Power distribution network day-ahead low-carbon scheduling method based on prediction optimization fusion learning

Abstract

A power distribution network day-ahead low-carbon scheduling method based on predictive optimization fusion learning comprises the steps of collecting and cleaning multi-source data through a data collecting and preprocessing step to form a training data set; predicting node carbon potential by using a sequence neural network model, wherein the model adopts a Seq2Seq structure and a GRU unit to process time sequence data; constructing an adjustable load low-carbon scheduling model to minimize carbon emission and meet power grid operation constraints; predicting, optimizing and fusing learning and model training, and optimizing by constructing a mixed decision loss function and utilizing a back propagation algorithm to realize collaborative optimization of prediction and scheduling decisions; and generating an adjustable load optimization scheduling strategy, and ensuring that actual requirements are met through parameter adjustment and optimization, thereby realizing low-carbon and efficient scheduling of the power distribution network. According to the method, based on prediction optimization fusion learning, low carbon emission and efficient operation of a power system are realized.

Description

Technical Field

[0001] The present invention relates to a distribution network dispatching technology, and in particular to a distribution network day-ahead low-carbon dispatching method based on prediction optimization fusion learning. Background Art

[0002] Traditional low-carbon dispatching methods mainly adopt the framework of "prediction first, optimization later", that is, first predict the load, power generation and carbon emissions based on historical data and statistical models, and then use mathematical optimization methods to formulate dispatching plans to minimize carbon emissions while ensuring power supply security and economy. The significance of this method is to achieve a balance between system operation indicators and carbon emission reduction targets as much as possible by establishing prediction models and optimization algorithms. However, this phased processing method fails to fully consider the impact of prediction errors on the final dispatching decision, which is prone to decision-making bias. When dealing with new energy uncertainties and load fluctuations, existing technologies are prone to prediction errors and inaccurate dispatching due to model limitations. Traditional dispatching methods usually rely on open-loop prediction and optimization strategies, which are difficult to achieve optimal carbon emission control in complex distribution network environments. With the access of new energy and the increase in system complexity, how to comprehensively consider prediction errors and decision optimization in the dispatching process has become an important technical challenge that needs to be solved in current low-carbon dispatching.

[0003] In traditional low-carbon scheduling methods, the prediction stage and the scheduling optimization stage are often run independently. This phased processing mode fails to achieve information sharing and collaborative feedback. When predicting load, power generation and carbon emissions, the prediction model usually only relies on historical data and statistical models, but fails to fully integrate the lower-level decision-making requirements and actual operating conditions into the model construction, resulting in large deviations at local critical moments even if the overall prediction error is low. This kind of separation not only makes it difficult for the prediction results to be directly reflected in the optimization decision, but also in the subsequent optimization process, small prediction errors may be amplified, resulting in a significant deviation between the final scheduling plan and the actual demand, thereby affecting the economy and reliability of the system operation. In addition, with the large-scale access of renewable energy and the increasing complexity of load demand, traditional background technologies seem to be unable to cope with uncertainty management. Most existing prediction methods rely on historical data statistics and fixed model parameters. It is often difficult to capture the dynamic characteristics of new energy generation (such as wind power and solar power) caused by multiple factors such as climate conditions and equipment performance, as well as the random changes in user load. When faced with extreme climate or emergencies, the reliability and accuracy of the prediction results of this method are greatly reduced, resulting in the inability of scheduling decisions to adapt to on-site operating conditions in a timely manner, affecting the overall effect of the system's low-carbon scheduling.

[0004] It should be noted that the information disclosed in the above background technology section is only used for understanding the background of the present application, and therefore may include information that does not constitute prior art known to ordinary technicians in the field. Summary of the invention

[0005] The main purpose of the present invention is to overcome the defects existing in the above-mentioned background technology and provide a day-ahead low-carbon scheduling method for distribution network based on prediction optimization fusion learning.

[0006] To achieve the above object, the present invention adopts the following technical solutions: A method for low-carbon dispatching of distribution network based on prediction optimization fusion learning, comprising the following steps: S1. Data collection and preprocessing: Collect multi-source data of the power system, including historical load data, renewable energy output data, meteorological data and carbon emission data, and preprocess the data to form a high-quality training data set; S2. Sequential neural network node carbon potential prediction: A node carbon potential prediction method based on a sequence-to-sequence (Seq2Seq) model is used to process multi-source data through an encoder-decoder structure, manage complex time dependencies, and use a gated recurrent unit (GRU) as a computing unit to achieve accurate prediction of the future carbon emission intensity of nodes; S3. Construction of adjustable load low-carbon dispatch model: Construct an adjustable load low-carbon dispatch model with the goal of minimizing the carbon emissions of system operation while meeting constraints such as power balance, equipment operation restrictions, and congestion of the power grid; S4. Prediction optimization fusion learning and model training: Combine prediction error and decision error to construct a hybrid decision loss function, and use neural network training to achieve overall optimization of the low-carbon scheduling strategy. Use the back propagation algorithm to perform gradient descent optimization on the hybrid decision loss function to reduce the errors in both prediction and decision-making, and dynamically adjust the weight coefficient according to the actual system requirements to improve the prediction accuracy and decision-making effect; S5. Generation of adjustable load optimization scheduling strategy: Generate the adjustable load optimization scheduling strategy based on the predicted node carbon potential and the adjustable load low-carbon scheduling model, check whether the scheduling decision meets the requirements, and tune the parameters until they are met.

[0007] S6. Result acquisition: Obtain the optimal dispatching strategy for adjustable loads and predict the carbon potential of nodes to achieve low-carbon dispatching of the distribution network on the day before.

[0008] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the day-ahead low-carbon dispatching method for a distribution network based on predictive optimization fusion learning.

[0009] A computer program product includes a computer program, and when the computer program is executed by a processor, the method for day-ahead low-carbon dispatching of a distribution network based on predictive optimization fusion learning is implemented.

[0010] The present invention has the following beneficial effects: The present invention proposes a day-ahead low-carbon dispatching method for distribution networks based on prediction optimization fusion learning. The prediction optimization fusion learning method effectively solves the dispatching decision deviation problem caused by prediction error transmission in the traditional "prediction first, optimization later" framework. The prediction error and decision error are collaboratively optimized by constructing a hybrid decision loss function. The multi-source time series data is modeled by combining the sequence-to-sequence model (Seq2Seq) and the gated recurrent unit (GRU). The complex dynamic characteristics of the node carbon potential are accurately captured, and the carbon emission prediction accuracy is significantly improved. At the same time, the dynamic weight coefficient is used to balance the prediction and decision-making objectives. The adjustable load low-carbon dispatching model is combined to perform multi-dimensional collaborative optimization of power balance, flow constraints and operating boundaries. Carbon emissions are minimized while ensuring the safe and stable operation of the power grid. The real-time feedback mechanism dynamically adapts to the fluctuations of new energy and load uncertainties, and enhances the robustness and responsiveness of the system in a complex distribution network environment. This provides an innovative solution for building a low-carbon smart grid that combines high-precision prediction, adaptive optimization and multi-constraint collaboration.

[0011] The prediction optimization fusion learning method in the present invention effectively solves the defect that the prediction error has a great influence on the decision result in the traditional prediction-first-then-optimization framework, and provides a dynamic adaptive, accurate and efficient decision optimization solution for the low-carbon dispatch of power systems, which has significant technological advancement and broad application prospects.

[0012] Other beneficial effects of the embodiments of the present invention will be further described below. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 This is an overall flow chart of a method for low-carbon dispatching of a distribution network based on predictive optimization fusion learning according to an embodiment of the present invention.

[0014] Figure 2 This is a flow chart of a day-ahead low-carbon scheduling framework based on prediction optimization fusion learning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0015] The following is a detailed description of the embodiments of the present invention. It should be emphasized that the following description is only exemplary and is not intended to limit the scope and application of the present invention.

[0016] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0017] See also Figure 1 and Figure 2 The embodiment of the present invention provides a method for low-carbon dispatching of a distribution network based on prediction optimization fusion learning, comprising the following steps: Step S1. Data collection and preprocessing: Collect multi-source data of the power system, including historical load data, renewable energy output data, meteorological data and carbon emission data, and perform preprocessing such as cleaning, normalization and feature extraction on the data to form a high-quality training data set; Step S2. Sequential neural network node carbon potential prediction: Utilize the node carbon potential prediction method based on the sequence-to-sequence (Seq2Seq) model, process multi-source data through the encoder-decoder structure, manage complex time dependencies, and use the gated recurrent unit (GRU) as the computing unit to achieve accurate prediction of the future carbon emission intensity of the node.

[0018] In a preferred embodiment, in step S2, a prediction method based on a sequence-to-sequence (Seq2Seq) model is adopted, wherein the model includes an encoder and a decoder structure, wherein the encoder is used to extract features from the input multi-source data and generate a context vector of a fixed length, and the decoder generates a node carbon potential prediction sequence in the future period based on the context vector. The model uses a gated recurrent unit (GRU) as a calculation unit, which can effectively capture the long-term and short-term dependencies in the time series data and simulate the nonlinear relationship between different prediction factors. Specifically, the encoder processes the input sequence by updating the gate vector and resetting the gate vector to generate a candidate hidden state, and updates the current hidden state based on the updated gate vector and the candidate hidden state, and finally generates a context vector; the decoder initializes the hidden state through the context vector, and gradually generates an output sequence, and the output sequence is subjected to probability distribution calculation by the softmax function, and finally generates a node carbon potential prediction result.

[0019] More specifically, the processing of the encoder may include the following steps: dynamically control the retention and forgetting of historical information in the input sequence through the update gate and the reset gate, generate the update gate vector and the reset gate vector; calculate the candidate hidden state based on the current input, the reset gate vector and the previous hidden state; use the update gate vector to fuse the candidate hidden state with the previous hidden state, and update the current hidden state time by time; use the hidden state of the final time step as the context vector to pass the global features of the input sequence to the decoder. The processing of the decoder may include the following steps: initialize the decoder hidden state based on the context vector generated by the encoder; update the current hidden state by combining the previous hidden state, the previous prediction output and the context vector time by time; map the current hidden state to the output probability distribution, and generate the prediction result of the current time step by maximizing the probability; iteratively generate the complete output sequence to achieve the time series prediction of the node carbon potential.

[0020] Step S3. Construction of adjustable load low-carbon dispatch model: Construct an adjustable load low-carbon dispatch model with the goal of minimizing the carbon emissions of system operation while meeting constraints such as power balance of the power grid, equipment operation restrictions, and congestion.

[0021] In some embodiments, in step S3, the objective function is to minimize the carbon emissions of the distribution network, and dynamic optimization scheduling is achieved through the following constraints: a) adjustable load constraints, ensuring that the adjusted load power meets power invariance, adjustment unidirectionality, adjustment upper and lower bounds, and time series continuity; b) flow constraints, using a linearized model to characterize the relationship between active and reactive power injection of the power grid bus; c) power balance constraints, ensuring the balance of supply and demand of power generation, power purchase, new energy grid connection and load power in the distribution network; d) operating boundary constraints, limiting branch flow, voltage amplitude, phase angle and generator output range to meet the requirements of safe and stable operation of the power grid. In a further embodiment, the adjustable load constraint optimizes the load distribution by adjusting the translation amount of the load power, and limits the cumulative deviation of power adjustment, the mutual exclusivity of adjustment directions and the power change rate in adjacent time periods; the flow constraint constructs a linearized active and reactive power injection relationship based on the line impedance parameters and the bus voltage amplitude and phase angle; the operating boundary constraint includes upper and lower limit constraints on branch flow, voltage amplitude, phase angle, thermal power unit output and new energy grid-connected power to ensure that the system operates within a safe threshold.

[0022] Step S4. Prediction optimization fusion learning and model training: Combine the prediction error and decision error to construct a hybrid decision loss function, and use the neural network training method to achieve the overall optimization of the low-carbon scheduling strategy. Use the back propagation algorithm to perform gradient descent optimization on the hybrid decision loss function to reduce the errors in both prediction and decision-making, and dynamically adjust the weight coefficient according to the actual system requirements to improve the prediction accuracy and decision-making effect; In a preferred embodiment, in step S4, the prediction error and the decision error are weightedly fused by constructing a hybrid decision loss function, the hybrid decision loss function includes a prediction error term and a decision error term, and the loss function is gradient-descent optimized by a back-propagation algorithm to reduce the errors in both prediction and decision-making, and the weight coefficient is dynamically adjusted according to the actual system requirements to improve the prediction accuracy and decision-making effect. Specifically, the collaborative optimization of prediction and scheduling decisions is achieved through the following mechanisms: a) constructing a hybrid decision loss function, weighting and fusion the prediction error term and the decision error term according to a dynamic weight coefficient, the prediction error term characterizes the deviation between the predicted value and the actual value of the node carbon potential, and the decision error term characterizes the deviation between the scheduling decision based on the predicted value and the optimal decision based on the actual value; b) using the back-propagation algorithm to perform gradient descent optimization on the hybrid decision loss function, synchronously updating the parameters of the prediction model and the scheduling decision model, and gradually reducing the prediction error and decision error; c) dynamically adjusting the weight coefficient according to system requirements to balance the optimization goals of prediction accuracy and decision effect. More preferably, the decision error term is gradient solvable by constructing a substitute function, wherein the substitute function is constructed based on the difference between the optimal decision and the actual decision, and the nonlinear decision error is converted into a differentiable form through a maximization operation to support back-propagation training of the neural network.

[0023] Step S5. Generate an optimal scheduling strategy for adjustable loads: Generate an optimal scheduling strategy for adjustable loads based on the predicted node carbon potential and the adjustable load low-carbon scheduling model, check whether the scheduling decision meets the requirements, and perform parameter tuning until it meets the requirements.

[0024] Step S6. Result acquisition: Obtain the adjustable load optimization scheduling strategy and predict the node carbon potential to achieve low-carbon scheduling of the distribution network on the day before.

[0025] The following further describes an algorithm example and experimental verification of a specific embodiment of the present invention.

[0026] The present invention proposes a day-ahead low-carbon dispatching method for distribution networks based on predictive optimization fusion learning. Affected by the dual uncertainty of source and load of new energy access on the power generation side and flexible resources on the load side, the trend of node carbon potential has complex characteristics. Considering the lower-level adjustable load dispatching model, a reasonable arrangement and optimization dispatching strategy is made for the factors affecting the carbon emission trajectory, and a carbon potential prediction model for the lower-level low-carbon dispatching nodes is constructed. The carbon emission trend is captured by a sequential neural network, and the day-ahead low-carbon dispatching based on predictive optimization fusion learning is realized.

[0027] The method specifically comprises the following steps: 1. Establish a node carbon potential prediction model based on sequential neural network to realize the day-ahead node carbon potential output.

[0028] Node carbon potential refers to the carbon emission intensity of each node in the power system, reflecting the carbon emission level associated with a specific node during power transmission and use. Accurately predicting node carbon potential is crucial for formulating low-carbon scheduling strategies, which helps optimize power resource allocation and reduce overall carbon emissions.

[0029] To this end, a node carbon potential prediction method based on the sequence-to-sequence (Seq2Seq) model was established. By collecting multi-source data in the power system and forming input features, the future carbon emission intensity of each node can be accurately predicted, thereby providing a reliable data basis for low-carbon scheduling decisions.

[0030] According to an embodiment of the present invention, a Seq2Seq model with an encoder-decoder structure is used to process multi-source data such as historical load data, renewable energy output data, meteorological data, and carbon emission data, and is able to manage complex time dependencies, effectively integrate heterogeneous input variables, and simulate nonlinear relationships between different prediction factors. In order to show the relationship between variables, the general form of the prediction model is:

[0031] Among them, the future node carbon potential is determined by the prediction model generate; is the selected input feature sequence, with a length of ; is the generated output sequence, with length .

[0032] Specifically, the prediction model consists of an encoder and a decoder. The encoder extracts features from the input sequence and converts it into a fixed-length context vector; the decoder generates a predicted output sequence based on the context vector to describe the carbon potential of the node in the future period. The model structure can fully capture the long-term and short-term dependencies in time series data, overcome the limitations of traditional methods when dealing with inconsistent input and output sequence lengths, and thus achieve accurate prediction of the carbon potential of the node. The gated recurrent unit (GRU) is used as the computing unit in the implementation.

[0033] In the encoder, the input sequence Converted to hidden state The updating process of the computing unit is as follows:

[0034] in, is the update gate vector, is the reset gate vector; is a candidate hidden state. , , is the input weight matrix; , , is the weight matrix of the hidden state; , , is the bias vector. represents Hadamard multiplication; It is a sigmoid activation function; is the hyperbolic tangent activation function. The final hidden state of the encoder is Often used as context vector .

[0035] In the decoder, the decoder decodes the context vector into a length of The output sequence The decoder uses the context vector to initialize the hidden state and then gradually generates the output sequence. The calculation formula is as follows:

[0036] Among them, the decoder GRU uses the previous hidden state , previous output and context vector At each time step Update its hidden state. is the weight matrix of the output layer; is the bias vector of the output layer. Final node carbon potential prediction Depend on express.

[0037] In the present invention, data collection includes comprehensive collection of historical loads, renewable energy generation, meteorological conditions and carbon emission records in the power system, followed by cleaning, normalization and feature extraction of the data to construct a high-quality training data set.

[0038] 2. Establish an adjustable load low-carbon scheduling model and optimize the load scheduling strategy based on predicted carbon potential.

[0039] A low-carbon dispatch model for adjustable loads achieves the minimization of carbon emissions by dynamically optimizing the dispatch of adjustable load resources in the power system while meeting the requirements of safe, stable and economical operation of the power grid. The model makes full use of the control capability of adjustable loads, optimizes the overall load distribution of the system by rationally allocating power demand, thereby achieving the goal of reducing fossil energy consumption and carbon emissions, with significant environmental and economic benefits.

[0040] In one embodiment of the present invention, adjustable loads are defined as power consumption equipment and energy storage devices that can actively adjust power consumption according to dispatch instructions within a certain range. First, the basic parameters of various adjustable loads in the power system (including regulation range requirements, etc.) are comprehensively collected, and a low-carbon dispatch optimization model is constructed. Its objective function is to minimize the carbon emissions of system operation while meeting the constraints of power balance, equipment operation restrictions, and congestion of the power grid.

[0041] To achieve the above objectives, the present invention constructs an adjustable load optimization scheduling model with minimizing carbon emissions as the objective function, and its formula is as follows:

[0042] in, represents the carbon emissions of the distribution network, is the total number of distribution network nodes, To predict the nodal carbon potential, is the adjustable load power after adjustment.

[0043] The adjustable load optimization scheduling model is composed of the following constraints, including adjustable load constraints, power flow constraints, power balance constraints, and operation boundary constraints, as follows: (a) Adjustable load constraints The adjustable load represents the power relationship of the adjustable load before and after adjustment, ensuring the invariance of the adjustable load power, the unidirectionality of the adjustment, the upper and lower limits of the adjustment, and the continuity of the time sequence, as follows:

[0044]

[0045] in, is the adjustable load power before adjustment, Allowable power deviation, It is a set of adjustments A single load adjustment action in It is the adjustment parameter that constrains the upper and lower limits of load adjustment. On the one hand, it limits the power deviation and alleviates the mismatch of user demand. On the other hand, it also limits the user's tolerance. It is the maximum power change to avoid step or sudden changes.

[0046] (b) Power flow constraints The power flow constraint is expressed by a linearized power flow model, as follows:

[0047] in, , For bus Linear expressions for active and reactive power injection. , is the voltage amplitude and voltage phase angle; is the line impedance coefficient.

[0048] (c) Power balance constraints

[0049] in, Purchase electricity from the main grid for distribution network operators. The amount of electricity connected to the grid by new energy sources, The power generation of thermal power units. It is the collection of inflow and outflow nodes of the busbar.

[0050] (d) Operational boundary constraints

[0051] Among them, upper and lower limit constraints on branch currents, voltage phase angles, generator sets and new energy units are agreed upon.

[0052] 3. Through the prediction optimization fusion learning method, the carbon potential prediction model for the lower-level low-carbon scheduling nodes is trained to realize the day-ahead low-carbon scheduling based on prediction optimization fusion learning.

[0053] Based on the prediction optimization fusion learning method, by combining the prediction error and decision error at the same time, a hybrid decision loss function is constructed, and the overall optimization of the low-carbon dispatch strategy is achieved by neural network training. This method aims at the prediction error transmission problem in the traditional "prediction first, then optimization" process. By introducing decision error feedback in the model training stage, the coordinated optimization of both prediction and dispatch decision is achieved, thereby improving the accuracy and adaptability of the power grid dispatch plan.

[0054] In one embodiment of the present invention, the historical data, real-time data and external environmental variables in the power system are first collected, cleaned and normalized to form a high-quality training data set. Then, the node carbon potential is preliminarily predicted using a node carbon potential prediction method based on a sequence-to-sequence (Seq2Seq) model, and the prediction error is calculated. Subsequently, the prediction result is input into an adjustable load optimization scheduling model constructed with minimizing carbon emissions as the objective function, and the scheduling decision error is calculated by comparing the preliminary scheduling scheme generated based on the optimization algorithm with the preset target scheduling result. Based on the above two errors, the present invention constructs a hybrid decision loss function, which is composed of a weighted sum of a prediction error term and a decision error term according to a certain weight, wherein the weight coefficient can be dynamically adjusted according to the actual system requirements, thereby taking into account both prediction accuracy and decision effect during the training process.

[0055] First, define the decision error as described above, and obtain the decision error by the difference between the optimal decision based on the observed value and the actual decision based on the predicted value, as shown below:

[0056] During the neural network training process, the hybrid decision loss function is optimized by gradient descent through the back propagation algorithm, so that the model parameters are continuously updated, and the errors in both prediction and decision-making are gradually reduced. After training, the neural network can not only output high-precision prediction results, but also provide real-time feedback and correct prediction errors in scheduling decisions, ensuring the robustness and economy of the low-carbon scheduling strategy in actual operation.

[0057] The decision error is unsolvable in gradient descent and is replaced by a class decision error. The calculation method is as follows:

[0058] Secondly, for the traditional mean square error loss function, its expression is as follows:

[0059] The hybrid decision loss function is a weighted sum of the above-mentioned class decision error loss function and the mean square error loss function, and its expression is as follows:

[0060] in, is the decision mixing coefficient.

[0061] In summary, the prediction optimization fusion learning method of the present invention effectively solves the defect that the prediction error has a great influence on the decision results in the traditional prediction-first-then-optimization framework, and provides a dynamic, adaptive, accurate and efficient decision optimization solution for the low-carbon dispatch of power systems, which has significant technological advancement and broad application prospects.

[0062] Based on the above process, an overall low-carbon scheduling framework based on prediction optimization fusion learning is proposed, such as Figure 2 shown.

[0063] The present invention proposes a method for low-carbon dispatching of distribution network based on prediction optimization fusion learning. The present invention realizes the coordinated optimization of high-precision prediction and dispatching decision of key parameters in power system by adopting the method based on prediction optimization fusion learning. Specifically, the present invention constructs a hybrid decision loss function, integrates the prediction error and decision error according to a certain weight, and uses neural network for training, thereby effectively reducing the dispatching error caused by prediction deviation in the traditional "prediction first, optimization later" method. Compared with the prior art, the node carbon potential prediction method based on sequence to sequence model of the present invention shows obvious advantages in capturing complex time series dynamics, can provide more accurate and timely data support for low-carbon dispatching decision, and improve the accuracy and robustness of low-carbon dispatching of power system. Through the real-time feedback mechanism, the model can dynamically adapt to the changes of system operation status and constraints, ensure that in the complex operation environment with large-scale access of new energy and large load fluctuations, it still maintains a high response speed and robustness, and provides reliable data support and technical guarantee for the safe and economic operation of power grid. Furthermore, due to the wide application range of the present invention, it can be promoted and applied in a variety of power systems and dispatching scenarios, thus providing an innovative and efficient solution for building a green and low-carbon smart grid.

[0064] An embodiment of the present invention further provides a storage medium for storing a computer program, which at least performs the above method when executed.

[0065] An embodiment of the present invention further provides a control device, comprising a processor and a storage medium for storing a computer program; wherein the processor is configured to execute at least the method described above when executing the computer program.

[0066] An embodiment of the present invention further provides a processor, wherein the processor executes a computer program and at least executes the method described above.

[0067] The storage medium can be implemented by any type of non-volatile storage device, or a combination thereof. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic random access memory (FRAM), a ferromagnetic random access memory (Flash Memory), a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD-ROM); the magnetic surface memory can be a disk memory or a tape memory. The storage medium described in the embodiments of the present invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0068] In the several embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0069] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0070] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0071] A person skilled in the art can understand that: all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), disks or optical disks, etc. Various media that can store program codes.

[0072] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

[0073] The methods disclosed in the several method embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new method embodiments.

[0074] The features disclosed in several product embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new product embodiments.

[0075] The features disclosed in several method or device embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new method embodiments or device embodiments.

[0076] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art of the present invention, several equivalent substitutions or obvious variations can be made without departing from the concept of the present invention, and the performance or use is the same, which should be regarded as belonging to the protection scope of the present invention.

Claims

1. A method for low-carbon dispatching of distribution network based on prediction optimization fusion learning, characterized in that: The following steps are involved: S1. Data collection and preprocessing: Collect multi-source data of the power system, including historical load data, renewable energy output data, meteorological data and carbon emission data, and preprocess the data to form a high-quality training data set; S2. Sequential neural network node carbon potential prediction: A node carbon potential prediction method based on a sequence-to-sequence (Seq2Seq) model is used to process multi-source data through an encoder-decoder structure, manage complex time dependencies, and use a gated recurrent unit (GRU) as a computing unit to achieve accurate prediction of the future carbon emission intensity of nodes; S3. Construction of adjustable load low-carbon dispatch model: Construct an adjustable load low-carbon dispatch model with the goal of minimizing the carbon emissions of system operation while meeting the power balance, equipment operation restrictions and congestion constraints of the power grid; S4. Prediction optimization fusion learning and model training: Combine prediction error and decision error to construct a hybrid decision loss function, and use neural network training to achieve overall optimization of the low-carbon scheduling strategy. Use the back propagation algorithm to perform gradient descent optimization on the hybrid decision loss function to reduce the errors in both prediction and decision-making, and dynamically adjust the weight coefficient according to the actual system requirements to improve the prediction accuracy and decision-making effect; S5. Generate the optimal scheduling strategy for adjustable loads: Generate the optimal scheduling strategy for adjustable loads based on the predicted node carbon potential and the adjustable load low-carbon scheduling model, check whether the scheduling decision meets the requirements, and perform parameter tuning until the requirements are met; S6. Result acquisition: Obtain the optimal dispatching strategy for adjustable loads and predict the carbon potential of nodes to achieve low-carbon dispatching of the distribution network on the day before.

2. The method for low-carbon dispatching of distribution network based on prediction optimization fusion learning according to claim 1 is characterized in that: In step S2, a prediction method based on a sequence-to-sequence (Seq2Seq) model is adopted. The model includes an encoder and a decoder structure. The encoder is used to extract features from the input multi-source data and generate a context vector of a fixed length. The decoder generates a node carbon potential prediction sequence in the future time period based on the context vector. The model uses a gated recurrent unit (GRU) as a computing unit, which can effectively capture the long-term and short-term dependencies in time series data and simulate the nonlinear relationship between different prediction factors.

3. The method for low-carbon dispatching of distribution network based on prediction optimization fusion learning according to claim 2 is characterized in that: The encoder processes the input sequence by updating the gate vector and resetting the gate vector to generate candidate hidden states, and updates the current hidden state based on the update gate vector and the candidate hidden state, and finally generates a context vector; the decoder initializes the hidden state through the context vector and gradually generates an output sequence, and the output sequence is calculated by the softmax function for probability distribution, and finally generates a node carbon potential prediction result.

4. The method for low-carbon dispatching of distribution network based on prediction optimization fusion learning according to claim 1 is characterized in that: In step S3, the objective function is to minimize the carbon emissions of the distribution network, and dynamic optimization scheduling is achieved through the following constraints: a) Adjustable load constraints to ensure that the adjusted load power meets power invariance, adjustment unidirectionality, adjustment upper and lower bounds, and time sequence continuity; b) Power flow constraints, using a linearized model to characterize the relationship between active and reactive power injection into the power grid bus; c) Power balance constraints to ensure the balance of power supply and demand of power generation, power purchase, new energy grid connection and load power in the distribution network; d) Operation boundary constraints, which limit branch currents, voltage amplitudes, phase angles and generator output ranges to meet the requirements for safe and stable operation of the power grid.

5. The method for low-carbon dispatching of distribution network based on prediction optimization fusion learning according to claim 4 is characterized by: The adjustable load constraint optimizes the load distribution by adjusting the load power translation amount, and limits the cumulative deviation of power adjustment, the mutual exclusivity of adjustment directions, and the power change rate in adjacent time periods; The power flow constraint constructs a linear active and reactive power injection relationship based on line impedance parameters and bus voltage amplitude and phase angle; The operation boundary constraints include upper and lower limit constraints on branch currents, voltage amplitudes, phase angles, thermal power unit outputs, and renewable energy grid-connected power to ensure that the system operates within a safe threshold.

6. The method for low-carbon dispatching of distribution network based on prediction optimization fusion learning according to claim 1 is characterized in that: In step S4, the prediction error and the decision error are weightedly fused by constructing a hybrid decision loss function, which includes a prediction error term and a decision error term. The loss function is gradient-descent optimized by a back-propagation algorithm to reduce the errors in both prediction and decision-making. The weight coefficient is dynamically adjusted according to actual system requirements to improve prediction accuracy and decision-making effect.

7. The method for low-carbon dispatching of distribution network based on prediction optimization fusion learning according to claim 6 is characterized in that: In step S4, the collaborative optimization of prediction and scheduling decisions is achieved through the following mechanisms: a) constructing a hybrid decision loss function, weighting and fusing the prediction error term and the decision error term according to a dynamic weight coefficient, wherein the prediction error term represents the deviation between the predicted value and the actual value of the node carbon potential, and the decision error term represents the deviation between the scheduling decision based on the predicted value and the optimal decision based on the actual value; b) using a back propagation algorithm to perform gradient descent optimization on the hybrid decision loss function, synchronously updating the parameters of the prediction model and the scheduling decision model, and gradually reducing the prediction error and the decision error; c) Dynamically adjusting the weight coefficient according to system requirements to balance the optimization goals of prediction accuracy and decision-making effect.

8. The method for low-carbon dispatching of distribution network based on prediction optimization fusion learning according to claim 7 is characterized in that: The decision error term is gradient solvable by constructing a substitute function, which is constructed based on the difference between the optimal decision and the actual decision, and converts the nonlinear decision error into a differentiable form through a maximization operation to support back-propagation training of the neural network.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for day-ahead low-carbon dispatching of a distribution network based on predictive optimization fusion learning as described in any one of claims 1 to 8 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for day-ahead low-carbon dispatching of a distribution network based on predictive optimization fusion learning as described in any one of claims 1 to 8 is implemented.

Images