Wind power distribution method based on adaptive adjustment, electronic device and medium

Abstract

The application discloses a wind power distribution method based on adaptive adjustment, an electronic device and a medium, and relates to the technical field of wind power generation. The method comprises the following steps: obtaining a wind farm ultra-short-term regulation instruction; predicting multi-unit power and health characteristics, and constructing a unit characteristic grid; performing attention-driven distribution based on a regulation target and the grid to obtain an initial scheme; constructing a twin wind farm model and deducing a wake effect to obtain a corrected group; and calling a cost model to combine the twin model to progressively optimize and obtain a power distribution parameter optimization result. The method solves the technical problem that the existing wind farm power distribution method is difficult to comprehensively consider the health state of a wind turbine generator, the wake effect and the regulation cost, resulting in low power distribution precision, insufficient overall operation efficiency and safety of the wind farm, and achieves the technical effects of realizing accurate distribution of wind farm power under the premise of meeting the grid ultra-short-term regulation demand, reducing unit fatigue loss and wake interference, and improving the overall operation efficiency and reliability of the wind farm.

Description

Technical Field

[0001] This invention relates to the field of wind power generation technology, specifically to a wind power distribution method, electronic equipment, and medium based on adaptive adjustment. Background Technology

[0002] With the continuous expansion of new energy power generation, wind farms are gradually undertaking more grid regulation tasks. In the operational scenario where wind farms participate in grid power regulation and active power control, the large number of wind turbines, their complex spatial distribution, and their significant influence from environmental factors such as wind speed and direction lead to obvious fluctuations and uncertainties in the overall power output of the wind farm. At the same time, the varying degrees of equipment health differences and wake effects among the individual wind turbines during long-term operation result in strong spatiotemporal correlations in wind power allocation.

[0003] In the ultra-short-term power regulation and power allocation control scenario of wind farms, due to the high randomness of wind resources and the obvious wake interference and differences in operating status among wind turbines, existing methods have shortcomings in performance indicators such as power allocation rationality, regulation efficiency and equipment operation reliability. The technical root cause is that existing technologies usually allocate power based only on the available power of the unit or a simple scheduling strategy, making it difficult to comprehensively consider multi-dimensional factors such as unit health status, wake effect and regulation cost, thus making it difficult to achieve coordinated optimization control among multiple units in a wind farm. Summary of the Invention

[0004] This application provides an adaptive adjustment-based wind power allocation method, electronic equipment, and medium. The key aspect is addressing the technical obstacle of wind power allocation that makes it difficult to balance regulation accuracy, equipment operating status, and overall operating efficiency in scenarios where wind farms participate in ultra-short-term power regulation and multi-unit collaborative control of the power grid. This is due to the strong spatiotemporal correlation and dynamic fluctuations in wind turbine output power caused by the randomness of wind resources, differences in unit health status, and wake coupling. By introducing an attention-driven power allocation mechanism and combining it with a twin wind farm model for wake effect deduction and progressive optimization of regulation costs, along with a data processing flow based on ultra-short-term unit characteristic grid construction using unit available power prediction and health assessment, adaptive optimization of multi-unit power allocation is achieved. This improves the accuracy of wind farm power regulation, reduces regulation costs, and enhances the overall operating efficiency and reliability of the wind farm.

[0005] The first aspect of this application provides a wind power allocation method based on adaptive adjustment, the method comprising:

[0006] The process involves: obtaining ultra-short-term regulation instructions for a wind farm, including active power regulation targets corresponding to future ultra-short-term time zones; predicting the power and health characteristics of multiple wind turbines in the wind farm based on the future ultra-short-term time zones to construct an ultra-short-term turbine characteristic grid; performing attention-driven wind power allocation on the multiple wind turbines based on the active power regulation targets and the ultra-short-term turbine characteristic grid to obtain an initial power allocation scheme; constructing a twin wind farm model and performing wake effect deduction and adjustment on the initial power allocation scheme based on the twin wind farm model to obtain a power allocation correction group; and calling a wind power regulation cost model and combining it with the twin wind farm model to progressively optimize the regulation cost target of the power allocation correction group to obtain the power allocation parameter optimization result.

[0007] A second aspect of this application provides an electronic device comprising: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute the adaptive adjustment-based wind power allocation method provided in this application.

[0008] A third aspect of this application provides a computer-readable storage medium storing a computer program for executing the adaptive adjustment-based wind power allocation method provided in this application.

[0009] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0010] First, the power regulation commands for the wind farm in the ultra-short term are obtained, and the corresponding active power regulation targets are determined. Then, based on the future timeframe, the available power and operational health status of each wind turbine in the wind farm are predicted, and a characteristic grid reflecting the turbine's operational capacity and status is constructed. Next, based on the power regulation targets and the turbine characteristic grid, power is allocated to each turbine using an attention mechanism to form an initial power allocation scheme. Then, a digital twin model of the wind farm is constructed, and the initial allocation scheme is simulated and corrected for wake effects, resulting in multiple optimized candidate schemes. Finally, combined with a wind power regulation cost model, progressive optimization calculations are performed on the candidate schemes to determine the optimal wind turbine power allocation parameters, thereby achieving adaptive and precise control of wind power allocation and effectively meeting the dynamic regulation needs of the wind farm. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a schematic diagram of the wind power allocation method based on adaptive adjustment provided in the embodiments of this application.

[0013] Figure 2 A schematic diagram of the structure of the electronic device provided in this application.

[0014] Explanation of reference numerals in the attached drawings: Bus 300, Receiver 301, Processor 302, Transmitter 303, Memory 304, Bus Interface 305. Detailed Implementation

[0015] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0016] Example 1, as Figure 1 As shown, this application provides a wind power allocation method based on adaptive adjustment, the method including:

[0017] Obtain ultra-short-term adjustment instructions from the wind farm, wherein the ultra-short-term adjustment instructions include active power adjustment targets corresponding to future ultra-short-term time zones.

[0018] In this embodiment, the ultra-short-term regulation instruction from the wind farm is first obtained. This instruction indicates the power regulation requirements that the wind farm needs to meet within a short future timeframe. The ultra-short-term timeframe can be a period of 0 to 4 hours, or it can be configured as other ultra-short-term windows according to scheduling requirements. Within this timeframe, the grid dispatch system will issue power control requests to the wind farm. The regulation instruction also includes an active power regulation target for the corresponding time period, representing the expected total active power output value that the wind farm needs to achieve within the ultra-short-term timeframe. Upon receiving this regulation instruction, it is used as the overall control basis for subsequent power prediction and power allocation calculations for each wind turbine, ensuring that the wind farm can conduct multi-unit coordinated control while meeting grid dispatch requirements.

[0019] Based on the future ultra-short-term time zone, the power and health characteristics of multiple wind turbines in the wind farm are predicted, and an ultra-short-term turbine characteristic grid is constructed.

[0020] In one embodiment, based on the future ultra-short-term time zone in the ultra-short-term regulation command, the operating capacity and equipment status of multiple wind turbines within the wind farm are comprehensively predicted within that time range. Specifically, since the power generation capacity of wind turbines is affected by random fluctuations in wind resources such as wind speed and direction, as well as differences in turbine operating status, direct power allocation can easily lead to difficulties in timely reflecting the differences in the future available power generation capacity and health status of each turbine, resulting in unreasonable power allocation or uneven turbine load. Therefore, firstly, by combining historical operating data, real-time monitoring data, and environmental parameters, the available power of each wind turbine in the future ultra-short-term time zone is predicted, thereby estimating the power generation capacity of each turbine in that time period. At the same time, the operating status parameters of each turbine are analyzed and evaluated, and its health status is predicted to obtain health characteristic information reflecting the reliability and operational risk level of the turbines. After obtaining the available power prediction results and health assessment results of each unit, and combining the spatial distribution of wind turbines in the wind farm, the relevant characteristic data are gridded, and the power capacity and health status of each unit are mapped to the corresponding spatial grid cells. This constructs an ultra-short-term unit characteristic grid that can reflect the overall operating capacity and health status of the wind farm units, providing a data basis for subsequent power allocation calculations, thereby improving the rationality of wind power allocation and the reliability of operation regulation.

[0021] Furthermore, based on the aforementioned future ultra-short-term time zone, power and health characteristics predictions are performed on multiple wind turbine units of the wind farm to construct an ultra-short-term unit characteristic grid, including:

[0022] Environmental sensing parameters and operating status parameters are collected for each wind turbine to obtain a state characteristic sequence for each turbine. Based on the future ultra-short-term time zone, the available power of the multiple wind turbines is predicted according to the state characteristic sequence of each turbine, to obtain a set of available power for each turbine. Based on the future ultra-short-term time zone, the multiple wind turbines are health assessed and predicted according to the state characteristic sequence of each turbine, to obtain a set of turbine health assessments. The turbine distribution characteristic data of the wind farm is gridded to generate a turbine distribution characteristic grid. The set of available power and the set of turbine health assessments are mapped to the turbine distribution characteristic grid to generate the ultra-short-term turbine characteristic grid.

[0023] Preferably, environmental sensing parameters and operational status parameters are first collected through environmental sensors and a unit monitoring system installed on the wind turbine. Environmental sensing parameters include data reflecting wind resource conditions such as wind speed, wind direction, air density, ambient temperature, and turbulence intensity. Operational status parameters include data reflecting the unit's operating status such as turbine speed, blade pitch angle, generator current, voltage, nacelle vibration value, bearing temperature, unit output power, and operating load. The collected multi-source data are then time-aligned and serialized according to a unified time scale to form a sequence of state characteristics for each wind turbine in a continuous time dimension.

[0024] For example, taking wind turbine No. 1 of a wind farm as an example, data is collected at 10-minute sampling intervals over a certain period of time, and the following state characteristic sequence can be obtained: At 08:00, the ambient wind speed is 7.8 m / s, the wind direction is 215°, and the air density is 1.19 kg / m³. 3 The ambient temperature was 16.3℃, the unit speed was 13.6 rpm, the blade pitch angle was 4.2°, the generator current was 485A, the nacelle vibration was 0.36 mm / s, the bearing temperature was 63.4℃, and the unit output power was 2.05 MW. At 08:10, the ambient wind speed was 8.1 m / s, the wind direction was 218°, and the air density was 1.18 kg / m³. 3 The ambient temperature was 16.5℃, the unit speed was 13.9 rpm, the blade pitch angle was 4.0°, the generator current was 502A, the nacelle vibration was 0.38 mm / s, the bearing temperature was 63.9℃, and the unit output power was 2.16 MW. At 08:20, the ambient wind speed was 8.4 m / s, the wind direction was 220°, and the air density was 1.18 kg / m³. 3 The ambient temperature was 16.6℃, the unit speed was 14.1 rpm, the blade pitch angle was 3.9°, the generator current was 518A, the nacelle vibration value was 0.37 mm / s, the bearing temperature was 64.2℃, and the unit output power was 2.28MW.

[0025] After acquiring the state characteristic sequences of each unit, the prior duration required for trend prediction is matched based on the time range of the future ultra-short-term time zone. Parameters corresponding to the available power prediction are then extracted from the state characteristic sequences of each unit according to this prior duration. These parameters include wind speed, wind direction, air density, ambient temperature, turbulence intensity, unit speed, blade pitch angle, nacelle vibration value, and bearing temperature. These parameters are then summarized to form the characteristic prediction sequences of each unit. Subsequently, the normalized characteristic prediction sequences of each unit are input into a second network for predicting available power of the units, pre-built based on a BP neural network. This second network has been trained based on historical unit characteristic datasets and historical unit available power datasets through forward propagation, loss calculation, backpropagation, and parameter optimization. It can map the newly received characteristic prediction sequences of each unit according to the learned knowledge, generating the available power of each unit, thus forming a set of available power for each unit. At the same time, based on the state characteristic sequence of each unit, the second unit health assessment network is used to perform health assessment prediction on each wind turbine, outputting the unit health assessment value corresponding to each unit, thereby obtaining the unit health assessment set, which is used to reflect the operational reliability and health level of each unit.

[0026] Subsequently, based on the spatial coordinates, arrangement direction, and spacing of each wind turbine within the wind farm, the wind farm area is divided into grids. For example, the wind farm area can be divided into multiple spatial grid units according to a preset spatial resolution, and each wind turbine can be mapped to its corresponding grid unit according to its geographical coordinates, thereby generating a turbine distribution characteristic grid. This allows each grid unit to represent the spatial distribution of turbines within a specific area. Then, the available power set and the health assessment set of each turbine are mapped to the turbine distribution characteristic grid according to the spatial location of each wind turbine. This ensures that each grid unit simultaneously contains the available power prediction information and health status assessment information of the corresponding turbine, thus forming an ultra-short-term turbine characteristic grid that reflects the power generation capacity and operational health status of each turbine in the wind farm. This provides fundamental data support for subsequent wind turbine power allocation and coordinated regulation.

[0027] For example, taking a wind farm as an example, the wind farm covers an area of ​​approximately 2km × 2km. Based on the geographical coordinates of the turbines, the wind farm area is divided into a 4×4 spatial grid. Each grid unit corresponds to a spatial region, and the predicted available power and health assessment value of the wind turbines within that region are recorded. Wind turbine unit 1, located in grid unit G(1,1), has a predicted available power of 2.32MW and a health assessment value of 0.93; wind turbine unit 2, located in grid unit G(1,2), has a predicted available power of 2.18MW and a health assessment value of 0.91; wind turbine unit 3, located in grid unit G(2,1), has a predicted available power of 2.45MW and a health assessment value of 0.95; wind turbine unit 4, located in grid unit G(2,2), has a predicted available power of 2.27MW and a health assessment value of 0.90; and wind turbine unit 5, located in grid unit G(3,2), has a predicted available power of 2.36MW and a health assessment value of 0.92.

[0028] Furthermore, based on the aforementioned future ultra-short-term time zone, a health assessment and prediction is performed on the multiple wind turbine units according to the state characteristic sequence of each unit, resulting in a unit health assessment set, including:

[0029] Based on the future ultra-short-term time zone, trend prediction is performed on the state characteristic sequences of each unit to obtain the predicted sequence of each unit characteristic; the historical unit characteristic dataset and the historical unit health assessment dataset are cleaned, aligned, and partitioned to obtain the unit health assessment training set and the unit health assessment validation set; the BP neural network is trained under supervision based on the unit health assessment training set to obtain the first network for unit health assessment; the first network for unit health assessment is validated and optimized based on the unit health assessment validation set to obtain the second network for unit health assessment; the predicted sequence of each unit characteristic is input into the second network for unit health assessment to generate the unit health assessment set.

[0030] Preferably, based on the future ultra-short-term time zone, the same method described above is used to extract the characteristic prediction sequence of each unit from the state characteristic sequence of each unit for trend prediction. While obtaining the predicted sequences of each unit's characteristics, historical unit characteristic datasets and historical unit health assessment datasets are extracted from historical operating data. The historical unit characteristic data then undergoes data cleaning, including removing abnormal sensor data, eliminating duplicate records, filling in missing data, and smoothing abnormal data abrupt changes. Next, data from different sources are timestamped to establish a one-to-one correspondence between unit characteristic data and corresponding time-based unit health assessment data. The cleaned data is then normalized to ensure all feature variables are within a uniform numerical range. The processed historical unit characteristic dataset and historical unit health assessment dataset are then divided according to a preset ratio, such as 7:3 or 8:2, to generate a unit health assessment training set and a unit health assessment validation set. Both the training and validation sets include normalized wind speed, wind direction, air density, ambient temperature, turbulence intensity, unit speed, blade pitch angle, generator current, generator voltage, output power, nacelle vibration value, bearing temperature, and previously assigned unit health assessment values.

[0031] Subsequently, a backpropagation (BP) neural network structure was designed and its parameters initialized. This BP neural network includes an input layer, several hidden layers, and an output layer. The number of nodes in the input layer is equal to the dimension D of the input vector, i.e., the dimension after concatenating all features. The hidden layers are set to a multi-layer fully connected structure, for example, three hidden layers with 256, 128, and 64 nodes respectively, used to extract nonlinear features from low-order characteristics to high-order health representations layer by layer. The activation function of the hidden layers can be ReLU to alleviate gradient vanishing and accelerate convergence. The output layer selects different configurations according to the label format: when the health assessment label is a continuous unit health assessment value, the output layer has one node and uses linear activation; when the label is a health level classification, the number of nodes in the output layer is equal to the number of levels C and Softmax is used to output the probability of each level. During the parameter initialization stage, the weights of each fully connected layer are initialized using Xavier or He to ensure the stability of the variance of the forward propagation, and the bias term is initialized to 0, thereby improving the convergence stability in the early stage of training. Next, the BP neural network is trained under supervision using the unit health assessment training set. Specifically, the training samples are input into the network in batches, for example, batch size B=64. For any batch of input vectors, the network performs forward propagation sequentially. During this process, the first hidden layer performs a linear transformation on the input and applies activation to obtain the hidden representation. The second hidden layer continues to... Mapped to more abstract health-related features The third hidden layer is based on Obtaining compression characterization The output layer outputs predicted values, where the regression task outputs the unit health assessment value, and the classification task outputs the probability vector.

[0032] After obtaining the predicted output, if it is a continuous value, mean squared error loss is used; if it is a health level classification, cross-entropy loss is used. The loss value is propagated forward using the gradient according to the chain rule, starting from the output layer. The gradient of the weight matrix and bias with respect to the loss of each layer is calculated sequentially. The gradient reflects how each parameter should be fine-tuned to reduce the loss of the current batch. The optimizer can use Adam or SGD+momentum. For example, using Adam with a learning rate of 0.001, each parameter is adaptively updated based on the first and second moment estimates, so that the network gradually learns the mapping law from state characteristics to health assessment. The above process of forward propagation, loss calculation, backpropagation, and parameter update is repeated for all batches to complete one epoch. Training continues until the maximum number of training epochs is reached or the convergence condition is met, thus obtaining the first network for crew health assessment. During training, the performance of the first network for crew health assessment is monitored and overfitting is prevented based on the crew health assessment validation set, so as to optimize the first network to obtain the second network for crew health assessment. Specifically, after every certain number of epochs, the validation set is input into the current network, and the validation loss and key evaluation metrics are calculated. For example, MAE or RMSE is calculated for regression tasks, and accuracy, F1 score, or AUC is calculated for classification tasks. If the validation loss does not decrease for P consecutive epochs, such as 30 epochs, the network is considered to have entered an overfitting or learning stagnation phase. At this point, a network optimization strategy is triggered. On the one hand, regularization is enabled or enhanced, for example, by adding Dropout (e.g., 0.2) to the hidden layers to randomly deactivate some neurons, or by increasing the L2 regularization coefficient to suppress excessive weights. On the other hand, training hyperparameters are adjusted, for example, by proportionally decaying the learning rate (e.g., multiplying it by 0.5) to refine convergence. If necessary, the network capacity is adjusted, for example, by reducing the number of nodes in the first hidden layer from 256 to 192 to reduce complexity, or by using an early stopping mechanism to directly retain the network parameters with the minimum validation loss as the final network, thereby obtaining a second network for crew health assessment with better and more stable generalization performance on the validation set.

[0033] After training, the characteristic prediction sequence of each wind turbine at each time node in the future ultra-short time zone is sequentially input into the second health assessment network. The network calculates the health assessment results for the corresponding time nodes. These health assessment results include the unit health assessment values ​​of each unit, providing unit health constraints and optimization basis for the subsequent power allocation process, so as to achieve a health constraint allocation that takes into account both regulation objectives and unit reliability.

[0034] Based on the active power regulation target and the ultra-short-term unit characteristic grid, attention-driven wind power allocation is performed on the multiple wind turbine units to obtain an initial power allocation scheme.

[0035] In one embodiment, after obtaining the active power regulation target and the ultra-short-term unit characteristic grid for the future ultra-short-term time zone, the regulation target is used as a constraint on the overall output of the wind farm, and the unit available power prediction information and health assessment information contained in the unit characteristic grid are used as the allocation basis to perform attention-driven power allocation calculations on multiple wind turbines in the wind farm. Since the available power generation capacity and operational health status of each wind turbine in the wind farm differ, and wind resource conditions have significant time-varying and uncertainties, if only average allocation or fixed-ratio allocation is used for power regulation, it is easy to cause some units to bear excessive loads or units with poor health status to operate excessively, thereby leading to unreasonable power allocation and increased unit operation risks. Therefore, adaptive power allocation is required in combination with unit operation characteristics. Specifically, firstly, the available power and health status of each wind turbine in the future ultra-short-term time zone are extracted from the ultra-short-term unit characteristic grid and encoded into unit feature vectors to characterize the power generation capacity, operational risks, and adjustment potential of each unit in that time period. Subsequently, an attention calculation mechanism is constructed, weighting the contribution of available power and the contribution of health and reliability separately. This allows the attention weights to adaptively reflect the priority allocation of different units under the current regulation task. Units with higher available power and better health status receive higher attention weights, while those with lower available power or poorer health status receive lower weights or are subject to restricted allocation. After obtaining the attention weights of each unit, the weights of multiple units are normalized and combined with the active power regulation target of the wind farm. The target total power is then allocated to each wind turbine according to the weight ratio, generating the initial active power command value for each unit in the future ultra-short-term time zone, thus forming the initial power allocation scheme. Through this method, the power allocation process can fully reflect the differences in generating capacity and health status of the units, achieving adaptive and collaborative allocation among multiple units. This improves the rationality and efficiency of power allocation while meeting the overall regulation target of the wind farm, and reduces the operational risks of the units. It also provides a basic input for subsequent wake extrapolation correction and cost optimization.

[0036] Furthermore, based on the active power regulation target and the ultra-short-term unit characteristic grid, attention-driven wind power allocation is performed on the multiple wind turbines to obtain an initial power allocation scheme, including:

[0037] Based on the ultra-short-term unit characteristic grid, the available power attention of the multiple wind turbines is allocated to obtain an available power attention grid; based on the ultra-short-term unit characteristic grid, the healthy attention of the multiple wind turbines is allocated to obtain a healthy attention grid; based on the available power attention weight and the healthy attention weight, the available power attention grid and the healthy attention grid are jointly calculated to obtain a collaborative attention grid; based on the active power regulation target, the power of the multiple wind turbines is allocated according to the collaborative attention grid to generate the initial power allocation scheme.

[0038] Preferably, the predicted available power of each wind turbine in the future ultra-short time zone is first extracted from the ultra-short-term unit characteristic grid. Let the available power of the i-th wind turbine be... Next, the total available power of all units is calculated, and the available power attention value of each unit is determined by a percentage calculation method. This calculation reflects the contribution of each unit to the overall power generation capacity. Subsequently, based on the spatial coordinates of each wind turbine in the wind farm, the corresponding available power attention value is mapped to the corresponding grid cell in the unit distribution grid, thus forming the available power attention grid. Then, the same process is used for the unit health assessment values ​​corresponding to each wind turbine in the ultra-short-term unit characteristic grid; that is, the sum of all unit health assessment values ​​is calculated, and the health attention value of each unit is obtained by a percentage calculation method to reflect the impact of unit operating health on power distribution. Finally, based on the spatial location of the units, the health attention value of each unit is mapped to the corresponding grid cell, thus forming the health attention grid. After obtaining the available power attention grid and the health attention grid, the two types of attention information are calculated collaboratively. That is, based on the available power attention weight and the health attention weight, the available power attention value and the health attention value corresponding to each unit are weighted and fused to generate a collaborative attention value. The available power attention weight and the health attention weight are preset weight coefficients, and the sum of the available power attention weight and the health attention weight is 1. By adjusting the values ​​of the available power attention weight and the health attention weight, the influence ratio of power generation capacity factors and health status factors in power allocation can be controlled. By mapping the collaborative attention values ​​calculated by each unit to the unit distribution grid according to the spatial location of the unit, a collaborative attention grid is formed. Then, the sum of the collaborative attention values ​​of all units in the collaborative attention grid is calculated. Based on the proportion of each unit's collaborative attention value in the total collaborative attention, the active power regulation target is proportionally allocated. This allows the calculation of the power allocation results for each unit in the future ultra-short time zone. The allocated power of all units is then combined to form an initial power allocation scheme, providing a basic scheme for subsequent wake effect correction and regulation cost optimization. This enables the power allocation process to simultaneously consider the unit's power generation capacity and operational health status, improving the rationality of wind power allocation and the reliability of operation regulation.

[0039] A twin wind farm model is constructed, and the initial power allocation scheme is adjusted based on the wake effect of the twin wind farm model to obtain the power allocation correction group.

[0040] In one embodiment, after obtaining the initial power allocation scheme for the wind turbines, a twin wind farm model corresponding to the actual operating state of the wind farm is constructed to further consider the wake effect caused by airflow interference between the wind turbines. Since the wind turbines in a wind farm are typically distributed in an array in space, the operation of upstream turbines will generate a wake effect on downstream turbines, leading to a decrease in inflow wind speed and an increase in turbulence intensity, thus affecting their actual output power. If the wake coupling relationship between the turbines is not fully considered during the power allocation process, it can easily lead to deviations between the initial power allocation scheme and the actual operating conditions, resulting in insufficient output from some turbines or a decrease in overall power regulation accuracy. Therefore, it is necessary to analyze and correct the wake effect of the power allocation scheme through a simulation model. Specifically, based on fundamental information such as the spatial location, spacing, arrangement, and terrain of each wind turbine within the wind farm, and combined with real-time or predicted environmental parameters such as wind speed, wind direction, air density, and turbulence intensity, a twin wind farm model corresponding to the actual wind farm's operating characteristics is established in the computing platform. This model dynamically reflects the coupling relationship between wind energy flow and turbine operating status within the wind farm. After completing the twin wind farm model construction, the initial power allocation scheme is loaded into the twin wind farm model as control input, and the wind farm operation process is simulated and extrapolated within the model environment. The wake effect of upstream wind turbines on downstream turbines is calculated, thus generating wake effect simulation data. Subsequently, the initial power allocation scheme is adjusted based on the turbine wake propagation network constructed based on the wake effect simulation data. For example, for downstream turbines affected by strong wake effects, their power allocation values ​​can be appropriately reduced, while for turbines less affected by wake effects and with sufficient available power, their power allocation values ​​can be appropriately increased. By performing multiple simulations and adjustments to the power allocation parameters in a twin wind farm model, several power allocation schemes corrected for wake effects can be generated. Finally, these multiple power allocation schemes corrected for wake effects are summarized to form a power allocation correction group. This makes the power allocation schemes more consistent with the actual operating characteristics of wind farms, improving the accuracy and feasibility of power allocation, and providing a set of candidate schemes for subsequent optimization based on the adjustment cost model.

[0041] Furthermore, based on the twin wind farm model, the initial power allocation scheme is adjusted using wake effect deduction to obtain a power allocation correction group, including:

[0042] The initial power allocation scheme is simulated using the twin wind farm model to obtain wake effect simulation data. The wake effect simulation data is then used to identify the influence characteristics of the multiple wind turbines, resulting in a wake turbine influence characteristic set. Based on this wake turbine influence characteristic set, spatial topology associations are performed between the turbines to construct a wake propagation network. Finally, the initial power allocation scheme is corrected using this wake propagation network to generate the power allocation correction group.

[0043] Preferably, the target active power of each unit in the initial power allocation scheme is first determined. As the control input to the twin wind farm model, it performs time-series simulations at preset time steps within a future ultra-short-term time zone. At each time step, the twin wind farm model outputs wake-related simulation data, including the downstream wake velocity or wind speed field distribution for each turbine, the wake diffusion boundary, and the inflow wind speed for each turbine. Compared to the free-flow wind speed when not affected by the wake The simulation outputs for each time step are summarized to obtain the wake effect simulation data. Then, for each upstream unit i, the wake diffusion boundary is extracted from the wake effect simulation data according to a preset key name, serving as the wake influence range. This refers to the spatial region where the wake velocity decreases beyond a preset threshold; from which the inflow velocity of each unit is extracted. Compared to the free-flow wind speed when not affected by the wake The difference between the values ​​is taken as the wake wind speed attenuation; the achievable power limit of each unit under wake conditions is extracted from this, serving as the available power limit. These influencing characteristics are summarized to form a wake turbine influence characteristic set. Then, the spatial topology association of the units is performed based on this wake turbine influence characteristic set. Specifically, each wind turbine in the wind farm is considered a network node. When the wake influence characteristic identification results indicate the wake influence range of upstream unit i... When the rotor sweep area of ​​downstream unit j is covered, or when the wake velocity attenuation exceeds a preset threshold, a directed edge is established between node i and node j, indicating that the wake propagates from the upstream unit to the downstream unit. Furthermore, a weight is assigned to the directed edge to quantify the wake influence intensity; this weight can be obtained by weighted fusion of wind velocity attenuation and power loss. Through the construction of the nodes, edges, and weights described above, a unit wake propagation network reflecting the wake propagation path, influence direction, and influence strength is obtained.

[0044] Then, based on the out-degree, in-degree, and edge weight distribution of each node in the wake propagation network, the units with strong wake influence and those with strong wake impact are identified, and the initial power allocation scheme is adjusted accordingly. Specifically, for units with strong wake influence, for example, those with high-weight outgoing edges to multiple downstream units, their allocated power is reduced according to a preset reduction ratio or according to the edge weight, resulting in the reduction amount. To mitigate the wake effect, for heavily affected turbines, such as those with high-weighted incoming edges and power losses greater than zero, their allocated power is limited to their actual generating capacity. Next, the reduced power and the reduced power demand from affected turbines are redistributed according to a strategy to turbines with less wake impact, larger actual generating capacity margins, and better health conditions, ensuring that the corrected total power still meets the overall active power regulation target of the wind farm. Finally, to form a power allocation correction group, multiple candidate schemes can be generated using different combinations of correction strategy parameters for the above feedback correction process. For example, different reduction ratios, different edge weight thresholds, and different power backfill allocation rules can be set. Each generated corrected turbine power allocation result constitutes a power allocation correction scheme. The collection of multiple power allocation correction schemes forms a power allocation correction group, providing a candidate solution space for subsequent progressive optimization of the regulation cost target. Through the above process, the wake propagation relationship between wind turbine units and its impact on the actual power generation capacity of the units are fully considered on the basis of the initial power allocation scheme. This enables the power allocation scheme to more accurately reflect the airflow coupling characteristics and unit operating status within the wind farm, thereby effectively reducing the negative impact of wake effect on the overall power generation efficiency of the wind farm and improving the consistency between the power allocation results and the actual operating conditions.

[0045] The wind power regulation cost model is invoked, and the regulation cost objective of the power allocation correction group is progressively optimized by combining the twin wind farm model to obtain the optimization results of the power allocation parameters.

[0046] In one embodiment, after obtaining the power allocation correction group, to further determine the optimal wind turbine power allocation scheme, the wind power regulation cost model is invoked, and the regulation cost objective of the power allocation correction group is progressively optimized using a twin wind farm model. Since different power allocation correction schemes will generate varying degrees of regulation costs such as turbine fatigue loss, power loss, and wind curtailment loss during actual wind farm operation, selecting a power allocation scheme based solely on a single indicator or simple rule may result in a scheme that, while meeting power regulation requirements, is not optimal in terms of overall operating cost and turbine collaborative operation efficiency. Therefore, a comprehensive cost evaluation method is needed to optimize and screen candidate schemes. Specifically, each power allocation correction scheme in the power allocation correction group is input as a candidate control scheme into the twin wind farm model, and simulation calculations are used to obtain the overall operating status of the wind farm and the operating data of each wind turbine under that power allocation scheme. Subsequently, the simulation results are input into the wind power regulation cost model to evaluate the regulation cost of each candidate power allocation scheme. This model calculates the comprehensive cost incurred by the wind farm when implementing the corresponding power regulation strategy. This cost comprehensively considers factors such as changes in unit operating load, power loss, wind curtailment loss, and unit operating stability, thereby obtaining the corresponding regulation cost. After completing the cost evaluation of all candidate schemes, the power allocation correction group is progressively optimized and screened according to the preset regulation cost target. For example, a set of candidate schemes that meet the constraints can be selected first based on the basic cost constraints, and then the comprehensive regulation cost is further compared among the candidate schemes that meet the constraints to select the power allocation scheme with the lowest cost. Finally, the optimal power allocation scheme obtained after progressive optimization is determined as the power allocation parameter optimization result and used as the final power regulation command for each wind turbine in the future ultra-short time zone, thereby realizing the optimized control of wind farm power allocation and improving the scientific nature of power allocation decision-making and overall operating efficiency.

[0047] Furthermore, the wind power regulation cost model is invoked, and combined with the twin wind farm model, the regulation cost objective of the power allocation correction group is progressively optimized to obtain the power allocation parameter optimization results, including:

[0048] A first regulation cost objective and a second regulation cost objective are obtained. The first regulation cost objective includes wind power fatigue cost constraints, active power loss cost constraints, and wind curtailment loss cost constraints. The second regulation cost objective includes minimizing the wind power regulation coordination cost. Based on the first regulation cost objective, the power allocation correction group is traversed and optimized according to the wind power regulation cost model and the twin wind farm model to establish an allocation correction optimization group. The coordination cost of the allocation correction optimization group is calculated according to the wind power regulation coordination cost weight to obtain the wind power regulation coordination cost distribution. Based on the wind power regulation coordination cost distribution, the allocation correction optimization group is iteratively optimized according to the second regulation cost objective to generate the power allocation parameter optimization result.

[0049] Preferably, a first adjustment cost objective and a second adjustment cost objective are first obtained. The first adjustment cost objective is used to constrain and screen candidate power allocation schemes, including wind power fatigue cost constraints, active power loss cost constraints, and wind curtailment loss cost constraints. The second adjustment cost objective is used to further achieve comprehensive optimization based on satisfying the constraints, including minimizing the wind power regulation coordination cost. Subsequently, candidate power allocation correction schemes are selected one by one from the power allocation correction group and input into a twin wind farm model for simulation. The corresponding unit operation data and wind farm operation data are obtained, and then the simulation data are input into the wind power regulation cost model to calculate the wind power fatigue cost, active power loss cost, and wind curtailment loss cost corresponding to the scheme. When the three costs satisfy the constraint conditions of the first adjustment cost objective, i.e., all do not exceed the corresponding threshold, the candidate power allocation correction scheme is retained and added to the allocation correction optimization group, thereby forming a set of candidate schemes that satisfy the basic cost constraints. After obtaining the allocation correction optimization group, for each candidate scheme in the allocation correction optimization group, based on its corresponding fatigue cost, active power loss cost, and wind curtailment loss cost, a weighted fusion is performed according to the preset coordination cost weight to obtain the wind power regulation coordination cost coefficient. This wind power regulation coordination cost coefficient can be expressed as: ;in, The cost of wind power fatigue; Cost of active power loss; The cost of wind curtailment; As a weighting factor for wind power fatigue costs, Weighting of active power loss cost. As the weight of wind curtailment loss cost, and By statistically organizing the wind power regulation coordination cost coefficients calculated from all candidate schemes within the allocation correction optimization group, a wind power regulation coordination cost distribution is formed, which reflects the comprehensive regulation cost level of different candidate schemes.

[0050] Subsequently, in the allocation correction optimization group that satisfies the first regulation cost objective constraint, the optimization criterion is to minimize the cooperative cost. Candidate schemes are sorted according to their cooperative costs from high to low or low to high, and the scheme with the lowest cooperative cost is taken as the current optimal solution. Simultaneously, this optimal solution can be iteratively adjusted within the power allocation parameter space. For example, small-step perturbations are applied to the allocation power of key units to generate new candidate schemes, and their costs are calculated again using the twin wind farm model and the regulation cost model. If the cooperative cost is lower and still satisfies the first regulation cost objective constraint, the optimal solution is updated. After multiple rounds of iterative optimization, the power allocation scheme with the minimum cooperative cost that satisfies the constraints is finally output, and its corresponding unit power allocation parameters are determined as the power allocation parameter optimization result. This achieves the minimization of the comprehensive regulation cost of the wind turbine power allocation scheme under the condition of satisfying operational constraints, improving the economy and operational coordination efficiency of wind farm power regulation.

[0051] Furthermore, based on the first adjustment cost objective, the power allocation correction group is traversed and optimized according to the wind power adjustment cost model and the twin wind farm model to establish an allocation correction optimization group, including:

[0052] The q-th power allocation correction scheme is extracted from the power allocation correction group, where q is a positive integer; the twin wind farm model is simulated and controlled according to the q-th power allocation correction scheme to obtain the q-th power allocation simulation data; the wind power regulation cost model is activated, and the wind power regulation cost analysis index of the wind power regulation cost model includes wind power fatigue cost, active power loss cost, and wind curtailment loss cost; the q-th power allocation simulation data is input into the wind power regulation cost model to obtain the q-th wind power regulation cost sequence; if the q-th wind power regulation cost sequence satisfies the first regulation cost objective, the q-th power allocation correction scheme is added to the allocation correction optimization group.

[0053] Optionally, the candidate schemes in the power allocation correction group are first numbered sequentially, and the q-th candidate power allocation correction scheme is selected. This scheme includes the power allocation parameters of each wind turbine in the future ultra-short time zone. Then, the target power of each turbine at each time node in the q-th power allocation correction scheme is loaded into the twin wind farm model as a control variable. Calculations are performed within a preset simulation step size to obtain the simulation operation data of the wind farm when implementing the scheme. This q-th power allocation simulation data includes the actual output power sequence of each turbine, key operating state sequences such as turbine speed and pitch angle, effective wind speed sequence under the influence of wind farm inflow and wake, total wind farm output power sequence, and relevant data on electrical side active power loss, providing an input basis for cost calculation. Next, the wind power regulation cost model is activated. This model is used to convert the simulation operation data after the scheme is implemented into quantifiable cost analysis indicators, specifically including wind power fatigue cost, active power loss cost, and wind curtailment loss cost.

[0054] When calculating the fatigue cost of wind power, the unit output power in the q-th power distribution simulation data is first used to calculate the change in unit power at adjacent time points. The absolute value of the power change amplitude is used to approximate the change in mechanical load. Then, the fatigue damage amount within the time period is obtained by accumulating the change in mechanical load throughout the entire adjustment cycle. Next, the fatigue damage amount of each unit is accumulated and multiplied by the fatigue cost conversion factor to obtain the fatigue cost of the wind farm, which is used to reflect the equipment life consumption caused by unit adjustment.

[0055] When calculating the cost of active power loss, the unit output power, line voltage, and line resistance are first extracted from the q-th power distribution simulation data. The line current is calculated by dividing the unit output power by the line voltage. The line power loss is then obtained by multiplying the square of the line current by the line resistance. The line power loss is then accumulated to obtain the total wind farm loss. The time-period loss is obtained by accumulating the product of the total wind farm loss and the duration over the entire regulation cycle. The time-period loss is then multiplied by the loss cost coefficient to obtain the cost of active power loss, which reflects the power loss generated by transmission lines, electrical equipment, etc. within the wind farm.

[0056] When calculating the cost of wind curtailment losses, the total available power and actual grid-connected power of the wind farm at each moment within the regulation period are first extracted from the power distribution simulation data of the qth power generation period. The total available power is obtained by summing the available power of each unit at that moment, and the actual grid-connected power is obtained by summing the actual output power of each unit at that moment. Then, the difference between the total available power and the actual grid-connected power is calculated, and the part of the difference that is not less than zero is taken to obtain the wind curtailment power at each moment. Next, the product of the wind curtailment power at each moment within the entire regulation period and the corresponding time step is accumulated to obtain the wind curtailment electricity within the regulation period. Finally, the wind curtailment electricity is multiplied by the wind curtailment economic loss coefficient to obtain the cost of wind curtailment losses, which reflects the loss caused by the unutilized generated power due to regulation-limited generation or operational constraints.

[0057] After analyzing and calculating the simulated data for the q-th power allocation, the wind power regulation cost model outputs the q-th wind power regulation cost sequence, which corresponds to the wind power fatigue cost coefficient, active power loss cost coefficient, and wind curtailment loss cost coefficient, respectively. These coefficients characterize the comprehensive cost level of the candidate scheme in three dimensions: equipment lifespan consumption, electrical losses, and wind curtailment losses. Finally, each coefficient in the q-th wind power regulation cost sequence is compared item by item with the preset fatigue cost constraint threshold, active power loss cost constraint threshold, and wind curtailment loss cost constraint threshold in the first regulation cost objective. When all three cost coefficients meet the constraints, the scheme is determined to be a feasible candidate scheme and added to the allocation correction optimization group for subsequent collaborative cost calculation and iterative optimization. This achieves effective screening of candidate power allocation schemes, reduces the number of schemes that do not meet the constraints from entering the subsequent optimization process, and improves the efficiency and reliability of power allocation optimization results.

[0058] Furthermore, the wind power regulation coordination cost weight includes wind power fatigue cost weight, active power loss cost weight, and wind curtailment loss cost weight.

[0059] Optionally, the wind power regulation coordination cost weight is used to characterize the importance of different regulation cost indicators in the comprehensive regulation cost assessment. When calculating the comprehensive cost of the power allocation scheme, it is necessary to consider multiple factors such as the fatigue level of wind turbine operation, active power loss on the electrical side, and wind curtailment loss. However, the impact of different factors on the wind farm's operating objectives varies. Therefore, corresponding weight parameters are set to weight various costs. Among them, the wind power fatigue cost weight is used to reflect the impact of turbine fatigue loss on the comprehensive cost. When this weight is high, it indicates that more emphasis is placed on reducing turbine operating load changes and equipment fatigue during the power allocation optimization process to extend the service life of wind turbines. The active power loss cost weight is used to reflect the importance of electrical losses within the wind farm on the comprehensive cost. When this weight is high, it indicates that more attention is paid to reducing active power losses in lines and equipment during the power allocation process to improve the overall power generation efficiency of the wind farm. The wind curtailment loss cost weight is used to reflect the impact of wind curtailment loss on the overall regulation cost. When this weight is high, it indicates that more emphasis is placed on reducing the waste of usable wind energy and improving wind energy utilization during the power allocation process. In practical applications, the three types of weights mentioned above can be set according to the wind farm's operation strategy or dispatch requirements, so that the weights of wind power fatigue cost, active power loss cost, and wind curtailment loss cost meet preset normalization conditions, such as the sum of the three being 1. Through this method, a balance can be achieved among multiple adjustment objectives in the comprehensive cost assessment, thereby providing a more reasonable evaluation basis for subsequent power allocation optimization.

[0060] In summary, the embodiments of this application have at least the following technical effects:

[0061] First, the ultra-short-term regulation instructions for the wind farm are obtained, including the active power regulation target corresponding to the future ultra-short-term time zone. Then, based on the future ultra-short-term time zone, power and health characteristics of multiple wind turbines in the wind farm are predicted, and an ultra-short-term turbine characteristic grid is constructed. Next, attention-driven wind power allocation is performed on the multiple wind turbines based on the active power regulation target and the ultra-short-term turbine characteristic grid to obtain an initial power allocation scheme. Then, a twin wind farm model is constructed, and the wake effect is extrapolated and adjusted based on the twin wind farm model to obtain a power allocation correction group. Finally, the wind power regulation cost model is invoked, and combined with the twin wind farm model, the regulation cost target of the power allocation correction group is progressively optimized to obtain the power allocation parameter optimization result. This invention addresses the technical problem that existing wind farm power allocation methods struggle to comprehensively consider the health status of wind turbine units, wake effects, and regulation costs, resulting in low power allocation accuracy and insufficient overall operating efficiency and safety of wind farms. It achieves the technical effect of accurately allocating wind farm power while meeting the ultra-short-term regulation needs of the power grid, reducing turbine fatigue losses and wake interference, and improving the overall operating efficiency and reliability of wind farms.

[0062] Example 2: Based on the inventive concept of the adaptive adjustment-based wind power allocation method in the foregoing embodiments, this application also provides an electronic device, including: at least one processor; a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the adaptive adjustment-based wind power allocation method described in any one of Examples 1 above.

[0063] Appendix Figure 2 This is a schematic diagram of the structure of an exemplary electronic device of this application. Figure 2 In this document, the bus architecture is represented by bus 300. Bus 300 may include any number of interconnected buses and bridges, and bus 300 connects various circuits including one or more processors represented by processor 302 and memory represented by memory 304. Bus 300 may also connect various other circuits such as peripheral electronics, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. Bus interface 305 provides an interface between bus 300 and receiver 301 and transmitter 303. Receiver 301 and transmitter 303 may be the same element, i.e., a transceiver, providing a unit for communicating with various other devices over a transmission medium. Processor 302 is responsible for managing bus 300 and general processing, while memory 304 can be used to store data used by processor 302 during operation.

[0064] In Embodiment 3, based on the adaptive adjustment-based wind power allocation method in the foregoing embodiments and using the same inventive concept, this application also provides a computer-readable storage medium storing a computer program, which, when executed, implements the steps of the adaptive adjustment-based wind power allocation method described in any one of Embodiment 1.

[0065] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A wind power allocation method based on adaptive adjustment, characterized in that, The method includes: Obtain ultra-short-term adjustment instructions from the wind farm, wherein the ultra-short-term adjustment instructions include active power adjustment targets corresponding to future ultra-short-term time zones; Based on the future ultra-short-term time zone, the power and health characteristics of multiple wind turbines in the wind farm are predicted, and an ultra-short-term turbine characteristic grid is constructed. Based on the active power regulation target and the ultra-short-term unit characteristic grid, attention-driven wind power allocation is performed on the multiple wind turbine units to obtain an initial power allocation scheme; Construct a twin wind farm model, and perform wake effect deduction and adjustment on the initial power allocation scheme based on the twin wind farm model to obtain the power allocation correction group; The wind power regulation cost model is invoked, and combined with the twin wind farm model, the regulation cost objective of the power allocation correction group is progressively optimized to obtain the optimization results of the power allocation parameters. This includes: obtaining a first regulation cost objective and a second regulation cost objective. The first regulation cost objective is used to constrain and screen candidate power allocation correction schemes, including wind power fatigue cost constraints, active power loss cost constraints, and wind curtailment loss cost constraints. The second regulation cost objective is used to further achieve comprehensive optimization based on satisfying the constraints, including minimizing the wind power regulation coordination cost coefficient. Candidate power allocation correction schemes are selected one by one from the power allocation correction group and input into the twin wind farm model for simulation and deduction to obtain the corresponding unit operation data and... Wind farm operation data is used, and simulation data is then input into the wind power regulation cost model to calculate the wind power fatigue cost, active power loss cost, and wind curtailment loss cost corresponding to the proposed scheme. When the three costs meet the constraint conditions of the first regulation cost objective, i.e., none of them exceed the corresponding threshold, the candidate power allocation correction scheme is retained and added to the allocation correction optimization group, thus forming a set of candidate schemes that meet the basic cost constraints. After obtaining the allocation correction optimization group, for each candidate scheme in the allocation correction optimization group, based on its corresponding wind power fatigue cost, active power loss cost, and wind curtailment loss cost, a weighted fusion is performed according to the preset collaborative cost weight to obtain the wind power regulation collaborative cost coefficient. This wind power regulation collaborative cost coefficient can be expressed as: ;in, The cost of wind power fatigue; Cost of active power loss; The cost of wind curtailment; As a weight for wind power fatigue costs, Weighting of active power loss cost. As the weight of wind curtailment loss cost, and .

2. The wind power allocation method based on adaptive adjustment as described in claim 1, characterized in that, Based on the aforementioned future ultra-short-term time zone, power and health characteristics of multiple wind turbines in the wind farm are predicted, and an ultra-short-term turbine characteristic grid is constructed, including: Environmental sensing parameters and operating status parameters are collected for each wind turbine to obtain the status characteristic sequence of each unit; Based on the future ultra-short-term time zone, the available power of the multiple wind turbines is predicted according to the state characteristic sequence of each unit, and the set of available power of the units is obtained. Based on the future ultra-short-term time zone, a health assessment and prediction is performed on the multiple wind turbines according to the state characteristic sequence of each unit to obtain a set of unit health assessments. The wind farm's turbine distribution characteristic data is processed into a grid to generate a turbine distribution characteristic grid. The available power set of the units and the health assessment set of the units are mapped to the distributed characteristic grid of the units to generate the ultra-short-term unit characteristic grid.

3. The wind power allocation method based on adaptive adjustment as described in claim 2, characterized in that, Based on the aforementioned future ultra-short-term time zone, a health assessment and prediction is performed on the multiple wind turbine units according to the state characteristic sequence of each unit, resulting in a unit health assessment set, including: Based on the future ultra-short-term time zone, the trend of the state characteristic sequence of each unit is predicted to obtain the characteristic prediction sequence of each unit. The historical unit characteristic dataset and the historical unit health assessment dataset are cleaned, aligned and divided to obtain the unit health assessment training set and the unit health assessment validation set. The BP neural network is trained under supervision based on the unit health assessment training set to obtain the first network for unit health assessment. The first network for unit health assessment is validated and optimized based on the unit health assessment validation set to obtain the second network for unit health assessment. The predicted sequence of each unit characteristic is input into the second network of the unit health assessment to generate the set of unit health assessments.

4. The wind power allocation method based on adaptive adjustment as described in claim 1, characterized in that, Based on the active power regulation target and the ultra-short-term unit characteristic grid, attention-driven wind power allocation is performed on the multiple wind turbines to obtain an initial power allocation scheme, including: Based on the ultra-short-term unit characteristic grid, the available power attention of the multiple wind turbine units is allocated to obtain the available power attention grid; Based on the ultra-short-term unit characteristic grid, healthy attention is allocated to the multiple wind turbine units to obtain a healthy attention grid. The available power attention grid and the healthy attention grid are jointly calculated based on the available power attention weight and the healthy attention weight to obtain a collaborative attention grid; Based on the active power regulation target, the power is allocated to the multiple wind turbine units according to the collaborative attention grid, and the initial power allocation scheme is generated.

5. The wind power allocation method based on adaptive adjustment as described in claim 1, characterized in that, Based on the twin wind farm model, the initial power allocation scheme is analyzed and adjusted for wake effects to obtain a power allocation correction group, including: Based on the twin wind farm model, the wake effect of the initial power allocation scheme is simulated to obtain wake effect simulation data; The wake effect simulation data is used to identify the influence characteristics of the multiple wind turbine units and obtain the wake effect influence characteristic set of the turbine units. Based on the set of wake unit influence characteristics, spatial topology association of the units is performed to construct the wake propagation network of the units. The initial power allocation scheme is corrected by feedback based on the unit wake propagation network to generate the power allocation correction group.

6. An electronic device, characterized in that, The electronic device includes: processor; Memory used to store the processor's executable instructions; The processor is used to execute the adaptive adjustment-based wind power allocation method according to any one of claims 1 to 5.

7. A computer-readable storage medium, characterized in that, The storage medium stores a computer program for executing the adaptive adjustment-based wind power allocation method according to any one of claims 1 to 5.

Images