A virtual power plant power market transaction agent method and device

Abstract

The application provides a virtual power plant power market transaction agent method and device, comprising: for the identification conclusion that the site layout is arranged in a band shape, determining the predicted duration of the aggregated output decline section and the valley bottom depth according to the cloud cluster travel trajectory and the irradiation measured data; whether the preset retreat trigger threshold is reached is analyzed through the predicted duration of the aggregated output decline section and the valley bottom depth, forming an activation signal of retreat quotation adjustment; the retreat quotation trigger threshold is updated according to the activation signal and the irradiation measured data, whether the updated trigger threshold breaks through the preset slope limit is evaluated, and the advance locking time of the quotation retreat is determined; the advance locking time of the quotation retreat and the fitting form of the aggregated output decline section are optimized, the locking time point set of the market retreat quotation is determined; the real-time transaction execution instruction is generated through the locking time point set of the market retreat quotation and the cloud cluster travel trajectory, and the optimization output of the virtual power plant power market transaction is confirmed.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method and apparatus for virtual power plant electricity market transaction agency. Background Technology

[0002] In the electricity market, virtual power plants (VPS) serve as a crucial platform for integrating distributed energy resources, playing an irreplaceable role in ensuring the stability and economic efficiency of power supply. With the widespread application of new energy sources such as distributed photovoltaic (PV), effectively aggregating and managing the output of these dispersed resources has become a core issue for improving market trading efficiency and grid security. VPS need to complete precise bidding and dispatch within a very short timeframe to cope with rapid changes in electricity demand, and the volatility of distributed PV output directly impacts the achievement of this goal. Currently, many solutions, when dealing with distributed PV aggregation output, often neglect the impact of external environmental factors on output volatility, especially the complex changes under dynamic weather conditions. Existing methods mostly focus on static geographical distribution optimization, assuming that higher site dispersion leads to smaller output volatility, but fail to fully consider the interference of the spatiotemporal evolution of weather phenomena on output smoothness. This limitation results in output volatility not only failing to decrease in certain scenarios but also exhibiting a more drastic trend, affecting the stability of bidding and execution effectiveness. Cloud clusters, as a major weather factor affecting PV power generation, directly impact sites in different geographical locations with their movement trajectories and coverage areas, causing output volatility to exhibit spatiotemporal correlation. If the interaction between the cloud's movement direction and the spatial layout of power plants is not accurately grasped, it is difficult to predict the downward trend of the power output curve. For example, when a cloud moves along a specific direction, if multiple power plants happen to be arranged in a strip along this path, a continuous shading effect may occur, causing the aggregated power output curve to have a significant trough in a short period of time and a prolonged trough, thus posing a significant risk to pricing strategies. Therefore, how to accurately identify the duration and depth of the downward phase of distributed photovoltaic aggregated power output under the dynamic influence of cloud movement, and how to adjust trading strategies in a timely manner based on the spatiotemporal relationship between power plant layout and cloud trajectory, has become a critical issue that urgently needs to be addressed in virtual power plant electricity market trading. Summary of the Invention

[0003] This invention provides a virtual power plant electricity market transaction agency method, mainly including: Obtain cloud cluster outline information and trajectory from satellite cloud image data; The projection distribution of each site is calculated based on the cloud's trajectory and the set of geographical coordinates of the distributed photovoltaic sites, and the arrangement of the site layout is determined. Based on the arrangement of the site layout and the trajectory of the cloud cluster, combined with actual irradiation data, the characteristics of the decrease in aggregate output are predicted. Based on the comparison between the aggregate output decline characteristics and the preset threshold, a trading adjustment signal is generated; The timing of the transaction adjustment is determined by updating the trigger threshold based on the transaction adjustment signal and the measured irradiation data. The transaction timing set is optimized by combining the locking timing with the aggregated output decline characteristics. Based on the set of transaction times and the trajectory of the cloud cluster, a transaction execution instruction is generated, and the optimized results of the virtual power plant electricity market transaction are output.

[0004] Furthermore, cloud contour information and trajectory are obtained from satellite cloud image data, including: Cloud contour image data is acquired from the satellite cloud image receiving terminal via a data channel; Perform grayscale gradient calculation on the cloud contour image data, extract the grayscale transition region between the cloud and the background, and mark the current contour boundary point set of the cloud; The cloud centroid position is calculated based on the cloud contour image sequence within a continuous time frame. The displacement direction and distance of the cloud are determined by the centroid position offset. The cloud trajectory is then fitted to obtain the cloud trajectory. Obtain the measured irradiance data and geographic coordinate set of the distributed photovoltaic site, and perform spatial location matching between the geographic coordinate set and the cloud trajectory to identify the correspondence between the cloud trajectory and the irradiance data.

[0005] Furthermore, based on the cloud cluster's trajectory and the set of geographical coordinates of the distributed photovoltaic sites, the projected distribution status of each site is calculated, and the arrangement of the site layout is determined, including: Extract the direction vector of the cloud's trajectory; The vertical distance from each station to the trajectory is calculated based on the direction vector and the geographical coordinates of each station, and the projected distance is obtained. Determine whether the projected distance falls within a preset range, and count the proportion of stations that fall within the range; If the ratio exceeds a preset threshold, the projection distance is determined to be concentrated. Based on the centralized distribution status and the arrangement order of the stations along the trajectory direction, the projected position of each station in the trajectory direction is calculated to determine the arrangement pattern of the station layout.

[0006] Furthermore, based on the arrangement of the site layout and the trajectory of the cloud cluster, combined with measured irradiance data, the characteristics of the decrease in aggregation output are predicted, including: If the sites are arranged in a strip, a site occlusion sequence is established based on the direction of the cloud's trajectory. The occlusion transmission time interval between adjacent sites is calculated based on the occlusion sequence and the cloud movement speed, and the aggregation output decrease time is obtained by summing them up. Historical irradiance sequences of each site were obtained, and the irradiance attenuation magnitude of each site under shading was predicted by a time-series fitting model. The irradiation attenuation amplitude is weighted and summed according to the installed capacity weight of each site to obtain the polymerization output decline curve, and the valley depth of the decline curve is extracted.

[0007] Furthermore, based on the comparison result between the aggregation output decline characteristic and the preset threshold, a transaction adjustment signal is generated, including: Obtain the duration of the decrease in aggregate output and the depth of the valley; The descent duration is compared with a preset duration threshold, and the valley depth is compared with a preset depth threshold. If the descent duration exceeds a duration threshold and the valley depth exceeds a depth threshold, then a preset threshold is determined to have been reached. Based on the reach determination result, a transaction adjustment signal is output.

[0008] Furthermore, based on the transaction adjustment signal and the updated trigger threshold of the measured irradiation data, the timing of the transaction adjustment lockout is determined, including: The rate of change of irradiance intensity is calculated based on the measured irradiance data at the time the transaction adjustment signal is triggered. The trigger threshold is adjusted proportionally based on the rate of change and a preset correction coefficient to obtain the updated threshold. Calculate the slope of the threshold change before and after the update, and compare it with the preset slope threshold; If the slope of the change exceeds a preset threshold, it is determined that the slope limit has been exceeded, and the current moment is determined as the lock-in time for the transaction adjustment.

[0009] Furthermore, optimizing the transaction timing set through the locking timing and the aggregation output decline characteristic includes: The fitting shape of the decline curve is analyzed based on the characteristics of the decline in polymerization output. The fitting shape consists of the starting point of the decline, the trough point, and the recovery point. The inflection point of the descending curve is detected by a sliding window algorithm, and the time of the feature point and the output coordinates are recorded. Calculate the time offset between the locking timing and the descent start point, and adjust the locking timing according to the offset to obtain the optimized locking timing; The optimized locking timing is combined with the feature points of the descending curve to form a set of transaction timing points.

[0010] Furthermore, after generating the transaction execution instruction based on the set of transaction timestamps and the cloud trajectory, the process includes: The price change rate is obtained by calculating the ratio of the price adjustment difference between adjacent executed instructions to the time interval.

[0011] Furthermore, after obtaining the price change rate, the process includes: If the rate of change in the quoted price is lower than the preset stable threshold, then the smoothness of the polymerization output is determined to meet the requirements. Based on the smoothness determination result, the transaction execution instruction is marked as a valid instruction and output to the scheduling platform.

[0012] The present invention also provides a virtual power plant electricity market trading agency device, the device comprising: The cloud outline and trajectory recognition module is used to obtain cloud outline information and movement trajectory from satellite cloud image data; The morphology judgment module is used to calculate the projection distribution status of each site based on the cloud's trajectory and the set of geographical coordinates of the distributed photovoltaic sites, and to determine the arrangement of the site layout. The feature prediction module is used to predict the characteristics of decreased aggregate output based on the arrangement of the site layout and the trajectory of the cloud cluster, combined with actual irradiation data. The signal generation module is used to generate a trading adjustment signal based on the comparison result between the aggregate output decline characteristics and a preset threshold; The timing determination module is used to update the trigger threshold based on the transaction adjustment signal and the measured irradiation data to determine the locking timing of the transaction adjustment; The set optimization module is used to optimize the set of transaction timings by combining the locking timing with the aggregated output decline characteristics; The instruction generation module is used to generate transaction execution instructions based on the set of transaction times and the trajectory of the cloud cluster, and output the optimized results of the virtual power plant electricity market transaction.

[0013] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses a virtual power plant electricity market transaction agency method, addressing the issue of aggregated output fluctuations caused by the impact of cloud movement on irradiance data. It integrates cloud trajectory prediction, site layout analysis, and bid retracement optimization into a logically coherent solution. This invention obtains the cloud outline and movement direction from satellite cloud images, combines site geographic coordinates with measured irradiance data, identifies the banded arrangement characteristics of the cloud trajectory and site layout, and then uses a spatiotemporal correlation prediction model to extract the predicted duration and valley depth of the aggregated output decline segment. It assesses whether a bid retracement adjustment signal has been triggered, dynamically updates the trigger threshold, optimizes the early locking timing of bid retracement, and finally generates real-time transaction execution instructions, ensuring that the smoothness of aggregated output meets the requirements for ultra-short-term bid stability. The core innovation of this invention lies in its significant improvement in the stability and profitability of virtual power plants in electricity market transactions through accurate prediction of cloud trajectory and irradiance data and intelligent optimization of bid timing. Attached Figure Description

[0014] Figure 1 This is a flowchart of a virtual power plant electricity market transaction agency method according to the present invention.

[0015] Figure 2 This is a schematic diagram of the structure of a virtual power plant electricity market trading agency device according to the present invention. Detailed Implementation

[0016] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0017] like Figure 1 This embodiment of a virtual power plant electricity market transaction agency method and apparatus may specifically include: Step S101: Obtain cloud contour image data and movement direction information from the satellite cloud image receiving terminal, obtain station-by-station irradiation measurement data and station geographic coordinate set, and identify the current contour boundary point set and cloud trajectory.

[0018] A data channel is established between the scheduling platform and the satellite cloud image receiving terminal. Cloud contour image data is acquired from the receiving terminal. This cloud contour image data includes the grayscale distribution and edge morphology of the cloud at the current moment. The Sobel operator is used to perform gradient calculations on the grayscale distribution in the horizontal and vertical directions to extract the grayscale transition regions between the cloud and the background. The current contour boundary point set of the cloud is marked based on the continuity of the grayscale transition regions. A sequence of cloud contour images within consecutive time frames is acquired from the satellite cloud image receiving terminal. The centroid position of the cloud is calculated based on the coordinate mean of the current contour boundary point set. Vector difference calculations are performed on the offset of the cloud centroid position in adjacent time frames to obtain the displacement direction and displacement distance of the cloud centroid. The cloud trajectory is then fitted based on the displacement direction and displacement distance. The system obtains the measured data of the irradiance sensor and the set of geographical coordinates of each distributed photovoltaic site from the scheduling platform. The measured data of the irradiance sensor includes the solar irradiance intensity value of each site in the current sampling period, such as the solar irradiance intensity value of each site per hour in the past 24 hours. The system performs spatial location matching based on the set of geographical coordinates of the sites and the trajectory of the cloud cluster. Combines the current contour boundary point set of the cloud cluster with the trajectory of the cloud cluster, the system identifies the correspondence between the trajectory of the cloud cluster and the measured irradiance data of each site.

[0019] In one embodiment, the scheduling platform establishes a data channel with the satellite cloud image receiving terminal through a preset communication protocol. The satellite cloud image receiving terminal transmits cloud outline image data to the scheduling platform at fixed time intervals. The cloud outline image data is stored in grayscale image format, wherein the cloud area presents a higher grayscale value and the background sky area presents a lower grayscale value.

[0020] Specifically, the Sobel operator is an edge detection method based on the first derivative, which identifies edge locations by calculating the gray-level gradient values ​​of each pixel in the horizontal and vertical directions.

[0021] For example, for any pixel in the cloud contour image data, the Sobel operator performs a weighted summation operation on the neighborhood of that pixel using both horizontal and vertical convolution templates to obtain the horizontal and vertical gradient components of that pixel. The sum of the squares of the horizontal and vertical gradient components is then taken as the square root to obtain the gradient magnitude of that pixel. When the gradient magnitude exceeds a preset gradient threshold, the pixel is determined to belong to a gray-level transition region between the cloud and the background. The coordinates of pixels satisfying the continuity condition within this gray-level transition region are extracted to form the current contour boundary point set of the cloud.

[0022] In one possible implementation, the centroid position of the cloud cluster is calculated based on the average abscissa and ordinate of all pixels in the current contour boundary point set of the cloud cluster. For the sequence of cloud cluster contour images transmitted by the satellite cloud image receiving terminal within consecutive time frames, the centroid position of the cloud cluster corresponding to each time frame is calculated. Vector difference operations are performed on the centroid positions of adjacent time frames to obtain the displacement direction and distance of the cloud cluster centroid between adjacent time frames. Based on the displacement direction and distance of multiple consecutive time frames, a linear fitting method is used to determine the trajectory of the cloud cluster.

[0023] It should be noted that the scheduling platform synchronously acquires the measured data from the irradiance sensors at each distributed photovoltaic site and the set of site geographic coordinates. The measured data from the irradiance sensors records the solar irradiance intensity value received by each site within the current sampling period. Based on the latitude and longitude information of each site in the site geographic coordinate set and the spatial location of the cloud cluster's trajectory, the platform matches this information with the coverage area of ​​the cloud cluster's current contour boundary point set to identify the correspondence between the cloud cluster's trajectory and the measured irradiance data at each site.

[0024] Step S102: Calculate the projection distance of each station relative to the cloud's path based on the cloud's trajectory and the set of station coordinates, evaluate whether the projection distances of multiple consecutive stations are concentrated within a preset strip, and identify whether the station layout is arranged in a strip.

[0025] The trajectory direction vector is extracted based on the cloud's movement trajectory. This trajectory direction vector is determined by the displacement direction of the cloud's centroid within a continuous time frame. The latitude and longitude coordinates of each station in the station's geographic coordinate set are obtained. A perpendicular line is drawn from each station's coordinates to the trajectory direction vector, and the vertical distance from each station to the cloud's movement path is calculated to obtain the projected distance of each station. A preset upper and lower limit value for a strip-shaped interval is set. It is then determined whether the projected distance of each station falls within the preset strip-shaped interval. The proportion of stations falling within the preset strip-shaped interval is counted. If this proportion exceeds a preset concentration threshold, the projected distances of the stations are determined to be concentrated, thus obtaining a concentrated distribution indicator for the station projected distances. Based on the concentrated distribution of the projected distances of the stations, and combined with the order of the stations along the cloud path, the projected position pi of each station in the trajectory direction is calculated, where pi is the distance from the station coordinates to the starting point of the trajectory after projection onto the trajectory vector. All pi are sorted from smallest to largest, and the Spearman rank correlation coefficient r between the sorted sequence and the original index sequence of the stations is calculated. If the concentrated distribution is true and r > 0.9, the station layout is determined to be in a strip arrangement; otherwise, it is not in a strip arrangement. The classification conclusion of whether the station layout is in a strip arrangement is output.

[0026] In one embodiment, the trajectory direction vector is extracted based on the displacement direction of the cloud's centroid within consecutive time frames. The trajectory direction vector represents the spatial orientation of the cloud's movement along a specific direction. For each station in the set of station geographic coordinates, a perpendicular line is drawn from the station's coordinates to the line containing the trajectory direction vector. The length of this perpendicular line is the vertical distance from the station to the cloud's path, and this vertical distance is recorded as the projected distance of the station.

[0027] Specifically, the preset strip interval is a distance range pre-set according to the spatial distribution characteristics of the distributed photovoltaic site cluster. The upper and lower limits of the strip interval define a narrow spatial domain.

[0028] For example, if photovoltaic power stations in a certain area are mainly distributed along both sides of a road or river, then the width of the strip interval is related to the width of the road or river. During the judgment process, the projected distance value of each station is read sequentially, and each projected distance is compared with the upper and lower limits of the strip interval. If the projected distance is between the upper and lower limits, the station is marked as falling into the strip interval. After all stations are judged, the number of stations falling into the strip interval is counted, and the proportion of this number to the total number of stations is calculated.

[0029] It should be noted that the concentration threshold is a pre-set proportional threshold. When the proportion of stations falling into the strip exceeds this concentration threshold, it indicates that the projection distance distribution of most stations is relatively concentrated. At this time, the concentrated distribution label is output as true; otherwise, the concentrated distribution label is output as false.

[0030] In one possible implementation, when the centralized distribution flag is true, the order in which each site is arranged along the cloud's path is further determined.

[0031] Specifically, the stations are sorted according to their projected positions on the trajectory direction vector. If the sorted stations show a progressive spatial distribution, the station layout is determined to be in a strip-like arrangement, and the output classification conclusion is strip-like arrangement; if the station distribution is discrete or disordered, the output classification conclusion is non-strip-like arrangement.

[0032] It is understandable that the classification conclusion of the strip arrangement reflects the spatial layout characteristics of the site group in the direction of cloud movement. When the sites are arranged in a strip, the time interval between the cloud passing through each site is short, and the fluctuation of the aggregation output has temporal continuity.

[0033] Step S103: Based on the identification conclusion that the site layout is arranged in a strip, determine the predicted duration and valley depth of the aggregation output decline segment according to the cloud trajectory and actual irradiation measurement data.

[0034] The system identifies a strip-shaped arrangement of monitoring stations. If this is the case, a station shading sequence is established based on the station arrangement along the cloud's trajectory. This sequence is ordered according to the order in which the cloud passes each station, determining the shading order of each station. The ratio of the spatial distance between adjacent stations to the cloud's movement speed is calculated based on the station shading sequence and the cloud's trajectory speed, yielding the shading transmission time interval between adjacent stations. The shading transmission time intervals of all adjacent stations in the shading sequence are summed to obtain the predicted duration of the aggregation output decline segment. Measured irradiance data for each station is obtained, and hourly irradiance intensity data for the past 24 hours is extracted from the monitoring systems of each station to form a historical irradiance intensity sequence. Based on the shading order of each station in the shading sequence, a Long Short-Term Memory (LSTM) network is used to perform time-series fitting on the measured irradiance data of each station. The LSM network's input is the historical irradiance intensity sequence for each station, with a sequence length of 24, and its output is the irradiance attenuation amplitude of each station during the shading period. Installed capacity data Ci (e.g., 50MW for site A) is extracted from the monitoring systems of each site. The weight wi = Ci / ΣCj is calculated, where ΣCj is the sum of the installed capacity of all sites. The irradiance attenuation amplitude of each site is weighted and summed according to the temporal order of the site shading sequence to obtain the aggregated power output decline curve. The start and end times of the power output decline segment are identified based on the decline curve. The minimum point of aggregated power output is searched within the decline segment. The difference between the power output value corresponding to the minimum point and the power output value at the start of the decline segment is calculated to obtain the valley depth of the aggregated power output decline segment.

[0035] In one embodiment, when the identification conclusion that the site layout is arranged in a strip indicates that multiple distributed photovoltaic sites are distributed sequentially along the direction of cloud movement, the sites are sorted according to their projection positions on the cloud movement trajectory. Sites with earlier projection positions are arranged at the front of the shading sequence, and sites with later projection positions are arranged at the back of the shading sequence, thereby establishing a complete site shading sequence.

[0036] Specifically, the calculation of the occlusion transmission time interval is based on the cloud's movement speed and the spatial distance between adjacent sites.

[0037] For example, assuming the cloud cluster moves in a northeast direction, and the shading sequence includes several photovoltaic stations arranged sequentially along this direction, the projected distance between two adjacent stations in the cloud cluster's direction of movement is measured. Dividing this projected distance by the cloud cluster's moving speed yields the time it takes for the cloud cluster to move from one station to the next; this time is the shading transmission time interval between adjacent stations. The cloud cluster's moving speed is determined by the ratio of the displacement distance of the cloud cluster's centroid within a continuous time frame to the time frame interval.

[0038] It should be noted that the predicted duration of the aggregation output decline segment is obtained by summing up the occlusion transmission time intervals of all adjacent sites in the site occlusion sequence. This predicted duration reflects the total time span of the cloud cluster moving from the starting site to the ending site of the occlusion sequence.

[0039] In one possible implementation, a Long Short-Term Memory (LSTM) network is a gated recurrent neural network that models long-term dependencies in time-series data through input, forget, and output gates. For the measured irradiance data at each site, the LSM network's input is a sequence of irradiance intensity values ​​at multiple historical sampling points, arranged chronologically to form a one-dimensional time-series vector. The LSM network uses the forget gate to determine which historical irradiance information should be retained and which should be discarded, uses the input gate to incorporate the current irradiance data into the network state, and uses the output gate to extract feature information related to the shading period, outputting the irradiance attenuation magnitude of the site during the cloud shading period. Further, the irradiance attenuation magnitude characterizes the change in irradiance intensity at a single site before and after cloud shading, determined by the difference between the site's normal irradiance intensity before shading and the lowest irradiance intensity during shading. Based on the irradiance change patterns in historical shading events, the LSM network predicts the irradiance attenuation magnitude of each site under the current cloud shading condition.

[0040] Preferably, the irradiance attenuation amplitudes of each site are weighted and summed according to the temporal sequence of the site shading sequence to form a decreasing curve of aggregate output. The specific process of the weighted summation is as follows: determine the weight coefficient of each site in the aggregate output based on its installed capacity, multiply the irradiance attenuation amplitude of each site by the corresponding weight coefficient, and then sequentially superimpose the weighted attenuation amplitudes of each site on the time axis according to the shading transmission time interval to form a decreasing curve of aggregate output that changes with time. The horizontal axis of this decreasing curve is time, and the vertical axis is the change in aggregate output.

[0041] Understandably, the identification of the output decline segment is achieved by detecting the slope change of the aggregate output decline curve. When the curve slope changes from a positive value to a negative value, it is marked as the start time of the decline segment. When the curve slope returns from a negative value to a positive value or zero value, it is marked as the end time of the decline segment.

[0042] In one embodiment, the valley depth is calculated within the identified power output decline segment. The aggregated power output values ​​at each time point within the decline segment are traversed, and the minimum point is searched for. The power output value corresponding to this minimum point is recorded as the valley depth. The power output value corresponding to the start time of the decline segment is recorded as the starting power output value. The valley depth is obtained by subtracting the starting power output value from the valley depth. Through this process, based on the temporal fitting of the site occlusion sequence and the Long Short-Term Memory network, the predicted duration and valley depth of the aggregated power output decline segment can be accurately extracted, providing a quantitative basis for the virtual power plant's pricing decisions.

[0043] Step S104: Analyze whether the preset pullback trigger threshold is reached by analyzing the predicted duration of the pullback decline segment and the valley depth, thus generating an activation signal for pullback price adjustment.

[0044] Specifically, when analyzing whether a preset pullback trigger threshold has been reached by combining the predicted duration and valley depth of the aggregate output decline segment to generate an activation signal for pullback price adjustment, the following steps are taken: The decline duration and valley depth of the aggregate output decline segment are acquired. The decline duration is compared with a duration threshold in the preset pullback trigger threshold, and the valley depth is compared with a depth threshold in the preset pullback trigger threshold. If the predicted duration exceeds the duration threshold and the valley depth exceeds the depth threshold, it is determined that the preset pullback trigger threshold has been reached, resulting in a trigger determination result. Based on the trigger determination result, if the trigger determination result indicates that the preset pullback trigger threshold has been reached, an activation signal for pullback price adjustment is output.

[0045] In one embodiment, the preset pullback trigger threshold consists of two parts: a duration threshold and a depth threshold. The duration threshold represents the upper limit of the duration of aggregated output decline that the virtual power plant can withstand in electricity market transactions, and the depth threshold represents the maximum allowable decline in aggregated output relative to the normal level. The predicted duration of the aggregated output decline segment is compared with the duration threshold, and the valley depth is compared with the depth threshold. When the predicted duration exceeds the duration threshold and the valley depth exceeds the depth threshold, it is determined that the current decline in aggregated output has reached the preset pullback trigger threshold.

[0046] Specifically, the reach determination result serves as the trigger condition for subsequent price adjustment. When the reach determination result indicates that the preset withdrawal trigger threshold has been reached, the scheduling platform outputs an activation signal for withdrawing the price adjustment. The activation signal is used to notify the virtual power plant's price execution unit to initiate the price withdrawal process.

[0047] It should be noted that the output of the activation signal signifies that the virtual power plant has switched from the regular bidding state to the bid withdrawal preparation state, thus reserving an adjustment window before cloud cover causes a decrease in aggregate output.

[0048] Step S105: Update the bid retraction trigger threshold based on the activation signal and actual irradiation measurement data, evaluate whether the updated trigger threshold exceeds the preset slope threshold, and determine the early locking time for bid retraction.

[0049] The activation signal for the pullback price adjustment and the measured irradiance data of each site are obtained. Based on the measured irradiance data corresponding to the activation signal trigger time, the rate of change of the current irradiance intensity is calculated. The rate of change is the ratio of the difference in irradiance intensity between adjacent sampling times to the sampling time interval. The duration threshold and depth threshold in the pullback price trigger threshold are proportionally scaled based on the product of the rate of change and a preset rate correction coefficient to obtain the updated pullback price trigger threshold. The threshold change is calculated based on the updated and previous pullback price trigger thresholds. The ratio of the threshold change to the update time interval is used as the threshold change slope. The threshold change slope is compared with a preset slope threshold. If the threshold change slope exceeds the preset slope threshold, a slope breach determination indicator is obtained. Based on the slope breach determination indicator, if the determination indicator indicates that the preset slope threshold has been breached, the current moment is determined as the early locking opportunity for price pullback.

[0050] In one embodiment, when the scheduling platform receives an activation signal for withdrawing the price adjustment, it synchronously acquires the measured irradiance data of each site within several sampling periods before and after the activation signal trigger time. The measured irradiance data includes the solar irradiance intensity value of each site and its corresponding sampling timestamp.

[0051] Specifically, the rate of change of irradiance is calculated based on the difference in irradiance between adjacent sampling times.

[0052] For example, the irradiance value at the current sampling time and the irradiance value at the previous sampling time are read, and the difference between the two is calculated as the change in irradiance. The change in irradiance is then divided by the time interval between the two sampling times to obtain the rate of change in irradiance. The rate correction coefficient is a pre-set scaling factor used to convert the rate of change in irradiance into a threshold adjustment range. When the rate of change in irradiance is large, the product of the rate correction coefficient and the rate of change increases accordingly, thereby exerting a greater adjustment effect on the pullback bid trigger threshold. Further, the scaling process of the threshold update adjusts the duration threshold and depth threshold in the pullback bid trigger threshold by using the product of the rate correction coefficient and the rate of change as a scaling factor. When the rate of decrease in irradiance accelerates, the scaling factor shortens the duration threshold and lowers the depth threshold, thereby obtaining a more sensitive updated pullback bid trigger threshold.

[0053] It should be noted that the threshold change slope reflects the degree of drastic change in the pullback trigger threshold during continuous updates. The threshold change is obtained by calculating the difference between the updated and the previous pullback trigger thresholds. The threshold change slope is obtained by dividing the threshold change slope by the update time interval. The threshold change slope is then compared with a preset slope threshold. If the threshold change slope exceeds the preset slope threshold, a slope breach determination indicator is output.

[0054] Understandably, when the slope breakout indicator shows that the preset slope threshold has been exceeded, the current moment is determined as the early locking opportunity for price retracement, and the virtual power plant initiates the price retracement locking operation at this time.

[0055] Step S106: Optimize the timing of locking in price retracements and the fitting pattern of the declining aggregate output segment to determine the set of locking points for market retracement prices.

[0056] The process involves obtaining the early locking opportunity for price retracement and the downward curve of aggregated output. Based on the curve trend before and after the early locking opportunity, a fitting shape of the downward segment is extracted. This fitting shape consists of three feature points: the starting point, the trough point, and the recovery point. The extraction process uses a sliding window algorithm to detect turning points: the window size is 10 time points, and the slope change threshold is 0.05. If the slope changes from positive to negative, it is the starting point; from negative to zero, it is the trough point; and from zero to positive, it is the recovery point. For example, the starting point coordinates are (t1, p1), the trough point is (t2, p2), and the recovery point is (t3, p3). The time coordinates and output coordinates of these three feature points are recorded to form the morphological characteristics of the downward segment. The morphological characteristics of the downward segment are then time-aligned with the early locking opportunity to calculate the early locking opportunity t.lock The time offset Δt = t from the starting point t1 of the descent segment lock -t1, if the time offset is positive and exceeds the preset offset threshold of 5 minutes, the locking timing will be adjusted forward by Δt; if the time offset is negative and exceeds the preset offset threshold of 5 minutes, the locking timing will be adjusted backward by |Δt|, resulting in an optimized locking timing. Based on the optimized locking timing and the morphological characteristics of the downward segment, the time points corresponding to the optimized locking timing are combined with the starting point, trough point, and recovery point of the downward segment to determine the set of locking timing points for market retracement quotations.

[0057] In one embodiment, the scheduling platform obtains the early locking timing of the bid pullback and the decline curve of the aggregated output. The decline curve records the complete trend of the aggregated output changing over time during the cloud cover period, including all data points of the output decline phase and the output recovery phase.

[0058] Specifically, the extraction of the fitted morphology is achieved by identifying three feature points on the descent curve. The starting point is the moment when the output in the descent curve begins to decrease, corresponding to the turning point where the aggregated output begins to decay from its normal level. The trough point is the moment when the output in the descent curve reaches its lowest value, corresponding to the aggregated output level when the cloud shading effect is strongest. The recovery point is the moment when the output in the descent curve recovers to near normal levels, corresponding to the point where the aggregated output rebounds after the cloud leaves the site area. The time coordinates and output coordinates of the starting point, trough point, and recovery point are recorded respectively to form the morphological features of the descent segment. Furthermore, the time offset is calculated by subtracting the time value of the starting point of the descent segment from the time value of the advance locking timing.

[0059] For example, if the calculation result is positive, it indicates that the early locking timing is after the starting point. If the positive value exceeds a preset offset threshold, the early locking timing will be adjusted forward to a suitable position before the starting point. If the calculation result is negative, it indicates that the early locking timing is before the starting point. If the absolute value of the negative value exceeds a preset offset threshold, the early locking timing will be adjusted backward to reduce the lead time, thereby obtaining an optimized locking timing.

[0060] It should be noted that the set of locking points is composed of the optimized locking timing and the three characteristic points of the downward segment. The optimized locking timing is marked as the retracement start point, the starting point is marked as the output decline point, the trough point is marked as the output minimum point, and the recovery point is marked as the output recovery point. The four points are arranged in chronological order to form the set of locking points for market retracement quotes.

[0061] Step S107: Generate real-time transaction execution instructions by combining the locked time set of market pullback quotes with the cloud cluster's trajectory, and confirm the optimized output of the virtual power plant's electricity market transaction.

[0062] The system acquires a set of locked-in time points for market pullback quotes and the cloud cluster's trajectory. Based on the time information corresponding to each time point in the locked-in time point set and the estimated arrival time of the cloud cluster at each station in the cloud cluster's trajectory, it generates real-time transaction execution instructions corresponding to each locked-in time point. These instructions include the quote adjustment time and amount. Based on the quote adjustment amount and time in the real-time transaction execution instructions, the ratio of the difference in quote adjustment amount to the time interval is calculated for each adjacent execution instruction to obtain the quote change rate. The maximum value is taken as the overall quote change rate. This overall quote change rate is compared with a preset ultra-short-term quote stability threshold. If the overall quote change rate is lower than the threshold, the smoothness of aggregated output is determined to meet the ultra-short-term quote stability requirements because the quote change rate reflects the fluctuation amplitude of aggregated output. A smoothness satisfaction determination result is obtained. Based on the smoothness satisfaction determination result, if the determination result indicates that the smoothness of aggregated output meets the ultra-short-term quote stability requirements, the real-time transaction execution instruction is marked as a valid instruction and output to the dispatching platform to identify the optimized output of virtual power plant electricity market transactions.

[0063] In one embodiment, the scheduling platform obtains a set of locked time points for market pullback quotes and cloud trajectory prediction data. The set of locked time points includes four key moments: the pullback start time, the power output decrease time, the power output minimum time, and the power output recovery time. The cloud trajectory prediction data includes the estimated time when the cloud will arrive at each station in the future.

[0064] Specifically, the process of forming real-time transaction execution instructions is based on matching the time information of each time point in the locked time point set with the expected time of cloud arrival at each site.

[0065] For example, the pullback initiation point is compared with the expected time when the cloud cluster will reach the starting site in the site occlusion sequence to determine the corresponding price adjustment time. For the output decrease point, it is compared with the time when the cloud cluster covers half of the sites in the sequence; the lowest output point is compared with the time when the cloud cluster covers all the sites in the sequence; and the output recovery point is compared with the time when the cloud cluster leaves the ending site in the sequence. For each point in the locked point set, the price adjustment amount is determined according to the output change magnitude corresponding to that point on the aggregate output decrease curve. The aggregate output decrease curve is obtained by fitting historical cloud occlusion data using the least squares method. The input is a time series and output measurement values, and the output is a curve function, for example, the magnitude is calculated as P. t =P max -P min , where P tLet P be the change in output at time t. max For a peak output of 100kW, P min The minimum output is 20kW. The price adjustment time and amount are combined to form the real-time transaction execution instruction corresponding to that point in time. This operation is performed on all points in the locked time set to form a complete real-time transaction execution instruction sequence. Further, the price change rate is calculated by performing a differential operation on adjacent execution instructions. The price adjustment amounts of two adjacent execution instructions are read, and the difference between them is calculated as the price adjustment amount difference. The price adjustment times of two adjacent execution instructions are read, and the difference between them is calculated as the time interval. The price adjustment amount difference is divided by the time interval to obtain the price change rate. The price change rate is compared with a preset ultra-short-term price stability threshold of 0.5 yuan per kWh per minute. If the price change rate is lower than 0.5, the aggregated output smoothness is determined to meet the ultra-short-term price stability requirements.

[0066] It should be noted that when the smoothness determination result indicates that the aggregated output smoothness meets the requirements for ultra-short-term price stability, the dispatching platform will mark the real-time transaction execution instruction sequence as a valid instruction and output the valid instruction to the power market transaction execution unit, thus completing the optimized output identification of virtual power plant power market transactions.

[0067] like Figure 2 This invention provides a virtual power plant electricity market trading agency device, mainly comprising: The cloud outline and trajectory recognition module is used to obtain cloud outline information and movement trajectory from satellite cloud image data; The morphology judgment module is used to calculate the projection distribution status of each site based on the cloud's trajectory and the set of geographical coordinates of the distributed photovoltaic sites, and to determine the arrangement of the site layout. The feature prediction module is used to predict the characteristics of decreased aggregate output based on the arrangement of the site layout and the trajectory of the cloud cluster, combined with actual irradiation data. The signal generation module is used to generate a trading adjustment signal based on the comparison result between the aggregate output decline characteristics and a preset threshold; The timing determination module is used to update the trigger threshold based on the transaction adjustment signal and the measured irradiation data to determine the locking timing of the transaction adjustment; The set optimization module is used to optimize the set of transaction timings by combining the locking timing with the aggregated output decline characteristics; The instruction generation module is used to generate transaction execution instructions based on the set of transaction times and the trajectory of the cloud cluster, and output the optimized results of the virtual power plant electricity market transaction.

[0068] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

Claims

1. A virtual power plant electricity market transaction agency method, characterized in that, include: Obtaining cloud cluster outline information and cloud cluster trajectory from satellite cloud image data specifically includes: acquiring cloud cluster outline image data from a satellite cloud image receiving terminal via a data channel; performing grayscale gradient calculation on the cloud cluster outline image data to extract grayscale transition regions between the cloud cluster and the background, and marking the current outline boundary point set of the cloud cluster; calculating the cloud cluster centroid position based on the cloud cluster outline image sequence within consecutive time frames, determining the cloud cluster displacement direction and distance through the cloud cluster centroid position offset, and fitting the cloud cluster trajectory. The projected distribution of each site is calculated based on the cloud's trajectory and the geographical coordinates of the distributed photovoltaic sites. The arrangement of the sites is then determined. Specifically, this includes: extracting the direction vector of the cloud's trajectory; calculating the vertical distance from each site to the trajectory based on the direction vector and the geographical coordinates of each site to obtain the projected distance; determining whether the projected distance falls within a preset range and calculating the proportion of sites falling within the range; if the proportion of sites falling within the range exceeds a preset threshold, the projected distance is determined to be in a concentrated distribution state; and based on the concentrated distribution state and the arrangement order of the sites along the trajectory direction, the projected position of each site along the trajectory direction is calculated to determine the arrangement of the site layout. Based on the arrangement of the site layout and the trajectory of the cloud cluster, combined with measured irradiance data, the characteristics of aggregate output decline are predicted. Specifically, this includes: if the sites are arranged in a strip, establishing a site shading sequence based on the direction of the cloud cluster's trajectory; calculating the shading transmission time interval between adjacent sites based on the site shading sequence and the cloud cluster's movement speed, and accumulating them to obtain the aggregate output decline duration; obtaining the historical irradiance intensity sequence of each site, and predicting the irradiance attenuation amplitude when each site is shaded using a time-series fitting model; weighting and summing the irradiance attenuation amplitude according to the installed capacity weight of each site to obtain the aggregate output decline curve, and extracting the valley depth of the aggregate output decline curve. Based on the comparison result between the aggregated output decline characteristics and the preset threshold, a trading adjustment signal is generated, specifically including: obtaining the aggregated output decline duration and valley depth; comparing the aggregated output decline duration with a preset duration threshold, and comparing the valley depth with a preset depth threshold; if the aggregated output decline duration exceeds the duration threshold and the valley depth exceeds the depth threshold, it is determined that the preset threshold has been reached, and a reaching determination result is obtained; based on the reaching determination result, a trading adjustment signal is output. The timing of the transaction adjustment is determined by updating the trigger threshold based on the transaction adjustment signal and the measured irradiance data. Specifically, this includes: calculating the rate of change of irradiance intensity based on the measured irradiance data at the time the transaction adjustment signal is triggered; proportionally adjusting the trigger threshold based on the rate of change of irradiance intensity and a preset correction coefficient to obtain the updated threshold; calculating the slope of the threshold before and after the update, and comparing the slope with a preset slope threshold; if the slope of the threshold before and after the update exceeds the preset threshold, it is determined that the slope limit has been exceeded, and the current moment is determined as the timing of the transaction adjustment. Optimizing the trading timing set by combining the locking timing with the aggregated output decline characteristics specifically includes: analyzing the fitting shape of the decline curve based on the aggregated output decline characteristics, wherein the fitting shape consists of the decline start point, the trough point, and the recovery point; detecting the turning point of the decline curve using a sliding window algorithm, and recording the characteristic point time and output coordinates; calculating the time offset between the locking timing and the decline start point, adjusting the locking timing according to the time offset to obtain the optimized locking timing; and combining the optimized locking timing with the characteristic points of the decline curve to form a trading timing set. Based on the set of transaction times and the trajectory of the cloud cluster, a transaction execution instruction is generated, and the optimized results of the virtual power plant electricity market transaction are output.

2. The virtual power plant electricity market transaction agency method as described in claim 1, characterized in that, After generating the transaction execution instruction based on the set of transaction times and the trajectory of the cloud cluster, the process includes: The price change rate is obtained by calculating the ratio of the price adjustment difference between adjacent executed instructions to the time interval.

3. The virtual power plant electricity market transaction agency method as described in claim 2, characterized in that, After obtaining the price change rate, the process includes: If the rate of change in the quoted price is lower than the preset stable threshold, then the smoothness of the polymerization output is determined to meet the requirements. Based on the smoothness determination result, the transaction execution instruction is marked as a valid instruction and output to the scheduling platform.

4. A virtual power plant electricity market trading agency device, characterized in that, The device includes: The cloud contour and trajectory recognition module is used to obtain cloud contour information and cloud trajectory from satellite cloud image data. Specifically, it includes: acquiring cloud contour image data from a satellite cloud image receiving terminal through a data channel; performing grayscale gradient calculation on the cloud contour image data to extract grayscale transition regions between the cloud and the background, and marking the current contour boundary point set of the cloud; calculating the cloud centroid position based on the cloud contour image sequence within a continuous time frame, determining the cloud displacement direction and distance through the cloud centroid position offset, and fitting the cloud trajectory. The morphology determination module is used to calculate the projection distribution state of each site based on the cloud's trajectory and the set of geographical coordinates of the distributed photovoltaic sites, and to determine the arrangement of the site layout. Specifically, it includes: extracting the direction vector of the cloud's trajectory; calculating the vertical distance from each site to the trajectory based on the direction vector and the geographical coordinates of each site to obtain the projection distance; determining whether the projection distance falls within a preset interval and counting the proportion of sites falling within the interval; if the proportion of sites falling within the interval exceeds a preset threshold, it is determined that the projection distance is in a concentrated distribution state; and calculating the projection position of each site in the trajectory direction based on the concentrated distribution state and the arrangement order of the sites along the trajectory direction to determine the arrangement of the site layout. The feature prediction module is used to predict the aggregation output decline characteristics based on the arrangement of the site layout and the cloud trajectory, combined with measured irradiance data. Specifically, it includes: if the sites are arranged in a strip, establishing a site shading sequence based on the cloud trajectory direction; calculating the shading transmission time interval between adjacent sites based on the site shading sequence and the cloud movement speed, and accumulating them to obtain the aggregation output decline duration; obtaining the historical irradiance intensity sequence of each site, and predicting the irradiance attenuation magnitude of each site during shading using a time-series fitting model; weighting and summing the irradiance attenuation magnitude according to the installed capacity weight of each site to obtain the aggregation output decline curve, and extracting the valley depth of the aggregation output decline curve. The signal generation module is used to generate a trading adjustment signal based on the comparison result of the aggregate output decline characteristics and a preset threshold. Specifically, it includes: obtaining the aggregate output decline duration and valley depth; comparing the aggregate output decline duration with a preset duration threshold and comparing the valley depth with a preset depth threshold; if the aggregate output decline duration exceeds the duration threshold and the valley depth exceeds the depth threshold, it is determined that the preset threshold has been reached, and a reaching determination result is obtained; based on the reaching determination result, a trading adjustment signal is output. The timing determination module is used to update the trigger threshold based on the transaction adjustment signal and the measured irradiance data, and determine the locking timing of the transaction adjustment. Specifically, it includes: calculating the rate of change of irradiance intensity based on the measured irradiance data at the time the transaction adjustment signal is triggered; proportionally adjusting the trigger threshold based on the rate of change of irradiance intensity and a preset correction coefficient to obtain the updated threshold; calculating the slope of the threshold before and after the update, and comparing the slope with a preset slope threshold; if the slope of the threshold before and after the update exceeds the preset threshold, it is determined that the slope limit has been exceeded, and the current moment is determined as the locking timing of the transaction adjustment. The set optimization module is used to optimize the set of trading points based on the locking timing and the aggregated output decline characteristics. Specifically, it includes: analyzing the fitting shape of the decline curve based on the aggregated output decline characteristics, wherein the fitting shape consists of the decline start point, the trough point, and the recovery point; detecting the turning point of the decline curve using a sliding window algorithm, and recording the feature point time and output coordinates; calculating the time offset between the locking timing and the decline start point, adjusting the locking timing according to the time offset to obtain the optimized locking timing; and aggregating the optimized locking timing with the feature points of the decline curve to form a set of trading points. The instruction generation module is used to generate transaction execution instructions based on the set of transaction times and the trajectory of the cloud cluster, and output the optimized results of the virtual power plant electricity market transaction.

Images