A method and system for coordinated operation of a medium and long term wind and light field station

Abstract

The application belongs to the technical field of new energy power dispatching, and relates to a medium and long term wind and light field station coordinated operation method and system. The method comprises the following steps: first, acquiring regional medium and long term meteorological time series data, extracting time series characteristics through LSTM and performing amplitude verification to obtain first wind and light complementary characteristics; then, obtaining second wind and light complementary characteristics through random forest screening and purification. The characteristics are averaged and clustered to divide typical output forms, and congestion and vacancy two types of risk identification are generated by comparing upper and lower threshold values of the channel. The initial power plan is translated and optimized according to the risk identification to obtain a first power plan; the amount of abandoned electricity is calculated, and if it exceeds the standard, a second power plan is obtained through energy storage peak shaving and valley filling. The cross-month reference sequence of the plan is extracted and the fluctuation variance is calculated, and after the variance meets the standard, the medium and long term dispatching instruction is output. The application effectively adapts to the medium and long term power balance demand of the high proportion of new energy system, and improves the overall efficiency of the wind and light field station coordinated operation.

Description

Technical Field

[0001] This invention belongs to the field of new energy power dispatching technology, and in particular relates to a method and system for coordinated operation of medium- and long-term wind and solar power plants. Background Technology

[0002] With the continuous expansion of new energy installed capacity, intermittent new energy sources such as wind power and photovoltaics are gradually becoming the main power sources for the power system. Wind and solar resources naturally have strong randomness, temporal fluctuations, and reverse complementarity characteristics, resulting in significant differences between peak and valley output curves and frequent shifts in output patterns across months. This poses great challenges to the long-term power balance of the power grid, the utilization rate of transmission channels, and the coordinated operation of generating units.

[0003] Currently, existing medium- and long-term wind and solar power dispatching strategies mostly rely on historical average output or single meteorological forecast data to formulate power plans. These strategies can only achieve macro-level power balance constraints and are insufficient to deeply explore the complementary fluctuation patterns of wind and solar power over time. Traditional technical solutions generally suffer from the following technical defects: First, conventional meteorological time-series analysis methods have weak capabilities in extracting the characteristics of wind-solar inverse coupling fluctuations and cannot effectively distinguish between effective complementary fluctuations and random noise fluctuations, resulting in low accuracy in power output pattern identification. Second, existing clustering methods are mostly based on direct classification of raw power output data, which is easily affected by meteorological baseline deviations and abnormal jump points, causing distortion in the characterization of typical power output patterns and failing to adapt to dynamic scene changes at medium- and long-term scales. Third, traditional power plan optimization methods mostly focus on short-term intraday peak shaving and valley filling, lacking risk prediction and time-series shift control mechanisms for cross-monthly cycles, and are unable to take into account multiple constraints such as channel congestion, channel vacancy, and wind and solar curtailment.

[0004] In summary, existing medium- and long-term wind and solar power plant operation and control technologies have shortcomings such as insufficient feature mining, inaccurate power output pattern identification, weak multi-constraint collaborative optimization capabilities, and poor cross-cycle operation stability, making it difficult to meet the medium- and long-term, high-precision, and high-stability wind and solar collaborative dispatch requirements under a high proportion of new energy power systems. Summary of the Invention

[0005] Based on this, it is necessary to provide a method and system for the coordinated operation of medium- and long-term wind and solar power plants that can deeply explore the fluctuation patterns of medium- and long-term wind-solar hybrid power generation, accurately identify typical combined power output patterns, take into account grid channel constraints and absorption constraints, and achieve two-level optimization of power generation plans and cross-monthly stability verification.

[0006] Firstly, this application provides a method for coordinated operation of medium- and long-term wind and solar power stations, including:

[0007] The historical meteorological evolution sequence and historical meteorological mean sequence of the target area are obtained at a long-term time scale. The historical meteorological evolution sequence and historical meteorological mean sequence are input into a long short-term memory network for time series feature extraction. After fluctuation amplitude verification, the first wind-solar complementary feature sequence is output.

[0008] The first wind-solar complementary feature sequence is input into the random forest algorithm model to extract the wind-solar complementary fluctuation pattern features at medium and long scales, and the second wind-solar complementary feature sequence is output.

[0009] The mean clustering algorithm was used to cluster the second wind-solar complementary feature sequence to determine the typical joint power output pattern.

[0010] If the peak output of a typical combined output configuration exceeds the threshold of the transmission channel capacity, a channel congestion risk is identified, and a first risk flag is generated.

[0011] Obtain a preset channel vacancy threshold. If the output valley value of a typical combined output mode is less than the channel vacancy threshold, it is determined that there is a channel vacancy risk, and a second risk indicator is generated.

[0012] Based on the first risk identifier and the second risk identifier, the pre-constructed initial medium- and long-term electricity plan is time-shifted to obtain the first electricity plan.

[0013] Calculate the expected wind and solar power curtailment corresponding to the first power plan. If the expected wind and solar power curtailment is greater than the curtailment threshold, then the first power plan is processed for peak shaving and valley filling to obtain the second power plan.

[0014] Extract the cross-monthly operating benchmark sequence corresponding to the second power supply plan, and calculate the fluctuation variance of the cross-monthly operating benchmark sequence; if the fluctuation variance is less than the variance threshold, then the second power supply plan is determined as a medium- and long-term dispatch instruction.

[0015] In one embodiment, the historical meteorological evolution sequence and the historical meteorological mean sequence are input into a long short-term memory network for temporal feature extraction. After fluctuation amplitude verification, a first wind-solar complementary feature sequence is output, including:

[0016] The evolution gradient sequence was calculated based on the historical meteorological evolution sequence and the historical meteorological mean sequence:

[0017]

[0018] In the formula, For the first The gradient value of the evolution over a period of time; The first in the historical meteorological evolution sequence Meteorological element values ​​for a given time period; The first in the historical meteorological mean series Historical statistical averages of meteorological elements corresponding to the time period; This refers to the sequence number of the time period.

[0019] The historical meteorological evolution sequence includes measured evolution subsequences of wind speed and measured evolution subsequences of sunlight intensity, divided into preset time periods.

[0020] The historical meteorological mean series includes a wind speed historical mean subsequence and a light intensity historical mean subsequence, each corresponding to a preset time period.

[0021] Temporal correlation analysis was performed on the wind speed gradient subsequence and the illumination gradient subsequence in the evolution gradient sequence to extract the inverse fluctuation correlation characteristics of the two meteorological elements and obtain the wind-solar complementary fluctuation characteristics.

[0022] The wind-solar hybrid fluctuation features are input into a long short-term memory network in chronological order. The network is then used for temporal dependency modeling and feature fusion encoding to obtain the first wind-solar hybrid feature sequence.

[0023] The fluctuation amplitude characteristic value of the first wind-solar hybrid characteristic sequence is calculated according to the following formula:

[0024]

[0025] In the formula, This represents the characteristic value of the fluctuation amplitude. The first wind-solar complementary feature sequence is the first The characteristic values ​​of the time period; This represents the arithmetic mean of the first wind-solar complementary characteristic sequence over the entire time period;

[0026] The fluctuation amplitude feature value is compared with the fluctuation threshold. If the fluctuation amplitude feature value is greater than the fluctuation threshold, the first wind-solar complementary feature sequence is determined to be a valid complementary feature sequence, and the valid first wind-solar complementary feature sequence is output.

[0027] In one embodiment, the first wind-solar complementary feature sequence is input into a random forest algorithm model to extract the wind-solar complementary fluctuation pattern features at medium to long-term scales, and the second wind-solar complementary feature sequence is output, including:

[0028] The first wind-solar complementary feature sequence is divided into multiple time series slices by a sliding window with a preset window length and sliding step size.

[0029] Each time series slice is input into the random forest algorithm model. The importance of each feature dimension is calculated by the reduction of Gini impurity of each decision tree in the random forest algorithm. The average of the calculation results of all decision trees is taken, and the feature weight value corresponding to each feature dimension is output.

[0030] No. The feature weights of the dimensional features satisfy:

[0031]

[0032] In the formula, This represents the total number of decision trees in the random forest algorithm model. For the first The first decision tree The reduction in Gini impurity corresponding to the dimensional feature; For the first The feature weight values ​​of the dimensional features.

[0033] Based on the feature weight values, each feature dimension is filtered, and feature dimensions with feature weight values ​​higher than the weight threshold are retained and determined as valid feature dimensions, thus obtaining the wind-solar complementary fluctuation pattern features.

[0034] The characteristics of wind-solar complementary fluctuation patterns are arranged in chronological order to construct a complementary fluctuation matrix.

[0035] The complementary wave matrix satisfies:

[0036]

[0037] In the formula, The matrix consists of complementary wave patterns. This represents the total number of time periods; The number of effective feature dimensions; For the first Within the time period The values ​​of the effective wind-solar complementary fluctuation pattern characteristics.

[0038] The complementary wave matrix is ​​weighted and fused along the feature dimension to generate a second wind-solar complementary feature sequence.

[0039] No. The values ​​of the second wind-solar complementary feature sequence during the time period satisfy:

[0040]

[0041] In the formula, For the first The values ​​of the second wind-solar complementary feature sequence for the time period.

[0042] Calculate the fluctuation complementarity index of the second wind-solar complementary feature sequence, and compare the fluctuation complementarity index with the complementarity threshold.

[0043] If the fluctuation complementarity index is greater than the complementarity threshold, then the second wind-solar complementary feature sequence is output.

[0044] The volatility complementarity index satisfies:

[0045]

[0046] In the formula, It is a volatility complementarity index; The time-series component obtained by fusing wind speed-type fluctuation features in the complementary fluctuation matrix; The temporal component is obtained by fusing illumination-related wave characteristics; For covariance calculation; , These are the standard deviations of the wind speed and light intensity fluctuation component sequences, respectively.

[0047] In one embodiment, the mean clustering algorithm is used to cluster the second wind-solar complementary feature sequence to determine typical joint power output patterns, including:

[0048] Based on the preset number of clusters, select the corresponding number of initial cluster centers in the feature space of the second wind-solar complementary feature sequence.

[0049] Calculate the Euclidean distance between each temporal feature vector and each initial cluster center in the second wind-solar complementary feature sequence.

[0050] Based on the principle of minimum distance attribution, each temporal feature vector is assigned to the morphological cluster to which the corresponding initial cluster center belongs, based on the Euclidean distance metric, resulting in multiple initial morphological clusters.

[0051] Calculate the mean of all temporal feature vectors within each initial morphological cluster, and update the cluster center of the corresponding morphological cluster using the mean of the feature vectors to obtain the current cluster center.

[0052] Determine whether the current cluster center meets the preset convergence condition; if not, replace the previous cluster center with the current cluster center and update the current cluster center.

[0053] If the current cluster center meets the preset convergence condition, a unique morphological identifier is assigned to each converged morphological cluster, and the typical joint output morphology of the corresponding category is determined based on the feature vector of the converged cluster center.

[0054] Typical joint output patterns include morphological identifier encoding, cluster center feature vector, and joint output statistical feature parameters.

[0055] In one embodiment, based on a first risk identifier and a second risk identifier, a pre-constructed initial medium-to-long-term power consumption plan is time-shifted to obtain a first power consumption plan, including:

[0056] Based on the first risk identifier and the second risk identifier, identify the peak period of channel congestion and the trough period of channel vacancy under the corresponding typical joint output mode, and generate a joint output risk fluctuation sequence with time period attribute tags.

[0057] Based on the earliest start time and latest end time of all risk periods in the joint output risk fluctuation sequence, a target adjustment time window covering all risk periods is determined.

[0058] According to the target, the time range of the time window is adjusted, and the power output plan data for the corresponding period is extracted from the pre-constructed initial medium and long-term power plan to obtain the local power plan.

[0059] The initial medium- and long-term power generation plan is a baseline output plan pre-prepared based on the installed capacity of the power plants, medium- and long-term power generation indicators, and the rated capacity of the transmission channels.

[0060] Using the output values ​​of each time period of the local power plan as the fitting target, a time-series feature vector for the corresponding time period is constructed. The time-series feature vector is then input into a random forest regression model to fit and obtain the continuous output trend curve within the target adjustment window.

[0061] Based on the output trend curve, the extreme points of output are identified, and the peak and valley characteristics of output are extracted.

[0062] Output peak-valley characteristics include peak time Valley moment Peak output With the lowest output value .

[0063] Based on the peak and valley characteristics of power output and the preset transmission channel capacity threshold With channel vacancy threshold Calculate the timing translation step size.

[0064] The timing shift step size satisfies:

[0065]

[0066] in, The average rate of change of output between peak and valley periods, satisfying ; This represents the offset step size of the output curve along the time axis.

[0067] According to the time-series shift step, the output sequence of the initial medium- and long-term power plan within the target adjustment time window is shifted and adjusted to shift the output peak value towards the output trough value period, thus obtaining the first power plan.

[0068] In one embodiment, the expected wind and solar power curtailment corresponding to the first power plan is calculated. If the expected wind and solar power curtailment exceeds the curtailment threshold, peak shaving and valley filling are applied to the first power plan to obtain a second power plan, including:

[0069] Obtain the wind and solar power generation forecast sequence and grid load curve that match the time scale and time range of the first power generation plan.

[0070] Based on the wind and solar power generation forecast sequence and the grid load curve, the amount of wind and solar power output exceeding the transmission and absorption capacity is calculated for each time period, and the summation is used to obtain the expected amount of wind and solar power curtailment.

[0071] The expected amount of wind and solar power curtailment will be met:

[0072]

[0073] In the formula, This refers to the expected amount of wind and solar power curtailed. For the first Predicted output of wind and solar power generation during specific time periods; This is the preset outgoing channel capacity threshold; For the first Power absorption capacity corresponding to the grid load during a given time period; For the first electricity plan Planned output during the specified time period; The duration of a single time period; This represents the total number of time periods within the scheduling cycle.

[0074] If the expected amount of wind and solar power curtailment exceeds the curtailment threshold, then obtain the energy storage available capacity constraint sequence for the corresponding time scale.

[0075] The available capacity constraint sequence for energy storage includes the maximum charging power, maximum discharging power, and upper and lower limits of state of charge for each time period.

[0076] Using the output sequence of the first power plan, the wind and solar power generation prediction sequence, and the energy storage available capacity constraint sequence as input features, the random forest optimization model is used to fit the mapping relationship between the charging and discharging strategy and the curtailment suppression target, and the energy storage charging and discharging adjustment commands for each time period are output.

[0077] The output value of the first power plan is superimposed and corrected according to the energy storage charging and discharging adjustment command for the corresponding time period to obtain the second power plan.

[0078] The second power plan is satisfied:

[0079]

[0080] In the formula, For the second electricity plan Planned output during the specified time period; For the first The energy storage charging and discharging adjustment command for a given time period, with positive values ​​representing energy storage discharging and negative values ​​representing energy storage charging.

[0081] In one embodiment, the cross-monthly operating baseline sequence corresponding to the second electricity plan is extracted, and the variance of the cross-monthly operating baseline sequence is calculated, including:

[0082] The average output of the second power plan is statistically analyzed at the monthly time granularity. The benchmark output characteristic values ​​of each month are extracted and arranged in time sequence to obtain the cross-monthly operation benchmark sequence.

[0083] The monthly benchmark sequence is input into the Isolation Forest algorithm model. The Isolation Forest algorithm identifies anomalous jump data points in the sequence. The identified anomalous jump data points are corrected by neighborhood mean interpolation, and a smooth benchmark sequence is output.

[0084] Extract time-series interpolation features from the smoothed baseline sequence; the time-series interpolation features include the output amplitude of each baseline node, the time-series position, and the rate of change of output of adjacent nodes.

[0085] Piecewise linear fitting is performed based on time-series interpolation characteristics to construct a continuous cross-monthly baseline output curve.

[0086] Obtain the preset peak-shaving margin value of the power grid. Using the peak-shaving margin value as the normalization benchmark, perform normalization calculation on the time-by-time output deviation of the benchmark output curve to calculate the fluctuation variance.

[0087] The variance of the fluctuation satisfies:

[0088]

[0089] In the formula, The variance is the fluctuation. The first reference output curve Output value over a given period; The average output value over the entire cycle of the baseline output curve; This represents the peak-shaving margin of the power grid. This represents the total number of time periods in the scheduling cycle.

[0090] Secondly, this application also provides a medium- and long-term wind and solar power station coordinated operation system, the system comprising:

[0091] The wind and solar feature extraction module is used to obtain the historical meteorological evolution sequence and historical meteorological mean sequence of the target area at a long-term time scale. The historical meteorological evolution sequence and historical meteorological mean sequence are input into a long short-term memory network for time series feature extraction, and the effective first wind and solar complementary feature sequence is output. It is also used to input the first wind and solar complementary feature sequence into a random forest algorithm model to extract the wind and solar complementary fluctuation pattern features at a medium- and long-term time scale, and output the second wind and solar complementary feature sequence.

[0092] The typical morphology recognition module is used to cluster the second wind-solar complementary feature sequence using the mean clustering algorithm to determine the typical combined power output morphology; it is also used to obtain a preset transmission channel capacity threshold. If the peak output of the typical combined power output morphology is greater than the transmission channel capacity threshold, it is determined that there is a channel congestion risk and a first risk identifier is generated; it is also used to obtain a preset channel vacancy threshold. If the valley output of the typical combined power output morphology is less than the channel vacancy threshold, it is determined that there is a channel vacancy risk and a second risk identifier is generated.

[0093] The power plan optimization module is used to perform time-series shifting processing on the pre-constructed initial medium- and long-term power plan based on the first risk identifier and the second risk identifier to obtain the first power plan; it is also used to calculate the expected wind and solar power curtailment corresponding to the first power plan. If the expected wind and solar power curtailment is greater than the curtailment threshold, the first power plan is subjected to peak shaving and valley filling processing to obtain the second power plan.

[0094] The benchmark verification module is used to extract the cross-monthly benchmark sequence corresponding to the second power plan and calculate the fluctuation variance of the cross-monthly benchmark sequence; if the fluctuation variance is less than the variance threshold, the second power plan is determined as a medium- and long-term dispatch instruction.

[0095] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0096] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned method.

[0097] The aforementioned method, system, computer equipment, and storage medium for the coordinated operation of medium- and long-term wind and solar power stations acquire historical meteorological evolution sequences and historical meteorological mean sequences at medium- and long-term scales in the target area. A long short-term memory network is used to perform deep feature extraction on the two sets of time-series data, and invalid noise features are filtered out by combining fluctuation amplitude verification, outputting a valid first wind-solar complementary feature sequence. The first wind-solar complementary feature sequence is input into a random forest algorithm model, and the fluctuation pattern features of wind-solar complementary power are filtered and purified through feature weights to obtain a second wind-solar complementary feature sequence. A mean-based clustering algorithm is used to perform clustering convergence calculation on the second wind-solar complementary feature sequence, accurately classifying different operating conditions and determining the typical joint power output patterns of the wind and solar power stations. By preset transmission channel capacity thresholds and channel vacancy thresholds, threshold judgments are made for the peak and trough values ​​of the typical joint power output patterns, respectively. When the peak value exceeds the limit, a first risk indicator representing the channel congestion risk is generated; when the trough value exceeds the limit, a second risk indicator representing the channel vacancy risk is generated. Based on two types of risk indicators, the initial medium- and long-term power generation plan is optimized by time-series shifting, adjusting the power output time-series distribution to adapt to channel constraints, resulting in the first power generation plan. The expected curtailment of wind and solar power in the first power generation plan is calculated by combining the wind and solar power generation forecast sequence and the grid load curve. When the curtailment exceeds the curtailment threshold, energy storage resources are used to perform peak shaving and valley filling corrections on the first power generation plan, resulting in a second power generation plan that balances channel constraints and grid absorption constraints. The monthly operating baseline sequence of the second power generation plan is extracted, and the fluctuation variance is calculated. When the fluctuation variance is less than a preset variance threshold, the second power generation plan is determined as the final medium- and long-term dispatch instruction.

[0098] This invention addresses the problems of inaccurate feature representation, large deviations in output pattern identification, lack of channel risk prediction, weak adaptability of power plans, and insufficient long-term operational stability in traditional medium- and long-term wind and solar power dispatching methods. This invention effectively mines the temporal complementarity of wind and solar power through dual-layer intelligent feature extraction, improving the effectiveness of medium- and long-term output features; it accurately identifies typical output patterns through clustering, providing a reliable basis for grid risk assessment; it guides the temporal shift of power plans based on dual risk identification, effectively avoiding channel congestion and vacancy issues and improving channel utilization efficiency; it reduces renewable energy curtailment losses and adapts to grid absorption capacity through power curtailment verification and energy storage peak shaving and valley filling optimization; and it ensures long-term stable and reliable dispatching plans through cross-monthly fluctuation stability verification, effectively meeting the medium- and long-term power balance needs of high-proportion renewable energy systems and improving the overall efficiency of coordinated operation of wind and solar power plants. Attached Figure Description

[0099] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0100] Figure 1 A flowchart of a medium- and long-term wind and solar power station coordinated operation method provided in an embodiment of the present invention;

[0101] Figure 2 This is a structural block diagram of a medium- and long-term wind and solar power station coordinated operation system provided in an embodiment of the present invention. Detailed Implementation

[0102] 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.

[0103] In one embodiment, such as Figure 1 As shown, this application provides a method for coordinated operation of medium- and long-term wind and solar power stations, which may include the following steps:

[0104] Step S101: Obtain the historical meteorological evolution sequence and historical meteorological mean sequence of the target area at the long-term time scale. Input the historical meteorological evolution sequence and historical meteorological mean sequence into the long short-term memory network for time series feature extraction. After fluctuation amplitude verification, output the first wind-solar complementary feature sequence.

[0105] Specifically, by collecting temporal evolution data of wind speed and illumination within the corresponding scheduling cycle in the target area, as well as historical statistical average data for the corresponding time period, two sets of meteorological sequence samples with temporal correspondence are constructed. A Long Short-Term Memory (LSTM) network is used to model and encode the temporal correlation and dynamic evolution patterns of the two sets of sequences, mining the temporal fluctuation characteristics and complementary correlation characteristics between wind speed and illumination. Simultaneously, a fluctuation amplitude quantification verification method is used to eliminate stationary invalid features and random noise features, retaining only effective temporal features with significant fluctuation characteristics, ultimately forming a standardized and highly effective first wind-solar complementary feature sequence.

[0106] Step S102: Input the first wind-solar complementary feature sequence into the random forest algorithm model, extract the wind-solar complementary fluctuation pattern features at medium and long scales, and output the second wind-solar complementary feature sequence.

[0107] Based on the feature importance calculation capability of the random forest algorithm, a weight analysis is performed on the multi-dimensional temporal features of the first wind-solar complementary feature sequence to quantify the contribution of each feature dimension to the fluctuation of wind-solar joint output. By using weight thresholds, core fluctuation features with high correlation are retained, while redundant and low-contribution invalid feature dimensions are eliminated, accurately extracting the wind-solar complementary fluctuation patterns suitable for medium- and long-term scheduling scenarios. The selected effective features are then subjected to temporal reconstruction and feature fusion to generate a second wind-solar complementary feature sequence that is dimensionally simplified, more regular, and can truly reflect the medium- and long-term linkage fluctuation characteristics of wind and solar power.

[0108] Step S103: The mean clustering algorithm is used to cluster the second wind-solar complementary feature sequence to determine the typical joint power output pattern.

[0109] Based on the purified wind-solar hybrid features, a refined classification of operating conditions is achieved. The temporal feature vector of the second wind-solar hybrid feature sequence is used as the clustering sample. Cluster centers are initialized with a preset number of clusters, and sample assignment is completed according to the feature space distance criterion. The cluster centers are continuously corrected through iterative updates until convergence. After cluster convergence, multiple stable morphological clusters are formed, each corresponding to a fixed wind-solar fluctuation coupling mode. The operating conditions corresponding to each category are defined by combining the cluster center features and intra-cluster statistical parameters. Finally, multiple typical joint output modes covering medium- and long-term operating scenarios are obtained, achieving standardized and scenario-based classification of wind and solar power station operating conditions.

[0110] Step S104: Obtain the preset transmission channel capacity threshold. If the peak output of the typical combined output mode is greater than the transmission channel capacity threshold, it is determined that there is a risk of channel congestion, and a first risk identifier is generated.

[0111] Based on the combined output statistical parameters corresponding to each typical combined output mode, the maximum output peak value under each type of operating condition is extracted, and the output peak value is compared with the rated capacity threshold of the power grid transmission channel. When the ultimate output of a typical mode exceeds the transmission carrying capacity of the channel, it is determined that there is an operational risk of transmission channel congestion and power exceeding the limit for this type of operating condition, and a unique first risk identifier is attached to this operating condition to realize the structured marking of channel congestion risk.

[0112] Step S105: Obtain the preset channel vacancy threshold. If the output valley value of the typical combined output mode is less than the channel vacancy threshold, it is determined that there is a channel vacancy risk, and a second risk identifier is generated.

[0113] By extracting the minimum output trough value of each typical combined power output mode, the output trough value is compared with a preset channel vacancy judgment threshold. When the overall wind and solar combined power output is low and the channel transmission capacity cannot be effectively utilized, it is determined that there is an operational risk of channel vacancy and insufficient resource utilization in the current operating condition, and a corresponding second risk label is generated. The dual risk label can fully cover two types of adverse operating scenarios: extreme peak values ​​and extreme trough values ​​of wind and solar power output.

[0114] Step S106: Based on the first risk identifier and the second risk identifier, perform time-series shift processing on the pre-constructed initial medium- and long-term power plan to obtain the first power plan.

[0115] Based on the risk time intervals corresponding to the typical power output patterns identified by the two types of risk indicators, the target time window that needs adjustment is determined. A reasonable time-series shift step size is calculated based on the peak-valley variation characteristics of the power output curve within the window. The power output time-series distribution of the initial medium- and long-term power plan is then shifted and adjusted overall, changing the occurrence times of the original peaks and valleys, avoiding extreme conditions of channel congestion and channel idling. This optimizes the power output time-series structure without changing the overall medium- and long-term power volume, resulting in a first power plan adapted to grid channel constraints.

[0116] Step S107: Calculate the expected wind and solar power curtailment corresponding to the first power plan. If the expected wind and solar power curtailment is greater than the curtailment threshold, then perform peak shaving and valley filling on the first power plan to obtain the second power plan.

[0117] The power plan is optimized and corrected a second time based on grid absorption constraints. Combining real-time regional wind and solar power generation forecast data with grid load absorption data, the redundant power exceeding the grid absorption capacity and transmission capacity of the first power plan is calculated for each time period, and the expected wind and solar power curtailment within the cycle is statistically obtained. When the cumulative curtailment exceeds the preset control threshold, adjustable energy storage capacity resources are retrieved, and charging and discharging adjustment instructions for each time period are obtained through random forest model optimization. Peak values ​​are reduced and valley values ​​are filled in for the planned output, smoothing out local output redundancy and gaps, effectively reducing the level of new energy curtailment, and obtaining a second power plan that takes into account both transmission capacity constraints and grid absorption constraints.

[0118] Step S108: Extract the cross-monthly operating benchmark sequence corresponding to the second power plan, and calculate the fluctuation variance of the cross-monthly operating benchmark sequence; if the fluctuation variance is less than the variance threshold, then the second power plan is determined as a medium- and long-term dispatch instruction.

[0119] Based on the full-cycle time-series data of the second power generation plan, benchmark output characteristics are statistically analyzed at the monthly granularity, and a cross-monthly operational benchmark sequence is constructed. Abrupt changes in the sequence are removed through anomaly correction processing to form a stable benchmark output curve. The variance of the sequence fluctuation is quantitatively calculated using the grid peak-shaving margin to objectively evaluate the stability of the medium- and long-term power generation plan. When the variance of the fluctuation meets the preset stability threshold, the current power generation plan is deemed to have smooth time-series fluctuations and good cross-cycle operational stability, meeting the requirements for medium- and long-term dispatch operations, and is thus determined as the final executable medium- and long-term dispatch instruction.

[0120] The aforementioned method for coordinated operation of medium- and long-term wind and solar power plants first acquires historical meteorological evolution sequences and historical meteorological mean sequences for the target area at medium- and long-term scales. It then uses a Long Short-Term Memory (LSTM) network to extract temporal deep features and filters out invalid noise features using fluctuation amplitude verification, outputting a valid first wind-solar complementary feature sequence. This first feature sequence is input into a random forest algorithm model, where feature weights are used to refine the fluctuation patterns of wind-solar complementary power generation, resulting in a second wind-solar complementary feature sequence. Mean clustering convergence calculation is then performed on the second feature sequence to accurately classify differentiated operating conditions and determine typical joint power output patterns of wind and solar power plants. Using preset transmission channel capacity thresholds and channel vacancy thresholds, the peak and valley output values ​​of typical patterns are judged to exceed limits, generating a first risk indicator for channel congestion and a second risk indicator for channel vacancy. Based on these two risk indicators, the initial medium- and long-term power generation plan is time-shifted and optimized to adapt to grid channel constraints, resulting in a first power generation plan. By combining wind and solar power generation forecast sequences with grid load curves, the expected curtailment of wind and solar power in the first power generation plan is calculated. When the curtailment exceeds the curtailment threshold, peak shaving and valley filling corrections are performed using energy storage resources, resulting in a second power generation plan that balances channel constraints and grid absorption constraints. The monthly operating baseline sequence of the second power generation plan is extracted, and the fluctuation variance is calculated. When the fluctuation variance meets the threshold requirement, the second power generation plan is determined as the final medium- and long-term dispatch instruction. This invention solves the problems of inaccurate characterization of medium- and long-term wind and solar power dispatch features, large deviations in power output pattern identification, lack of channel risk prediction, poor adaptability of power generation plans, and insufficient long-term operational stability in traditional wind and solar power dispatching. This invention employs dual-layer intelligent feature extraction to accurately uncover the temporal complementarity of wind and solar power, enhancing the effectiveness of feature representation. It leverages clustering to accurately identify typical power output patterns, providing reliable support for risk assessment. Based on dual risk identification, it optimizes the temporal distribution of power plans, effectively avoiding channel congestion and vacancy issues and improving channel utilization. Through power curtailment verification and energy storage optimization, it reduces renewable energy curtailment and adapts to grid absorption capacity. Combined with cross-monthly fluctuation verification, it ensures the long-term stable operation of the dispatch plan, effectively adapting to the medium- and long-term power balance needs of high-proportion renewable energy systems and improving the overall efficiency of wind and solar power plant collaborative operation.

[0121] In one embodiment, the historical meteorological evolution sequence and the historical meteorological mean sequence are input into a long short-term memory network for temporal feature extraction. After fluctuation amplitude verification, the first wind-solar complementary feature sequence is output, which may include the following steps:

[0122] Step S201: Calculate the evolution gradient sequence based on the historical meteorological evolution sequence and the historical meteorological mean sequence.

[0123]

[0124] In the formula, For the first The gradient value of the evolution over a period of time; The first in the historical meteorological evolution sequence Meteorological element values ​​for a given time period; The first in the historical meteorological mean series Historical statistical averages of meteorological elements corresponding to the time period; This refers to the sequence number of the time period.

[0125] The historical meteorological evolution sequence includes measured evolution subsequences of wind speed and measured evolution subsequences of sunlight intensity, divided according to preset time periods.

[0126] Step S202: The historical meteorological mean sequence includes a wind speed historical mean subsequence and a light intensity historical mean subsequence that correspond one-to-one with the preset time period.

[0127] Step S203: Perform time-series correlation analysis on the wind speed gradient subsequence and the illumination gradient subsequence in the evolution gradient sequence, extract the inverse fluctuation correlation characteristics of the two types of meteorological elements, and obtain the wind-solar complementary fluctuation characteristics.

[0128] Step S204: Input the wind-solar hybrid fluctuation features into the long short-term memory network in temporal order, and perform temporal dependency modeling and feature fusion encoding through the long short-term memory network to obtain the first wind-solar hybrid feature sequence.

[0129] Step S205: Calculate the fluctuation amplitude characteristic value of the first wind-solar hybrid characteristic sequence according to the following formula:

[0130]

[0131] In the formula, This represents the characteristic value of the fluctuation amplitude. The first wind-solar complementary feature sequence is the first The characteristic values ​​of the time period; This represents the arithmetic mean of the first wind-solar complementary characteristic sequence over the entire time period;

[0132] Step S206: Compare the fluctuation amplitude feature value with the fluctuation threshold. If the fluctuation amplitude feature value is greater than the fluctuation threshold, the first wind-solar complementary feature sequence is determined to be a valid complementary feature sequence, and the valid first wind-solar complementary feature sequence is output.

[0133] Specifically, a medium- to long-term meteorological time-series dataset is first constructed. The historical meteorological evolution sequence refers to the time-series meteorological data obtained from actual observations of the target area, including measured evolution subsequences of wind speed and light intensity, used to characterize the real-time dynamic changes in wind and solar resources. The historical meteorological mean sequence refers to the meteorological baseline mean data for the corresponding period, statistically derived from multi-year historical meteorological data, including historical mean subsequences of wind speed and light intensity, used to provide a meteorological baseline reference. The time-series dimensions of the two types of sequences are strictly aligned to ensure that the data for each time period can be calculated accordingly. Based on the above two types of sequences, an evolution gradient sequence is calculated for each time period. The evolution gradient represents the relative deviation of the measured meteorological value for a single time period from the historical mean, used to eliminate the annual meteorological baseline deviation and truly reflect the dynamic fluctuation range of wind and solar resources. After obtaining the wind speed gradient subsequence and the illumination gradient subsequence, a time-series correlation analysis is performed on the two sets of gradient subsequences. This analysis compares the fluctuation trends of the two meteorological elements over time intervals to identify inverse patterns such as increasing wind speed and decreasing illumination, or vice versa, extracting wind-solar complementary fluctuation features that characterize resource complementarity. The continuously time-series wind-solar complementary fluctuation features are input into a Long Short-Term Memory (LSTM) network. The LTM network is used to mine long-period dependencies in the time-series data, performing time-series modeling and feature fusion encoding on the discrete complementary fluctuation features to generate a first wind-solar complementary feature sequence with uniform dimensions and continuous time sequence. To eliminate invalid stationary data and random observation noise, fluctuation amplitude feature values ​​are calculated for the first wind-solar complementary feature sequence. These fluctuation amplitude feature values ​​quantify the overall volatility of the entire time-series sequence. By comparing the fluctuation amplitude feature values ​​with a preset fluctuation threshold, effective sequences with significant fluctuation characteristics and practical complementary value are selected, ultimately outputting the effective first wind-solar complementary feature sequence.

[0134] This embodiment eliminates data bias caused by differences in meteorological baselines across different seasons and months by introducing historical mean benchmark data and gradient calculation methods, thus solving problems such as inaccurate feature representation and baseline drift caused by directly using raw meteorological data. Through time-series correlation analysis, it clearly extracts the inherent complementary mechanism of wind-solar inverse fluctuations, avoiding the shortcomings of traditional methods that merely stack meteorological data and fail to uncover inherent coupling patterns. Leveraging the time-series memory characteristics of Long Short-Term Memory (LSTM) networks, it effectively captures the slow fluctuation evolution patterns at medium- to long-term scales, compensating for the limitation of traditional statistical methods in modeling long-term time-series dependencies. By adding a fluctuation amplitude verification mechanism, it automatically filters out ineffective time-series segments that are stable and without fluctuations or lack complementary adjustment space, as well as random noise anomalies, achieving precise feature purification and effective screening.

[0135] In one embodiment, inputting the first wind-solar complementary feature sequence into a random forest algorithm model to extract the wind-solar complementary fluctuation pattern features at medium to long-term scales and outputting the second wind-solar complementary feature sequence may include the following steps:

[0136] Step S301: The first wind-solar complementary feature sequence is divided into sliding window segments according to a preset window length and sliding step size to generate multiple sets of time series slices.

[0137] Step S302: Input each group of time series slices into the random forest algorithm model, calculate the importance of each feature dimension by the reduction of Gini impurity of each decision tree in the random forest algorithm, take the average of the calculation results of all decision trees, and output the feature weight value corresponding to each feature dimension.

[0138] No. The feature weights of the dimensional features satisfy:

[0139]

[0140] In the formula, This represents the total number of decision trees in the random forest algorithm model. For the first The first decision tree The reduction in Gini impurity corresponding to the dimensional feature; For the first The feature weight values ​​of the dimensional features.

[0141] Step S303: Filter each feature dimension according to the feature weight value, retain the feature dimensions whose feature weight value is higher than the weight threshold and determine them as valid feature dimensions, and obtain the wind-solar complementary fluctuation pattern feature.

[0142] Step S304: Arrange the characteristics of wind-solar complementary fluctuation patterns in chronological order to construct a complementary fluctuation matrix.

[0143] The complementary wave matrix satisfies:

[0144]

[0145] In the formula, The matrix consists of complementary wave patterns. This represents the total number of time periods; The number of effective feature dimensions; For the first Within the time period The values ​​of the effective wind-solar complementary fluctuation pattern characteristics.

[0146] Step S305: Perform weighted fusion processing on the complementary wave matrix along the feature dimension to generate the second wind-solar complementary feature sequence.

[0147] No. The values ​​of the second wind-solar complementary feature sequence during the time period satisfy:

[0148]

[0149] In the formula, For the first The values ​​of the second wind-solar complementary feature sequence for the time period.

[0150] Step S306: Calculate the fluctuation complementarity index of the second wind-solar complementary feature sequence, and compare the fluctuation complementarity index with the complementarity threshold.

[0151] Step S307: If the fluctuation complementarity index is greater than the complementarity threshold, then output the second wind-solar complementary feature sequence.

[0152] The volatility complementarity index satisfies:

[0153]

[0154] In the formula, It is a volatility complementarity index; The time-series component obtained by fusing wind speed-type fluctuation features in the complementary fluctuation matrix; The temporal component is obtained by fusing illumination-related wave characteristics; For covariance calculation; , These are the standard deviations of the wind speed and light intensity fluctuation component sequences, respectively.

[0155] Specifically, the preprocessed first wind-solar complementary feature sequence undergoes refined feature selection and fusion optimization. First, a sliding window segmentation method is used to process the time series sequence. The preset window length is the time series interval span of a single feature extraction, and the preset sliding step size is the interval between each time series movement of the window. Multiple sets of equal-length time series slices are generated by sliding the window sequentially, achieving segmented and refined analysis of long-term features. All time series slices are input into a trained random forest algorithm model. The random forest is composed of multiple independent decision trees. By calculating the reduction in Gini impurity during the splitting process of each decision tree node, the contribution of each feature dimension to distinguishing wind and solar power output fluctuations is quantified. Gini impurity is an indicator used to measure the degree of disorder in sample data; the greater the reduction in Gini impurity after feature splitting, the stronger the feature's ability to distinguish differences in operating conditions. The arithmetic mean of the reduction in Gini impurity for the same dimension of features corresponding to all decision trees is calculated to obtain the feature weight value corresponding to each feature dimension. The feature weight value is used to objectively represent the effective contribution ratio of a single-dimensional feature.

[0156] Furthermore, all feature dimensions are filtered based on preset weight thresholds. Dimensions with feature weight values ​​higher than the threshold are retained as valid feature dimensions, while redundant feature dimensions with low contribution and strong interference are removed. This process purifies the features to obtain wind-solar complementary fluctuation patterns that truly reflect medium- and long-term variation patterns. The filtered valid features are arranged in chronological order to construct a structured complementary fluctuation matrix. This matrix uses the total scheduling period as the row dimension and the number of valid feature dimensions as the column dimension, storing the valid feature values ​​for each period. The complementary fluctuation matrix is ​​then subjected to weighted fusion of feature dimensions. Based on the sum of the products of the feature weight values ​​of each dimension and the feature values ​​of the corresponding period, a second wind-solar complementary feature sequence with unified dimensions and condensed information is generated for each period. To ensure the effectiveness of the complementary output features, a fluctuation complementarity index is introduced for final verification. This index is calculated based on covariance and standard deviation. Covariance measures the temporal correlation trend between wind speed and illumination components, while standard deviation characterizes the degree of fluctuation dispersion of the two components. The fluctuation complementarity index is obtained by calculating the negative correlation between wind speed and illumination, two independent fluctuation components. It is then compared with a preset complementarity threshold. Only when the fluctuation complementarity index is greater than the threshold is the current sequence determined to have significant wind-solar reverse complementarity characteristics, and the final effective second wind-solar complementary feature sequence is output.

[0157] This embodiment utilizes a Gini impurity weighting evaluation mechanism based on random forests to achieve objective quantitative screening of wind and solar time-series features. It eliminates the subjectivity of traditional manual feature selection, accurately removing low-contribution, irregular, redundant, and noisy features while retaining highly identifiable core fluctuation features, thus improving the accuracy and effectiveness of wind and solar feature data. By using a sliding window to segmentally parse long-term time-series sequences, it addresses the problem of large spans and difficulty in uncovering local patterns in medium- and long-term time-series data, taking into account both global time-series trends and local fluctuation details. Through constructing a complementary fluctuation matrix and weighted fusion, it achieves information aggregation and dimensionality reduction of multi-dimensional features, generating a fused feature sequence with stronger time-series consistency and more concentrated patterns. Simultaneously, it introduces a fluctuation complementarity verification mechanism based on negative correlation characteristics to quantitatively verify the reverse complementarity of wind and solar resources from a mathematical perspective, ensuring that the output second wind and solar complementary feature sequence has clear physical meaning and engineering value.

[0158] In one embodiment, the second wind-solar complementary feature sequence is clustered using the mean clustering algorithm to determine the typical combined power output pattern, which may include the following steps:

[0159] Step S401: Select the corresponding number of initial cluster centers in the feature space of the second wind-solar complementary feature sequence according to the preset number of clusters.

[0160] Step S402: Calculate the Euclidean distance between each temporal feature vector in the second wind-solar complementary feature sequence and each initial cluster center.

[0161] Step S403: Based on the principle of minimum distance attribution, each temporal feature vector is assigned to the morphological cluster to which the corresponding initial cluster center belongs based on the Euclidean distance metric, resulting in multiple initial morphological clusters.

[0162] Step S404: Calculate the feature mean of all temporal feature vectors in each initial morphological cluster, and update the cluster center of the corresponding morphological cluster with the feature mean to obtain the current cluster center.

[0163] Step S405: Determine whether the current cluster center meets the preset convergence condition; if not, replace the previous cluster center with the current cluster center and update the current cluster center.

[0164] Step S406: If the current cluster center meets the preset convergence condition, assign a unique morphological identifier to each converged morphological cluster, and determine the typical joint output morphology of the corresponding category based on the feature vector of the converged cluster center.

[0165] The typical joint output pattern includes morphological identifier encoding, cluster center feature vector, and joint output statistical feature parameters.

[0166] Specifically, firstly, based on a preset number of clusters, an equal number of initial cluster centers are selected in the high-dimensional feature space corresponding to the second wind-solar complementary feature sequence. These initial cluster centers serve as the initial feature reference points for each category of working conditions and are used for subsequent sample assignment determination. Then, for each time-series feature vector within the feature sequence, the Euclidean distance between it and each initial cluster center is calculated. The Euclidean distance quantifies the feature similarity between the time-series feature vector and the cluster center; the smaller the distance, the higher the feature similarity. Following the principle of minimum distance assignment, each time-series feature vector is assigned to the morphological cluster corresponding to the nearest cluster center, completing the initial classification of all samples and obtaining multiple initial morphological clusters. After each clustering iteration, the overall feature mean of all time-series feature vectors within a single morphological cluster is calculated. The cluster center of the corresponding morphological cluster is updated using this cluster mean, resulting in a new round of current cluster centers. The algorithm continuously checks whether the current cluster centers meet the preset convergence conditions. Convergence conditions are generally defined as the cluster center offset being less than a set threshold or reaching the maximum number of iterations. If the convergence conditions are not met, the algorithm iteratively executes distance calculation, sample classification, and cluster center update processes until the clustering results stabilize. Once the cluster centers meet the convergence conditions, the sample distribution of each morphological cluster no longer changes significantly from the cluster center, indicating that various wind and solar power fluctuation conditions have been stably distinguished. A unique morphological identifier code is assigned to each converged morphological cluster, and the feature vector of the finally converged cluster center is used as the baseline feature for that type of operation. Combined with the output parameters obtained from intra-cluster statistics, a typical joint output pattern is constructed, including the morphological identifier code, the cluster center feature vector, and joint output statistical feature parameters, thus completing the standardized classification of medium- and long-term wind and solar power operation conditions.

[0167] This embodiment achieves automatic classification of wind-solar hybrid characteristics through mean clustering, using objective feature distance metrics as the basis for distinguishing operating conditions. This avoids the problems of traditional operating condition classification relying on manual experience and inconsistent classification standards. By iteratively updating the cluster centers and introducing a convergence judgment mechanism, the stability and convergence accuracy of operating condition clustering are effectively improved, ensuring that the various typical joint output patterns obtained from the classification can truly cover the medium- and long-term fluctuation patterns of wind and solar resources in the target area. Through the structured definition of unified morphological coding, benchmark features, and statistical parameters, standardized and digital representation of different wind-solar operating conditions is achieved. This step can effectively distinguish the differences in output characteristics under different wind-solar hybrid coupling states, solving the problems of ambiguous output pattern distinction, mixed operating conditions, and inaccurate identification of extreme operating conditions in traditional methods. This improves the pertinence and reliability of subsequent grid channel risk identification and power plan optimization adjustments.

[0168] In one embodiment, the pre-constructed initial medium-to-long-term power plan is time-shifted based on the first risk identifier and the second risk identifier to obtain the first power plan, which may include the following steps:

[0169] Step S501: Based on the first risk identifier and the second risk identifier, identify the peak period of channel congestion and the trough period of channel vacancy under the corresponding typical joint output mode, and generate a joint output risk fluctuation sequence with time period attribute tags.

[0170] Step S502: Determine the target adjustment time window covering all risk periods based on the earliest start time and latest end time of all risk periods in the joint output risk fluctuation sequence.

[0171] Step S503: Adjust the time window according to the target, extract the power output plan data for the corresponding period from the pre-constructed initial medium- and long-term power plan, and obtain the local power plan.

[0172] The initial medium- and long-term power plan is a baseline output plan pre-prepared based on the installed capacity of the power plants, medium- and long-term power indicators, and the rated capacity of the transmission channels.

[0173] Step S504: Using the output value of each time period of the local power plan as the fitting target, construct the time series feature vector of the corresponding time period, input the time series feature vector into the random forest regression model, and fit to obtain the continuous output trend curve within the target adjustment window.

[0174] Step S505: Identify the extreme points of output based on the output trend curve and extract the peak and valley features of output.

[0175] Among them, the peak and valley characteristics of output include the peak time. Valley moment Peak output With the lowest output value .

[0176] Step S506: Based on the peak and valley characteristics of the output and the preset output channel capacity threshold... With channel vacancy threshold Calculate the timing translation step size.

[0177] The timing shift step size satisfies:

[0178]

[0179] in, The average rate of change of output between peak and valley periods, satisfying ; This represents the offset step size of the output curve along the time axis.

[0180] Step S507: According to the time-series shift step, the output sequence of the initial medium- and long-term power plan within the target adjustment time window is shifted to adjust the output peak value towards the output valley value period to obtain the first power plan.

[0181] Specifically, based on the generated first and second risk identifiers, the power supply plan timing optimization is carried out. By matching the typical joint output patterns corresponding to the two types of risk identifiers, the peak congestion periods and valley idle periods corresponding to each pattern are accurately located. All risk periods are integrated to generate a joint output risk fluctuation sequence with clear time period attributes. Based on the time boundaries of all risk periods in this sequence, the earliest start time and latest end time are extracted, and a target adjustment time window that can completely cover all congestion and idle risk intervals is defined, thus determining the optimization adjustment range of the power supply plan. The initial medium- and long-term power supply plan adopted in this scheme is a benchmark output plan pre-compiled by combining the installed capacity of wind and solar power plants, medium- and long-term power supply assessment indicators, and the rated capacity of transmission channels. Data is extracted from this initial plan within the preset time window to obtain a partial power supply plan that only includes risk periods. Using the time-by-time output value of the partial power supply plan as the fitting target, a corresponding time-series feature vector is constructed. The time-series feature vector is input into a random forest regression model to fit and generate a continuous and smooth output trend curve within the window, eliminating the discontinuity problem of discrete plan data. Based on the continuous output trend curve, precise output extreme points are identified, and complete output peak-valley features, including peak time, valley time, peak output value, and valley output value, are extracted. Combining the extracted peak-valley features with preset transmission channel capacity thresholds and channel idle thresholds, the average output change rate between peak and valley periods is first calculated. Then, a time-series shift step size adapted to the current output fluctuation characteristics is obtained through a quantification formula. Finally, using this shift step size as the adjustment benchmark, the output sequence of the initial medium- and long-term power plan is shifted as a whole within the target time window, shifting the output peak value to the output valley period, thus completing the output time-series structure optimization and obtaining the first power plan that avoids the risk of extreme channel operation.

[0182] This embodiment relies on risk identification to accurately pinpoint abnormal operating periods of power grid channels, optimizing only those sections with congestion or vacancy risks without ineffective adjustments to the full-cycle power plan, effectively reducing optimization computation and improving scheduling efficiency. By fitting continuous output trend curves using a random forest regression model, it solves the problems of inaccurate extreme value identification and unclear fluctuation patterns in discrete power plan data, ensuring the accuracy of peak-valley feature extraction. A quantitative shift step size calculation method that matches the rate of output change replaces the coarse adjustment mode of fixed-duration shifts, adapting to different wind and solar power output fluctuation conditions and ensuring the rationality and adaptability of time-series shift adjustments. Through time-series offset optimization from peak to valley periods, it effectively mitigates short-term overload congestion and long-term vacancy waste in transmission channels, optimizing the output time-series distribution and improving the overall utilization rate of transmission channels without changing the total medium- and long-term power volume.

[0183] In one embodiment, the expected wind and solar power curtailment corresponding to the first power plan is calculated. If the expected wind and solar power curtailment is greater than the curtailment threshold, the first power plan is subjected to peak shaving and valley filling to obtain the second power plan. This may include the following steps:

[0184] Step S601: Obtain the wind and solar power generation prediction sequence and grid load curve that match the time scale and time range of the first power generation plan.

[0185] Step S602: Based on the wind and solar power generation forecast sequence and the grid load curve, calculate the amount of wind and solar power output that exceeds the transmission and absorption capacity for each time period, and sum them up to obtain the expected amount of wind and solar power curtailment.

[0186] The expected amount of wind and solar power curtailment will be met:

[0187]

[0188] In the formula, This refers to the expected amount of wind and solar power curtailed. For the first Predicted output of wind and solar power generation during specific time periods; This is the preset outgoing channel capacity threshold; For the first Power absorption capacity corresponding to the grid load during a given time period; For the first electricity plan Planned output during the specified time period; The duration of a single time period; This represents the total number of time periods within the scheduling cycle.

[0189] Step S603: If the expected amount of wind and solar power curtailment is greater than the curtailment threshold, then obtain the energy storage available capacity constraint sequence for the corresponding time scale.

[0190] The available energy storage capacity constraint sequence includes the maximum charging power, maximum discharging power, and upper and lower limits of state of charge for each time period.

[0191] Step S604: Using the output sequence of the first power plan, the wind and solar power generation prediction sequence, and the energy storage available capacity constraint sequence as input features, input the random forest optimization model to fit the mapping relationship between the charging and discharging strategy and the curtailment suppression target, and output the energy storage charging and discharging adjustment instructions for each time period.

[0192] Step S605: Based on the energy storage charging and discharging adjustment command, the output value of the corresponding time period of the first power plan is superimposed and corrected to obtain the second power plan.

[0193] The second power plan is satisfied:

[0194]

[0195] In the formula, For the second electricity plan Planned output during the specified time period; For the first The energy storage charging and discharging adjustment command for a given time period, with positive values ​​representing energy storage discharging and negative values ​​representing energy storage charging.

[0196] Specifically, the first power plan obtained through time-shift optimization undergoes absorption constraint verification and secondary optimization of energy storage. First, a wind and solar power generation forecast sequence and grid load curve perfectly matching the time scale and range of the first power plan are obtained to characterize the theoretical power generation capacity of new energy sources and the actual absorption capacity of the grid within the scheduling cycle. The predicted output of wind and solar power, the transmission channel capacity threshold, and the grid load absorption power are compared time-by-time. The minimum value between the upper limit of channel transmission and the upper limit of load absorption is taken as the system's acceptable power. The redundant power exceeding the system's acceptable power by the predicted wind and solar power output is calculated, and the expected curtailment of wind and solar power over the entire cycle is accumulated time-by-time to quantify the degree of waste of new energy resources corresponding to the first power plan. A fixed curtailment threshold is set as the criterion for qualified absorption. When the expected curtailment of wind and solar power exceeds the curtailment threshold, it is determined that the current power plan has a significant problem of limited new energy absorption, requiring energy storage adjustment optimization. Furthermore, a synchronous time-scale energy storage available capacity constraint sequence is obtained. This sequence includes the maximum charging power, maximum discharging power, and upper and lower limits of the state of charge of energy storage devices in each time period, used to limit the adjustable boundaries of energy storage and ensure that the adjustment process complies with equipment operating constraints. The first planned power output sequence, the predicted wind and solar power generation sequence, and the energy storage available capacity constraint sequence are used as input features and fed into a random forest optimization model. The model fits the mapping relationship between energy storage charging and discharging behavior and curtailment suppression targets at each time period. The model optimizes and outputs energy storage charging and discharging adjustment commands that meet the constraints. Finally, the first planned power output is superimposed and corrected time-by-time based on the energy storage charging and discharging adjustment commands. Positive commands correspond to energy storage discharging to supplement power output, and negative commands correspond to energy storage charging to reduce power output. This achieves peak shaving and valley filling adjustment of the power plan, generating a second power plan that simultaneously satisfies channel constraints, absorption constraints, and energy storage operation constraints.

[0197] This embodiment calculates the expected curtailment of wind and solar power through multi-dimensional boundary constraints, objectively and accurately quantifying the deficiencies in renewable energy consumption in medium- and long-term power plans, avoiding inaccurate consumption assessments caused by single-constraint judgments. By relying on the full-time constraint sequence of energy storage to participate in model optimization, it can strictly ensure that the charging and discharging strategy conforms to the actual operating capacity of energy storage equipment, avoiding unreasonable adjustment behaviors such as over-power and over-charge states. Using a random forest model to fit the nonlinear mapping relationship between multiple input variables and the curtailment suppression target, compared to traditional fixed-strategy adjustment methods, it can adapt to different wind and solar fluctuations and load changes, improving the accuracy and adaptability of energy storage adjustment strategies. By correcting and optimizing the temporal output distribution of the power plan through peak shaving and valley filling, it effectively reduces redundant power during periods of renewable energy surplus and fills the power shortage during periods of output gap, significantly reducing the amount of wind and solar power curtailed during medium- and long-term scheduling cycles and improving the level of renewable energy consumption.

[0198] In one embodiment, extracting the cross-monthly operating baseline sequence corresponding to the second electricity plan and calculating the variance of the cross-monthly operating baseline sequence may include the following steps:

[0199] Step S701: Perform output average statistics on the second power plan at the monthly time granularity, extract the benchmark output characteristic values ​​for each month and arrange them in time sequence to obtain the cross-monthly operation benchmark sequence.

[0200] Step S702: Input the cross-monthly benchmark sequence into the isolated forest algorithm model, identify anomalous jump data points in the sequence through the isolated forest algorithm, correct the identified anomalous jump data points by neighborhood mean interpolation, and output a smooth benchmark sequence.

[0201] Step S703: Extract time-series interpolation features from the smoothed reference sequence; the time-series interpolation features include the output amplitude, time-series position, and output change rate of each reference node and adjacent nodes.

[0202] Step S704: Based on the time-series interpolation features, perform piecewise linear fitting to construct a continuous cross-monthly benchmark output curve.

[0203] Step S705: Obtain the preset power grid peak-shaving margin value, use the peak-shaving margin value as the normalization benchmark, perform normalization calculation on the time-by-time output deviation of the benchmark output curve, and calculate the fluctuation variance.

[0204] The variance of the fluctuation satisfies:

[0205]

[0206] In the formula, The variance is the fluctuation. The first reference output curve Output value over a given period; The average output value over the entire cycle of the baseline output curve; This represents the peak-shaving margin of the power grid. This represents the total number of time periods in the scheduling cycle.

[0207] Specifically, the optimized second power supply plan undergoes cross-monthly stability verification and fluctuation quantification assessment. First, using a monthly statistical time granularity, the output data for each time period of the second power supply plan is averaged, and the baseline output characteristic values ​​for each month are extracted. All monthly baseline output characteristic values ​​are arranged chronologically to construct a cross-monthly operational baseline sequence, used to characterize the monthly output baseline variation pattern of the medium- and long-term power supply plan. The cross-monthly operational baseline sequence is input into an isolated forest algorithm model. Leveraging the isolated forest's ability to identify anomalous discrete data, abnormal jump data points caused by local output mutations are screened. For identified abnormal jump data points, neighborhood mean interpolation is used to correct the data. After removing abnormal interference data, a smoothed baseline sequence is output, ensuring the continuity of the overall temporal variation of the sequence. Multi-dimensional temporal interpolation features are extracted from the smoothed baseline sequence, specifically including the output amplitude, temporal position, and output change rate of adjacent nodes for each baseline node, fully covering the numerical and temporal variation characteristics of the baseline sequence. Piecewise linear fitting is performed based on the extracted time-series interpolation features to generate a continuous, uninterrupted cross-monthly benchmark output curve, thus achieving continuous modeling of discrete monthly benchmark data. A preset grid peak-shaving margin value is obtained and used as a normalized benchmark for output fluctuation assessment. The time-period output deviation of the continuous benchmark output curve is normalized and calculated. The cross-monthly output fluctuation variance is obtained by quantifying and solving the fluctuation variance formula, thereby objectively measuring the overall fluctuation stability of the medium- and long-term power plan.

[0208] This embodiment employs a monthly granular statistical approach to construct a baseline sequence, focusing on the cross-cycle operational characteristics of medium- and long-term scheduling, avoiding interference from short-term intraday fluctuations on overall stability assessment, and aligning with the operational assessment needs of medium- and long-term wind and solar scheduling. The isolated forest algorithm identifies and corrects abnormal transition points in the sequence, effectively eliminating localized abnormal data generated during power planning optimization and avoiding stability assessment distortion caused by anomalies. By fitting a continuous baseline output curve with multi-dimensional time-series features, the temporal correlation of discrete monthly data is improved, enhancing the completeness and accuracy of fluctuation quantification assessment. The grid peak-shaving margin is introduced as a normalized benchmark to calculate fluctuation variance, and combined with the actual peak-shaving capacity of the grid, the output fluctuation level is quantified. This differs from the traditional single mathematical variance calculation method, resulting in assessment results that better reflect the actual operational constraints of the grid.

[0209] In one embodiment, such as Figure 2 As shown, this application also provides a medium- and long-term wind and solar power station coordinated operation system, which may include:

[0210] The wind and solar feature extraction module 801 is used to obtain the historical meteorological evolution sequence and historical meteorological mean sequence of the target area at a long-term time scale, input the historical meteorological evolution sequence and historical meteorological mean sequence into a long short-term memory network for time series feature extraction, and output an effective first wind and solar complementary feature sequence; it is also used to input the first wind and solar complementary feature sequence into a random forest algorithm model to extract the wind and solar complementary fluctuation pattern features at a medium- and long-term scale, and output a second wind and solar complementary feature sequence.

[0211] The typical morphology recognition module 802 is used to cluster the second wind-solar complementary feature sequence using the mean clustering algorithm to determine the typical combined power output morphology; it is also used to obtain a preset transmission channel capacity threshold. If the peak output of the typical combined power output morphology is greater than the transmission channel capacity threshold, it is determined that there is a channel congestion risk and a first risk identifier is generated; it is also used to obtain a preset channel vacancy threshold. If the valley output of the typical combined power output morphology is less than the channel vacancy threshold, it is determined that there is a channel vacancy risk and a second risk identifier is generated.

[0212] The power planning optimization module 803 is used to perform time-series shift processing on the pre-constructed initial medium- and long-term power plan based on the first risk identifier and the second risk identifier to obtain the first power plan; it is also used to calculate the expected wind and solar power curtailment corresponding to the first power plan. If the expected wind and solar power curtailment is greater than the curtailment threshold, the first power plan is subjected to peak shaving and valley filling processing to obtain the second power plan.

[0213] The benchmark verification module 804 is used to extract the cross-monthly benchmark sequence corresponding to the second power plan and calculate the fluctuation variance of the cross-monthly benchmark sequence; if the fluctuation variance is less than the variance threshold, the second power plan is determined as a medium- and long-term dispatch instruction.

[0214] The aforementioned medium-to-long-term wind and solar power station coordinated operation system comprises four functional modules: a wind and solar feature extraction module, a typical morphology recognition module, a power generation plan optimization module, and an operation benchmark verification module. The wind and solar feature extraction module is responsible for basic data processing and two-layer intelligent feature mining. It acquires the historical meteorological evolution sequence and historical meteorological mean sequence of the target area at a medium-to-long-term scale. It then uses a long short-term memory network to extract temporal features and verify their validity, outputting a first wind-solar complementary feature sequence. Further, it uses a random forest algorithm to filter feature weights and refine patterns, outputting a second wind-solar complementary feature sequence representing the medium-to-long-term wind and solar fluctuation characteristics. The typical morphology recognition module is responsible for operating condition clustering and channel risk identification. It performs mean clustering calculation on the second wind-solar complementary feature sequence, dividing it into multiple typical joint output morphologies. Based on preset transmission channel capacity thresholds and channel vacancy thresholds, it performs limit judgments on the peak and valley values ​​of typical output morphologies, generating a first risk identifier representing channel congestion risk and a second risk identifier representing channel vacancy risk based on the judgment results. The power planning optimization module is responsible for the two-layer progressive optimization of the power plan. Based on two types of risk indicators, it performs time-series shift adjustments on the initial medium- and long-term power plan to optimize the output time-series distribution and obtain the first power plan. By calculating the expected curtailment of wind and solar power in the first power plan, peak shaving and valley filling corrections are performed based on energy storage constraints when the curtailment index exceeds the limit, resulting in a second power plan that takes into account both channel constraints and absorption constraints. The operation benchmark verification module is responsible for the stability assessment of the final scheduling scheme. It extracts the cross-monthly operation benchmark sequence of the second power plan, completes sequence smoothing correction and fluctuation variance calculation, and determines the stable and compliant second power plan as the final executable medium- and long-term scheduling instruction when the fluctuation variance meets the preset threshold condition.

[0215] This embodiment adopts a modular, layered architecture design, with each module progressively linked through preceding and following modules. This addresses the problems of traditional scheduling systems, such as shallow feature mining, limited operational condition identification, lack of risk prediction, single optimization dimensions, and lack of long-term stability verification. Each functional module is specifically responsible for its own technical aspects, avoiding functional redundancy and logical overlap, thus improving the stability and execution efficiency of the system. The overall system can adapt to the fluctuation characteristics of medium- and long-term wind and solar resources and multiple constraints of grid channels and absorption. It can automatically generate medium- and long-term scheduling instructions that are stable, adaptable to grid operation requirements, and have better renewable energy absorption effects, effectively improving the standardization, economy, and reliability of the coordinated operation of power stations in scenarios with high proportions of wind and solar access.

[0216] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0217] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the medium- and long-term wind and solar power station coordinated operation method as described above.

[0218] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0219] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0220] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, 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 the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for coordinated operation of medium- and long-term wind and solar power stations, characterized in that, The method includes: The historical meteorological evolution sequence and historical meteorological mean sequence of the target area at a long-term time scale are obtained. The historical meteorological evolution sequence and the historical meteorological mean sequence are input into a long short-term memory network for time series feature extraction. After fluctuation amplitude verification, the first wind-solar complementary feature sequence is output. The first wind-solar complementary feature sequence is input into the random forest algorithm model to extract the wind-solar complementary fluctuation pattern features at medium and long scales, and the second wind-solar complementary feature sequence is output. The mean clustering algorithm is used to cluster the second wind-solar complementary feature sequence to determine the typical joint power output pattern; Obtain a preset transmission channel capacity threshold. If the peak output of the typical combined output mode is greater than the transmission channel capacity threshold, it is determined that there is a risk of channel congestion, and a first risk identifier is generated. Obtain a preset channel vacancy threshold. If the output valley value of the typical combined output mode is less than the channel vacancy threshold, it is determined that there is a channel vacancy risk, and a second risk identifier is generated. Based on the first risk identifier and the second risk identifier, the pre-constructed initial medium- and long-term power plan is time-shifted to obtain the first power plan; Calculate the expected wind and solar power curtailment corresponding to the first power plan. If the expected wind and solar power curtailment is greater than the curtailment threshold, then perform peak shaving and valley filling on the first power plan to obtain the second power plan. Extract the cross-monthly operating benchmark sequence corresponding to the second power plan, and calculate the fluctuation variance of the cross-monthly operating benchmark sequence; if the fluctuation variance is less than the variance threshold, then the second power plan is determined as a medium- to long-term dispatch instruction.

2. The method according to claim 1, characterized in that, The step of inputting the historical meteorological evolution sequence and the historical meteorological mean sequence into a long short-term memory network for time-series feature extraction, and outputting a first wind-solar complementary feature sequence after fluctuation amplitude verification, includes: An evolution gradient sequence is calculated based on the historical meteorological evolution sequence and the historical meteorological mean sequence; The historical meteorological evolution sequence includes a wind speed evolution subsequence and a light intensity evolution subsequence divided into preset time periods. The historical meteorological mean sequence includes a wind speed historical mean subsequence and a light intensity historical mean subsequence that correspond one-to-one with the preset time period; A time-series correlation analysis was performed on the wind speed gradient subsequence and the illumination gradient subsequence in the evolution gradient sequence to extract the inverse fluctuation correlation characteristics of the two meteorological elements and obtain the wind-solar complementary fluctuation characteristics. The wind-solar hybrid fluctuation features are input into a long short-term memory network in temporal order. Temporal dependency modeling and feature fusion encoding are performed through the long short-term memory network to obtain the first wind-solar hybrid feature sequence. The fluctuation amplitude characteristic value of the first wind-solar complementary characteristic sequence is calculated according to the following formula: In the formula, The characteristic value of the fluctuation amplitude; The first wind-solar complementary feature sequence is the first The characteristic values ​​of the time period; This is the arithmetic mean of the first wind-solar complementary feature sequence over the entire time period; The fluctuation amplitude feature value is compared with the fluctuation threshold. If the fluctuation amplitude feature value is greater than the fluctuation threshold, the first wind-solar complementary feature sequence is determined to be a valid complementary feature sequence, and the valid first wind-solar complementary feature sequence is output.

3. The method according to claim 1, characterized in that, The step of inputting the first wind-solar complementary feature sequence into a random forest algorithm model to extract the wind-solar complementary fluctuation pattern features at medium and long-term scales and outputting the second wind-solar complementary feature sequence includes: The first wind-solar complementary feature sequence is divided into multiple time series slices by a sliding window segmentation according to a preset window length and sliding step size. Each time series slice is input into the random forest algorithm model. The importance of each feature dimension is calculated by the reduction of Gini impurity of each decision tree in the random forest algorithm. The average of the calculation results of all decision trees is taken, and the feature weight value corresponding to each feature dimension is output. No. The feature weights of the dimensional features satisfy: In the formula, The total number of decision trees contained in the random forest algorithm model; For the first The first decision tree The reduction in Gini impurity corresponding to the dimensional feature; For the first The feature weight values ​​of the dimensional feature; Based on the feature weight values, each feature dimension is filtered, and feature dimensions with feature weight values ​​higher than the weight threshold are retained and determined as valid feature dimensions, thus obtaining the wind-solar complementary fluctuation pattern characteristics. The characteristics of the wind-solar complementary fluctuation pattern are arranged in chronological order to construct a complementary fluctuation matrix; The complementary wave matrix is ​​weighted and fused along the feature dimension to generate a second wind-solar complementary feature sequence; No. The values ​​of the second wind-solar complementary feature sequence during the time period satisfy: In the formula, For the first The values ​​of the second wind-solar complementary characteristic sequence during the time period; For the first Within the time period The values ​​of the effective wind-solar complementary fluctuation pattern characteristics; The number of effective feature dimensions; Calculate the fluctuation complementarity index of the second wind-solar complementary feature sequence, and compare the fluctuation complementarity index with the complementarity threshold; If the fluctuation complementarity index is greater than the complementarity threshold, then the second wind-solar complementary feature sequence is output; The fluctuation complementarity index satisfies: In the formula, It is a volatility complementarity index; The time-series component obtained by fusing wind speed-type fluctuation features in the complementary fluctuation matrix; The temporal component is obtained by fusing illumination-related wave characteristics; For covariance calculation; , These are the standard deviations of the wind speed and light intensity fluctuation component sequences, respectively.

4. The method according to claim 1, characterized in that, The step of using the mean clustering algorithm to cluster the second wind-solar complementary feature sequence and determine typical joint power output patterns includes: Based on the preset number of clusters, a corresponding number of initial cluster centers are selected in the feature space of the second wind-solar complementary feature sequence; Calculate the Euclidean distance between each temporal feature vector in the second wind-solar complementary feature sequence and each of the initial cluster centers; Based on the principle of minimum distance attribution, each temporal feature vector is assigned to the morphological cluster to which the corresponding initial cluster center belongs, based on the Euclidean distance metric, to obtain multiple initial morphological clusters; Calculate the mean feature value of all the temporal feature vectors in each initial morphological cluster, and update the cluster center of the corresponding morphological cluster with the mean feature value to obtain the current cluster center; Determine whether the current cluster center meets the preset convergence condition; if not, replace the previous cluster center with the current cluster center and update the current cluster center. If the current cluster center satisfies the preset convergence condition, a unique morphological identifier is assigned to each converged morphological cluster, and the typical joint output morphology of the corresponding category is determined based on the feature vector of the converged cluster center. The typical joint output pattern includes morphological identifier encoding, cluster center feature vector, and joint output statistical feature parameters.

5. The method according to claim 1, characterized in that, Based on the first risk identifier and the second risk identifier, the pre-constructed initial medium- and long-term power consumption plan is time-shifted to obtain a first power consumption plan, including: Based on the first risk identifier and the second risk identifier, identify the peak period of channel congestion and the valley period of channel vacancy under the corresponding typical joint output mode, and generate a joint output risk fluctuation sequence with time period attribute tags; Based on the earliest start time and latest end time of all risk periods in the joint output risk fluctuation sequence, a target adjustment time window covering all risk periods is determined; Adjust the time window range according to the target, extract the output plan data for the corresponding period from the pre-constructed initial medium- and long-term power plan, and obtain the local power plan; The initial medium- and long-term power plan is a baseline output plan pre-prepared based on the installed capacity of the power station, medium- and long-term power indicators, and the rated capacity of the channel; Using the output values ​​of each time period of the local power plan as the fitting target, construct the time series feature vector of the corresponding time period, input the time series feature vector into the random forest regression model, and fit to obtain the continuous output trend curve within the target adjustment window; Based on the output trend curve, the extreme points of output are identified, and the peak and valley features of output are extracted. The output peak-valley characteristics include the peak time. Valley moment Peak output With the lowest output value ; Based on the output peak and valley characteristics and the preset output channel capacity threshold With channel vacancy threshold Calculate the timing shift step size; The time-series translation step size satisfies: in, The average rate of change of output between peak and valley periods, satisfying ; This represents the offset step size of the output curve along the time axis; According to the time-series shift step size, the output sequence of the initial medium- and long-term power plan within the target adjustment time window is shifted and adjusted to shift the output peak value towards the output trough period to obtain the first power plan.

6. The method according to claim 1, characterized in that, The calculation of the expected wind and solar power curtailment corresponding to the first power plan, and if the expected wind and solar power curtailment is greater than the curtailment threshold, then the first power plan is subjected to peak shaving and valley filling to obtain a second power plan, including: Obtain the wind and solar power generation forecast sequence and grid load curve that match the time scale and time range of the first power plan; Based on the wind and solar power generation prediction sequence and the grid load curve, the amount of wind and solar power output exceeding the transmission and absorption capacity is calculated for each time period, and the expected amount of wind and solar power curtailment is obtained by summing them up. The expected amount of wind and solar power curtailment meets the following requirements: In the formula, The expected amount of wind and solar power curtailment; For the first Predicted output of wind and solar power generation during specific time periods; This is the preset outgoing channel capacity threshold; For the first Power absorption capacity corresponding to the grid load during a given time period; For the first power plan Planned output during the specified time period; The duration of a single time period; This represents the total number of time periods within the scheduling cycle. If the expected amount of wind and solar power curtailment is greater than the curtailment threshold, then obtain the energy storage available capacity constraint sequence for the corresponding time scale. The available energy storage capacity constraint sequence includes the maximum charging power, maximum discharging power and upper and lower limits of the state of charge for each time period. Using the output sequence of the first power plan, the wind and solar power generation prediction sequence, and the energy storage available capacity constraint sequence as input features, the random forest optimization model is used to fit the mapping relationship between the charging and discharging strategy and the curtailment suppression target, and the energy storage charging and discharging adjustment command for each time period is output. The output value of the first power plan is superimposed and corrected according to the energy storage charging and discharging adjustment command for the corresponding time period to obtain the second power plan.

7. The method according to claim 1, characterized in that, The step of extracting the cross-monthly operating benchmark sequence corresponding to the second electricity plan and calculating the variance of the cross-monthly operating benchmark sequence includes: The second power plan is statistically analyzed by monthly time granularity to obtain the average output value, the benchmark output characteristic value of each month is extracted and arranged in time sequence to obtain the cross-month operation benchmark sequence; The cross-monthly benchmark sequence is input into the isolated forest algorithm model. The isolated forest algorithm identifies anomalous jump data points in the sequence. The identified anomalous jump data points are corrected by neighborhood mean interpolation, and a smoothed benchmark sequence is output. Temporal interpolation features are extracted from the smoothed reference sequence; the temporal interpolation features include the output amplitude, temporal position, and output change rate of adjacent nodes of each reference node; Based on the aforementioned time-series interpolation features, piecewise linear fitting is performed to construct a continuous cross-monthly benchmark output curve; Obtain a preset peak-shaving margin value for the power grid. Using the peak-shaving margin value as a normalization benchmark, perform a normalization operation on the time-by-time output deviation of the benchmark output curve to calculate the fluctuation variance. The variance of the fluctuation satisfies: In the formula, Let Variance be the fluctuation variance; The reference output curve is the first Output value over a given period; The average output value over the entire cycle of the reference output curve; This refers to the peak-shaving margin value of the power grid; This represents the total number of time periods in the scheduling cycle.

8. A medium- to long-term wind and solar power station coordinated operation system, characterized in that, The system includes: The wind and solar feature extraction module is used to obtain the historical meteorological evolution sequence and historical meteorological mean sequence of the target area at a long-term time scale, input the historical meteorological evolution sequence and the historical meteorological mean sequence into a long short-term memory network for time series feature extraction, and output an effective first wind and solar complementary feature sequence; it is also used to input the first wind and solar complementary feature sequence into a random forest algorithm model to extract the wind and solar complementary fluctuation pattern features at a medium- and long-term time scale, and output a second wind and solar complementary feature sequence. The typical morphology recognition module is used to cluster the second wind-solar complementary feature sequence using the mean clustering algorithm to determine the typical combined power output morphology; it is also used to obtain a preset transmission channel capacity threshold. If the peak output of the typical combined power output morphology is greater than the transmission channel capacity threshold, it is determined that there is a channel congestion risk and a first risk identifier is generated; it is also used to obtain a preset channel vacancy threshold. If the valley output of the typical combined power output morphology is less than the channel vacancy threshold, it is determined that there is a channel vacancy risk and a second risk identifier is generated. The power plan optimization module is used to perform time-series shift processing on a pre-constructed initial medium- and long-term power plan based on the first risk identifier and the second risk identifier to obtain a first power plan; it is also used to calculate the expected wind and solar power curtailment corresponding to the first power plan. If the expected wind and solar power curtailment is greater than the curtailment threshold, the first power plan is subjected to peak shaving and valley filling processing to obtain a second power plan. The benchmark verification module is used to extract the cross-monthly benchmark sequence corresponding to the second power plan and calculate the fluctuation variance of the cross-monthly benchmark sequence; if the fluctuation variance is less than the variance threshold, the second power plan is determined as a medium- and long-term dispatch instruction.

9. A computer device 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 method of 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 a processor, it implements the method of any one of claims 1 to 7.

Images