory network with attention mechanism constructed by the method exhibits irreplaceable advantages in prediction accuracy, feature capturing ability and dynamic adaptability, far exceeding traditional time series prediction methods. Traditional load forecasting mostly uses simple regression analysis or basic neural networks, which are difficult to handle complex time series features and multi-factor correlation in power grid load, resulting in large prediction errors, especially in special periods such as sudden changes in weather conditions and holidays, the error can exceed 10%, which seriously affects the scheduling effect. The present application extracts multi-level time series features from the preprocessed power grid load data, including hourly load trend, daily cycle fluctuation feature, weekly regularity feature and meteorological factor correlation feature. These features can fully reflect the internal law of load change, for example, the hourly trend captures the rising or falling momentum of the load within a day, the daily cycle feature reflects the peak and valley changes within 24 hours, the weekly regularity feature distinguishes the load difference between weekdays and weekends, and the meteorological factor correlation feature correlates environmental factors such as temperature, humidity and wind speed with load changes, achieving accurate quantification of weather impact. On this basis, the long short-term memory network model with attention mechanism further improves the feature processing capability. The traditional long short-term memory network model has fixed weight distribution for historical data, which cannot highlight the influence of key period data, while the attention mechanism can dynamically adjust the weight of different historical period data, layer by layer excavate high-order features from the original data, and finally output high-precision prediction results through the weighted integration of the attention layer. The Adam optimizer is used to minimize the mean square error between the predicted value and the actual value, and through dynamic adjustment of the learning rate and early stopping mechanism, the model is prevented from overfitting or insufficient convergence. This high-precision load forecasting result provides a reliable input for the subsequent optimization scheduling model, which can effectively reduce the load shortage or redundant power generation caused by prediction bias and reduce the overall cost. For example, during the load peak period, accurate prediction can schedule thermal power units to increase output and release the stored energy of energy storage devices in advance, avoiding power supply shortage; during the low valley period, the units can be arranged to shut down or the energy storage devices can be charged, reducing energy waste. It can be said that the load forecasting model of the present method realizes the prediction accuracy that traditional models cannot achieve through deep analysis of complex features and dynamic weight distribution.

[0040] The process of constructing the optimization scheduling model includes: C1: determining the decision variables and time scale of the optimization scheduling model; the decision variables include the output variables of thermal power units and the charge and discharge variables of energy storage devices; the output variables of the thermal power units refer to the output of each thermal power unit in each scheduling period, and the charge and discharge variables of the energy storage devices refer to the charging power or discharging power of each energy storage device in each scheduling period; It needs to be understood that the traditional thermal power unit scheduling model often only focuses on a single target, ignores the synergy of energy storage devices and grid safety constraints, resulting in a scheduling scheme that is either economically insufficient or has safety hazards. However, the present method explicitly takes the thermal power unit output variable and the energy storage device charging and discharging variable as the core decision variable, covering the running state of each unit, each energy storage device in each scheduling period, and achieving global regulation of multiple devices and multiple periods. This fine decision variable setting allows the model to accurately allocate tasks to each device, avoiding resource waste or load imbalance.

[0041] C2: Construct a target function for minimizing the comprehensive cost based on the decision variables, and the target function expression is: wherein, represents the fuel cost of the thermal power unit, represents the start-stop cost of the thermal power unit, represents the charging and discharging loss cost of the energy storage device, represents the start-stop cost of the thermal power unit, represents the charging and discharging loss cost of the energy storage device, represents the cycle life loss cost of the energy storage device, represents the grid safety and load shortage penalty cost; Further, the fuel cost is calculated based on the coal consumption characteristic curve of each thermal power unit, which is a relationship curve obtained by nonlinear fitting of the output and corresponding coal consumption in historical operation data. The specific form of the relationship curve is: Then the fuel cost can be calculated as: wherein, represents, represents the fitting coefficient of the coal consumption characteristic curve of the i th thermal power unit, which is obtained by least squares regression from the historical operation data of the unit, represents the output of the i th thermal power unit in period t represents the fuel cost of the th thermal power unit in period i represents the fitting coefficient of the coal consumption characteristic curve of the t th thermal power unit in period represents the length of the scheduling period, represents the unit price of coal, A represents the total number of thermal power units in the system, and T represents the scheduling period.

[0042] Further, the start-stop cost is calculated according to the number of start-stop events of the unit. For example, if the unit switches from a shutdown state to a startup state in period t , it generates a startup cost , and if it switches from a running state to a shutdown state, it generates a shutdown cost , then the start-stop cost is and The weighted sum of is obtained.

[0043] Furthermore, the charge and discharge loss cost is calculated based on the charge and discharge efficiency of the energy storage device and the real-time electricity price. Specifically, the charge and discharge loss cost is calculated as follows: ; in, Indicates the j Energy storage equipment in the period t The charge and discharge power loss, Indicates time period t The electricity price, and Represents charging and discharging power respectively, and They represent the charging and discharging efficiency respectively, and the range is [0, 1]. D represents the total number of energy storage devices.

[0044] Furthermore, the cycle life loss cost is calculated based on the total cycle life of the equipment, the initial purchase cost and the charge and discharge depth. The specific formula is: ; in, Indicates the j The life loss cost of a single charge and discharge cycle of an energy storage device, Indicates time period t The number of cycles, Represents the equipment purchase cost, Indicates the rated cycle life, Indicates rated capacity.

[0045] Furthermore, the penalty cost includes the sum of the line flow over-limit penalty and the load shortage penalty, where the line flow over-limit penalty is: ; The load shortfall penalty is: ; in, represents the penalty coefficient for exceeding the limit, Indicates line l Time t Transmission power, H represents the number of lines, Indicates time period t The thermal stability limit, Indicates a non-negative value. k represents the deficiency penalty constant, Indicates time period t The loss cost, Indicates time period t Network loss.

[0046] It needs to be explained that in the design of the objective function, the comprehensive cost minimization objective constructed by the method covers the fuel cost, start-stop cost of thermal power units, charging and discharging loss cost, cycle life loss cost of energy storage devices, and grid safety and load shortage penalty cost, forming a multi-dimensional cost control system. The traditional model often ignores the life loss of energy storage devices and the grid safety penalty, leading to problems such as excessive charging and discharging of energy storage devices and grid overload, while the method makes these implicit costs explicit, ensuring the economy of the dispatching scheme in the whole life cycle. For example, for the cycle life loss of energy storage devices, the model calculates the cost according to the charging and discharging depth, avoiding excessive consumption of energy storage life in pursuit of short-term benefits; for load shortage, higher penalty coefficients are set in peak periods to prioritize power supply stability in peak periods.

[0047] C3: incorporating the constructed thermal power unit output constraint sub-model, energy storage device charging and discharging constraints, and grid safety constraints in the objective function; Further, the construction of the constraint condition is the key to ensure the safety of the model, and its functionality is reflected in the all-round constraint coverage of thermal power units, energy storage devices, and the grid. The thermal power unit output constraint sub-model establishes a nonlinear relationship between output and coal consumption through the coal consumption characteristic curve, combined with parameters such as climbing rate and start-stop time, to ensure that the unit output adjustment is within the technical allowable range, avoiding equipment loss due to frequent start-stop or rapid climbing; the energy storage device charging and discharging constraint clearly defines the upper limit of charging and discharging power and the state of charge interval to prevent overcharging and overdischarging damage to the battery, while requiring the state of charge to return to the initial value at the end of the dispatching period to ensure sustainable use of energy storage devices; the grid safety constraint strictly limits power balance, line flow, and node voltage to ensure the stability of power transmission and avoid safety accidents such as line overload and voltage abnormalities.

[0048] Further, the coal consumption curve is associated as a basic constraint, which can obtain the upper and lower limits of the output, while the climbing rate is a dynamic adjustment condition, which is associated with the upper and lower limits of the output, and finally the start-stop time constraint is a state switching constraint, which is associated with the climbing rate and the upper and lower limits of the output, which is a progressive relationship. Through the hierarchical association of coal consumption curve association-climbing rate-start-stop time, the organic combination of coal consumption characteristic curve and climbing rate and start-stop time is realized, ensuring that the unit output adjustment is both economical and safe.

[0049] It also needs to be clear that these constraints are not isolated, but are integrated through the model to form a network of interrelated constraints. For example, the power balance constraint links the output of thermal power units, the charging and discharging of energy storage, the load, and the network loss, ensuring real-time balance between supply and demand. The line flow constraint limits the transmission power of each line, indirectly affecting the output allocation of thermal power units and energy storage devices. The coordinated action of these multiple constraints ensures that the dispatching scheme meets the economic requirements while complying with the safety operation standards, which cannot be achieved by traditional single constraint models.

[0050] C4: integrating the objective function and all the above constraints to form an optimal dispatching model; the optimal dispatching model takes the output variables of thermal power units and the charging and discharging variables of energy storage devices as the solution objects, and outputs a dispatching scheme parameter that simultaneously satisfies the minimum comprehensive cost and all the constraints.

[0051] Further, the thermal power unit output constraint sub-model, the energy storage device charging and discharging constraint, and the power grid safety constraint in the optimal dispatching model are not isolated, but form a constraint network through decision variable coupling, state variable transmission, and objective function feedback. The power balance constraint in the power grid safety constraint is the core hub of the constraint network, directly coupling the thermal power unit output variable and the energy storage charging and discharging variable. The state of charge of the energy storage device is indirectly related to the thermal power unit output through the charging and discharging power. The unit start-stop state affects the energy storage charging and discharging demand through output adjustment, forming a state-variable-state transmission chain, and ultimately forming a transmission relationship between unit start-stop, energy storage charging and discharging, state of charge, and unit early output adjustment.

[0052] The process of constructing the thermal power unit output constraint sub-model includes: D1: collecting historical operation data of thermal power units, and extracting coal consumption characteristic curves and output response rate parameters of the thermal power units; The coal consumption characteristic curve is obtained by nonlinear fitting of historical output values and corresponding coal consumption using a cubic polynomial regression method. Its expression is a function of coal consumption with respect to output. The specific steps include: (1) Obtain the historical operation data of the thermal power unit and perform preprocessing; (2) Divide the unit output range into M intervals with a certain granularity to balance the fitting accuracy and computational efficiency. Each interval is a data group, and the grouping purpose is to ensure that the coal consumption characteristics of each output segment can be fully fitted, avoiding excessive local deviation due to excessively wide output range; (3) Calculate statistical values for each data group, including the average output, average coal consumption, and coal consumption standard deviation in the interval to determine the coal consumption stability of the interval. If the coal consumption standard deviation of any interval exceeds 10% of the average, it indicates that the interval has large fluctuations in operating state, and the interval needs to be further subdivided, and whether there are any unprocessed outliers needs to be checked; (4) Based on the physical law of coal consumption characteristics, a cubic polynomial is selected as the fitting model. For example, if the unit coal consumption changes in a U-shaped or V-shaped curve with the output, there is an economic output point. Compared with a linear or quadratic polynomial, the cubic polynomial can more accurately capture the different trends of coal consumption at high, medium and low output levels; (5) The cubic polynomial obtained by fitting is used to calculate the coal consumption at each historical output point, and compared with the actual coal consumption, the fitting effect is verified by multiple indicators.

[0053] The output response rate parameter is determined by the climbing rate and the descending rate in the historical operation data, and the minimum start-up time from shutdown to start-up completion and the minimum shutdown time from operation to shutdown are recorded to form start-stop constraints; the climbing rate is the maximum output increment per unit time of the unit; the descending rate is the maximum output decrement per unit time of the unit.

[0054] D2: Establish a nonlinear mapping relationship between the output of the thermal power unit and the coal consumption based on the coal consumption characteristic curve, and obtain the coal consumption calculation value corresponding to the output level; D3: Combine the output response rate parameter to set the upper and lower limits of the output adjustment per unit time of the thermal power unit, and form the climbing constraints of the thermal power unit; D4: Based on the output response rate parameter and the upper and lower limits of the output adjustment, a thermal power unit output constraint sub-model is constructed; the thermal power unit output constraint sub-model includes output upper and lower limit constraints, climbing constraints and start-stop constraints.

[0055] The charge and discharge constraints of the energy storage device include charge and discharge power constraints and state of charge constraints; The charge and discharge power constraint means that the charging power does not exceed the rated charging power, the discharging power does not exceed the rated discharging power, and the charging and discharging are not performed at the same time; The state of charge constraint means that the state of charge is maintained within a predetermined safe interval and restored to the initial value at the end of the dispatching period. The state of charge is the ratio of the current stored energy to the rated capacity, which is calculated based on the state of charge in the last period, the charge and discharge amount in the current period, and the charge and discharge efficiency.

[0056] The grid safety constraints include power balance constraints, line flow constraints and node voltage constraints; The power balance constraint means that the sum of the output of the thermal power unit and the effective discharge amount of the energy storage is equal to the sum of the load, the network loss and the charging consumption of the energy storage; The line flow constraint means that the line transmission power does not exceed the thermal stability limit; The node voltage constraint means that the node voltage is maintained between 95% and 105% of the rated voltage.

[0057] The improved intelligent optimization algorithm is used to solve the optimization scheduling model to obtain an optimal scheduling scheme, including: E1: initializing a particle swarm, each particle representing a scheduling scheme, the dimension of the particle corresponding to the number of total scheduling periods, and the dimension value of each dimension respectively representing the output of all thermal power units and the charging and discharging power of all energy storage devices in the period; Further, the initialization range of the particle swarm is determined based on the constraint boundary of the decision variable, the initialization range of the thermal power unit output being the minimum output to the maximum output, and the initialization range of the charging and discharging power of the energy storage device being the rated discharging power to the rated charging power, wherein a negative value represents discharging; the initialized particle swarm is used as the initial population for the particle update operation.

[0058] E2: performing a particle update operation based on the initialized particle swarm to generate a new particle swarm, including: randomly selecting two particles as parent particles, randomly selecting 2-3 continuous or non-continuous scheduling periods in all dimensions of the particles, exchanging the dimension values of the two particles in the selected periods to form two child particles; for each child particle, randomly selecting a dimension corresponding to a scheduling period, and randomly generating a new value within the constraint range of the dimension corresponding to the scheduling period to replace the original value, to obtain an initial updated particle swarm; performing constraint checking on the initial updated particle swarm, if there are values that do not meet the thermal power unit output constraint, the energy storage device charging and discharging constraint, or the power grid safety constraint, then correcting through a constraint repair mechanism to finally form a new particle swarm; the constraint range of the dimension corresponding to the scheduling period refers to the allowed range of the thermal power unit output in the period and the allowed range of the charging and discharging power of the energy storage device; the constraint repair mechanism refers to adjusting the out-of-limit output value to the nearest constraint boundary; Further, in the constraint repair mechanism, if the thermal power unit output is out of limit, it is adjusted to the nearest upper and lower limit of the output, i.e., when the output is greater than the maximum output, it is adjusted to the maximum output, and when the output is less than the minimum output, it is adjusted to the minimum output; the adjusted output value is used to generate a new particle that meets the constraint.

[0059] E3: calculating the fitness value of each particle in the new particle swarm; the fitness value is obtained based on the objective function of the optimization scheduling model, wherein the calculation process of the fitness value is the prior art content in the field and is not the creative scheme of the present application, and is not described here. E4: comparing the fitness value of each particle in the new particle group with its historical optimal fitness value, if the fitness value of each particle in the current new particle group is less than the historical optimal fitness value, updating the historical optimal position of the particle as the current particle position, and selecting the position with the minimum fitness value from the historical optimal positions of all particles as a candidate of the global optimal position, if the fitness value of the candidate position is less than the fitness value of the current global optimal position, updating the global optimal position; E5: selecting the particle corresponding to the global optimal position as the optimized scheduling scheme after iteration to a preset number of times.

[0060] It should be understood that the traditional optimization algorithm often faces problems such as slow convergence and easy falling into local optimum when dealing with multi-constraint and high-dimensional scheduling problems, resulting in too long solving time or the obtained scheme being not globally optimal. The improved intelligent optimization algorithm of the method is optimized in terms of particle initialization, updating operation and constraint processing, which ensures that the optimal solution meeting all constraint conditions can be quickly found.

[0061] Further, in the particle group initialization stage, the algorithm determines a reasonable initialization range according to the constraint boundary of the decision variable, the thermal power output is limited between the minimum and maximum output, and the charging and discharging power of the energy storage is limited within the rated range. This constraint-based initialization method avoids the generation of invalid solutions and improves the quality of the initial population. At the same time, the number of particles is set to 5-10 times the dimension of the decision variable, which ensures the diversity of the population and lays a foundation for global optimization.

[0062] Further, after the particle is updated, if the thermal power output exceeds the limit, the energy storage charging and discharging is illegal, or the grid safety constraint is not satisfied, the algorithm will automatically adjust the exceeding value to the nearest constraint boundary, for example, the thermal power output exceeding the maximum output is reduced to the maximum value, and the state of charge exceeding the standard of the energy storage device is adjusted to the safety interval. This real-time repair function ensures that all particles always meet the constraint conditions, avoids the interference of invalid solutions on the optimization process, and improves the algorithm efficiency.

[0063] Further, compared with the traditional algorithm, the improved intelligent optimization algorithm of the method improves the solving speed and can maintain stable optimization ability in large-scale scheduling problems, and reduces the comprehensive cost of the scheduling scheme. This efficient and accurate solving ability enables the optimization scheduling model to be transformed from theoretical design to practical scheme, providing a feasible operation basis for real-time scheduling, and its functional advantages are unmatched by traditional algorithms.

[0064] The real-time scheduling of the multi-energy storage thermal power unit according to the optimal scheduling scheme comprises: F1: the optimized dispatching scheme is disassembled into the output instructions of the thermal power units and the charging and discharging instructions of the energy storage devices in each time period according to the dispatching time period; the output instruction of the thermal power unit is the target output value of each thermal power unit in the corresponding time period; the charging and discharging instruction of the energy storage device is the target charging power or discharging power of each energy storage device in the corresponding time period; Further, the process of disassembling the optimized dispatching scheme into the output instructions of the thermal power units in each time period includes: (1) determining the time granularity of the dispatching time period, such as 15 minutes, 30 minutes, 1 hour, etc., which is determined by the dispatching accuracy requirement; (2) clearly defining the time boundaries of each time period, such as time period 1 being 0:00-1:00, time period 2 being 1:00-2:00, etc.; (3) combing the overall time range covered by the optimized dispatching scheme, such as 24 hours, 48 hours, etc., and dividing the overall time range into multiple time periods that are continuous and non-overlapping according to the time period granularity determined in the first step, forming a clear time sequence framework, for example, 24 hours of dispatching can be divided into 24 time periods according to 1 hour granularity, at the same time, the unique identification of each time period needs to be marked, such as time period numbers 1 to N, to facilitate the matching of the subsequent instructions and time periods; (4) in the optimized dispatching scheme, finding the target output value part of each thermal power unit in each time interval, which contains the output sequence of each thermal power unit changing over time; (5) for each thermal power unit, matching each output value in its output sequence with the time period divided in the second step, if the time interval corresponding to the output value completely coincides with the time boundary of a certain time period, such as a 1-hour time interval, which is consistent with the time period granularity, then directly determining the output value as the target output value of the corresponding time period; if the time interval is different from the time period granularity, such as a 30-minute time interval in the scheme and a 1-hour time period granularity, then according to the time range covered by the time period, the output values of all time intervals in the time period need to be integrated as the target output value of the time period; (6) for each thermal power unit, repeating the matching process to obtain its target output value in each time period, which is the output instruction of the unit in the corresponding time period.

[0065] Further, the process of disassembling the optimized dispatching scheme into the charging and discharging instructions of the energy storage devices in each time period includes: (1) in the optimized dispatching scheme, finding the charging and discharging power curve part of each energy storage device, which is the power sequence of each energy storage device changing over time, including the charging and discharging directions, usually represented by positive values for charging and negative values for discharging, or distinguished by text labels; (2) For each energy storage device, split its charging and discharging power curve according to the time period divided in the second step. If the time resolution of the power curve is consistent with the time period granularity, such as one power value per hour, directly extract the power value corresponding to each time period as the charging and discharging power of that time period. If the resolution is different, such as a power curve with one value every 15 minutes and a time period of 1 hour, integrate the power values within the time period, for example, calculate the average power within the time period by time proportion weighting or take the average power within the time period, while retaining the identification of the charging and discharging direction; (3) For each energy storage device, repeat the above matching process to obtain its target charging power or discharging power in each time period, which is the charging and discharging instruction of the energy storage device in the corresponding time period.

[0066] F2: Send the decomposed thermal power unit output instruction to the thermal power unit control system and send the energy storage device charging and discharging instruction to the energy storage management system. Control the thermal power unit to adjust the real-time output according to the target output value through the control system, and control the energy storage device to charge or discharge according to the target power through the energy storage management system; F3: Monitor the operating state of the multi-energy storage thermal power unit and the actual value of the grid load in real time during the operation of the thermal power unit and the energy storage device according to the instructions; F4: Call the corresponding time period load prediction result output by the load prediction model, and compare the monitored actual value of the grid load with the corresponding load prediction result to calculate the deviation value; When the deviation value does not exceed the preset threshold, the current optimal scheduling scheme corresponding instruction is maintained, and the device continues to operate through the control system; When the deviation value exceeds the preset threshold, the optimal scheduling scheme is dynamically adjusted based on the real-time operation data to generate an adjusted scheduling scheme; F5: Convert the adjusted scheduling scheme into thermal power unit output adjustment instructions and energy storage device charging and discharging adjustment instructions for the corresponding time period, and send them to the corresponding control system to control the thermal power unit and the energy storage device to operate according to the adjusted instructions.

[0067] Further, the specific process of converting the adjusted scheduling scheme into thermal power unit output adjustment instructions and energy storage device charging and discharging adjustment instructions for the corresponding time period includes: (1) Receive and analyze the adjusted scheduling scheme; (2) Determine the original instruction reference value and the adjustment interval; It needs to be explained that the original instruction reference value is the output or charging and discharging instruction value before adjustment. For the thermal power unit, the original instruction reference value is the target output value of the time period in the last round of scheduling, and if it is the first adjustment, it is the instruction value of the initial optimal scheduling scheme. For the energy storage device, the original reference value is the target charging and discharging power of the time period in the last round of instruction.

[0068] (3) The output adjustment amount of the thermal power unit is obtained by calculating the difference between the target output value and the original output instruction value of the period in the adjusted scheme; (4) The charge and discharge adjustment amount of the energy storage device is obtained by calculating the difference between the target power and the original charge and discharge instruction value of the period in the adjusted scheduling scheme; (5) According to the output adjustment amount of the thermal power unit and the charge and discharge adjustment amount of the energy storage device, the output adjustment amount of the thermal power unit and the charge and discharge adjustment amount of the energy storage device are converted into standardized instructions containing device identification, period, target value, adjustment amount, priority, and check code; (6) The encrypted instructions are transmitted to the control system according to the priority.

[0069] Further, receiving and analyzing the adjusted scheduling scheme includes: First, the time range of the adjusted scheduling scheme is determined, and it is split according to the original scheduling period granularity, and each to-be-adjusted period number is labeled. Second, two types of core data are extracted from the adjusted scheduling scheme: one is the adjustment information of the thermal power unit, including the new target output value of each unit in each to-be-adjusted period; the other is the adjustment information of the energy storage device, including the new target charging power or discharging power of each device in the corresponding period.

[0070] The embodiments of the present application are described above in combination with the drawings, but the present application is not limited to the above specific embodiments, and the above specific embodiments are only illustrative and not limiting. Those skilled in the art can make changes, modifications, replacements and variations to the above embodiments without departing from the purpose and scope of the present application, which are all within the protection scope of the present application.

Claims

1. A method for optimizing the scheduling of multiple energy storage thermal power units, characterized in that: include: Collect operating data of multiple energy storage thermal power units and grid load data; Building a load forecasting model based on the operation data and the grid load data, and obtaining a load forecasting result through the load forecasting model; With the goal of minimizing comprehensive costs, an optimal dispatch model is constructed by combining the thermal power unit output constraint sub-model, energy storage device charging and discharging constraints, and grid security constraints. An improved intelligent optimization algorithm is used to solve the optimization scheduling model to obtain an optimal scheduling plan, and multiple energy storage thermal power units are scheduled in real time according to the optimal scheduling plan.

2. The method for optimizing the scheduling of multiple energy storage thermal power units according to claim 1, characterized in that: The optimal scheduling plan includes the target output value and start-stop status instructions of each thermal power unit in each time interval, the charge and discharge power curve of each energy storage device, the SOC target trajectory, the allocation of spinning reserve capacity between each thermal power unit and energy storage device, as well as an accompanying plan economic analysis report and a network loss control strategy formulated based on load forecast results and grid structure.

3. The method for optimizing and dispatching multiple energy storage thermal power units according to claim 1, characterized in that: The collection of operating data of multiple energy storage thermal power units and grid load data includes: The real-time output, coal consumption rate, start-stop status data of thermal power units, as well as the charge status, charge and discharge power, and charge and discharge efficiency data of energy storage devices are collected through sensors and data transmission networks as operating data of multi-energy storage thermal power units; Collect historical load curves, real-time load values, load fluctuation amplitudes, peak and valley time divisions, and load forecast error intervals as grid load data; The operating data of multiple energy storage thermal power units and the grid load data are aligned in time and space through edge computing nodes, and stored through blockchain technology.

4. The method for optimizing and dispatching multiple energy storage thermal power units according to claim 1, wherein: The step of constructing a load forecasting model based on the operation data and the grid load data, and obtaining a load forecasting result through the load forecasting model, includes: Extracting time series features from pre-processed grid load data; the time series features include hourly load trends, daily cycle fluctuation characteristics, weekly regularity characteristics, and meteorological factor correlation characteristics; Based on the time series features, a long short-term memory network integrated with an attention mechanism is constructed as a load forecasting model; the attention mechanism is used to dynamically assign weights to data from different historical periods; The dataset containing historical time series features and corresponding historical load values ​​is used as the training set. The Adam optimizer is used to minimize the mean square error between the predicted value and the actual value to train the constructed load forecasting model and obtain the trained load forecasting model. The real-time time series features obtained from the real-time collected power grid load data are input into the trained load forecasting model to obtain the load forecast results for the preset time period.

5. The method for optimizing and dispatching multiple energy storage thermal power units according to claim 1, characterized in that: The process of constructing the optimization scheduling model includes: Clarify the decision variables and time scale of the optimization scheduling model; the decision variables include the output variables of the thermal power units and the charge and discharge variables of the energy storage devices; the output variables of the thermal power units refer to the output of each thermal power unit in each scheduling period, and the charge and discharge variables of the energy storage devices refer to the charging power or discharging power of each energy storage device in each scheduling period; Constructing an objective function for minimizing comprehensive costs based on the decision variables; the comprehensive costs include fuel costs of thermal power units, start-up and shutdown costs of thermal power units, charging and discharging loss costs of energy storage devices, cycle life loss costs of energy storage devices, and grid security and load shortfall penalty costs; Incorporating the constructed thermal power unit output constraint sub-model, energy storage device charging and discharging constraints, and power grid security constraints into the objective function; The objective function is integrated with all the above constraints to form an optimization scheduling model; the optimization scheduling model takes the output variables of the thermal power units and the charge and discharge variables of the energy storage equipment as the solution objects, and outputs the scheduling plan parameters that simultaneously meet the minimum comprehensive cost and all constraints.

6. The method for optimizing and dispatching multiple energy storage thermal power units according to claim 5, characterized in that: The process of constructing the thermal power unit output constraint sub-model includes: Collect historical operating data of thermal power units and extract coal consumption characteristic curves and output response rate parameters of the thermal power units; the coal consumption characteristic curve is obtained by nonlinearly fitting historical output values ​​and corresponding coal consumption using a cubic polynomial regression method, and its expression is a function of coal consumption with respect to output; the output response rate parameter is determined by statistically analyzing the ramp-up rate and ramp-down rate in historical operating data, while also recording the minimum start-up time from shutdown to startup and the minimum shutdown time from operation to shutdown of the unit to form a start-stop constraint; the ramp-up rate is the maximum output increase of the unit per unit time; the ramp-down rate is the maximum output decrease of the unit per unit time; Based on the coal consumption characteristic curve, a nonlinear mapping relationship between the output of the thermal power unit and the coal consumption is established to obtain a calculated value of the coal consumption at the corresponding output level; Combined with the output response rate parameter, the upper and lower limits of the output adjustment per unit time of the thermal power unit are set to form the climbing constraint of the thermal power unit; Based on the output response rate parameter and the output adjustment upper and lower limits, a thermal power unit output constraint submodel is constructed; the thermal power unit output constraint submodel includes output upper and lower limit constraints, climbing constraints and start-stop constraints.

7. The method for optimizing and dispatching multiple energy storage thermal power units according to claim 6, characterized in that: The energy storage device charging and discharging constraints include charging and discharging power constraints and state of charge constraints; The charge and discharge power constraints mean that the charging power does not exceed the rated charging power, the discharging power does not exceed the rated discharging power, and charging and discharging are not performed at the same time; The state of charge constraint means that the state of charge remains within a preset safety range and returns to the initial value at the end of the scheduling cycle. The state of charge is the ratio of the current stored energy to the rated capacity, which is calculated based on the state of charge in the previous period, the charge and discharge amount in the current period, and the charge and discharge efficiency.

8. The method for optimizing and dispatching multiple energy storage thermal power units according to claim 7, characterized in that: The grid security constraints include power balance constraints, line flow constraints and node voltage constraints; The power balance constraint means that the sum of the output of the thermal power units and the effective discharge of the energy storage is equal to the sum of the load, network loss and energy storage charging consumption; The line power flow constraint means that the line transmission power does not exceed the thermal stability limit; The node voltage constraint means that the node voltage is maintained between 95% and 105% of the rated voltage.

9. The method for optimizing and dispatching multiple energy storage thermal power units according to claim 1, characterized in that: The improved intelligent optimization algorithm is used to solve the optimization scheduling model to obtain the optimal scheduling solution, including: Initialize the particle swarm. Each particle represents a scheduling scheme. The dimension of the particle corresponds to the total number of scheduling periods, and the value of each dimension represents the output of all thermal power units and the charge and discharge power of all energy storage devices in that period. A particle update operation is performed based on the initialized particle swarm to generate a new particle swarm, including: randomly selecting two particles as parent particles, randomly selecting 2-3 continuous or discontinuous scheduling periods in all dimensions of the particles, exchanging the dimension values ​​corresponding to the two particles in the selected periods, and forming two child particles; for each child particle, randomly selecting a dimension corresponding to a scheduling period, and randomly generating a new value within the constraint range of the dimension corresponding to the scheduling period to replace the original value, to obtain an initially updated particle swarm; performing constraint verification on the initially updated particle swarm, and if there are values ​​that do not meet the output constraints of the thermal power unit, the charging and discharging constraints of the energy storage device, or the safety constraints of the power grid, correcting them through a constraint repair mechanism, and finally forming a new particle swarm; the constraint range of the dimension corresponding to the scheduling period refers to the allowable range of the output of the thermal power unit and the allowable range of the charging and discharging power of the energy storage device in the period; the constraint repair mechanism refers to adjusting the output value that exceeds the limit to the nearest constraint boundary; Calculating the fitness value of each particle in the new particle swarm; the fitness value is obtained based on the weighted objective function of the optimization scheduling model; Compare the fitness value of each particle in the new particle swarm with its historical optimal fitness value. If the fitness value of each particle in the current new particle swarm is less than the historical optimal fitness value, then update the historical optimal position of the particle to the current particle position. At the same time, select the position with the smallest fitness value from the historical optimal positions of all particles as the candidate for the global optimal position. If the fitness value of the candidate position is less than the fitness value of the current global optimal position, then update the global optimal position. After iterating to a preset number of times, the particle corresponding to the global optimal position is selected as the optimized scheduling solution.

10. The method for optimizing and dispatching multiple energy storage thermal power units according to claim 1, characterized in that: The real-time scheduling of multiple energy storage thermal power units according to the optimal scheduling scheme includes: Decomposing the optimized scheduling plan into thermal power unit output instructions and energy storage device charge and discharge instructions for each time period according to the scheduling period; the thermal power unit output instruction is the target output value of each thermal power unit in the corresponding time period; the energy storage device charge and discharge instruction is the target charging power or discharging power of each energy storage device in the corresponding time period; The disassembled thermal power unit output command is sent to the thermal power unit control system, and the energy storage device charge and discharge command is sent to the energy storage management system. The control system controls the thermal power unit to adjust the real-time output according to the target output value, and the energy storage management system controls the energy storage device to charge or discharge according to the target power; When the thermal power units and energy storage equipment are operating according to instructions, the operating status of multiple energy storage thermal power units and the actual value of the grid load are monitored in real time; Call the load forecast result of the corresponding period output by the load forecast model, compare the actual value of the monitored power grid load with the corresponding load forecast result, and calculate the deviation value; When the deviation value does not exceed the preset threshold, the instructions corresponding to the current optimal scheduling solution are maintained, and the operation of the equipment continues to be controlled by the control system; When the deviation value exceeds a preset threshold, the optimal scheduling plan is dynamically adjusted based on the real-time operation data to generate an adjusted scheduling plan; The adjusted dispatch plan is converted into the thermal power unit output adjustment instructions and energy storage device charge and discharge adjustment instructions for the corresponding time period, and sent to the corresponding control system to control the thermal power unit and energy storage device to operate according to the adjusted instructions.

Images