Fan blade deicing power regulation method based on sparse bayesian control

Abstract

The application discloses a fan blade deicing power regulation method based on sparse Bayesian control, comprising the following steps: the application synchronously merges the partition surface temperature value, the partition heating power value, the blade vibration response value, the environmental temperature value and the incoming flow wind speed value of each heating partition, obtains a partition ice condition observation set, and thus improves the integrity of the partition ice condition representation; then the application outputs the icing state probability, the deicing response confidence and the deicing power boundary value through a sparse Bayesian ice condition identification model, and thus improves the pertinence of the ice condition determination and the power boundary determination; further, the application obtains a target power interval through candidate power interval contraction, and thus considers the deicing demand and the material temperature rise constraint; finally, the application completes target power issuing through partition adjustment, and thus reduces the invalid heating, the local residual ice and the local overheating risk.

Description

Technical Field

[0001] This invention relates to the technical field of wind power generation equipment control and wind turbine blade electric heating de-icing control, and particularly to a wind turbine blade de-icing power regulation method based on sparse Bayesian control. Background Technology

[0002] Wind power generation technology is being widely adopted in cold and high-altitude regions, mountain wind farms, cold and humid coastal areas, and areas with complex meteorological conditions. However, icing on wind turbine blades has become a key factor restricting the power generation efficiency, operational safety, and maintenance economy of the units. When ice forms on the blade surface, the blade shape and surface roughness change, easily causing aerodynamic lift-drag characteristics to deviate, output power to decrease, vibration loads to increase, and abnormal stress on yaw and pitch components. In severe cases, it can even induce shutdown protection, ice spillage, and structural fatigue accumulation. To address this problem, existing technologies have developed into various approaches, including mechanical de-icing, surface coating anti-icing, hot gas de-icing, and electric heating de-icing. Simultaneously, with the development of sensing technology, control technology, and data-driven methods, some technologies have begun to integrate operating parameters such as blade temperature, ambient temperature, humidity, wind speed, and vibration to adjust the heating circuit in a coordinated manner, improving the targeting of de-icing actions and reducing unnecessary energy consumption. Based on this, combining icing state identification with zoned power scheduling is gradually becoming an important development direction for active de-icing control technology for wind turbine blades.

[0003] However, existing wind turbine blade de-icing control schemes still have many shortcomings. First, existing technologies often use fixed threshold triggering, single-parameter discrimination, or empirical rules to set heating power, which is insufficient to adapt to the uncertainties in the formation, growth, loosening, and refreezing of blade icing. They are easily affected by local sensing errors, environmental fluctuations, and the switching of icing stages, leading to heating power settings that are too high or too low. Second, blade icing exhibits significant zonal differences. Different heating zones have inconsistent incoming air velocity, local temperature fields, material thermal diffusion conditions, and ice adhesion states. If a uniform power control is used, local residual ice and local overheating can coexist. Third, while existing data-driven methods can address blade icing... While the current approach identifies the ice conditions in each heating zone, it lacks a stable and clear mapping relationship between the identification results and the zoned power adjustment. This makes it difficult to further provide an executable de-icing power boundary and a target power range corrected for temperature rise constraints, thus leaving shortcomings at the control implementation level. Fourth, the existing scheme does not pay enough attention to the temperature rise constraints of the blade materials. Under continuous heating or high-power pulse conditions, the blade surface material and heating components may bear a large heat load, which is not conducive to long-term service stability or energy consumption control under cold conditions. Therefore, how to probabilistically identify the ice conditions of each heating zone under multi-source observation conditions and generate zoned target heating power with boundary constraints is a problem that urgently needs to be solved.

[0004] In summary, existing wind turbine blade de-icing control technologies suffer from insufficient accuracy in identifying icing conditions in different zones, unclear power setting boundaries, and inadequate integration of thermal safety constraints. Our invention addresses the problem of jointly identifying the icing state, de-icing response, and de-icing power boundaries of each heating zone under multi-source observation conditions, and generating the target heating power for each zone under the constraint of blade material temperature rise. Summary of the Invention

[0005] The purpose of this section is to outline some aspects of the embodiments of the present invention and to briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this section, the abstract and title of the invention. Such simplifications or omissions shall not be used to limit the scope of the present invention.

[0006] In view of the aforementioned existing problems, the present invention is proposed.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for regulating the de-icing power of wind turbine blades based on sparse Bayesian control, comprising: synchronously collecting the surface temperature value, heating power value, blade vibration response value, ambient temperature value, and incoming wind speed value of each heating zone of the wind turbine blade, merging them according to the zone number and the same sampling identifier to generate a zone ice condition observation set; The sectional ice condition observation set is input into the sparse Bayesian ice condition identification model, which outputs the freezing state probability, ice removal response confidence, and ice removal power boundary value of each heated zone. Candidate power intervals are determined based on the icing state probability, de-icing response confidence, and de-icing power boundary value of each heating zone. The candidate power intervals are then narrowed down by combining the zone surface temperature value, the zone heating power value, and the blade material temperature rise constraint to obtain the target power interval. The target heating power of each heating zone is determined according to the target power range of each heating zone, and the target heating power of each zone is sent to the blade heating controller for zone adjustment.

[0008] In a second aspect, the present invention provides a computer device, comprising: One or more processors; The memory stores operable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations, including the process described above for the wind turbine blade de-icing power regulation method based on sparse Bayesian control.

[0009] Thirdly, the present invention provides a computer-readable medium for storing software, the software including instructions executable by one or more computers, the instructions causing the one or more computers to perform operations, the operations including the process of the wind turbine blade de-icing power regulation method based on sparse Bayesian control as described above.

[0010] The beneficial effects of this invention are as follows: By synchronously merging the surface temperature, heating power, blade vibration response, ambient temperature, and incoming wind speed values ​​of each heating zone, this invention obtains a set of zoned ice condition observations, thereby improving the completeness of the zoned ice condition characterization; furthermore, by outputting the ice-forming state probability, de-icing response confidence, and de-icing power boundary value through a sparse Bayesian ice condition identification model, the invention improves the pertinence of ice condition judgment and power boundary determination; furthermore, by narrowing the candidate power range, the target power range is obtained, thus taking into account both de-icing requirements and material temperature rise constraints; finally, the target power is distributed through zoned adjustment, thereby reducing the risks of ineffective heating, local residual ice, and local overheating. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the 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. Wherein: Figure 1 This is a schematic flowchart of the wind turbine blade de-icing power regulation method based on sparse Bayesian control as shown in this invention. Detailed Implementation

[0012] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0013] Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort should fall within the scope of protection of this invention.

[0014] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0015] According to an embodiment of the present invention, in combination Figure 1The flowchart shown illustrates a method for regulating the de-icing power of wind turbine blades based on sparse Bayesian control, which specifically includes the following steps: S1. Simultaneously collect surface temperature, heating power, blade vibration response, ambient temperature, and incoming wind speed values ​​for each heating zone of the wind turbine blades. Merge these values ​​according to zone number and sampling identifier to generate a zone-based ice condition observation set. Note that the following should be noted in this step: In a preferred embodiment, each heating zone is arranged along the chord and span of the fan blade. Taking the outer surface heating configuration of a single blade as an example, the blade is divided into a leading edge heating zone, a middle heating zone, and a trailing edge heating zone from the leading edge to the trailing edge, and into a root section, a middle section, and a tip section from the root to the tip, thereby forming multiple heating zones with fixed spatial positions. In this embodiment, taking 999 heating zones as an example, they can be sequentially set as zone 111 to zone 999, where zone 111, zone 222, and zone 333 are located in the root section, zone 444, zone 555, and zone 666 are located in the middle section, and zone 777, zone 888, and zone 999 are located in the tip section. Within each section, they are further arranged in the order of leading edge, middle, and trailing edge. The above spatial division method ensures that each heating zone corresponds to a unique layout position, which facilitates the unique assignment of the zone surface temperature value and zone heating power value in the subsequent process.

[0016] S1.1. Arrange zone surface temperature acquisition units and zone power acquisition units in each heating zone of the wind turbine blade, arrange vibration acquisition units at the blade root, and arrange ambient temperature acquisition units and wind speed acquisition units on the outside of the nacelle; each acquisition unit outputs zone surface temperature value, zone heating power value, blade vibration response value, ambient temperature value and incoming wind speed value at the same acquisition time.

[0017] S1.2 Assign corresponding partition numbers to the surface temperature value and heating power value of each heating partition according to their spatial location, and assign the blade vibration response value, ambient temperature value and incoming wind speed value output at the same acquisition time to the same sampling mark, forming a partition observation unit with partition number and same sampling mark.

[0018] In a preferred embodiment, a partition location lookup table is established based on the installation position of the blade heating circuit. Then, each heating partition is assigned a fixed partition number in sequence from the blade root to the blade tip, and from the leading edge to the trailing edge in the same spanwise position. For example, the heating partition on the leading edge side of the root section is assigned partition number 1, the heating partition in the middle of the root section is assigned partition number 2, and the heating partition on the trailing edge side of the root section is assigned partition number 3. The corresponding partitions in the middle section are assigned partitions 4, 5, and 6 in sequence, and the corresponding partitions in the tip section are assigned partitions 7, 8, and 9 in sequence. When the partition surface temperature acquisition unit outputs the partition surface temperature value of the corresponding measuring point, the partition number of the heating partition to which the measuring point belongs is directly appended. When the partition power acquisition unit outputs the partition heating power value of the corresponding heating circuit, the partition number of the heating partition to which the heating circuit belongs is directly appended. Thus, each partition surface temperature value and each partition heating power value has a unique partition number, and the same value will not correspond to multiple partitions.

[0019] Furthermore, the acquisition controller issues a unified acquisition trigger command to each acquisition unit, obtains the acquisition results within the same acquisition trigger window, and uses the time information corresponding to the acquisition trigger as the same sampling identifier. For example, during unit operation, synchronous acquisition is triggered every 2 seconds. When the trigger time is 10:00:08 on March 17, 2025, the blade vibration response value, ambient temperature value, and incoming air velocity value acquired in this acquisition are uniformly assigned the same sampling identifier 20250317-100008. Subsequently, this same sampling identifier is copied and attached to the zone surface temperature value and zone heating power value corresponding to each zone number, thereby forming multiple zone observation units with zone numbers and same sampling identifiers.

[0020] S1.3. Using the same sampling identifier as the merging criterion, extract the zone observation units with the same identifier; then, using the zone number as the merging order, combine the zone surface temperature value, zone heating power value, blade vibration response value, ambient temperature value, and incoming wind speed value in the zone observation units with the same identifier to generate a zone ice condition observation subset.

[0021] Specifically, an observation buffer is established in the acquisition controller, sorted by the same sampling identifier. When each zone observation unit is received, its same sampling identifier is compared first. When the same sampling identifier is consistent, it is classified into the same sampling batch; when the same sampling identifier is inconsistent, it is classified into different sampling batches. Taking the same sampling identifier 20250317-100008 as an example, all zone observation units with the same sampling identifier from zone 1 to zone 9 are extracted as objects to be merged in the same batch. This processing method is based on the unique relationship between the same sampling identifier and the sampling time, so that each observation in the same batch corresponds to the blade thermal state and external environmental state at the same time.

[0022] Furthermore, under the same sampling identifier, each zone observation unit is expanded one by one in a fixed order from zone one to zone nine. The zone surface temperature value, zone heating power value, blade vibration response value, ambient temperature value, and incoming wind speed value in each zone observation unit are combined as the complete observation content of that zone to form a zone ice condition observation subset corresponding to a heating zone.

[0023] Preferably, since each ice condition observation subset carries a partition number, subsequent model inputs can call up the partitions one by one in a fixed order.

[0024] S1.4 Arrange the ice condition observation subsets of each zone according to the zone number, from the leading edge heating zone to the trailing edge heating zone, to generate the zone ice condition observation set.

[0025] Specifically, within the same span, the ice condition observation subsets corresponding to the leading edge heating zone are arranged first, then the ice condition observation subsets corresponding to the middle heating zone are arranged, and finally the ice condition observation subsets corresponding to the trailing edge heating zone are arranged. After completing one span, the arrangement continues in the order of root segment, middle segment, and tip segment. For the above 999 heating zones, zones 1, 2, and 3 can be arranged as root segment subsequences, zones 4, 5, and 6 as middle segment subsequences, and zones 7, 8, and 9 as tip segment subsequences, which are finally combined into a set of ice condition observations.

[0026] It should be noted that by merging the multi-source observations at the same acquisition time according to the partition number to form a partition ice condition observation set, each heated partition corresponds to a unique and complete observation content, thereby providing a structurally consistent and temporally consistent input basis for subsequent sparse Bayesian ice condition identification, thereby improving the accuracy of partition ice condition identification, reducing the deviation of partition power regulation, and enhancing the targeting of blade de-icing.

[0027] S2. Input the sectional ice condition observation set into the sparse Bayesian ice condition identification model, and output the freezing state probability, ice removal response confidence, and ice removal power boundary value for each heated zone. Note that the following should be noted in this step: In a preferred embodiment, the sparse Bayesian ice condition identification model is trained using historical operating samples before the unit is put into operation. The historical operating samples include different ambient temperature values, different incoming wind speed values, different zone heating power values, and zone surface temperature values ​​and blade vibration response values ​​under different icing conditions. The icing status is labeled by manual inspection records, icing image records, or shutdown inspection results. After training, a sparse Bayesian ice condition identification model is obtained, which includes an observation component screening unit, a posterior probability solution unit, and a power boundary solution unit. This model is then deployed in the wind turbine blade heating controller. The model inputs are the zone surface temperature value, zone heating power value, blade vibration response value, ambient temperature value, and incoming wind speed value from the same zone ice condition observation subset. The outputs are the icing status probability, de-icing response confidence, and de-icing power boundary value for the corresponding zone.

[0028] The training method for the sparse Bayesian ice condition identification model described above can be carried out using existing technologies and methods, and will not be elaborated further in this example.

[0029] S2.1 Input the subsets of ice condition observations in the set of ice condition observations into the sparse Bayesian ice condition identification model in sequence according to the partition number, and extract the observation components corresponding to the partition surface temperature value, partition heating power value, blade vibration response value, ambient temperature value and incoming wind speed value from the subsets of ice condition observations to form the ice condition identification input sequence.

[0030] Specifically, the sparse Bayesian ice condition identification model is a probabilistic identification model built based on the sparse Bayesian learning algorithm. The sparse Bayesian ice condition identification model includes an observation component screening unit, a posterior probability solution unit, and a power boundary solution unit. The observation component screening unit calculates the target weights for the zone surface temperature, zone heating power, blade vibration response, ambient temperature, and incoming wind speed in the zone ice condition observation set. Based on the target weights of each observation component, it removes observation components with zero target weights and retains those with non-zero target weights to generate a sparse observation combination. The posterior probability calculation unit generates the icing state probability and de-icing response confidence based on the sparse observation combination. The power boundary calculation unit generates the de-icing power boundary value based on the icing state probability, de-icing response confidence, and zone heating power value.

[0031] In a preferred embodiment, the ice condition observation subsets of zones one to nine are sequentially retrieved from the zone ice condition observation set according to the zone numbering order, and the observation values ​​are extracted in a unified arrangement so that each zone forms an ice condition identification input sequence.

[0032] In this embodiment, taking Zone 2 as an example, under a certain sampling marker, the corresponding zone surface temperature is -6.8℃, the zone heating power is 3.2 kW, the blade vibration response is 0.18 g, the ambient temperature is -9.5℃, and the incoming wind speed is 11.4 m / s. The above five values ​​are arranged in sequence to form the ice condition identification input sequence for Zone 2. Each zone is processed in the same order to ensure that the input structure received by the sparse Bayesian ice condition identification model is consistent.

[0033] S2.2 The sparse Bayesian ice condition identification model filters the observation components in the ice condition identification input sequence, removes the observation components with a target weight value of zero, and retains the observation components with a target weight value of non-zero, generating a sparse observation combination.

[0034] It should be noted that the observation component screening unit first evaluates the contribution of each type of observation component to the icing state discrimination based on the sample correlation obtained during the training phase, and assigns a corresponding target weight value to each type of observation component during the model iteration process. When an observation component is insufficient to distinguish the icing state under the current partition and current working condition, its target weight value gradually converges to zero. When an observation component has a stable correlation with the icing state or the thawing change state, its target weight value is retained as a non-zero value. Thus, under the current input conditions, the observation components with a target weight value of zero are removed from the ice condition identification input sequence, and only the observation components with a target weight value of non-zero are retained as sparse observation combinations.

[0035] S2.3 Input the sparse observation combination into the posterior probability solution unit of the sparse Bayesian ice condition identification model to solve the probability of the freezing state and generate the freezing state probability; and generate the freezing response confidence degree according to the freezing change state under the current input conditions.

[0036] In a preferred embodiment, after completing the observation component screening in step S2.2, the sparse observation combination corresponding to the current partition is first input into the posterior probability solving unit. The sparse observation combination consists of the non-zero target weight observation components retained by the current partition under the current sampling identifier. The non-zero target weight observation components include at least one of the partition surface temperature value, partition heating power value, blade vibration response value, ambient temperature value, and incoming wind speed value. At the same time, the posterior probability solving unit calls the icing sample group and non-icing sample group established during the training phase. The icing sample group and non-icing sample group each save the corresponding sample distribution interval, sample concentration interval, and sample fluctuation interval according to the observation component category. The posterior probability solving unit compares the observation components item by item according to the observation component category. The comparison rule is: first determine whether the value of the current observation component falls into the observation corresponding to the icing sample group. The sample set interval of the component is determined; if it falls into the interval, the observed component is determined to have a strong supporting effect on the icing state; if it does not fall into the interval, it is further determined whether it falls into the sample distribution interval of the observed component corresponding to the icing sample group; if it falls into the interval, the observed component is determined to have a general supporting effect on the icing state; if it still does not fall into the interval, the observed component is determined to have no direct supporting effect on the icing state; subsequently, the same rule is used to determine whether the current observed component falls into the sample set interval of the observed component corresponding to the non-icing sample group; if it falls into the sample set interval of the non-icing sample group, the observed component is determined to have a strong supporting effect on the non-icing state; if it only falls into the sample distribution interval of the non-icing sample group, the observed component is determined to have a general supporting effect on the non-icing state; if it does not fall into either interval, the observed component is determined to have no direct supporting effect on the non-icing state.

[0037] Furthermore, after completing the item-by-item comparison of all observed components, the comparison results of each observed component are jointly evaluated. The joint evaluation rules are as follows: when the number of observed components with strong support is large, and the number of observed components with general support is not less than half of the total number of retained observed components, the partition is determined to be the biased corresponding state; when the number of support for the icing state and the non-icing state is similar, the number of observed components with strong support is further compared, and the side with more strong support is taken as the current biased state; when the number of strong support on both sides is still the same, the comparison results corresponding to the surface temperature value and the blade vibration response value of the current partition are compared, and the comparison results of the partition surface temperature value and the blade vibration response value are used as the priority judgment basis; finally, the posterior probability solution unit converts the degree of support for the icing state and the degree of support for the non-icing state of the current partition into the relative proportion within the same numerical range, and determines the relative proportion corresponding to the icing state as the probability of the icing state.

[0038] Taking a certain heating zone as an example, if its sparse observation combination includes the zone surface temperature value of -7.2℃, the blade vibration response value of 0.23 g, and the incoming wind speed value of 12.1 m / s, and the zone surface temperature value and the blade vibration response value both fall within the sample set interval of the icing sample group, and the incoming wind speed value falls within the sample distribution interval of the icing sample group, while none of the above three observation components fall within the sample set interval of the non-icing sample group, then the posterior probability solving unit determines that the zone has a high degree of support for the icing state, and outputs a high icing state probability after conversion, for example, 0.89; if most of the observation components of the current zone fall within the sample set interval or sample distribution interval of the non-icing sample group, then a lower icing state probability is output, for example, 0.18.

[0039] In another preferred embodiment, after obtaining the probability of the current partition's icing state, the posterior probability solving unit continues to retrieve the partition surface temperature value, partition heating power value, and blade vibration response value corresponding to the current sampling time and the previous sampling time, and performs corresponding mapping according to the partition number to form the current partition's de-icing change judgment sequence; the posterior probability solving unit sequentially performs temperature rise comparison, power state comparison, and vibration change comparison; the temperature rise comparison rule is as follows: compare the difference between the partition surface temperature value at the current sampling time and the partition surface temperature value at the previous sampling time; when the difference is greater than or equal to 1.5℃, the current partition is judged to be in a state of significant temperature rise; when the difference is greater than or equal to 0.5℃ and less than 1.5℃, the current partition is judged to be in a state of slow temperature rise; when the difference is less than 0.5℃, the current partition is judged to be in a state of insufficient temperature rise.

[0040] Furthermore, the power state comparison rule is as follows: the partition heating power value at the current sampling time is compared with the de-icing power boundary value corresponding to that partition; when the partition heating power value at the current sampling time is greater than or equal to the lower power boundary value and less than or equal to the upper power boundary value, the current partition is determined to be a valid heating input state; when the partition heating power value at the current sampling time is less than the lower power boundary value, the current partition is determined to be a weak heating input state; when the partition heating power value at the current sampling time is greater than the upper power boundary value, the current partition is determined to be a high power input state, and is processed as a valid heating input state in subsequent confidence outputs, but in step S3, the power range is further narrowed in conjunction with the blade material temperature rise constraint.

[0041] Furthermore, the rules for comparing vibration changes are as follows: the blade vibration response value at the current sampling time is compared with the blade vibration response value at the previous sampling time; when the difference is greater than or equal to 0.03 g and the blade vibration response value at the current sampling time is less than or equal to 80% of the unit vibration alarm limit, the current zone is determined to be in a state of de-icing disturbance; when the difference is greater than or equal to 0.01 g and less than 0.03 g, the current zone is determined to be in a state of slight disturbance; when the difference is less than 0.01 g, the current zone is determined to be in a state of no obvious disturbance.

[0042] After completing the above three comparisons, the posterior probability solving unit determines the de-icing change state according to a fixed combination rule. The fixed combination rule is as follows: when the current partition simultaneously meets the conditions of significant temperature rise, effective heating input or high power input, and de-icing disturbance, the current partition is determined to be in a strong de-icing response state; when the current partition simultaneously meets the conditions of slow temperature rise, effective heating input or high power input, and slight disturbance, the current partition is determined to be in a moderate de-icing response state; when the current partition meets the conditions of insufficient temperature rise or weak heating... When the current partition is in a hot input state or meets any of the conditions of no significant disturbance, it is determined to be in a weak de-icing response state. Then, the corresponding de-icing response confidence score is output according to the de-icing change state. The output rules are as follows: when the current partition is in a strong de-icing response state, the de-icing response confidence score is limited to 0.80 to 0.95; when the current partition is in a medium de-icing response state, the de-icing response confidence score is limited to 0.55 to 0.79; when the current partition is in a weak de-icing response state, the de-icing response confidence score is limited to 0.20 to 0.54.

[0043] To ensure the uniqueness of the de-icing response confidence score, after determining the response state of the current partition, specific values ​​are given according to the following comparison rules: In a strong de-icing response state, if the temperature increment is greater than or equal to 2.5℃ and the vibration increment is greater than or equal to 0.05 g, the score is between 0.88 and 0.95; if the temperature increment is greater than or equal to 1.5℃ and less than 2.5℃, and the vibration increment is greater than or equal to 0.03 g and less than 0.05 g, the score is between 0.80 and 0.87. In a moderate de-icing response state, if the temperature increment is greater than or equal to 1.0℃ and the vibration increment is greater than or equal to 0.02 g, the score is between 0.68 and 0.79; if the temperature increment is greater than or equal to 0.5℃ and less than 1.0℃, and the vibration increment is greater than or equal to 0.01 g and less than 0.02 g, the score is between 0.55 and 0.67. In a weak de-icing response state, if the temperature increment is greater than 0℃ and less than 0.5℃, and the vibration increment is greater than 0... If the increment is less than 0.01 g, the value is between 0.35 and 0.54; if the temperature increment is less than or equal to 0℃, or the vibration increment is less than or equal to 0 g, the value is between 0.20 and 0.34.

[0044] S2.4 Based on the icing state probability, the confidence level of the de-icing response, and the regional heating power value, the sparse Bayesian ice condition identification model is used to solve the boundary of the de-icing power range and generate the de-icing power boundary value.

[0045] Specifically, the power boundary solution unit calls the partition de-icing power sample interval formed during the training phase. After inputting the icing state probability, de-icing response confidence, and current partition heating power value, it determines the power range that can cause ice loosening under the current ambient temperature and incoming wind speed. If the icing state probability is high and the de-icing response confidence is low, it indicates that there is significant icing in the current partition and the applied heating power has not yet formed sufficient de-icing feedback. In this case, the power boundary solution unit gives a higher de-icing power range. If the icing state probability is high and the de-icing response confidence is already high, a narrower and lower de-icing power range is given.

[0046] For example, under the conditions of an ambient temperature of -9.5℃, an incoming wind speed of 11.4 m / s, a current zone heating power of 3.2 kW, an icing probability of 0.87, and a de-icing response confidence level of 0.78, the power boundary solution unit can output a de-icing power boundary value of 3.0 kW to 3.8 kW; this range is the de-icing power range under the current operating conditions.

[0047] It should be noted that this embodiment uses a sparse Bayesian ice condition identification model to filter, solve probabilistically, and solve boundary conditions for multiple observation components. This can transform the original observation values ​​into ice-freezing state probabilities, ice-freeing response confidence, and ice-freeing power boundary values, thereby improving the targeting of ice condition identification, reducing redundant observation interference, and providing reliable boundary basis for subsequent power interval contraction.

[0048] S3. Based on the icing state probability, de-icing response confidence level, and de-icing power boundary value of each heating zone, determine the candidate power range. Then, by combining the zone surface temperature value, zone heating power value, and blade material temperature rise constraint, narrow down the candidate power range to obtain the target power range. Note that the following should be noted in this step: S3.1 Match the icing state probability, de-icing response confidence and de-icing power boundary value corresponding to each heating zone, extract the lower power boundary value and the upper power boundary value, and generate candidate power intervals.

[0049] Specifically, the partition number is used as the corresponding index to match the icing state probability, de-icing response confidence, and de-icing power boundary value obtained in step S2 for each heating partition; then, the lower power boundary value and the upper power boundary value are extracted from the de-icing power boundary value of each partition; finally, the lower power boundary value and the upper power boundary value of the same partition are combined to form a candidate power interval.

[0050] Preferably, this matching method uses the partition number as the sole basis, preventing the boundary mixing between different heating partitions.

[0051] S3.2. Combine the corresponding surface temperature value of each heating zone with the heating power value of the zone to obtain the current temperature rise state; then compare the current temperature rise state with the temperature rise constraint of the blade material to determine the power compression boundary.

[0052] In a preferred embodiment, under the same partition number, the partition surface temperature value at the current moment is arranged in correspondence with the partition heating power value at the current moment, and then the current heating state of the partition is determined by combining the changing trend of the partition surface temperature value in two adjacent samplings.

[0053] As an example, the current temperature rise states include: low temperature and low power state, low temperature and high power state, medium temperature and stable state, and high temperature and near-constraint state.

[0054] For example, if the current partition surface temperature is -3.9℃, the current partition heating power is 3.2 kW, and the temperature has increased by 2.9℃ compared to the previous sampling time, then the state is determined to be a medium-temperature stable state; if the current partition surface temperature has risen to 14.2℃, and the current partition heating power is still 3.6 kW, then it can be determined to be a high-temperature approach constraint state.

[0055] In another preferred embodiment, based on the blade shell material, adhesive layer material, and heating diaphragm covering position, a temperature rise constraint for the blade material corresponding to each heating zone is given. Taking a glass fiber composite blade and surface adhesive structure as an example, the temperature rise constraint for the blade material in the leading edge heating zone can be set to a zone surface temperature value not exceeding 18°C, the temperature rise constraint for the blade material in the middle heating zone can be set to a zone surface temperature value not exceeding 16°C, and the temperature rise constraint for the blade material in the trailing edge heating zone can be set to a zone surface temperature value not exceeding 14°C. When the current temperature rise status of a certain zone shows that its zone surface temperature value is close to the corresponding blade material temperature rise constraint, the candidate power range of that zone is compressed downward. When the current temperature rise status is in a low temperature and low power state, the candidate power range is not compressed.

[0056] S3.3. Based on the power compression boundary, the lower and upper power boundary values ​​of the candidate power interval are shrunk to obtain the constrained power interval.

[0057] Specifically, when the current temperature rise is close to the high temperature constraint state, the upper boundary value of the candidate power range is adjusted downward to a level that no longer causes the surface temperature of the zone to continue to rise rapidly, based on the power compression boundary. If necessary, the lower boundary value of the power is then moved up or down synchronously so that the contracted range is still within the range of the de-icing power boundary value.

[0058] For example, if the candidate power range is 3.0 kW to 3.8 kW, the current partition surface temperature reaches 14.2°C, and the blade material temperature rise constraint for this partition is 16°C, then the upper power boundary value can be reduced from 3.8 kW to 3.4 kW, while the lower power boundary value is maintained at 3.0 kW, thus obtaining a constrained power range of 3.0 kW to 3.4 kW; if the current partition surface temperature is closer to the blade material temperature rise constraint, then the candidate power range can be further reduced to 3.1 kW to 3.2 kW.

[0059] S3.4 Boundary verification of the constrained power range: When the lower boundary value of the power range is greater than the upper boundary value, the lower boundary value and the upper boundary value are simultaneously corrected to the same boundary position within the de-icing power boundary value to generate the target power range; when the lower boundary value of the power range is less than or equal to the upper boundary value, the constrained power range is determined as the target power range.

[0060] It should be noted that the lower boundary value and the upper boundary value of the power constraint range are compared first. When the two cross in opposite directions, it means that the compression of the candidate power range by the blade material temperature rise constraint exceeds the current available de-icing power space. At this time, the original shrinkage result is no longer retained. Instead, the de-icing power boundary value corresponding to the partition is returned, and the boundary closer to the current partition heating power value is selected as the unified correction position. Then, the lower boundary value and the upper boundary value of the power are simultaneously adjusted to the unified correction position to form a single-point target power range.

[0061] For example, if the de-icing power boundary value is between 3.0 kW and 3.8 kW, and after temperature rise constraint compression, the lower power boundary value is 3.35 kW and the upper power boundary value is 3.20 kW, then the two overlap. In this case, if the current zone heating power value is 3.2 kW, then between the two boundary positions of 3.0 kW and 3.8 kW, the 3.0 kW value, which is closer to 3.2 kW, is selected as the unified correction position, and the lower power boundary value and the upper power boundary value are simultaneously corrected to 3.0 kW, thus generating a single-point target power range.

[0062] Preferably, this embodiment incorporates the zone surface temperature value, zone heating power value, and blade material temperature rise constraint into the zone contraction process, and performs synchronous correction when the zones intersect. This limits the available power range to a more suitable boundary, thereby reducing local thermal shock, suppressing overheating risk, and improving the robustness of zone power regulation.

[0063] S4. Determine the target heating power for each heating zone based on the target power range, and then send the target heating power to the blade heating controller for zone adjustment. Note that the following should be noted in this step: S4.1 Extract the lower and upper power boundary values ​​in the target power range, and determine the range width based on the difference between the upper and lower power boundary values.

[0064] It should be noted that the method for determining the interval width includes: first, extracting the lower boundary value and the upper boundary value of the power from the target power interval, and then subtracting the lower boundary value from the upper boundary value to obtain the interval width.

[0065] S4.2 When the interval width is greater than zero, the midpoint of the interval between the lower power boundary value and the upper power boundary value is determined as the target heating power of the zone; when the interval width is equal to zero, the lower power boundary value is determined as the target heating power of the zone; when the interval width is less than zero, the smaller value between the lower power boundary value and the upper power boundary value is determined as the target heating power of the zone.

[0066] S4.3 Match the target heating power of each zone with the zone number, generate a power adjustment command and send it to the blade heating controller to adjust each heating zone.

[0067] In an optional implementation, the target heating power of each heating zone is registered as a set of adjustment correspondences using the zone number as an index. These correspondences are then organized into a power adjustment command sequence ordered by the zone number. For example, if the target heating power of zone 1 is 3.6 kW, zone 2 is 3.2 kW, and zone 3 is 2.8 kW, then power adjustment commands can be generated in the form of zone 1 - 3.6 kW, zone 2 - 3.2 kW, and zone 3 - 2.8 kW, and sent to the blade heating controller. After receiving the power adjustment commands, the blade heating controller calls the corresponding target heating power of each zone according to the zone number, preventing power mismatch between different zones.

[0068] Specifically, the method for zonal adjustment of each heating zone includes: the blade heating controller determines the power output level based on the power adjustment command and the rated parameters of the heating circuit corresponding to each heating zone; according to the power output level, the heating circuit corresponding to each heating zone is independently controlled on and off to form the heating power output; the heating power output is zonal status checked; when the heating power output is consistent with the power output level, the current power output level is maintained; when the heating power output is inconsistent with the power output level, the power output level of the corresponding heating zone is downgraded by one level, and independent on and off control is performed according to the downgraded power output level.

[0069] Furthermore, a corresponding rated parameter table for the heating circuit is first established for each heating zone. The rated parameters of the heating circuit include rated voltage, rated current, and rated power. Taking the heating zone in the middle of a fan blade as an example, its rated voltage is 690 V and its rated power is 4.0 kW. Based on this, multiple power output levels are divided according to the discrete proportion of the rated power, such as level 0, level 1, level 2, level 3, and level 4. Level 0 corresponds to 0 kW, level 1 corresponds to 25% of the rated power, level 2 corresponds to 50% of the rated power, level 3 corresponds to 75% of the rated power, and level 4 corresponds to 100% of the rated power. When the target heating power of zone 2 is 3.2 kW, the blade heating controller compares it with the rated power of 4.0 kW of the corresponding heating circuit in zone 2 to determine that it corresponds to between level 3 and level 4. Further, combined with the control discrete level, the closest level 3 or level 4 is selected as the current power output level. If the controller adopts the nearest value method, then 3.2 kW can correspond to level 3, i.e., 3.0 kW. kW level; if the upper-level method is adopted, it can correspond to level four, that is, 4.0 kW level.

[0070] Preferably, the nearest value is used in the de-icing condition to reduce the instantaneous power surge.

[0071] It should be noted that the heating circuit corresponding to each heating zone is configured as a control branch that can be individually turned on and off. The blade heating controller issues on or off commands to each control branch, so that the heating circuits of each zone can present different conduction states within the same control cycle. Independent on / off control includes: individually turning on the heating circuit of zone 1, individually turning off the heating circuit of zone 2, and intermittently turning on the heating circuit of zone 3, etc., rather than uniformly turning on or off all heating circuits of the entire blade. Taking a certain de-icing scenario as an example, the power output level corresponding to zone 1 is level 4, the power output level corresponding to zone 2 is level 3, and the power output level corresponding to zone 3 is level 2. Then, the blade heating controller can keep the heating circuit of zone 1 continuously conducting, the heating circuit of zone 2 conducting at a higher duty cycle, and the heating circuit of zone 3 conducting at a medium duty cycle within the same control cycle, thereby forming a heating power output corresponding to the target heating power of each zone.

[0072] In a preferred embodiment, the method for zonal status verification of heating power output is as follows: after each heating zone completes the on / off control of the current control cycle, the zone heating power value is collected again and compared with the power output level corresponding to the zone; when the measured zone heating power value falls within the allowable deviation range corresponding to the power output level, it is determined that the heating power output is consistent with the power output level; when the measured zone heating power value is higher or lower than the allowable deviation range, it is determined that the heating power output is inconsistent with the power output level; in the latter case, the power output level of the corresponding heating zone is downgraded by one level, and then independent on / off control is performed again at the downgraded power output level.

[0073] Preferably, this embodiment extracts the target power range boundary, determines the target heating power of each zone, generates zoned power adjustment commands, and combines the rated parameters of the heating circuit to determine the power output level and independent on / off control mode. This can convert the continuous power boundary into specific control actions, thereby improving the implementation of zoned de-icing adjustment, enhancing the consistency of controller execution, and reducing zoned output deviation.

[0074] Other aspects disclosed in the embodiments of the present invention also provide a computer device including one or more processors and a memory.

[0075] The memory is used to store operable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations, including the flow of the wind turbine blade de-icing power regulation method based on sparse Bayesian control in the foregoing embodiments, especially... Figure 1 The flowchart of the method is shown.

[0076] Other aspects disclosed in the embodiments of the present invention also propose a computer-readable medium for storing software including instructions executable by one or more computers, which, upon execution, cause the one or more computers to perform operations including the flow of the wind turbine blade de-icing power regulation method based on sparse Bayesian control of the foregoing embodiments, particularly... Figure 1 The flowchart of the method is shown.

[0077] It should be recognized that embodiments of the present invention may be implemented or carried out by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable storage medium.

[0078] The method can be implemented using standard programming techniques, including a non-transitory computer-readable storage medium configured with a computer program in the computer program, wherein the storage medium is configured such that the computer operates in a specific and predefined manner.

[0079] Each program can be implemented in a high-level procedural or object-oriented programming language to communicate with the computer system; however, if necessary, the program can be implemented in assembly or machine language.

[0080] In any case, the language can be either compiled or interpreted.

[0081] Furthermore, for this purpose, the program can run on a programmed application-specific integrated circuit.

[0082] The processes described herein (or variations and / or combinations thereof) can be executed under the control of one or more computer systems configured with executable instructions, and can be implemented by hardware or a combination thereof as code (e.g., executable instructions, one or more computer programs, or one or more applications) that commonly executes on one or more processors. The computer program includes a plurality of instructions executable by one or more processors.

[0083] Furthermore, the method can be implemented in any suitable computing platform, including but not limited to personal computers, minicomputers, mainframes, workstations, networked or distributed computing environments, standalone or integrated computer platforms, or in communication with charged particle tools or other imaging devices.

[0084] Various aspects of the present invention can be implemented in machine-readable code stored on a non-transitory storage medium or device, whether portable or integrated into a computing platform, such as a hard disk, optical read and / or write storage medium, RAM, ROM, etc., such that it can be read by a programmable computer, and when the storage medium or device is read by the computer, it can be used to configure and operate the computer to perform the processes described herein.

[0085] Furthermore, machine-readable code, or parts thereof, can be transmitted via wired or wireless networks.

[0086] When such media includes instructions or programs that combine with a microprocessor or other data processor to implement the steps described above, the invention described herein includes these and other different types of non-transitory computer-readable storage media.

[0087] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for regulating the de-icing power of a fan blade based on sparse Bayesian control, characterized in that, include: Simultaneously collect the zone surface temperature, zone heating power, blade vibration response, ambient temperature, and incoming wind speed values ​​in each heating zone of the wind turbine blades. Then, merge them according to the zone number and the same sampling identifier to generate a zone ice condition observation set. The sectional ice condition observation set is input into the sparse Bayesian ice condition identification model, which outputs the freezing state probability, ice removal response confidence, and ice removal power boundary value of each heated zone. Candidate power intervals are determined based on the icing state probability, de-icing response confidence, and de-icing power boundary value of each heating zone. The candidate power intervals are then narrowed down by combining the zone surface temperature value, the zone heating power value, and the blade material temperature rise constraint to obtain the target power interval. The target heating power of each heating zone is determined according to the target power range of each heating zone, and the target heating power of each zone is sent to the blade heating controller for zone adjustment; The target power range is generated based on the candidate power range, including: The surface temperature value of each heating zone is combined with the heating power value of the zone to obtain the current temperature rise state; then the current temperature rise state is compared with the temperature rise constraint of the blade material to determine the power compression boundary. Based on the power compression boundary, the lower and upper power boundary values ​​of the candidate power interval are shrunk to obtain the constrained power interval; Boundary verification is performed on the constrained power range; when the lower boundary value of the power in the constrained power range is greater than the upper boundary value, the lower boundary value and the upper boundary value are synchronously corrected to the same boundary position within the de-icing power boundary value to generate the target power range; when the lower boundary value of the power in the constrained power range is less than or equal to the upper boundary value, the constrained power range is determined as the target power range.

2. The method of claim 1, wherein the method is based on sparse Bayesian control of fan blade de-icing power regulation. The method for generating the regional ice condition observation set includes: Each heating zone of the wind turbine blade is equipped with a zone surface temperature acquisition unit and a zone power acquisition unit. A vibration acquisition unit is installed at the blade root, and an ambient temperature acquisition unit and a wind speed acquisition unit are installed on the outside of the nacelle. Each acquisition unit outputs the zone surface temperature value, zone heating power value, blade vibration response value, ambient temperature value, and incoming wind speed value at the same acquisition time. According to the spatial location of each heating zone, the surface temperature value and heating power value of each zone are assigned corresponding zone numbers, and the blade vibration response value, ambient temperature value and incoming wind speed value output at the same acquisition time are assigned the same sampling identifier, forming a zone observation unit with the zone number and the same sampling identifier; Using the same sampling identifier as the merging criterion, extract the zone observation units with the same identifier; then, using the zone number as the merging order, combine the zone surface temperature value, zone heating power value, blade vibration response value, ambient temperature value, and incoming wind speed value in the zone observation units with the same identifier to generate a zone ice condition observation subset. The ice condition observation subsets of the partitions are arranged in the order of the partition numbers from the leading edge heating partition to the trailing edge heating partition to generate the partition ice condition observation set.

3. The Sparse Bayesian Control based fan blade de-icing power regulation method of claim 1 or 2, wherein, Inputting the aforementioned regional ice condition observation set into the sparse Bayesian ice condition identification model includes: The subsets of ice condition observations in the partition ice condition observation set are sequentially input into the sparse Bayesian ice condition identification model according to the partition number. The observation components corresponding to the partition surface temperature value, partition heating power value, blade vibration response value, ambient temperature value and incoming wind speed value are extracted from the subsets of ice condition observations to form the ice condition identification input sequence. The sparse Bayesian ice condition identification model filters the observation components in the ice condition identification input sequence, removes the observation components with a target weight value of zero, and retains the observation components with a target weight value of non-zero, generating a sparse observation combination. The sparse observation combination is input into the posterior probability solving unit of the sparse Bayesian ice condition identification model to solve the probability of the freezing state and generate the freezing state probability; and the ice removal response confidence is generated based on the ice removal change state under the current input conditions. Based on the icing state probability, the de-icing response confidence, and the zone heating power value, the sparse Bayesian ice condition identification model is used to solve the boundary of the de-icing power range and generate de-icing power boundary values.

4. The method of claim 3, wherein the method is based on sparse Bayesian control. The sparse Bayesian ice condition identification model is a probabilistic identification model constructed based on the sparse Bayesian learning algorithm. The sparse Bayesian ice condition identification model includes an observation component screening unit, a posterior probability solution unit, and a power boundary solution unit. Specifically, the observation component filtering unit calculates the target weights for the zone surface temperature, zone heating power, blade vibration response, ambient temperature, and incoming wind speed in the zone ice condition observation set. Based on the target weight values ​​of each observation component, it removes observation components with a target weight value of zero and retains observation components with a target weight value of non-zero, generating a sparse observation combination. The posterior probability solving unit generates the icing state probability and the de-icing response confidence level based on the sparse observation combination. The power boundary solving unit generates the de-icing power boundary value based on the icing state probability, the de-icing response confidence level, and the zone heating power value.

5. The method of claim 3, wherein the method is based on sparse Bayesian control. The method for generating the candidate power range includes: The probability of freezing state, the confidence level of de-icing response, and the boundary value of de-icing power corresponding to each heating zone are matched accordingly, and the lower boundary value and upper boundary value of power are extracted to generate candidate power intervals.

6. The method of claim 1, wherein the method is implemented in a wind turbine controller. The target heating power for each heating zone is determined based on the target power range of each zone, including: Extract the lower and upper power boundary values ​​from the target power range, and determine the range width based on the difference between the upper and lower power boundary values. When the interval width is greater than zero, the midpoint of the interval between the lower power boundary value and the upper power boundary value is determined as the target heating power of the zone; when the interval width is equal to zero, the lower power boundary value is determined as the target heating power of the zone; when the interval width is less than zero, the smaller value between the lower power boundary value and the upper power boundary value is determined as the target heating power of the zone. The target heating power of each zone is mapped to the zone number, a power adjustment command is generated and sent to the blade heating controller to adjust each heating zone.

7. The method of claim 6, wherein the method is based on sparse Bayesian control of fan blade de-icing power regulation. The method for zonal adjustment includes: The blade heating controller determines the power output level based on the power adjustment command and the rated parameters of the heating circuit corresponding to each heating zone; According to the power output level, the heating circuits corresponding to each heating zone are independently switched on and off to form heating power output; The heating power output is checked for zone status; when the heating power output is consistent with the power output level, the current power output level is maintained; when the heating power output is inconsistent with the power output level, the power output level of the corresponding heating zone is downgraded by one level, and independent on / off control is performed according to the downgraded power output level.

8. A computer device, comprising: include: One or more processors; The memory stores operable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations, including the flow of the wind turbine blade de-icing power regulation method based on sparse Bayesian control as described in any one of claims 1 to 7.

9. A computer readable medium storing software, characterized in that: The software includes instructions executable by one or more computers, which, through execution, cause the one or more computers to perform operations including the flow of the wind turbine blade de-icing power regulation method based on sparse Bayesian control as described in any one of claims 1 to 7.

Images