Power resource scheduling method and system embedded with time sequence coupling constraint and storage medium

Abstract

The application provides a power resource scheduling method and system embedded with time sequence coupling constraints and a storage medium, relates to the technical field of energy management, and comprises the following steps: acquiring a classical security constraint economic scheduling model; performing recursive calculation on the classical security constraint economic scheduling model; wherein, in the forward propagation process of the recursive calculation process of the set neural network model, the parameters of the neural network at each time point remain consistent, and the output of the current time step calculation unit is taken as the input of the corresponding calculation unit of the next time step to connect the calculation units; the constraint condition is embedded into the output end of the set neural network model, and a gradient backpropagation information highway is constructed; the set neural network model is trained by using power training data, and the neural network parameters are adjusted to obtain a power scheduling model, so that a power resource scheduling scheme is output. The application significantly improves the calculation efficiency through coupling calculation units, parameter sharing and a recursive structure, and can quickly generate a power resource scheduling scheme.

Description

Technical Field

[0001] This application relates to the field of energy management technology, and in particular to a power resource scheduling method, system, medium, device and program product with embedded time-series coupling constraints. Background Technology

[0002] With the advancement of new power system construction, new energy sources are gradually becoming the dominant power source. The complexity and variability of their output increase the difficulty of maintaining power balance in the power system, leading to a continuous increase in the frequency of Security Constrained Economic Dispatch (SCED) calls, which places higher demands on its solution speed and the feasibility of results. However, existing technologies have the following shortcomings: First, deep reinforcement learning has low training efficiency, making it difficult to guarantee model results. Although applicable to continuous sequence decision problems, practical applications suffer from difficulties in recreating complex real-world scenarios in simulation environments, challenges in designing reward functions, and low training efficiency due to disordered exploration mechanisms. Second, end-to-end deep learning models are mostly used for single-timescale optimal power flow problems, making it difficult to handle classic security constrained economic dispatch models with multi-timescale and temporally coupled constraints, thus limiting their application scope. Third, RNN models suffer from gradient explosion / vanishing problems, leading to model training failures. Therefore, an efficient and high-quality SCED solution method is urgently needed to meet the complex and variable operational requirements of new power systems. Summary of the Invention

[0003] The purpose of this application is to provide a power resource scheduling method, system, computer-readable storage medium, and electronic device with embedded temporal coupling constraints, which can effectively realize the safe and economical scheduling of power resources.

[0004] To address the aforementioned technical problems, this application provides a power resource scheduling method embedding time-series coupling constraints, the specific technical solution of which is as follows:

[0005] Obtain the classic security-constrained economic dispatch model corresponding to power resources;

[0006] The classical safety-constrained economic dispatch model is recursively calculated using a defined neural network model. During the forward propagation of the recursive calculation process, the parameters of the neural network remain consistent at each time point. Simultaneously, the output of the calculation unit at the current time step is used as the input to the corresponding calculation unit at the next time step, thus connecting the calculation units of the defined neural network model. For each calculation unit at each time step, its input is the load demand at the current time step and the generator active power output at the previous time step, and its output is the generator's active power at the current time step.

[0007] The constraints of the classical safety-constrained economic scheduling model are embedded into the output of the set neural network model, and a gradient backhaul information highway is constructed without going through the neural network path.

[0008] The set neural network model is trained using power training data, and the neural network parameters are adjusted using the backpropagation algorithm to obtain the power dispatch model;

[0009] Based on the power dispatching model and real-time power resource data, a power resource dispatching scheme is output.

[0010] Optionally, the security-constrained scheduling model for obtaining power resources includes:

[0011] Construct an objective function that minimizes the total cost of electricity generation;

[0012] Obtain the constraints; the constraints include power balance constraints, generator active power output constraints, ramping constraints, and transmission line thermal stability limit constraints.

[0013] Optionally, after embedding the constraints of the classical security-constrained economic scheduling model into the computational units of the defined neural network model, the method further includes:

[0014] Control factors are generated through a neural network, and these control factors are used to indicate that the output of the power dispatch model satisfies the constraints.

[0015] Optionally, generating control factors via neural networks includes:

[0016] Calculate the maximum possible output of the generator at the current time step; the maximum possible output is the smaller value between the generator's maximum output and the first unit's ramp-up capability; the first unit's ramp-up capability is the sum of the generator's output at the previous time step and the allowable ramp-up amount;

[0017] Calculate the minimum possible output of the generator at the current time step; the minimum possible output is the larger of the generator's minimum output and the unit's ramp-up capability; the second unit's ramp-up capability is the difference between the generator's output at the previous time step and the allowable ramp-up amount;

[0018] Calculate the actual active power output of the generator at the current time step; the actual active power output falls between the minimum possible output and the maximum possible output, and is adjusted by a control factor generated by a loss function;

[0019] Calculate the relaxation power at the current time step; the relaxation power is the sum of all load demands minus the sum of the output of all locked generators.

[0020] Optionally, adjusting the neural network parameters using the backpropagation algorithm includes:

[0021] The correlation characteristics between adjacent time steps are determined based on the maximum possible output, the minimum possible output, and the control factor.

[0022] The gradient information of the loss function corresponding to the set neural network model is generated based on the information features of the time step and the correlation features between adjacent time steps.

[0023] Optional, also includes:

[0024] Calculate the first partial derivative between the maximum possible output and the generator output of the previous time step;

[0025] Calculate the second partial derivative between the minimum possible output and the generator output of the previous time step;

[0026] The control factor is calculated based on the difference between the first partial derivative and the second partial derivative.

[0027] This application also provides a power resource scheduling system with embedded time-series coupling constraints, including:

[0028] The scheduling model acquisition module is used to acquire the classic security-constrained economic scheduling model corresponding to power resources;

[0029] The recursive calculation module is used to recursively calculate the classic safety-constrained economic dispatch model using a defined neural network model. During the forward propagation of the recursive calculation process of the defined neural network model, the parameters of the neural network remain consistent at each time point. Simultaneously, the output of the calculation unit at the current time step is used as the input of the corresponding calculation unit at the next time step, thus connecting the calculation units of the defined neural network model. For each calculation unit at each time step, its input is the load demand at the current time step and the generator active power output at the previous time step, and its output is the generator's active power at the current time step.

[0030] The constraint setting module is used to embed the constraints of the classic safety constraint economic scheduling model into the output of the set neural network model, and to construct a gradient backhaul information highway that does not pass through the neural network path.

[0031] The model training module is used to train the set neural network model using power training data, and adjust the neural network parameters through the backpropagation algorithm to obtain the power dispatch model;

[0032] The power resource scheduling module is used to output a power resource scheduling scheme based on the power scheduling model and real-time power resource data.

[0033] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0034] This application also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method described above when it invokes the computer program in the memory.

[0035] This application also provides a computer program product, including a computer program that implements the steps of the method described above.

[0036] This application provides a power resource scheduling method with embedded time-series coupling constraints, comprising: acquiring a classical security-constrained economic scheduling model corresponding to the power resources; recursively calculating the classical security-constrained economic scheduling model using a set neural network model; wherein, during the forward propagation process of the set neural network model's recursive calculation, the parameters of the neural network at each time point remain consistent, and the output of the current time step calculation unit is used as the input of the corresponding calculation unit in the next time step to connect the calculation units of the set neural network model; for each time step calculation unit, its input is the load demand of the current time step and the generator active power output of the previous time step, and its output is the generator's active power in the current time step; embedding the constraints of the classical security-constrained economic scheduling model into the output of the set neural network model, and constructing a gradient backhaul information highway that does not pass through the neural network path; training the set neural network model using power training data, and adjusting the neural network parameters through a backpropagation algorithm to obtain a power scheduling model; and outputting a power resource scheduling scheme based on the power scheduling model and real-time power resource data.

[0037] This application employs a neural network model to recursively calculate the classic security-constrained economic dispatch model. During forward propagation, the neural network parameters remain consistent across time points, and the output of the current time step is used as the input for the next time step. This reduces redundant calculations and significantly improves computational efficiency through parameter sharing and recursive structure, enabling rapid generation of power resource dispatch schemes to meet the complex and ever-changing operational needs of new power systems. By constructing a high-speed gradient backhaul information path that bypasses the neural network, the vanishing or exploding gradient problems that may occur during traditional neural network training are effectively avoided, further improving the model's training efficiency and stability, and ensuring good convergence performance even on long-term time series problems. Furthermore, this application embeds the constraints of the classic security-constrained economic dispatch model into the output of the neural network model, ensuring that the generated dispatch scheme strictly meets the security constraints of the power system. This allows the model to fully consider the physical characteristics and operational requirements of the power system when optimizing the dispatch scheme, avoiding system operational risks caused by violation of constraints and improving the feasibility and reliability of the power resource dispatch scheme.

[0038] This application also provides a power resource scheduling system with embedded time-series coupling constraints, a computer-readable storage medium, an electronic device, and a computer program product, which have the above-mentioned beneficial effects, and will not be elaborated here. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of this application 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 embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0040] Figure 1 A flowchart illustrating a power resource scheduling method with embedded temporal coupling constraints, provided as an embodiment of this application;

[0041] Figure 2 This is a schematic diagram of the framework of a general-purpose loop connection and information superhighway provided in the embodiments of this application;

[0042] Figure 3 This is a schematic diagram of the internal structure of the computing unit in the neural network model provided in the embodiments of this application;

[0043] Figure 4 This is a schematic diagram of a power resource scheduling system structure with embedded temporal coupling constraints provided in an embodiment of this application;

[0044] Figure 5 This is a structural diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0046] This invention proposes a general mechanism framework that integrates cyclic linking and information superhighway. The cyclic linking mechanism enables the solution of continuous time-series problems, while the information superhighway mechanism constructs a gradient backpropagation information superhighway by embedding constraints into the neural network. The information superhighway mechanism enhances the model's understanding of the physical mechanism while ensuring that the cyclic linking mechanism does not suffer from gradient explosion / gradient vanishing problems.

[0047] See Figure 1 , Figure 1A flowchart illustrating a power resource scheduling method with embedded temporal coupling constraints provided in this application embodiment is included. The method comprises:

[0048] S101: Obtain the classic security-constrained economic dispatch model corresponding to power resources;

[0049] S102: The classical safety-constrained economic dispatch model is recursively calculated using a set neural network model; wherein, during the forward propagation process of the recursive calculation of the set neural network model, the parameters of the neural network at each time point remain consistent, and the output of the calculation unit at the current time step is used as the input of the corresponding calculation unit at the next time step to connect the calculation units of the set neural network model; for each calculation unit at each time step, its input is the load demand at the current time step and the active power output of the generator at the previous time step, and its output is the active power of the generator at the current time step;

[0050] S103: Embed the constraints of the classical safety-constrained economic scheduling model into the output of the set neural network model, and construct a gradient backhaul information highway that does not pass through the neural network path.

[0051] S104: The set neural network model is trained using power training data, and the neural network parameters are adjusted through the backpropagation algorithm to obtain the power dispatch model;

[0052] S105: Output a power resource scheduling scheme based on the power dispatching model and real-time power resource data.

[0053] Classic safety-constrained economic dispatch models can be obtained from relevant literature, industry standards, or materials provided by power system dispatching agencies. These models typically consider both the economic (e.g., minimizing generation costs) and safety (e.g., line power flow constraints, generator output limits, and reserve capacity constraints) factors of the power system. It is an optimization model with numerous constraints, used to determine how to allocate the output of each generator to achieve economic dispatch while ensuring the safe operation of the power system.

[0054] One feasible approach involves constructing an objective function that minimizes the total power generation cost, followed by obtaining and adding constraints. These constraints include power balance constraints, generator active power output constraints, ramping constraints, and transmission line thermal stability limit constraints.

[0055] The classical safety constraints are expressed by the scheduling model as follows:

[0056] ;

[0057] The constraints are as follows:

[0058] Constraint 1: ;

[0059] Constraint 2: ;

[0060] Constraint 3: ;

[0061] Constraint 4: ;

[0062] The objective of the classic safety-constrained scheduling model is to minimize the total generation cost; that is, the specific expression of the objective function for minimizing the total generation cost is:

[0063] ;

[0064] The above formula indicates that the generator is in Active power at any given time; ensuring the power grid operates within constraints. Power balance at any given moment Indicates the node at time [time]. The load; the generator's active power output is limited by constraint two. and These represent the minimum and maximum active power output of the generator, respectively; the generator in The ramp constraint (uphill / downhill constraint) at any time is shown in constraint 3; the thermal stability limit of the branch power flow should meet the requirements of constraint 4. It is the injection transfer distribution factor. This indicates the connection relationship between the generator and the node; and These represent the minimum and maximum thermal stability limits for transmission lines, respectively.

[0065] Subsequently, a neural network model suitable for power dispatching problems is constructed. During the recursive computation of the neural network model, it is ensured that the parameters of the neural network remain consistent at each time point during forward propagation. This means that at each time step, the parameters such as the weights and biases of the neural network are the same, thus guaranteeing the uniformity and stability of the model over time.

[0066] The output of the calculation unit at the current time step is used as the input of the corresponding calculation unit at the next time step. Specifically, for each calculation unit at each time step, its input includes the load demand at the current time step (i.e., the total power consumption that the power system needs to meet at this moment) and the active power output of the generators at the previous time step (i.e., the actual power output of each generator at the previous moment). In this way, the neural network model can pass on the information of the time series, forming a continuous calculation process.

[0067] The output of the computation unit at each time step is the active power of the generator at the current time step. This output is the power value that each generator should output at the current moment, calculated by the neural network model based on the input load demand and the generator output at the previous time step, in order to meet the dispatching needs of the power system.

[0068] The above process can be viewed as an application of the recurrent connection mechanism to extend the existing end-to-end optimal power flow (E2E OPF) model in solving a multi-timescale security-constrained economic scheduling model with temporal coupling constraint 3. This application's embodiment adopts a general recurrent connection and information superhighway (RCIS) framework. This framework includes two core mechanisms that can adapt to existing E2E OPF models and alleviate the gradient vanishing / exploding problem: the recurrent connection mechanism and the information superhighway mechanism. The proposed overall RCIS framework is as follows: Figure 2 As shown, Figure 2 This is a schematic diagram of the framework of a general-purpose loop connection and information superhighway provided in the embodiments of this application.

[0069] like Figure 2 As shown, the recurrent connection mechanism has two characteristics: recurrence and connectivity. Recurrence refers to the neural network model in the RCIS framework maintaining consistency across time steps (i.e., full-process reuse). For example... Figure 2 As shown, during the forward propagation process, the parameters of the neural network at each time point... They are all the same; this cyclical nature is equivalent to using the same model repeatedly. (Total number of time steps), therefore the model size does not increase with the increase of time step length. Connectivity means that the output of each time step will be used as the input of the next time step, thereby connecting two consecutive time models together to build a power dispatching model that can learn the temporal coupling relationship.

[0070] To solve the model described above, this invention uses an existing end-to-end model. This involves a recursive computation, where the output of each time step serves as the input for the next time step, thus constructing a loop connection mechanism. At time step... Input variables The vector consists of two parts: the load demand at the current time step. The active power output of the generator at the previous time point The output variable is the generator's active power at the current time point. With this architecture, the model can dynamically model the temporal characteristics of the dataset during forward propagation and learn time-related information patterns through backpropagation.

[0071] In step S103, the constraints of the classic safety-constrained economic dispatch model are embedded into the output of the set neural network model. This can be achieved by adding an additional constraint processing mechanism to the output layer of the neural network. For example, a penalty function can be used to adjust the loss function when the generator output or other results output by the neural network violate the constraints in the safety-constrained economic dispatch model, thereby guiding the neural network model to output results that meet the constraints.

[0072] Simultaneously, a high-speed gradient backpropagation information channel is constructed that bypasses the neural network pathway. This is to quickly transmit constraint-related gradient information back to the parameter update stage of the neural network during backpropagation. This avoids problems such as gradient vanishing or gradient exploding caused by improper constraint handling in traditional neural network training, improving the efficiency and stability of model training. Specifically, this can be achieved by designing special network structures or optimization algorithms that allow constraint gradient information to directly affect the update of neural network parameters without going through complex internal neural network pathways.

[0073] In step S104, power training data is first collected. This data can include historical power load demand, generator output, and corresponding dispatch schemes. The collected power training data is then used to train the established neural network model. During training, the neural network parameters are adjusted using the backpropagation algorithm. Specifically, the error between the dispatch result output by the neural network model and the actual dispatch scheme is first calculated (this error can be measured by defining an appropriate loss function). Then, the backpropagation algorithm is used to propagate the error back to each parameter of the neural network. The parameters are updated using optimization methods such as gradient descent, continuously optimizing the performance of the neural network model to obtain a power dispatch model that better reflects the power dispatch patterns.

[0074] After the model training is completed, a power resource scheduling scheme is output based on the obtained power dispatching model and real-time power resource data (such as current load demand, generator status, etc.). Real-time data is input into the trained neural network model, which, based on the learned scheduling rules and constraints, quickly outputs the optimal generator active power allocation scheme under the current conditions, thereby achieving effective power resource scheduling and ensuring the safe and economical operation of the power system.

[0075] This application's embodiments recursively calculate the classic security-constrained economic dispatch model using a neural network model, maintaining consistent neural network parameters at each time point during forward propagation. Simultaneously, the output of the current time step is used as the input for the next time step, reducing redundant calculations. The parameter sharing and recursive structure significantly improve computational efficiency, enabling rapid generation of power resource dispatch schemes to adapt to the complex and ever-changing operational needs of new power systems. By constructing a high-speed gradient backhaul information path that bypasses the neural network, the vanishing or exploding gradient problems that may occur during traditional neural network training are effectively avoided, further improving the model's training efficiency and stability, and ensuring good convergence performance even on long-term time series problems. Furthermore, this application embeds the constraints of the classic security-constrained economic dispatch model into the output of the neural network model, ensuring that the generated dispatch scheme strictly meets the power system's security constraints. This allows the model to fully consider the physical characteristics and operational requirements of the power system when optimizing the dispatch scheme, avoiding system operational risks caused by violation of constraints and improving the feasibility and reliability of the power resource dispatch scheme.

[0076] Based on the above embodiments, control factors can also be generated through neural networks, which are used to indicate that the output of the power dispatch model satisfies the constraints.

[0077] like Figure 2 As shown, the information superhighway mechanism constructs a path that does not pass through any neural network by embedding constraints one to three into the output of the neural network. The advantage is that the gradient backpropagation process does not pass through the internal structure of the neural network, and is therefore not affected by the parameters of the neural network, thus constructing an information superhighway for gradient backpropagation.

[0078] The RCIS framework embeds constraints within the computational units, as shown in the detailed diagram below. Figure 3 As shown, Figure 3 This is a schematic diagram of the internal structure of a computational unit in a neural network model provided in this application embodiment. It is assumed that the computational unit is a... There are 1 hidden layer, each layer contains 1 hidden layer. A neural network of neurons. Control factors are generated using the sigmoid function. At time step Given an input vector, the neural network generates the generator active power output at the current time step in the following way. :

[0079] Constraint 5: That is, to calculate the maximum possible output of the generator at the current time step. The maximum possible output force Maximum output of the generator The smaller of the climbing capacity of the first unit and the climbing capacity of the second unit; the climbing capacity of the first unit is the generator output of the previous time step. With the allowable uphill gradient sum.

[0080] Constraint Six: That is, to calculate the minimum possible output of the generator at the current time step. The minimum possible output power is the minimum output power of the generator. The larger of the climbing capacity of the second unit and the climbing capacity of the second unit; the climbing capacity of the second unit is the generator output of the previous time step. With the allowable downhill gradient difference.

[0081] Constraint 7: That is, to calculate the actual active power output of the generator at the current time step. The actual active power The output falls between the minimum possible output and the maximum possible output, and is adjusted by a control factor generated by the loss function.

[0082] Constraint 8: .

[0083] That is, calculate the relaxation power at the current time step. The relaxation power The sum of all load demands Subtract the sum of the output of all locked generators .

[0084] In the above formula, and The maximum and minimum output values ​​of the unit are redefined when both the maximum / minimum output of the unit and the up / down ramping constraints of the unit are met.

[0085] In this way, constraints five through eight are embedded in the defined neural network model, ensuring that the output of the model always satisfies these constraints. The information superhighway mechanism provided in this application brings two major advantages to the RCIS framework:

[0086] On the one hand, this mechanism integrates physical information into the neural network architecture, allowing the RNN to directly learn physical information through backpropagation, thereby improving the feasibility of decision-making schemes.

[0087] On the other hand, an information path that does not go through a specific neural network model is constructed. During the gradient backpropagation process, this information path does not go through a specific neural network model, ensuring that the gradient of this path is not affected by the neural network parameters. Therefore, this mechanism effectively alleviates the gradient vanishing / exploding problem.

[0088] The following mathematical derivation proves that the neural network model, based on the information highway mechanism, has the ability to mitigate gradient vanishing / exploding. Assume the loss function of the neural network is expressed as... The correlation characteristics between adjacent time steps can be determined based on the maximum possible output, minimum possible output, and control factors. Furthermore, based on the information characteristics of each time step and the correlation characteristics between adjacent time steps, the gradient information of the loss function corresponding to the set neural network model can be generated. The gradient calculation formula for this loss function is as follows:

[0089] ;

[0090] in, These are the parameters of a deep neural network.

[0091] The gradient calculation formula of the loss function shows that it consists of two components:

[0092] The first part is It reflects the information characteristics of each time step;

[0093] The second part is It includes the correlation between adjacent time steps.

[0094] Through this design, the RCIS framework proposed in this application can simultaneously learn cross-temporal relationships (climbing / slipping constraints) and temporal feature information in the training dataset through backpropagation in both the recurrent original network and the dual network.

[0095] For RNN models, The value of satisfies the constraints corresponding to the following gradient formula:

[0096] ,(1);

[0097] when With the total time Constantly increasing, When the value of approaches 0, the RNN model will experience the gradient vanishing problem; when As time goes by Adding more and more RNN models can lead to gradient explosion.

[0098] Regarding the proposed RCIS framework Satisfy the following cyclic gradient formula:

[0099] (2);

[0100] Where the formula , and The values ​​are represented as (3), (4), and (5) respectively:

[0101] (3);

[0102] Formula (3) defines and right The partial derivatives, The value is the value from the previous time step, and r is the scaling factor. This is the offset.

[0103] (4);

[0104] (5);

[0105] (6);

[0106] As can be seen from the above formula, first calculate the first partial derivative between the maximum possible output and the generator output of the previous time step, then calculate the second partial derivative between the minimum possible output and the generator output of the previous time step, and finally calculate the control factor based on the difference between the first and second partial derivatives. Control factor express and right The difference between the partial derivatives.

[0107] The cyclic gradient formula of RCIS differs fundamentally from that of traditional RNNs. In the gradient formula of the traditional neural network model shown in formula (1), the gradient at all time steps is either always greater than 1 or restricted to the interval [0,1], which is the root cause of the gradient explosion / vanishing problem. However, the RCIS framework applied in this application, i.e., the gradient formula corresponding to formula (2), can be greater than 1 at some time steps and break through the [0,1] range at other time steps. By controlling... and The partial derivatives of the gradient ensure that it does not tend to 0, thus avoiding the vanishing gradient problem. They also ensure that the gradient does not increase indefinitely, thus avoiding the exploding gradient problem.

[0108] As the time step approaches infinity, the recursive gradient does not necessarily converge to 0 or infinity mathematically. If the gradient begins to converge to zero, it can be made to tend towards 1 by setting the parameters to 1, thus avoiding rapid vanishing. All parameter values ​​are learned autonomously by the network through backpropagation. This mechanism enables the network to adaptively decide when to suppress and when to preserve the gradient. This application effectively alleviates the gradient explosion / vanishing problem in recurrent neural networks by dynamically adjusting the gradient range. The learning process of network parameters is entirely data-driven, requiring no manual setting of fixed thresholds.

[0109] See Figure 4 , Figure 4 This application provides a schematic diagram of a power resource scheduling system with embedded time-coupling constraints, comprising:

[0110] The scheduling model acquisition module is used to acquire the classic security-constrained economic scheduling model corresponding to power resources;

[0111] The recursive calculation module is used to recursively calculate the classic safety-constrained economic dispatch model using a defined neural network model. During the forward propagation of the recursive calculation process of the defined neural network model, the parameters of the neural network remain consistent at each time point. Simultaneously, the output of the calculation unit at the current time step is used as the input of the corresponding calculation unit at the next time step, thus connecting the calculation units of the defined neural network model. For each calculation unit at each time step, its input is the load demand at the current time step and the generator active power output at the previous time step, and its output is the generator's active power at the current time step.

[0112] The constraint setting module is used to embed the constraints of the classic safety constraint economic scheduling model into the output of the set neural network model, and to construct a gradient backhaul information highway that does not go through the neural network path.

[0113] The model training module is used to train the set neural network model using power training data, and adjust the neural network parameters through the backpropagation algorithm to obtain the power dispatch model;

[0114] The power resource scheduling module is used to output a power resource scheduling scheme based on the power scheduling model and real-time power resource data.

[0115] Based on the above embodiments, as a preferred embodiment, the scheduling model acquisition module includes:

[0116] Objective function building unit, used to construct an objective function that minimizes the total cost of electricity generation;

[0117] The constraint acquisition unit is used to acquire the constraint conditions; the constraint conditions include power balance constraints, generator active power output constraints, ramping constraints, and thermal stability limit constraints of transmission lines.

[0118] Based on the above embodiments, as a preferred embodiment, it further includes:

[0119] A control factor generation module is used to generate control factors through a neural network, wherein the control factors are used to indicate that the output of the power dispatch model satisfies the constraints.

[0120] Based on the above embodiments, as a preferred embodiment, the control factor generation module is a module for performing the following steps:

[0121] Calculate the maximum possible output of the generator at the current time step; the maximum possible output is the smaller value between the generator's maximum output and the first unit's ramp-up capability; the first unit's ramp-up capability is the sum of the generator's output at the previous time step and the allowable ramp-up amount;

[0122] Calculate the minimum possible output of the generator at the current time step; the minimum possible output is the larger of the generator's minimum output and the unit's ramp-up capability; the second unit's ramp-up capability is the difference between the generator's output at the previous time step and the allowable ramp-up amount;

[0123] Calculate the actual active power output of the generator at the current time step; the actual active power output falls between the minimum possible output and the maximum possible output, and is adjusted by a control factor generated by a loss function;

[0124] Calculate the relaxation power at the current time step; the relaxation power is the sum of all load demands minus the sum of the output of all locked generators.

[0125] Based on the above embodiments, as a preferred embodiment, it further includes:

[0126] The control factor calculation module is used to perform the following steps:

[0127] Calculate the first partial derivative between the maximum possible output and the generator output of the previous time step;

[0128] Calculate the second partial derivative between the minimum possible output and the generator output of the previous time step;

[0129] The control factor is calculated based on the difference between the first partial derivative and the second partial derivative.

[0130] This application also provides an embodiment corresponding to a computer-readable storage medium. The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method described in the above method embodiments.

[0131] It is understood that if the methods in the above embodiments are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and executes all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0132] The computer-readable storage medium provided in this embodiment includes the method mentioned above, and has the same effect.

[0133] This application also provides an electronic device, see [link to document]. Figure 5 The present application provides a structural diagram of an electronic device, such as... Figure 5 As shown, it may include a processor 1410 and a memory 1420.

[0134] The processor 1410 may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor 1410 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 1410 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor 1410 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, the processor 1410 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0135] The memory 1420 may include one or more computer-readable storage media, which may be non-transitory. The memory 1420 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In this embodiment, the memory 1420 is used to store at least the following computer program 1421, which, after being loaded and executed by the processor 1410, is capable of implementing the relevant steps in the methods executed by the electronic device side as disclosed in any of the foregoing embodiments. In addition, the resources stored in the memory 1420 may also include an operating system 1422 and data 1423, etc., and the storage method may be temporary storage or permanent storage. The operating system 1422 may include Windows, Linux, Android, etc.

[0136] In some embodiments, the electronic device may further include a display screen 1430, an input / output interface 1440, a communication interface 1450, a sensor 1460, a power supply 1470, and a communication bus 1480.

[0137] certainly, Figure 5 The structure of the electronic device shown does not constitute a limitation on the electronic device in the embodiments of this application. In practical applications, the electronic device may include more than [other components]. Figure 5 More or fewer components as shown, or combinations of certain components.

[0138] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. As the system provided in the embodiments corresponds to the method provided in the embodiments, the description is relatively simple; relevant parts can be found in the method section.

[0139] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. It should be noted that those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of this application.

[0140] It should also be noted that, in this specification, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. A power resource scheduling method embedding time-series coupling constraints, characterized in that, include: Obtain the classic security-constrained economic dispatch model corresponding to power resources; The classical safety-constrained economic dispatch model is recursively calculated using a defined neural network model. During the forward propagation of the recursive calculation process of the defined neural network model, the parameters of the neural network remain consistent at each time point. Simultaneously, the output of the calculation unit at the current time step is used as the input to the corresponding calculation unit at the next time step, thus connecting the calculation units of the defined neural network model. For each calculation unit at each time step, its input is the load demand at the current time step and the generator active power output at the previous time step, and its output is the generator's active power at the current time step. The constraints of the classical safety-constrained economic scheduling model are embedded into the output of the set neural network model, and a gradient backhaul information highway is constructed without going through the neural network path. The set neural network model is trained using power training data, and the neural network parameters are adjusted using the backpropagation algorithm to obtain the power dispatch model; Based on the power dispatching model and real-time power resource data, a power resource dispatching scheme is output. The method further includes, after embedding the constraints of the classical security-constrained economic scheduling model into the computational unit of the defined neural network model: Control factors are generated through a neural network, and these control factors are used to indicate that the output of the power dispatch model satisfies the constraints. Among them, the control factors generated through neural networks include: Calculate the maximum possible output of the generator at the current time step; the maximum possible output is the smaller value between the generator's maximum output and the first unit's ramp-up capability; the first unit's ramp-up capability is the sum of the generator's output at the previous time step and the allowable ramp-up amount; Calculate the minimum possible output of the generator at the current time step; the minimum possible output is the larger of the generator's minimum output and the unit's ramp-up capability; the second unit's ramp-up capability is the difference between the generator's output at the previous time step and the allowable ramp-up amount; Calculate the actual active power output of the generator at the current time step; the actual active power output falls between the minimum possible output and the maximum possible output, and is adjusted by a control factor generated by a loss function; Calculate the relaxation power at the current time step; the relaxation power is the sum of all load demands minus the sum of the output of all locked generators; Adjusting neural network parameters using the backpropagation algorithm includes: The correlation characteristics between adjacent time steps are determined based on the maximum possible output, the minimum possible output, and the control factor. The gradient information of the loss function corresponding to the set neural network model is generated based on the information features of the time step and the correlation features between adjacent time steps; Calculate the first partial derivative between the maximum possible output and the generator output of the previous time step; Calculate the second partial derivative between the minimum possible output and the generator output of the previous time step; The control factor is calculated based on the difference between the first partial derivative and the second partial derivative.

2. The power resource dispatching method according to claim 1, characterized in that, The security-constrained scheduling model for obtaining power resources includes: Construct an objective function that minimizes the total cost of electricity generation; Obtain the constraints; the constraints include power balance constraints, generator active power output constraints, ramping constraints, and transmission line thermal stability limit constraints.

3. A power resource scheduling system with embedded time-series coupling constraints, characterized in that, include: The scheduling model acquisition module is used to acquire the classic security-constrained economic scheduling model corresponding to power resources; The recursive calculation module is used to recursively calculate the classic safety-constrained economic dispatch model using a defined neural network model. During the forward propagation of the recursive calculation process of the defined neural network model, the parameters of the neural network remain consistent at each time point. Simultaneously, the output of the calculation unit at the current time step is used as the input of the corresponding calculation unit at the next time step, thus connecting the calculation units of the defined neural network model. For each calculation unit at each time step, its input is the load demand at the current time step and the generator active power output at the previous time step, and its output is the generator's active power at the current time step. The constraint setting module is used to embed the constraints of the classic safety constraint economic scheduling model into the output of the set neural network model, and to construct a gradient backhaul information highway that does not pass through the neural network path. The model training module is used to train the set neural network model using power training data, and adjust the neural network parameters through the backpropagation algorithm to obtain the power dispatch model; The power resource scheduling module is used to output a power resource scheduling scheme based on the power scheduling model and real-time power resource data; A control factor generation module is used to generate control factors through a neural network, wherein the control factors are used to indicate that the output of the power dispatch model satisfies the constraints. The control factor generation module is used to perform the following steps: Calculate the maximum possible output of the generator at the current time step; the maximum possible output is the smaller value between the generator's maximum output and the first unit's ramp-up capability; the first unit's ramp-up capability is the sum of the generator's output at the previous time step and the allowable ramp-up amount; Calculate the minimum possible output of the generator at the current time step; the minimum possible output is the larger of the generator's minimum output and the unit's ramp-up capability; the second unit's ramp-up capability is the difference between the generator's output at the previous time step and the allowable ramp-up amount; Calculate the actual active power output of the generator at the current time step; the actual active power output falls between the minimum possible output and the maximum possible output, and is adjusted by a control factor generated by a loss function; Calculate the relaxation power at the current time step; the relaxation power is the sum of all load demands minus the sum of the output of all locked generators; The control factor calculation module is used to perform the following steps: calculate the first partial derivative between the maximum possible output and the generator output of the previous time step; calculate the second partial derivative between the minimum possible output and the generator output of the previous time step; and calculate the control factor based on the difference between the first partial derivative and the second partial derivative.

4. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for implementing the steps of the method as described in claim 1 or 2 when executing the computer program.

5. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed, implements the steps of the method as described in claim 1 or 2.

6. A computer program product, characterized in that, It includes a computer program that, when executed, implements the steps of the method as described in claim 1 or 2.

Images