CN-121998200-A - Photovoltaic power generation multiscale power prediction transaction optimization method

CN121998200ACN 121998200 ACN121998200 ACN 121998200ACN-121998200-A

Abstract

The invention provides a photovoltaic power generation multiscale power prediction transaction optimization method, and relates to the technical field of power transaction. The method includes constructing a corresponding data set for each transaction day that is preselected. And simulating a preset transaction strategy based on the data set aiming at the data set corresponding to each transaction day to obtain a plurality of decision prediction tracks. And screening a plurality of high-risk tracks with transaction results meeting high-risk conditions from the decision prediction tracks, and determining at least one high-risk evolution mode based on the change rule of the power generation quantity prediction value in each high-risk track. For each identified high risk evolution mode, a corresponding transaction sub-policy is generated, and the transaction policy is updated based on the transaction sub-policies. The method and the system can identify and respond to transaction risks caused by specific prediction evolution modes, help reduce potential losses caused by prediction information cross-period evolution and decision constraint accumulation, and improve decision adaptability and risk resistance.

Inventors

BAI YAN
Dong Zengkai
LI YUAN
NI WANGDAN
CHEN LV
XIE FENGYU
PAN JINGNA
SONG JU

Assignees

华能澜沧江新能源有限公司

Dates

Publication Date: 20260508
Application Date: 20260210

Claims (10)

1. The photovoltaic power generation multiscale power prediction transaction optimization method is characterized by comprising the following steps of: Constructing a corresponding data set for each preselected trading day, wherein the data set comprises a group of daily forecast sequences, daily forecast sets and actual power sequences corresponding to the trading day; The method comprises the steps of carrying out simulation on a preset transaction strategy based on a data set corresponding to each transaction day to obtain a plurality of decision prediction tracks, wherein the transaction strategy is used for carrying out transaction decisions at a plurality of preset transaction moments based on the daily prediction sequence and the daily prediction set to obtain transaction contracts; Screening a plurality of high-risk tracks with transaction results meeting high-risk conditions from the decision prediction tracks, and determining at least one high-risk evolution mode based on the change rule of the power generation quantity prediction value in each high-risk track; Generating a corresponding transaction sub-strategy for each identified high-risk evolution mode, updating the transaction strategy based on the transaction sub-strategy, so that when the transaction strategy is executed, triggering the transaction sub-strategy corresponding to the high-risk evolution mode under the condition that the change of the real-time updated intra-day prediction set is monitored to be matched with any high-risk evolution mode.
2. The method of optimizing a photovoltaic power generation multiscale power forecast trade according to claim 1, wherein the pre-day forecast sequence comprises a plurality of chronologically arranged power generation capacity forecast values, each forecast value corresponding to one trade time in the trade day; the intra-day prediction set comprises a plurality of prediction subsequences which are rolled and released at different release time points on the same day of the transaction day, and each prediction subsequence comprises power generation amount prediction values corresponding to a plurality of continuous transaction moments; The actual power sequence comprises actual power generation amount values corresponding to all trade moments in the trade day.
3. The method of optimizing photovoltaic power generation multiscale power prediction transaction according to claim 2, wherein the step of constructing the data set comprises: And carrying out time alignment processing on the pre-day predicted sequence, the predicted subsequence and the actual power sequence according to the same group of transaction time preset for the transaction day.
4. The photovoltaic power generation multiscale power prediction trading optimization method of claim 2, wherein simulating a preset trading strategy based on the data set comprises: sequentially traversing each transaction moment in a day-ahead prediction sequence and a day-in prediction set of the data set according to a time sequence; Aiming at the transaction time traversed currently, executing a preset transaction strategy, generating at least one transaction decision according to the current available prediction information set and generating transaction contracts corresponding to each transaction decision based on the current power sequence and each transaction contract, wherein the current available prediction information set is formed based on the current prediction sequence and a prediction subsequence in the intra-day prediction set with the release time not later than the current time; and combining the transaction decisions, the transaction contracts and the transaction results corresponding to each transaction moment into a decision prediction track.
5. The method for optimizing photovoltaic power generation multi-scale power prediction trading according to claim 1, wherein the step of screening a plurality of high risk trajectories, the trading results of which meet a high risk condition, from the decision prediction trajectories comprises: And sequencing the decision prediction tracks according to the transaction result, and determining the decision prediction track with the preset proportion at the tail in the sequencing result as a high risk track.
6. The method of optimizing photovoltaic power generation multiscale power prediction trading of claim 1, wherein determining at least one high risk evolution mode based on a law of change in the predicted value of power generation in each high risk trajectory comprises: Aiming at each high-risk track, acquiring a intra-day prediction set corresponding to the high-risk track; analyzing a change trend formed by a plurality of power generation quantity predicted values aiming at the same future transaction time along with the release time of the power generation quantity predicted values in the intra-day prediction set; And defining at least one high-risk evolution mode by analyzing the similar variation trends identified in the plurality of high-risk tracks, and establishing corresponding trigger judgment conditions for each high-risk evolution mode.
7. The photovoltaic power generation multiscale power prediction transaction optimization method according to claim 6, wherein the matching manner of the high risk evolution mode comprises: For any future transaction time, if in the intra-day prediction set, there are multiple continuous prediction value updates for the time, and the prediction value of each update is consistent in change direction and exceeds a preset threshold value compared with the corresponding prediction value in the previous update value or the pre-day prediction sequence, determining that the evolution of the current prediction data matches the high risk evolution mode.
8. The photovoltaic power generation multiscale power prediction trading optimization method of claim 1, wherein the generating a corresponding trading sub-strategy for each identified high-risk evolution mode comprises: Extracting each high-risk track matched with the target high-risk evolution mode to form a training set; determining each decision time when the target high risk evolution mode is triggered for each high risk track in the training set; generating a training sample for each triggered decision time based on the system state of the time, the system state comprising the predicted data state of the time, the held related transaction contract state, and the accumulated transaction results from the time to the end of the transaction day; And carrying out strategy parameter optimization based on a plurality of training samples to obtain a transaction sub-strategy corresponding to the target high risk evolution mode.
9. The method for optimizing photovoltaic power generation multiscale power prediction transaction according to claim 8, wherein the obtaining policy parameters corresponding to the high risk evolution mode based on each training sample comprises: Constructing a strategy optimization model by taking the expected transaction loss after the target high risk evolution mode is triggered as an optimization target, wherein the input of the strategy optimization model is the system state in the training sample, and the output of the strategy optimization model is a group of strategy parameters; and solving the strategy optimization model to obtain a group of optimal strategy parameters.
10. A photovoltaic power generation multiscale power forecast transaction optimization system, comprising: The data acquisition module is used for constructing a corresponding data set for each transaction day, wherein the data set comprises a group of daily forecast sequences, daily intra-forecast sets and actual power sequences corresponding to the transaction day; The track generation module is used for simulating a preset transaction strategy based on a data set corresponding to each transaction day to obtain a plurality of decision prediction tracks; each decision prediction track comprises a transaction decision sequence sequenced according to transaction moments and transaction contracts corresponding to each transaction decision in the transaction decision sequence, and transaction results obtained based on the actual power sequence and each transaction contract; The risk screening module is used for screening a plurality of high-risk tracks with transaction results meeting high-risk conditions from the decision prediction tracks, and determining at least one high-risk evolution mode based on the change rule of the power generation quantity predicted value in each high-risk track; And the strategy adjustment module is used for generating a corresponding transaction sub-strategy for each identified high-risk evolution mode, updating the transaction strategy based on the transaction sub-strategy, so that when the transaction strategy is executed, the transaction sub-strategy corresponding to the high-risk evolution mode is triggered under the condition that the change of the real-time updated intra-day prediction set is monitored to be matched with any high-risk evolution mode.

Description

Photovoltaic power generation multiscale power prediction transaction optimization method Technical Field The application relates to the technical field of power trading, in particular to a photovoltaic power generation multiscale power prediction trading optimization method. Background In the context of participation of photovoltaic power generation in the power market, an operation subject needs to sequentially submit and adjust a transaction plan in multiple time scales (such as the day before and the day in time) according to a prediction result of future power generation power in order to obtain economic benefits, and finally bears settlement risks caused by actual power generation and plan deviation. Currently, independent prediction results based on corresponding time scales in each transaction period are generally relied on as decision bases. Widely used transaction optimization methods also tend to make independent, local optimal decisions at each decision time based on the most recent predicted segment at that time. This approach may be effective in a single time scale. However, such a technical path may not adequately evaluate the deeper, cross-cycle structural risk. This risk arises from the interaction between the dynamic evolution characteristics of the multi-scale generation forecast sequences and the rigid constraints formed by the serialization decisions (i.e. established trade contracts or market positions). When the predicted data exhibits a sustained and significant adverse evolution trend over a subsequent time window, contracts established earlier based on different prediction assumptions may limit the subsequent tuning space, possibly forcing the operator to correct under high cost conditions, thereby weakening the overall revenue. Thus, the current core challenge goes from managing "prediction errors" at a single point in time, to managing more complex "predicting high risk evolution patterns to interact with the decision chain", i.e. there is a limitation in systematically identifying such risks triggered by a particular prediction evolution pattern and conducted through the decision chain. Disclosure of Invention In order to overcome the above problems in the prior art, the present disclosure provides a photovoltaic power generation multiscale power prediction transaction optimization method, including: Constructing a corresponding data set for each preselected trading day, wherein the data set comprises a group of daily forecast sequences, daily forecast sets and actual power sequences corresponding to the trading day; The method comprises the steps of carrying out simulation on a preset transaction strategy based on a data set corresponding to each transaction day to obtain a plurality of decision prediction tracks, wherein the transaction strategy is used for carrying out transaction decisions at a plurality of preset transaction moments based on the daily prediction sequence and the daily prediction set to obtain transaction contracts; Screening a plurality of high-risk tracks with transaction results meeting high-risk conditions from the decision prediction tracks, and determining at least one high-risk evolution mode based on the change rule of the power generation quantity prediction value in each high-risk track; Generating a corresponding transaction sub-strategy for each identified high-risk evolution mode, updating the transaction strategy based on the transaction sub-strategy, so that when the transaction strategy is executed, triggering the transaction sub-strategy corresponding to the high-risk evolution mode under the condition that the change of the real-time updated intra-day prediction set is monitored to be matched with any high-risk evolution mode. Further, the day-ahead prediction sequence comprises a plurality of power generation amount predicted values which are arranged in time sequence, and each predicted value corresponds to one transaction time in the transaction day; the intra-day prediction set comprises a plurality of prediction subsequences which are rolled and released at different release time points on the same day of the transaction day, and each prediction subsequence comprises power generation amount prediction values corresponding to a plurality of continuous transaction moments; the actual power sequence comprises actual power generation amount values corresponding to all trade moments in the trade day. Further, the step of constructing the dataset includes: And carrying out time alignment processing on the pre-day predicted sequence, the predicted subsequence and the actual power sequence according to the same group of transaction time preset for the transaction day. Further, the simulating the preset transaction strategy based on the data set includes: sequentially traversing each transaction moment in a day-ahead prediction sequence and a day-in prediction set of the data set according to a time sequence; Aiming at the transaction time trav