Wind farm optimal scheduling method, system, electronic device and storage medium based on double power prediction system

Abstract

The present invention relates to the field of wind power generation technology, and specifically to a method, system, electronic device and storage medium for optimizing the dispatching of a wind farm based on a dual power prediction system. The method comprises: extracting a first power prediction data set and a second power prediction data set based on the dual power prediction system; extracting a target prediction reporting rate and a power prediction accuracy rate from the first power prediction data set and the second power prediction data set; constructing an accuracy evaluation model, extracting a power prediction accuracy evaluation result based on the impact of historical data of the same period on the prediction accuracy rate; and outputting an optimal dispatching strategy based on the power prediction accuracy evaluation result with the goal of minimizing the power prediction deviation. The purpose is to achieve optimized dispatching of a wind farm based on a dual power prediction system, and to improve the feasibility of the dispatching strategy of a wind farm under different power prediction results.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power generation, and particularly to a wind farm optimization scheduling method, system, electronic device and storage medium based on a dual power prediction system. Background Art

[0002] In the field of wind power generation, power dispatching agencies have strict assessments on the power prediction results reported by wind farm sites, covering various indicators such as the reporting rate, medium-term and short-term as well as ultra-short-term power prediction accuracy rates. At the same time, there are also detailed assessment requirements for the data quality of various automation devices, equipment availability, link status, and telemetry signals that interact between the site and the dispatching. For example, the medium-term power prediction reporting rate needs to reach 100%, and the ultra-short-term reporting rate also has high standards, and different assessment items correspond to different electricity assessment ratios or rules. However, the current centralized control center faces many dilemmas in the management process. On the one hand, the acquisition of assessment data is extremely passive and there is a delay. Since the dispatching issues assessment results monthly, it is difficult to detect anomalies in the reporting of prediction data and non-compliance with the accuracy rate in a timely manner, resulting in frequent assessments of the power prediction reporting rate and accuracy rate. On the other hand, the obtained assessment data is only the final summary result, and it is impossible to conduct refined stage-by-stage and time-by-time analysis of each stage of power prediction, which makes it difficult to deeply analyze the root causes of problems. Moreover, for power prediction manufacturers, due to the lack of refined management means, it is impossible to specifically require them to optimize the prediction model. Even if the manufacturers optimize regularly, the assessment results are still not ideal. Although some sites are equipped with dual power prediction systems, because the dispatching only provides a set of general assessment data, it is impossible to conduct refined accuracy analysis on the two systems, and it is difficult to discover their respective advantages and disadvantages and optimize them specifically. In addition, for important grid-connected equipment connected to the dispatching by the site, since the communication and data quality information are in the hands of the dispatching side, the centralized control center and the site are in a passive waiting state for notification and cannot actively monitor and analyze. Once the dispatching fails to detect in time, it will cause continuous assessment problems.

[0003] Currently, the optimization scheduling of wind farms based on dual power prediction systems aims to comprehensively utilize the two systems to improve scheduling efficiency and prediction accuracy. However, there is currently a key problem of insufficient feasibility of scheduling strategies. Existing scheduling strategies often lack sufficient consideration of the complex and changeable operating environment of wind farms and fail to comprehensively integrate various influencing factors. When formulating scheduling strategies, although there is data from dual power prediction systems, due to the inability to conduct refined analysis of their accuracy characteristics and change trends, it is difficult to dynamically adjust the strategy weights according to the actual prediction capabilities of the two systems under different time periods and different meteorological conditions, resulting in the strategy possibly over-relying on a certain system or evenly distributing weights, and unable to achieve the optimal decision. Summary of the Invention

[0004] The purpose of the present invention is to realize the optimal dispatching of wind farms based on the dual power prediction system, improve the feasibility of the dispatching strategy of wind farms under different power prediction results, and provide a method, system, electronic device and storage medium for optimizing the dispatching of wind farms based on the dual power prediction system. The technical solutions adopted are as follows:

[0005] The technical solution of the first aspect of the present invention provides a wind farm optimization scheduling method based on a dual power prediction system, the method comprising:

[0006] Extracting a first power prediction data set and a second power prediction data set based on a dual power prediction system;

[0007] Extracting a target prediction reporting rate and a power prediction accuracy rate from the first power prediction data set and the second power prediction data set;

[0008] Construct an accuracy evaluation model, extract the power prediction accuracy evaluation results based on the impact of historical data on the prediction accuracy;

[0009] Based on the power prediction accuracy evaluation results, the optimal scheduling strategy is output with the goal of minimizing the power prediction deviation.

[0010] Further, extracting the first power prediction data set and the second power prediction data set based on the dual power prediction system includes:

[0011] Extracting a first power prediction value sequence, a second power prediction value sequence, a first meteorological data sequence, and a second meteorological data sequence from the dual power prediction system according to a preset time resolution;

[0012] Performing data cleaning and alignment preprocessing on the first power prediction value sequence, the second power prediction value sequence, the first meteorological data sequence, and the second meteorological data sequence;

[0013] Associating the preprocessed first power prediction value sequence and the first meteorological data sequence to obtain a first power prediction data set;

[0014] The preprocessed second power prediction value sequence and the second meteorological data sequence are associated to obtain a second power prediction data set.

[0015] Further, extracting the target prediction reporting rate in the first power prediction data set and the second power prediction data set includes:

[0016] Traversing the timestamp and related record information corresponding to each power prediction value in the first power prediction data set and the second power prediction data set;

[0017] According to the preset reporting time interval, the actual reporting times of the dual power prediction system within the statistical period are extracted;

[0018] The predicted reporting rate is extracted based on the actual number of reports of the dual power prediction system.

[0019] Further, extracting the power prediction accuracy rates in the first power prediction data set and the second power prediction data set includes:

[0020] Based on the first power prediction data set and the second power prediction data set, the root mean square error between the power prediction value and the actual power measurement value at each time point is calculated respectively, and the expression is:

[0021]

[0022] In the formula, represents the root mean square error; Indicates the total number of data; Indicates The power prediction value at each time point; Indicates The actual power measurement value at a point in time.

[0023] Furthermore, an accuracy evaluation model is constructed, and based on the impact of historical data of the same period on the prediction accuracy, the power prediction accuracy evaluation results are extracted, including:

[0024] Acquire historical data of the dual power prediction system in the same period, extract features related to the power prediction accuracy from the historical data in the same period, and construct a first feature matrix and a second feature matrix, wherein the first feature matrix and the second feature matrix at least include: meteorological data, unit operation status data, and root mean square error;

[0025] The accuracy evaluation model is constructed using the random forest algorithm, and the accuracy evaluation model is trained using the first feature matrix and the second feature matrix respectively;

[0026] The trained accuracy evaluation model is used to output the accuracy evaluation results of the dual power prediction system.

[0027] Furthermore, the expression of the accuracy evaluation result of the dual power prediction system is:

[0028]

[0029] In the formula, Indicates The accuracy evaluation results of the power prediction system; Indicates The self-use weight adjustment factor of the power forecast system, ; Indicates The number of decision trees included in the accuracy evaluation model constructed by each power prediction system; Indicates In a power prediction system, the prediction accuracy value output by the k-th decision tree for the input feature vector; represents the number of features; represents the k-th element in the feature importance weight vector corresponding to the k-th power prediction system; represents the value of the k-th feature in the feature matrix of the

[0030] k-th power prediction system.

[0031] Furthermore, based on the power prediction accuracy evaluation results, the optimal scheduling strategy output with the goal of minimizing the power prediction deviation includes:

[0032] Establish the operating constraints of the wind farm units and the grid connection constraints;

[0033]

[0034] In the formula, represents the minimization operator; represents the time period of the scheduling decision; represents the first balance coefficient, calculated according to the self - use weight adjustment factor of the first power prediction system; represents the second balance coefficient, calculated according to the self - use weight adjustment factor of the second power prediction system; represents the power prediction value of the first power prediction system at time t; represents the power prediction value of the second power prediction system at time t; represents the actual power measurement value at time t; represents the weight given to the first power prediction system at time t based on the accuracy evaluation results, ; represents the weight given to the second power prediction system at time t based on the accuracy evaluation results, ; represents the accuracy evaluation result of the first power prediction system; represents the accuracy evaluation result of the second power prediction system.

[0035] The technical solution of the second aspect of the present invention provides a wind farm optimization scheduling system based on a dual power prediction system, which adopts the wind farm optimization scheduling method based on the dual power prediction system provided by the technical solution of the first aspect of the present invention. The system includes:

[0036] A data acquisition module configured to extract a first power prediction data set and a second power prediction data set based on the dual power prediction system;

[0037] A feature extraction module configured to extract the target prediction reporting rate and the power prediction accuracy rate in the first power prediction data set and the second power prediction data set;

[0038] An accuracy rate evaluation module configured to construct an accuracy rate evaluation model and extract the power prediction accuracy rate evaluation result based on the influence of historical data in the same period on the prediction accuracy rate;

[0039] A scheduling strategy optimization module configured to output an optimal scheduling strategy with the goal of minimizing the power prediction deviation based on the power prediction accuracy rate evaluation result.

[0040] The technical solution of the third aspect of the present invention provides an electronic device, which includes: a processor and a memory communicatively connected to the processor; wherein, the memory stores instructions executable by the processor, and when the instructions are executed by the processor, the processor can execute the steps of the wind farm optimization scheduling method based on the dual power prediction system described in the technical solution of the first aspect of the present invention.

[0041] The technical solution of the fourth aspect of the present invention provides a computer-readable storage medium, on which a program for implementing the wind farm optimization scheduling method based on the dual power prediction system is stored, and when the program for implementing the wind farm optimization scheduling method based on the dual power prediction system is executed by a processor, the steps of the wind farm optimization scheduling method based on the dual power prediction system described in the technical solution of the first aspect of the present invention are implemented.

[0042] The present invention has the following beneficial effects:

[0043] The wind farm optimization scheduling method based on the dual power prediction system provided by the present invention operates in a multi-step collaborative manner. The extracted dual power prediction data set ensures that the prediction information of the dual power prediction system can be fully considered; the constructed accuracy evaluation model uses historical contemporaneous data to deeply analyze the influencing factors, so that the output accuracy evaluation results can more scientifically reflect the reliability of the dual power prediction system under different working conditions; finally, based on the accuracy evaluation results, an optimal scheduling strategy with the goal of minimizing power prediction deviation is formulated, and the weight distribution can be dynamically adjusted according to the real-time accuracy of each system. When the accuracy of one system is high, its prediction results are more relied on to avoid decision-making errors caused by blind averaging or fixed weight distribution. At the same time, the method comprehensively considers the correlation between historical contemporaneous data and power prediction accuracy, so that the wind farm can more reasonably arrange the power generation plan and optimize the load distribution of the unit in a complex and changeable operating environment, and reduce the risks caused by inaccurate prediction or unreasonable scheduling, thereby significantly improving the stability of wind farm operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0045] Figure 1 A method flow chart of a wind farm optimization scheduling method based on a dual power prediction system provided by an embodiment of the present invention;

[0046] Figure 2 A schematic structural diagram of a wind farm optimization scheduling system based on a dual power prediction system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0047] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the wind farm optimization scheduling method, system, electronic device and storage medium based on the dual power prediction system proposed by the present invention, its specific implementation method, structure, features and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0048] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0049] The following specifically describes the specific solutions of a wind farm optimization scheduling method, system, electronic device, and storage medium provided by the present invention in conjunction with the accompanying drawings.

[0050] Please refer to Figure 1 , which shows a flowchart of a method for optimizing the scheduling of a wind farm based on a dual power prediction system according to an embodiment of the present invention. The method includes:

[0051] Step S100: Extract a first power prediction data set and a second power prediction data set based on the dual power prediction system; specifically, the dual power prediction system is two sets of independently operating power prediction systems configured in the wind farm. They each use different prediction algorithms, models, and data processing methods to predict the future power generation of the wind farm; usually, one system focuses on accurate prediction under a specific meteorological condition, and the other may have better power prediction at different time scales or different operating conditions. By using these two systems simultaneously, power prediction information can be provided for the scheduling decision of the wind farm from multiple perspectives, increasing the scientificity and reliability of the decision, and thus better coping with the complex and variable power generation changes of the wind farm and meeting the relevant requirements of the power dispatching agency;

[0052] Step S100 specifically includes:

[0053] Step S110: Extract a first power prediction value sequence, a second power prediction value sequence, a first meteorological data sequence, and a second meteorological data sequence from the dual power prediction system according to a preset time resolution; for the extraction of the meteorological data sequence, the meteorological data sources integrated or accessed by the dual power prediction system itself can also be relied on, and meteorological data such as wind speed, wind direction, air pressure, temperature, and humidity can be extracted respectively according to the same time resolution to form a first meteorological data sequence and a second meteorological data sequence;

[0054] Step S120: Perform data cleaning and alignment preprocessing on the first power prediction value sequence, the second power prediction value sequence, the first meteorological data sequence, and the second meteorological data sequence; specifically, for the power prediction value sequence, set a reasonable threshold range based on the installed capacity of the wind farm and the statistical characteristics of historical power data to judge outliers and remove the outliers from the sequence; for the meteorological data sequence, identify outliers according to meteorological common sense and the range of local long-term accumulated meteorological data, determine the abnormal data and remove it; for the missing values in the data, linear interpolation or spline interpolation can be used to fill them to ensure the integrity and reasonableness of the data; due to possible time synchronization errors or inconsistent data recording frequencies between the power prediction system and the meteorological data acquisition system, time alignment operations are required. Specifically, based on the time stamps of the power prediction value sequence, match the corresponding meteorological data sequence in chronological order; if the time points of the meteorological data do not exactly coincide with the time points of the power prediction values, the nearest meteorological data record to the power prediction time point can be selected according to the principle of proximity to ensure that each power prediction value can correspond to accurate meteorological data, so that different sequences are accurately aligned in time for subsequent correlation and analysis.

[0055] Step S130: Correlate the preprocessed first power prediction value sequence and the first meteorological data sequence to obtain the first power prediction data set; specifically, through database operations, each power prediction value in the first power prediction value sequence after cleaning and alignment preprocessing is associated one by one with the meteorological data items in the first meteorological data sequence corresponding to the same time point, forming a two-dimensional data table structure containing power prediction values and related meteorological data. Each row represents a data record at a specific time point, and the columns are the power prediction value and different meteorological element values respectively, thus constructing the first power prediction data set;

[0056] Step S140: Correlate the preprocessed second power prediction value sequence and the second meteorological data sequence to obtain the second power prediction data set; the extraction method of the second power prediction data set is the same as that of the first power prediction data set;

[0057] The sub-steps of Step S100 are closely coordinated, providing a high-quality and strongly correlated data basis for the optimal dispatching of the wind farm based on the dual power prediction system from data acquisition, processing to integration.

[0058] Step S200: Extract the target prediction reporting rate and power prediction accuracy rate from the first power prediction data set and the second power prediction data set;

[0059] Step S200 specifically includes:

[0060] Step S210: Extracting the target prediction reporting rate in the first power prediction dataset and the second power prediction dataset includes:

[0061] Step S211: Traverse the timestamps and related record information corresponding to each power prediction value in the first power prediction dataset and the second power prediction dataset; specifically, read the first power prediction dataset and the second power prediction dataset, both of which are structured data sets (in the form of two-dimensional data tables) containing power prediction values and related meteorological data and other information obtained from previous steps; for the first power prediction dataset, starting from the first row of data, sequentially read the timestamp field information corresponding to the power prediction value in each row, and at the same time obtain other possible related record information in this row of data, such as data source identification, data quality mark, etc.; in the same way, perform a traversal operation on the second power prediction dataset to ensure that the complete timestamps and related record information corresponding to all power prediction values in the two datasets are obtained; the above information will be used as the basis for determining the reporting situation in the follow-up. Through the timestamp, the specific time point corresponding to each power prediction value can be determined;

[0062] Step S212: Extract the actual reporting times of the dual power prediction system within the statistical period according to the preset reporting time interval; in this embodiment, according to the internal management requirements of the wind farm, determine the preset reporting time interval, preferably reporting power prediction data every 30 minutes. Then, taking a day as the statistical period, the number of times to be reported within a day should be 48 times; for the first power prediction system, after extracting the timestamp information from its dataset, according to the preset reporting time interval, sequentially check whether there is a corresponding power prediction value record at each reporting time point. If there is, count it once; after traversing all the data within the statistical period, obtain the actual reporting times of the first power prediction system within the statistical period. Use the same method to operate on the second power prediction system;

[0063] Step S213: Extract the prediction reporting rate based on the actual reporting times of the dual-power prediction system; specifically, after obtaining the actual reporting times of the first power prediction system and the second power prediction system within the statistical period, calculate the prediction reporting rate according to the following formula: Prediction reporting rate = (Actual reporting times / Should-report times) × 100%; Record and organize the two calculated prediction reporting rates. The reporting rate data will be used as one of the indicators to evaluate the operation stability and compliance of the dual-power prediction system, and can be used for comparative analysis with other wind farms or industry standards, and also provide a reference basis for the internal management decision-making of the wind farm; By systematically traversing, analyzing, and calculating the first power prediction data set and the second power prediction data set, the prediction reporting rate of the dual-power prediction system can be accurately extracted; It helps the wind farm to comprehensively and accurately grasp the reporting situation of the two power prediction systems within the specified statistical period, and timely discover possible problems such as untimely reporting and missed reporting, so as to effectively monitor the operation stability of the power prediction system and the integrity of data transmission; Based on the obtained prediction reporting rate, the wind farm can better evaluate its own performance in meeting the reporting requirements of the power dispatching agency, early warning potential assessment risks, and providing data support for subsequent in-depth analysis of the performance differences of the power prediction system and the formulation of targeted optimization measures, which helps to improve the operation management level and the scientific nature of dispatching decisions of the entire wind farm.

[0064] Step S220: Extracting the power prediction accuracy rates in the first power prediction data set and the second power prediction data set includes:

[0065] Step S221: Based on the first power prediction data set and the second power prediction data set, calculate the root mean square error between the power prediction value and the actual power measurement value at each time point respectively. Its expression is:

[0066]

[0067] In the formula, represents the root mean square error; represents the total number of data; represents the th power prediction value at the time point; represents the Actual power measurement values at each time point; this formula uses the root mean square error as a quantization index and accurately calculates the prediction accuracy of each power prediction system in different time periods based on the first power prediction data set and the second power prediction data set; it can intuitively and accurately measure the deviation degree between the two power prediction systems and the actual power, enabling the wind farm operators to clearly understand the reliability of the prediction results of each system under different working conditions; by comparing the root mean square errors of the two systems, it helps to discover the advantages and disadvantages of different systems in terms of prediction accuracy, and further provides an index for subsequent work such as selecting a better prediction result for scheduling decisions, evaluating the performance of the power prediction system, and specifically optimizing and improving the prediction model.

[0068] Step S300: Construct an accuracy evaluation model, and extract the power prediction accuracy evaluation result based on the influence of historical data in the same period on the prediction accuracy.

[0069] Step S300 specifically includes:

[0070] Step S310: Obtain the historical data in the same period of the dual power prediction system, extract the features related to the power prediction accuracy from the historical data in the same period, and construct a first feature matrix and a second feature matrix. The first feature matrix and the second feature matrix at least include: meteorological data, unit operation status data, and root mean square error; specifically, first, determine the time range of the historical data in the same period, and historical data corresponding to the current time period for power prediction within 3 - 5 years can be selected. For example, if predicting a certain period in the current winter, the data of the same period in previous winters will be extracted; the historical data in the same period should cover the historical power prediction value records generated by the dual power prediction system, and at the same time, it should also include the relevant records of the actual power measurement values at the corresponding moments; from the data storage system of the wind farm, the power prediction data of the first power prediction system and the second power prediction system in the corresponding historical period are respectively extracted through data query statements according to the time stamp and the corresponding power prediction system identifier; at the same time, other relevant auxiliary data during this period, such as meteorological data and unit operation status data, also need to be collected.

[0071] For the extraction of meteorological data features:

[0072] Basic meteorological element features: For the collected historical meteorological data, common meteorological elements are extracted as features, including but not limited to average wind speed, wind speed standard deviation, wind direction and its change frequency, daily maximum temperature, daily minimum temperature, and average air pressure; these basic features can intuitively reflect the potential impact of the meteorological conditions at that time on the power prediction accuracy.

[0073] Meteorological change trend characteristics: Extract the change rate of meteorological elements over time, including but not limited to the wind speed change rate and the temperature change rate; these change trend characteristics help to capture the impact of the meteorological dynamic change process on power prediction;

[0074] For the extraction of characteristics of the unit operation status data:

[0075] Characteristics of the basic operation parameters of the unit: From the operation records of the wind farm units, extract characteristics including but not limited to the cumulative operation duration, cumulative start-stop times, average operation duration per time, and average fault-free operation time interval of each unit; these characteristics can reflect the impact of the operation stability and fatigue degree of the unit on power prediction as a whole;

[0076] Fault-related characteristics: Conduct a detailed classification and statistics on the faults that have occurred in the history of the unit, and record the occurrence frequencies of different types of faults; at the same time, calculate the first operation power deviation and fault recovery time after each fault repair; through these characteristics, the model can learn the relationship between the fault factors and the power prediction accuracy rate;

[0077] For the extraction of root mean square error characteristics:

[0078] According to the calculation formula of the root mean square error (RMSE), calculate the root mean square error corresponding to the first power prediction system and the second power prediction system respectively for each historical same period; the root mean square error can intuitively reflect the deviation degree between the power prediction value and the actual value in this time period, and the accuracy evaluation model can directly learn the size of the prediction deviation in different situations;

[0079] Construct the first feature matrix and the second feature matrix based on the above meteorological data characteristics, unit operation status data characteristics, and fault-related characteristics: For the first feature matrix, arrange the eigenvalues corresponding to each historical same-period time period in sequence as rows. That is, the data in the first row are all the eigenvalues extracted from the first historical same-period time period, and fill them into the corresponding columns of the matrix in sequence according to the feature order, and so on to construct the entire first feature matrix; in the same way, construct the second feature matrix to ensure that the feature meanings at the corresponding positions in the two feature matrices are the same, but they correspond to different power prediction systems respectively; finally, the two feature matrices completely integrate multi-dimensional information related to the historical same-period power prediction accuracy of the dual power prediction system, providing a standardized data structure basis for the subsequent training of the accuracy evaluation model; Step S310 constructs the first feature matrix and the second feature matrix by systematically obtaining the historical same-period data of the dual power prediction system. This process provides a data basis for the subsequent construction of the accuracy evaluation model, can quantitatively integrate various key factors in the complex operation environment of the wind farm, enables the model to fully learn the mutual relationships between different factors and their comprehensive impact on the power prediction accuracy, so as to more accurately discover the laws in historical data, reliably evaluate the power prediction accuracy of the dual power prediction system under different working conditions, provide strong data support for the optimization scheduling decision of the wind farm, and effectively improve the rationality and adaptability of the scheduling strategy.

[0080] Step S320: Construct an accuracy evaluation model using the random forest algorithm, and train the accuracy evaluation model using the first feature matrix and the second feature matrix respectively. Specifically, construct an accuracy evaluation model using the random forest algorithm, and perform operations on the first power prediction system and the second power prediction system respectively. First, initialize two random forest regression models. Since what needs to be evaluated is accuracy, and accuracy can be regarded as a prediction problem of continuous numerical values, a regression model is preferably used in this embodiment. Then, set the value range of the key parameters. For example, the value range of the number of decision trees is preferably between 100 and 500, the maximum depth of each tree is preferably about 5 to 20 layers, and the minimum number of samples required for internal node splitting is preferably 2 to 10. It should be noted that the approximate ranges of these initial parameters can be determined by hyperparameter optimization to obtain the best values later. For the first feature matrix and the second feature matrix, divide their corresponding historical data in the same period into a training set, a validation set, and a test set. 70% of the data is used as the training set, 20% as the validation set, and 10% as the test set. At the same time, the division is carried out by random sampling, but it is necessary to ensure that the distribution of various features in different subsets is relatively balanced to avoid data skew affecting the model training effect. Then, use the training set to train the two random forest models respectively. Take the features in the feature matrix as the input (for the first power prediction system and the second power prediction system, they respectively correspond to each row of data in their respective feature matrices), and the corresponding historical power prediction accuracy as the output target value. During the training process, use the grid search hyperparameter optimization method combined with cross-validation to find the optimal combination of hyperparameters. In this way, an accuracy evaluation model for the first power prediction system is trained for the first feature matrix, and an accuracy evaluation model for the second power prediction system is trained for the second feature matrix. Each of the two models learns the complex relationship between the features and the accuracy in the corresponding system for subsequent accuracy evaluation.

[0081] Step S330: Use the trained accuracy evaluation model to output the accuracy evaluation result of the dual power prediction system. Specifically, the expression of the accuracy evaluation result of the dual power prediction system is:

[0082]

[0083] In the formula, represents the accuracy evaluation result of the th power prediction system; represents the self-use weight adjustment factor of the th power prediction system, ; represents the number of decision trees included in the accuracy evaluation model constructed by the th power prediction system; represents the In a power prediction system, the prediction accuracy value output by the k-th decision tree for the input feature vector; represents the number of features; represents the k-th element in the feature importance weight vector corresponding to the k-th power prediction system; represents the value of the k-th feature in the feature matrix of the

[0084] k-th power prediction system; Specifically, obtain the relevant data that needs to be predicted for power at the current moment, and extract the corresponding features from information sources such as real-time dual power prediction system data, meteorological data, and unit operation status data according to the same feature extraction method as when constructing the feature matrix to form the current feature vector; for example, when constructing the feature matrix in step S320, features such as average wind speed, cumulative unit operation duration, and root mean square error are extracted, then at this time, the specific values of these corresponding features need to be obtained to form the feature vector; input the current feature vector corresponding to the first power prediction system into the accuracy evaluation model of the trained first power prediction system. Each decision tree in the model will output a prediction accuracy value according to the input feature vector, and then calculate according to the above expression of the accuracy evaluation result; perform the same operation steps, input the current feature vector corresponding to the second power prediction system into its trained accuracy evaluation model, and calculate the current accuracy evaluation result of the second power prediction system according to the expression of the accuracy evaluation result; finally, the accuracy evaluation result of the dual power prediction system in the current situation can be obtained, which can intuitively understand the reliability of the prediction results of the two systems in the current environment and working conditions, and provide a key basis for formulating subsequent scheduling strategies;

[0085] Step S400: Based on the evaluation result of the power prediction accuracy, output the optimal scheduling strategy with the goal of minimizing the power prediction deviation;

[0086] Step S400 specifically includes:

[0087] Step S410: Establish the operating constraints of the wind farm units and the grid connection constraints; specifically, first obtain the rated power parameters of each unit in the wind farm and determine the minimum and maximum of the generated power of each unit; for all units in the wind farm, the lower and upper limits of the total generated power need to be satisfied at any time, which can be expressed as:

[0088]

[0089]

[0090] In the formula, represents the total number of wind turbines; represents the th unit's generated power at time; and are the lower and upper limits of the total generated power of the wind farm respectively;

[0091] Consult the technical manual of the unit or analyze the actual operation data to determine the upward ramp rate and downward ramp rate of each unit; the ramp rate represents the maximum amplitude that the unit's power can increase or decrease per unit time, so there are constraint conditions:

[0092]

[0093] This constraint condition is used to ensure that the unit power change is within a reasonable range, avoid damage to the unit equipment and ensure the stability of the grid connection, and prevent problems such as mechanical stress and electrical shock caused by too fast power adjustment of the unit;

[0094] According to the operating specifications of the grid and the requirements for the wind farm connection point, determine the allowable range of the power factor, which can be expressed as:

[0095]

[0096] Specifically, the grid requires the power factor at the wind farm connection point to be maintained between 0.9 and 1 to ensure the power quality and reactive power balance of the grid. By monitoring the active power and reactive power of the wind farm, calculate the power factor , and meet this constraint condition by adjusting the reactive power output of the unit or using reactive power compensation equipment; for example, when the power factor is lower than the lower limit When the power factor is lower than the lower limit, the reactive power compensation device can be added or the excitation current of the unit can be adjusted to increase the reactive power output, so that the power factor can rise back to the allowable range; conversely, when the power factor is higher than the upper limit the reactive power output is reduced.

[0097] Obtain the maximum power capacity that the power grid can accept from the wind farm at the current moment , which is usually determined comprehensively by the power grid dispatching department according to the overall load situation of the power grid, the capacity of the transmission line, the power generation situation of other power sources, etc. The constraint condition for the total power generation of the wind farm is:

[0098]

[0099] This constraint condition is used to prevent the safe and stable operation of the power grid from being affected due to excessive power generation of the wind farm.

[0100] Step S420: Construct an objective function and use the linear programming algorithm to solve the optimal scheduling strategy. The expression of the objective function is:

[0101]

[0102] In the formula, represents the minimization operator; represents the time period of the scheduling decision; represents the first balance coefficient, which is calculated according to the self-use weight adjustment factor of the first power prediction system; represents the second balance coefficient, which is calculated according to the self-use weight adjustment factor of the second power prediction system; represents the power prediction value of the first power prediction system at time; represents the power prediction value of the second power prediction system at time; represents the actual power measurement value at time; represents the weight given to the first power prediction system at time based on the accuracy evaluation result, ; represents the weight given to the second power prediction system at time based on the accuracy evaluation result, ; represents the accuracy evaluation result of the first power prediction system; represents the accuracy evaluation result of the second power prediction system; and ;

[0103] Finally, the objective function and the established operating constraints of the wind farm units and grid connection constraints are transformed into the standard form of a linear programming problem, and a suitable linear programming solver is selected, such as the PuLP library in Python or CPLEX and Gurobi linear programming software; the transformed objective function and constraint conditions are input into the solver, and according to the linear programming algorithm, the optimal solution that minimizes the objective function is searched within the feasible solution space that satisfies the constraint conditions, that is, the power generation power setting values of each unit at different time points are obtained, and these values constitute the optimal scheduling strategy. The obtained power setting values can maximize the actual output power of the wind farm approaching the predicted power and reduce the deviation impact caused by inaccurate prediction; in the process of constructing the wind farm optimal scheduling method based on the dual power prediction system, it involves multiple complex links such as the analysis of historical data, the application of the accuracy evaluation model, and the consideration of different constraint conditions. The ultimate goal is to generate an optimal decision after comprehensively weighing various factors; the power setting values present the results of these complex analyses and weighings in an intuitive and executable manner, facilitating the wind farm operators to directly arrange the operation of the units and conduct power control based on these values, and effectively implementing the results of the optimal scheduling into the daily operation of the wind farm to ensure the effective implementation of the efficient optimal scheduling of the entire wind farm.

[0104] In step S400, by establishing the operating constraints of the wind farm units and the grid connection constraints, and constructing an objective function based on the power prediction accuracy evaluation results, and using the linear programming algorithm to solve the optimal scheduling strategy, a scientific scheduling decision is realized under the consideration of various factors such as the operating characteristics of the internal units of the wind farm, the grid connection requirements, and the power prediction accuracy; by constructing the objective function and solving the optimal scheduling strategy, the prediction information of the dual power prediction system is fully utilized, and the weights are reasonably allocated according to its accuracy evaluation results, aiming at minimizing the power prediction deviation, so that the power generation plan of the wind farm can be closer to the actual power demand, improving the power generation efficiency of the wind farm and reducing problems such as wind curtailment or grid scheduling difficulties caused by inaccurate power prediction, thereby enhancing the competitiveness and reliability of the wind farm in the power market and promoting the effective utilization and sustainable development of wind power resources.

[0105] Please refer to Figure 2 As shown in the figure, the technical solution of the second aspect of the present invention provides a wind farm optimal scheduling system based on a dual power prediction system, which adopts the wind farm optimal scheduling method based on the dual power prediction system provided by the technical solution of the first aspect of the present invention. The system includes:

[0106] A data acquisition module configured to extract a first power prediction data set and a second power prediction data set based on the dual power prediction system;

[0107] A feature extraction module, configured to extract the target prediction reporting rate and the power prediction accuracy rate in the first power prediction dataset and the second power prediction dataset;

[0108] An accuracy evaluation module, configured to construct an accuracy evaluation model, and extract the power prediction accuracy evaluation result based on the influence of historical data in the same period on the prediction accuracy;

[0109] A scheduling strategy optimization module, configured to output an optimal scheduling strategy with the goal of minimizing the power prediction deviation based on the power prediction accuracy evaluation result.

[0110] The technical solution of the third aspect of the present invention provides an electronic device, which includes: a processor and a memory communicatively connected to the processor; wherein, the memory stores instructions executable by the processor, and when the instructions are executed by the processor, the processor can execute the steps of the wind farm optimization scheduling method based on the dual power prediction system according to the technical solution of the first aspect of the present invention.

[0111] The technical solution of the fourth aspect of the present invention provides a computer-readable storage medium, on which a program for implementing the wind farm optimization scheduling method based on the dual power prediction system is stored, and when the program for implementing the wind farm optimization scheduling method based on the dual power prediction system is executed by a processor, the steps of the wind farm optimization scheduling method based on the dual power prediction system according to the technical solution of the first aspect of the present invention can be implemented.

[0112] It should be noted that: the above-mentioned sequence of embodiments of the present invention is only for description and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0113] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other. Each embodiment focuses on the differences from other embodiments.

Claims

1. An optimization scheduling method for a wind farm based on a dual-power prediction system, characterized in that The method comprises: Extracting a first power prediction data set and a second power prediction data set based on a dual power prediction system; Extracting a target prediction reporting rate and a power prediction accuracy rate from the first power prediction data set and the second power prediction data set; Construct an accuracy evaluation model, extract the power prediction accuracy evaluation results based on the impact of historical data on the prediction accuracy; Based on the power prediction accuracy evaluation results, the optimal scheduling strategy is output with the goal of minimizing the power prediction deviation, including: Establish wind farm unit operation constraints and grid access constraints; Construct the objective function and use the linear programming algorithm to solve the optimal scheduling strategy. The expression of the objective function is: In the formula, represents the minimization operator; represents the time period of the scheduling decision; represents the first balance coefficient, which is calculated according to the self - use weight adjustment factor of the first power prediction system; represents the second balance coefficient, which is calculated according to the self - use weight adjustment factor of the second power prediction system; represents the power prediction value of the first power prediction system at time; represents the power prediction value of the second power prediction system at time; represents the actual power measurement value at time; represents the weight given to the first power prediction system at time based on the accuracy evaluation result, ; represents the weight given to the second power prediction system at time based on the accuracy evaluation result, ; represents the accuracy evaluation result of the first power prediction system; represents the accuracy evaluation result of the second power prediction system.

2. The wind farm optimal scheduling method based on a dual-power prediction system according to claim 1, characterized in that Extracting the first power prediction data set and the second power prediction data set based on the dual power prediction system includes: Extracting a first power prediction value sequence, a second power prediction value sequence, a first meteorological data sequence, and a second meteorological data sequence from the dual power prediction system according to a preset time resolution; Performing data cleaning and alignment preprocessing on the first power prediction value sequence, the second power prediction value sequence, the first meteorological data sequence, and the second meteorological data sequence; Associating the preprocessed first power prediction value sequence and the first meteorological data sequence to obtain a first power prediction data set; The preprocessed second power prediction value sequence and the second meteorological data sequence are associated to obtain a second power prediction data set.

3. The wind farm optimal scheduling method based on a dual-power prediction system according to claim 1, characterized in that Extracting the target prediction reporting rate from the first power prediction data set and the second power prediction data set includes: Traversing the timestamp and related record information corresponding to each power prediction value in the first power prediction data set and the second power prediction data set; According to the preset reporting time interval, the actual reporting times of the dual power prediction system within the statistical period are extracted; The predicted reporting rate is extracted based on the actual number of reports of the dual power prediction system.

4. The wind farm optimal scheduling method based on the dual power prediction system according to claim 3, characterized in that, Extracting the power prediction accuracy rates from the first power prediction data set and the second power prediction data set includes: Based on the first power prediction data set and the second power prediction data set, the root mean square error between the power prediction value and the actual power measurement value at each time point is calculated respectively, and the expression is: In the formula, represents the root mean square error; represents the total number of data; represents the power prediction value at the th time point; represents the actual power measurement value at the 5. The wind farm optimal scheduling method based on the dual power prediction system according to claim 1, characterized in that Construct an accuracy evaluation model based on the impact of historical data on prediction accuracy, and extract the power prediction accuracy evaluation results including: Acquire historical data of the dual power prediction system in the same period, extract features related to the power prediction accuracy from the historical data in the same period, and construct a first feature matrix and a second feature matrix, wherein the first feature matrix and the second feature matrix at least include: meteorological data, unit operation status data, and root mean square error; The accuracy evaluation model is constructed using the random forest algorithm, and the accuracy evaluation model is trained using the first feature matrix and the second feature matrix respectively; The trained accuracy evaluation model is used to output the accuracy evaluation results of the dual power prediction system.

6. The wind farm optimal scheduling method based on the dual power prediction system according to claim 5, wherein The expression of the accuracy evaluation result of the dual power prediction system is: Wherein, represents the accuracy evaluation result of the th power prediction system; represents the self - use weight adjustment factor of the th power prediction system, ; represents the number of decision trees included in the accuracy evaluation model constructed by the th power prediction system; represents the prediction accuracy value output by the th decision tree in the th power prediction system for the input feature vector ; represents the number of features; represents the th element in the feature importance weight vector corresponding to the th power prediction system; represents the value of the th feature in the feature matrix corresponding to the th power prediction system.

7. Wind farm optimal scheduling system based on double power prediction system, characterized in that The wind farm optimization scheduling method based on the dual power prediction system according to any one of claims 1 to 6 is adopted, and the system comprises: A data acquisition module, configured to extract a first power prediction data set and a second power prediction data set based on the dual power prediction system; A feature extraction module configured to extract a target prediction reporting rate and a power prediction accuracy rate from the first power prediction data set and the second power prediction data set; An accuracy evaluation module is configured to construct an accuracy evaluation model and extract a power prediction accuracy evaluation result based on the impact of historical contemporaneous data on prediction accuracy; The scheduling strategy optimization module is configured to output the optimal scheduling strategy based on the power prediction accuracy evaluation result with the goal of minimizing the power prediction deviation.

8. An electronic device, characterized in that, The electronic device includes: a processor and a memory communicatively connected to the processor; wherein the memory stores instructions executable by the processor, and the instructions are executed by the processor so that the processor can execute the steps of the wind farm optimization scheduling method based on the dual power prediction system as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program for implementing a wind farm optimization scheduling method based on a dual power prediction system, and the program for implementing a wind farm optimization scheduling method based on a dual power prediction system is executed by a processor to implement the steps of the wind farm optimization scheduling method based on a dual power prediction system as described in any one of claims 1 to 6.

Images