A source-load price joint prediction method, device and system of a new energy power system

Abstract

The application discloses a source-load-price joint prediction method, device and system of a new energy power system, and belongs to the technical field of smart grids. The joint prediction method calculates a gating activation value by using a symbolic causal strength and a source-load-price operation state vector; fuses historical time sequence characteristics, the source-load-price operation state vector and a coupling direction signal to obtain cross-task coupling information; modulates the cross-task coupling information by using the gating activation value; superimposes values of modulation results corresponding to all task pairs i and j to obtain three kinds of input characteristics; and inputs the three kinds of input characteristics into respective corresponding prediction models to obtain respective prediction results of new energy output, load and price. This design considers time-varying coupling relationships among the source, load and price, has the advantages that the transmission direction and strength of the dynamically modulated cross-task information can explicitly represent the coupling relationships and reduce invalid information transmission, and finally realizes simultaneous accurate prediction of the source, load and price.

Description

Technical Field

[0001] This application belongs to the field of smart grid technology, and more specifically, relates to a method, apparatus and system for joint prediction of source and load prices in a new energy power system. Background Technology

[0002] As the proportion of new energy sources such as wind power and photovoltaics in the power system continues to increase, the randomness and weather dependence of new energy output have significantly enhanced the uncertainty in power system dispatch, power market operation, and operational status assessment. New energy output, load, and electricity price are key interrelated variables in the new power system, and their short-term forecasts directly affect system safety operation, economic dispatch, market transactions, and demand response decisions.

[0003] Existing short-term forecasting methods typically treat renewable energy output forecasting, load forecasting, and electricity price forecasting as independent tasks, or employ multi-task learning, shared base networks, feature fusion, and attention mechanisms to construct joint forecasting models. Independent forecasting models struggle to fully utilize the complementary information between source, load, and price tasks; while traditional joint forecasting models can learn relationships between tasks through shared representations, their task-sharing process is usually implicitly driven by gradient backpropagation, lacking explicit constraints on the direction, strength, and effectiveness of cross-task information transmission.

[0004] In new power systems, the source-load-price coupling relationship is influenced by meteorological conditions, market mechanisms, system supply and demand status, and user response behavior, exhibiting significant time-varying and state-dependent characteristics. High renewable energy output suppresses electricity prices through low marginal cost supply, peak load pushes up market-clearing prices, and electricity price changes affect load through demand response. However, the direction and intensity of these effects are not constant across all operating states. When the model shares the coupling relationships across different operating states as fixed relationships, beneficial cross-task information in one state can become misleading information in another, leading to negative migration and reduced prediction reliability. Summary of the Invention

[0005] In view of the above-mentioned defects or improvement needs of the existing technology, this application provides a method, device and system for joint prediction of source and load price of new energy power system, which solves the technical problem that the existing technology is difficult to predict power output, load and electricity price at the same time.

[0006] To achieve the above objectives, according to one aspect of this application, a method for joint prediction of source-load prices in a new energy power system is provided, comprising: Construct a joint forecast sample of source-load price at multiple time points t within a preset period, including new energy output, load, electricity price and meteorological covariates; Feature extraction is performed on the renewable energy output, load, and electricity price in each of the source-load-price joint prediction samples to obtain the historical time-series features of renewable energy output, load, and electricity price. The source-load-price operating state vector is then fused from the three. ; Based on the meteorological covariates, identify the sign causality strength between all task pairs i and j in the source-load-price joint prediction sample. The symbolic causal strength and its corresponding causal edge symbol The coupling direction signal is obtained by fusion. ; Using the symbolic causal strength and the source load price running state vector Calculate the gating activation value ;Will The fusion process yields cross-task coupled information. ; using the gate activation value For the cross-task coupling information Modulation; The superposition values ​​of the modulation results corresponding to all task pairs i and j are superimposed with three historical time series features to obtain three input features; the three input features are then input into their respective prediction models to obtain the prediction results for new energy output, load, and electricity price.

[0007] Furthermore, the use of the symbolic causal strength and the source load price running state vector Calculate the gating activation value ,include: Will The modulus and source load cost operating state vector After being input into the first fully connected neural network, it is normalized using the Sigmoid activation function, and then... Multiplication is used as the gate activation value , .

[0008] Furthermore, the aforementioned The fusion process yields cross-task coupled information. ,include: Will Input a preset neural network to enable it to utilize the formula The fusion process yields cross-task coupled information. .

[0009] Furthermore, the use of the gating activation value For the cross-task coupling information Modulation includes: Utilize the gating activation value of each task pair Information coupling across tasks Modulation is performed, and then the modulation results are superimposed to obtain intermediate variables. .

[0010] Furthermore, the historical time sequence features The extraction and fusion process includes: The renewable energy output, load, and electricity price in each source-load-price joint prediction sample are combined with meteorological covariates and then input into the corresponding task feature extractor to obtain the historical time-series features of renewable energy output, load, and electricity price. The source-load-price operating state vector is then fused from the three. ; Will , and Input to the second fully connected neural network So that it can use the formula Calculate the source load price running state vector corresponding to time point t. .

[0011] Furthermore, the symbolic causal strength The acquisition process includes: Using formula Calculate the source-load price joint prediction sample task i Pointing to task j The symbolic causal strength is then used to obtain the time-varying causal strength matrix for all tasks: ; All tasks include forecasting renewable energy output, load forecasting, and electricity price forecasting. Indicates a point in time t By task i Pointing to task j The strength of causal sign, This indicates the corresponding task pair at a specific time point. t of F Statistic, This represents the p-value corresponding to the significance test. Indicates the causal edge symbol. Describes the minimum-maximum normalization function. Indicates the significance threshold; The coupling direction signal The acquisition process includes: and Input to a third fully connected neural network The task is obtained by merging the two. iPointing to task j The coupling direction signal within the time window t, .

[0012] Furthermore, the step of inputting the three input features into their respective corresponding prediction models to obtain prediction results for new energy output, load, and electricity price includes: Input power characteristics Input the power output prediction model to obtain the power output prediction results; Load input characteristics Input the load forecasting model to obtain the load forecasting results; Inputting electricity price features Input the electricity price prediction model to obtain the electricity price prediction results.

[0013] According to another aspect of this application, a source-load price joint forecasting device for a new energy power system is provided, comprising: The sample construction module is used to construct a joint prediction sample of source-load price at multiple time points t within a preset period, including new energy output, load, electricity price and meteorological covariates. The status acquisition module is used to extract features from the renewable energy output, load, and electricity price in each of the source-load-price joint prediction samples, and obtain the historical time-series features of renewable energy output, load, and electricity price respectively. The source-load-price operating state vector is then fused from the three. ; The causal acquisition module is used to identify the sign causal strength between all task pairs i and j in the source-load-price joint prediction sample based on the meteorological covariates. The symbolic causal strength and its corresponding causal edge symbol The coupling direction signal is obtained by fusion. ; Information modulation module, used to utilize the symbol causality strength and the source load price running state vector Calculate the gating activation value ;Will The fusion process yields cross-task coupled information. ; using the gate activation value For the cross-task coupling information Modulation; The task prediction module is used to superimpose the modulation results corresponding to all task pairs i and j with three historical time series features to obtain three input features; the three input features are then input into their respective prediction models to obtain the prediction results for new energy output, load, and electricity price.

[0014] According to another aspect of this application, a source-load price joint prediction system for a new energy power system is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the source-load price joint prediction method for the new energy power system.

[0015] According to another aspect of this application, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the source-load price joint prediction method for the new energy power system.

[0016] In summary, compared with the prior art, the above-described technical solutions conceived in this application can achieve the following beneficial effects: (1) This invention provides a method for joint prediction of source-load price in a new energy power system, utilizing the symbolic causality strength and the source load price running state vector Calculate the gating activation value ;Will The fusion process yields cross-task coupled information. ; using the gate activation value For the cross-task coupling information Modulation is performed; the superimposed values ​​of the modulation results corresponding to all task pairs i and j are superimposed with three historical time-series features to obtain three input features; the three input features are then input into their respective prediction models to obtain the prediction results for renewable energy output, load, and electricity price. This design, considering the time-varying coupling relationship between source, load, and price caused by supply and demand, market factors, has the advantage of dynamically modulating the transmission direction and intensity of cross-task information, explicitly representing this coupling relationship, reducing invalid information transmission, and avoiding negative information migration; ultimately achieving simultaneous and accurate prediction of output, load, and electricity price.

[0017] (2) This plan utilizes Calculate the gating activation value This design, taking into account the dynamic changes in the coupling strength between tasks under different operating states, has the advantage of combining the symbolic causal strength with the current source-load valence operating state to dynamically determine the information transmission strength of each directed task pair, and utilizes... Block the information transmission path corresponding to the inactive causal edge.

[0018] (3) This scheme utilizes the formula The fusion process yields cross-task coupling information. This design takes into account the differences in the impact of causal source tasks on the target task under different operating states and coupling directions. The advantage is that by fusing the historical time-series characteristics of causal source tasks, the source load valence operating state vector, and the coupling direction signal, cross-task coupling information that matches the current operating state and directed task pair can be generated.

[0019] (4) This scheme utilizes the gating activation value of each task pair. Information coupling across tasks Modulation is performed, and then the modulation results are superimposed to obtain intermediate variables. This design takes into account that the target task may be affected by multiple causal source tasks at the same time, and the influence of different transmission paths is different. The advantage is that by gating and modulating each cross-task coupling information separately and superimposing it, the effective coupling information can be aggregated and the interference of ineffective coupling information on the target task characteristics can be reduced.

[0020] (5) In this scheme, the new energy output, load and electricity price in each source-load-price joint prediction sample are combined with meteorological covariates and then input into the corresponding task feature extractor. With this design, the task features of new energy output, load and electricity price are learned respectively, the independence of each task's own features is preserved, and the correct transmission of coupling effects between different tasks is ensured through the causal guidance mechanism, thereby improving the accuracy of joint prediction.

[0021] (6) This scheme utilizes the formula Calculate the source-load price joint prediction sample task i Pointing to task j The sign of causal strength; this design takes into account the differences in the existence, direction and strength of causal coupling relationships between different tasks. The advantage is that it can uniformly represent the significance test results, the sign of the causal edge and the normalized causal strength, filter out insignificant task relationships, and provide a directed, signed and continuously changing coupling prior for cross-task information transmission. Attached Figure Description

[0022] Figure 1 A flowchart of a method for jointly predicting the source-load price of a new energy power system provided in an embodiment of this application.

[0023] Figure 2 The principle of the source-load price joint prediction method for a new energy power system provided in an embodiment of this application.

[0024] Figure 3 A flowchart of a local condition Granger causality identification module provided in an embodiment of this application.

[0025] Figure 4This is a schematic diagram of the recognition result of a time-varying causal graph provided in an embodiment of this application.

[0026] Figure 5 A comparison chart showing the performance advantages of this application's embodiment compared to other prediction methods. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. Furthermore, the technical features involved in the various embodiments described below can be combined with each other as long as they do not conflict with each other.

[0028] Example 1 This embodiment provides a method for joint prediction of source and load prices in a new energy power system, such as... Figure 1 and Figure 2 As shown, it includes the sample construction stage, state acquisition stage, causal acquisition stage, information modulation stage, and task prediction stage.

[0029] Regarding the sample construction phase: Construct a joint prediction sample of source-load price at multiple time points t within a preset time period, including new energy output, load, electricity price, and meteorological covariates.

[0030] Specifically, historical time-series data on renewable energy output, load, electricity price, and meteorological covariates are acquired, and a joint source-load-price forecast sample is constructed in chronological order. The meteorological covariates include temperature, humidity, wind speed, and wind direction. Each sample consists of a length of [length missing]. T +1 is formed in consecutive time slices, where the previous T The renewable energy output, load, electricity price, and meteorological covariates at each time point are used as model inputs. T +1 time point's renewable energy output, load, and electricity price as forecast targets.

[0031] Regarding the state acquisition phase: Features are extracted from the renewable energy output, load, and electricity price in each of the source-load-price joint prediction samples to obtain their respective historical time-series features. The source-load-price operating state vector is then fused from the three. .

[0032] Specifically, the architecture includes three independent task feature extractors, corresponding to renewable energy output forecasting, load forecasting, and electricity price forecasting, respectively. Each task feature extractor receives the historical sequence and meteorological covariates for the corresponding task and outputs the time-series features of the historical data; each task predictor is used to map the fused historical time-series features into predicted values.

[0033] As an optional implementation, the historical time series features The extraction and fusion process includes: combining the renewable energy output, load, and electricity price in each source-load-price joint prediction sample with meteorological covariates and inputting the results into the corresponding task feature extractor to obtain the historical time-series features of renewable energy output, load, and electricity price. The source-load-price operating state vector is then fused from the three. ;Will , and Input to the second fully connected neural network So that it can utilize Calculate the source load price running state vector corresponding to time point t. .

[0034] Specifically, the independent task feature extractors correspond to new energy output, load, and electricity price, respectively; each task feature extractor receives the historical sequence of the corresponding target variable and meteorological covariates, and outputs the historical data time-series features of the corresponding task; the task feature extractor is implemented by a time-series encoder, which is selected from one of the following: Long Short-Term Memory Network Encoder, Gated Recurrent Unit Encoder, Convolutional Neural Network Encoder, Transformer Encoder, and Informer Encoder. ; in For task characteristics i The extractor, For prediction tasks i Historical data time series characteristics For the task i The input history sequence.

[0035] Regarding the causality acquisition phase: Based on the meteorological covariates, the sign causality strength between all task pairs i and j in the source-load-price joint prediction sample is identified. The symbolic causal strength and its corresponding causal edge symbol The coupling direction signal is obtained by fusion. .

[0036] As an optional implementation, the symbolic causality strength The acquisition process includes: calculating the task in the joint prediction sample of source and load price using the following formula. i Pointing to task j The strength of causal sign, ; This leads to the time-varying causal strength matrix for all tasks: ; All tasks include forecasting renewable energy output, load forecasting, and electricity price forecasting. Indicates a point in time t By task i Pointing to task j The strength of causal sign, This indicates the corresponding task pair at a specific time point. t of F Statistic, This represents the p-value corresponding to the significance test. Indicates the causal edge symbol. Describes the minimum-maximum normalization function. Indicates the significance threshold.

[0037] Specifically, for any ordered task pair (i,j) in the source-load price, a restricted regression model and an unrestricted regression model are constructed. The input information of the restricted regression model includes historical information of the target task, historical information of the third task, and meteorological covariates. The unrestricted regression model adds historical information of the candidate causal source tasks to the restricted regression model. ; ; in, , , These represent the historical sequences of the candidate causal source task, the target task, and the third task, respectively. , , , and For model parameters, Representing meteorological covariates, P , Q , R and L These represent the corresponding lag order.

[0038] If we want to predict electricity prices and observe whether there is a causal relationship between power output and electricity prices, the input information of the restricted regression model includes: historical electricity price information, historical load information, and meteorological covariates, and the output is the prediction result of future electricity prices; the input of the unrestricted regression model is: historical electricity price information, historical load information, historical power output information, and meteorological covariates.

[0039] If we want to predict power output and observe whether there is a causal relationship between electricity price and power output, the input information of a restricted regression model includes: historical power output information, historical load information, and meteorological covariates, and the output is a predicted future electricity price. The input of an unrestricted regression model is: historical power output information, historical load information, historical electricity price information, and meteorological covariates.

[0040] Construct the sum of squared residuals from the outputs of the restricted regression model and the unrestricted regression model. F The statistical measure, and based on the significance test, determine whether the candidate causal source task provides additional predictive information to the target task. F The statistic is: ; in, For prediction tasks i To predict tasks j of F Statistic; and Let these represent the sum of squared residuals for the restricted regression model and the unrestricted regression model, respectively. and Representing the restricted regression model and the unrestricted regression model respectively for the first... n Prediction task in each sample j The fitted value, K R and K U These represent the number of parameters in the restricted regression model and the unrestricted regression model, respectively. N e This represents the number of valid samples within a local time window; according to the above... F Calculate the time-varying causality strength matrix using statistics, significance test results, and causal edge signs: ; in, Indicates a point in time t By task i Pointing to task j The strength of causal sign, This indicates the corresponding ordered task pair at a given time point. t of F Statistic, This represents the p-value corresponding to the significance test. The causal edge symbol is determined by the regression coefficients corresponding to the candidate causal source tasks in the unrestricted regression model. Describes the minimum-maximum normalization function. Indicates the significance threshold.

[0041] Repeat the above steps until the time-varying causality strength of all ordered task pairs has been calculated, forming a time-varying causality strength matrix. : ; In this context, the superscripts re, load, and price correspond to the forecasting tasks for renewable energy output, load, and electricity price, respectively.

[0042] In this embodiment, historical time-series data of renewable energy output, load, electricity price, and meteorological covariates are first acquired. Renewable energy output includes wind power output and photovoltaic power output, and meteorological covariates include temperature, humidity, wind speed, and wind direction. The data are divided according to time sequence, and standardized parameters are estimated using the training set. Then, the same set of standardized parameters is applied to the validation and test sets.

[0043] like Figure 3 As shown, for each prediction time, a local time window is constructed using historical observation data prior to that time, and a Granger causality test is performed on the ordered task pairs between source-load-price tasks within this local time window. (Task...) i To the task j Taking predictive information as an example, we can construct the following two types of models: Restricted models only include the target task. j Historical items, third task k Historical terms and meteorological covariates; the unrestricted model incorporates a candidate causal source task based on the restricted model. i Historical terms. If the unrestricted model significantly reduces the sum of squared residuals compared to the restricted model, then the task... i Recognized as a task j Granger causality.

[0044] Calculated from the sum of squared residuals of the constrained and unconstrained models F Statistics, F The statistical measure serves as the fundamental measure of causal strength, and the sign of the causal edge is determined based on the regression coefficients corresponding to the candidate causal source tasks. After min-max normalization, the signed local causal strength is obtained. This process is repeated for all ordered task pairs to form the time-varying causal strength matrix at the current prediction time. The time-varying causal matrix can be represented as follows: Figure 4 As shown.

[0045] Furthermore, the coupling direction signal The acquisition process includes: and Input to a third fully connected neural network The task is obtained by merging the two. i Pointing to task j Coupling direction signal in time window t : .

[0046] Regarding the information modulation stage: utilizing the aforementioned symbolic causality strength and the source load price running state vector Calculate the gating activation value ;Will The fusion process yields cross-task coupled information. ; using the gate activation value For the cross-task coupling information Modulation is performed.

[0047] As an optional implementation, the utilization of the symbolic causal strength and the source load price running state vector Calculate the gating activation value Including: The modulus and source load cost operating state vector After being input into the first fully connected neural network, it is normalized using the Sigmoid activation function, and then... Multiplication is used as the gate activation value : .

[0048] Specifically, the causal gating mechanism generates gating activation values ​​based on the time-varying causal strength matrix and the source load valence running state vector: ; in, Indicates a point in time t By task i Pointing to task j The gating activation value, This represents the Sigmoid function. This represents a gated state modulation network, composed of fully connected neural networks. The corresponding causal strength is indicated; when the corresponding causal strength is zero, the gating activation value is zero, which blocks the corresponding cross-task information transmission path, thereby explicitly controlling the information transmission process between different prediction tasks.

[0049] As an optional implementation, the [implementation method] will The fusion process yields cross-task coupled information. Including: Input a preset neural network to enable it to utilize the formula The fusion process yields cross-task coupled information. .

[0050] Specifically, cross-task coupling information Historical time-series features based on causal source task Source-load-price operating state vector and coupling direction signal Perform the calculation: ; ; ; in, , and These represent the targets for new energy output, load, and electricity price at specific points in time. t Historical data time series characteristics.

[0051] As an optional implementation, the use of the gating activation value For the cross-task coupling information Modulation is performed, including: utilizing the gating activation value of each task pair. Information coupling across tasks Modulation is performed, and then the modulation results are superimposed to obtain intermediate variables. .

[0052] Specifically, this stage uses the historical time-series features of the causal source task, the source-load-price operating state vector, and the coupling direction signal as inputs. The source-load-price operating state vector is obtained by concatenating the historical time-series features of the three tasks and then passing them through a state coding network, representing the current overall operating state of the source-load-price. The coupling direction signal is obtained by mapping the causal edge symbols and causal strengths through an edge representation network, representing the current cross-task coupling direction and strength.

[0053] After the shared coupled network outputs a vector of potential spatial coupling effects, it is modulated by a causal gating mechanism. This mechanism uses time-varying causal strength as a hard constraint and combines it with the source-load-priority running state vector to generate a state modulation value. The gating activation value is equal to the product of the causal strength and the state modulation value. When a causal edge is not identified or is physically shielded, the corresponding causal strength is zero, and the gating activation value is also zero, thus blocking the transmission path. In this way, cross-task information is transmitted along the effective causal direction, and the transmission strength changes with the current running state.

[0054] For the target task j Its final fusion feature is obtained by adding its own historical data time series features to the coupling effect vector from all valid causal source tasks.

[0055] Regarding the task prediction stage: The superimposed values ​​of the modulation results corresponding to all task pairs i and j are superimposed with three historical time series features to obtain three input features; the three input features are then input into their respective prediction models to obtain the prediction results for new energy output, load, and electricity price.

[0056] Specifically, the final predicted value for the target task is obtained by adding the historical time-series features of the target task data to the gating coupling effect vector of all valid causal source tasks, and then inputting the result into the target task predictor: ; in, Indicate the target task j At the point of timet The predicted value, Indicate the target task j Task predictor, Indicates a point in time t The time-varying causality intensity matrix, This represents all targets in the time-varying causality strength matrix. j The effective causal source task is summed.

[0057] As an optional implementation, the step of inputting the three input features into their respective corresponding prediction models to obtain prediction results for new energy output, load, and electricity price includes: Input power characteristics Input the power output prediction model to obtain the power output prediction results; Load input characteristics Input the load forecasting model to obtain the load forecasting results; Inputting electricity price features Input the electricity price prediction model to obtain the electricity price prediction results. The comparison results of this embodiment with other algorithms are as follows: Figure 5 As shown.

[0058] Example 2 This embodiment provides a source-load price joint forecasting device for a new energy power system, including: The sample construction module is used to construct a joint prediction sample of source-load price at multiple time points t within a preset period, including new energy output, load, electricity price and meteorological covariates. The status acquisition module is used to extract features from the renewable energy output, load, and electricity price in each of the source-load-price joint prediction samples, and obtain the historical time-series features of renewable energy output, load, and electricity price respectively. The source-load-price operating state vector is then fused from the three. ; The causal acquisition module is used to identify the sign causal strength between all task pairs i and j in the source-load-price joint prediction sample based on the meteorological covariates. The symbolic causal strength and its corresponding causal edge symbol The coupling direction signal is obtained by fusion. ; Information modulation module, used to utilize the symbol causality strength and the source load price running state vector Calculate the gating activation value ;Will The fusion process yields cross-task coupled information. ; using the gate activation value For the cross-task coupling information Modulation; The task prediction module is used to superimpose the modulation results corresponding to all task pairs i and j with three historical time series features to obtain three input features; the three input features are then input into their respective prediction models to obtain the prediction results for new energy output, load, and electricity price.

[0059] Example 3 This embodiment provides a source-load price joint prediction system for a new energy power system, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the source-load price joint prediction method for the new energy power system.

[0060] This system can be a computing device such as a desktop computer, laptop, handheld computer, or cloud server. The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The memory can be used to store computer programs and / or modules. The processor implements various system functions by running or executing the computer programs and / or modules stored in the memory, and by accessing data stored in the memory.

[0061] Example 4 This embodiment provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the source-load price joint prediction method for the new energy power system.

[0062] Specifically, the memory may include high-speed random access memory, as well as non-volatile memory, such as hard disks, RAM, plug-in hard disks, smart media cards (SMC), secure digital (SD) cards, flash cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0063] Example 5 This application provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps of the source-load price joint prediction method for the new energy power system provided in the above embodiments.

[0064] The technical features of the embodiments described above can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification. It should be noted that the terms "in one embodiment," "for example," and "as in another example" in this application are intended to illustrate the application and are not intended to limit the application.

[0065] The embodiments described above are merely examples of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these modifications and improvements all fall within the protection scope of this application.

Claims

1. A method for joint prediction of source-load prices in a new energy power system, characterized in that, include: Construct a joint forecast sample of source-load price at multiple time points t within a preset period, including new energy output, load, electricity price and meteorological covariates; Feature extraction is performed on the renewable energy output, load, and electricity price in each of the source-load-price joint prediction samples to obtain the historical time-series features of renewable energy output, load, and electricity price. The source-load-price operating state vector is then fused from the three. ; Based on the meteorological covariates, identify the sign causality strength between all task pairs i and j in the source-load-price joint prediction sample. The symbolic causal strength and its corresponding causal edge symbol The coupling direction signal is obtained by fusion. ; Using the symbolic causal strength and the source load price running state vector Calculate the gating activation value ;Will The fusion process yields cross-task coupled information. ; Using the gate activation value For the cross-task coupling information Modulation; The superposition values ​​of the modulation results corresponding to all task pairs i and j are superimposed with three historical time series features to obtain three input features; the three input features are then input into their respective prediction models to obtain the prediction results for new energy output, load, and electricity price.

2. The source-load price joint forecasting method for new energy power systems as described in claim 1, characterized in that, The use of the symbolic causal strength and the source load price running state vector Calculate the gating activation value ,include: Will The modulus and source load cost operating state vector After being input into the first fully connected neural network, it is normalized using the Sigmoid activation function, and then... Multiplication is used as the gate activation value , .

3. The source-load price joint prediction method for new energy power systems as described in claim 2, characterized in that, The The fusion process yields cross-task coupled information. ,include: Will Input a preset neural network to enable it to utilize the formula The fusion process yields cross-task coupled information. .

4. The source-load price joint prediction method for new energy power systems as described in claim 2, characterized in that, The use of the gating activation value For the cross-task coupling information Modulation includes: Utilize the gating activation value of each task pair Information coupling across tasks Modulation is performed, and then the modulation results are superimposed to obtain intermediate variables. .

5. The source-load price joint forecasting method for new energy power systems as described in claim 1, characterized in that, The historical time sequence characteristics The extraction and fusion process includes: The renewable energy output, load, and electricity price in each source-load-price joint prediction sample are combined with meteorological covariates and then input into the corresponding task feature extractor to obtain the historical time-series features of renewable energy output, load, and electricity price. The source-load-price operating state vector is then fused from the three. ; Will , and Input to the second fully connected neural network So that it can use the formula Calculate the source load price running state vector corresponding to time point t. .

6. The source-load price joint forecasting method for new energy power systems as described in claim 1, characterized in that, The symbolic causal strength The acquisition process includes: Using formula Calculate the source-load price joint prediction sample task i Pointing to task j The symbolic causal strength is then used to obtain the time-varying causal strength matrix for all tasks: ； All tasks include forecasting renewable energy output, load forecasting, and electricity price forecasting. Indicates a point in time t By task i Pointing to task j The strength of causal sign, This indicates the corresponding task pair at a specific time point. t of F Statistic, This represents the p-value corresponding to the significance test. Indicates the causal edge symbol. Describes the minimum-maximum normalization function. Indicates the significance threshold; The coupling direction signal The acquisition process includes: and Input to a third fully connected neural network The task is obtained by merging the two. i Pointing to task j The coupling direction signal within the time window t, .

7. The source-load price joint forecasting method for new energy power systems as described in claim 1, characterized in that, The step of inputting the three input features into their respective prediction models to obtain prediction results for new energy output, load, and electricity price includes: Input power characteristics Input the power output prediction model to obtain the power output prediction results; Load input characteristics Input the load forecasting model to obtain the load forecasting results; Inputting electricity price features Input the electricity price prediction model to obtain the electricity price prediction results.

8. A source-load price joint forecasting device for a new energy power system, characterized in that, include: The sample construction module is used to construct a joint prediction sample of source-load price at multiple time points t within a preset period, including new energy output, load, electricity price and meteorological covariates. The status acquisition module is used to extract features from the renewable energy output, load, and electricity price in each of the source-load-price joint prediction samples, and obtain the historical time-series features of renewable energy output, load, and electricity price respectively. The source-load-price operating state vector is then fused from the three. ; The causal acquisition module is used to identify the sign causal strength between all task pairs i and j in the source-load-price joint prediction sample based on the meteorological covariates. The symbolic causal strength and its corresponding causal edge symbol The coupling direction signal is obtained by fusion. ; Information modulation module, used to utilize the symbol causality strength and the source load price running state vector Calculate the gating activation value ;Will The fusion process yields cross-task coupled information. ; Using the gate activation value For the cross-task coupling information Modulation; The task prediction module is used to superimpose the modulation results corresponding to all task pairs i and j with three historical time series features to obtain three input features; the three input features are then input into their respective prediction models to obtain the prediction results for new energy output, load, and electricity price.

9. A source-load price joint forecasting system for a new energy power system, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the source-load price joint prediction method for the new energy power system as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the source-load price joint prediction method for the new energy power system as described in any one of claims 1 to 7.

Images