Fan running state intelligent control system based on multi-source data

Abstract

The invention provides an intelligent fan running state control system based on multi-source data, and relates to the field of wind power generation, and the intelligent fan running state control system comprises the steps: firstly, collecting a time sequence set of running parameter data of a target fan object and a time sequence set of electric field outlet power data, and then constructing a field domain-equipment two-stage information fusion mechanism; and performing mutual information guided deep interaction on the global time sequence characteristics of the electric field outlet power and the local time sequence characteristics of the single machine operation parameters, thereby optimizing the expression of the self-information of the fan operation state by using the time sequence characteristics of the electric field outlet power, determining the detection result of the fan operation state, and determining the early warning level and the control strategy. The method is used for breaking through the technical bottleneck of decoupling of global context information and local equipment state of wind power plant operation in a traditional method, and is beneficial to more accurate detection and control of the operation state of the fan.

Description

Technical Field

[0001] This application relates to the field of wind power generation, and more particularly, in the embodiments of this application, it relates to an intelligent control system for the operating state of a wind turbine based on multi-source data. Background Art

[0002] With the accelerating transformation of the global energy structure towards renewable energy, wind power generation, as an important part of clean energy, has an increasingly urgent need for large-scale and intelligent operation and maintenance. Modern wind farms usually consist of dozens to hundreds of wind turbines, and each wind turbine forms a dynamic energy system under the coupling action of complex environments. However, traditional wind turbine state monitoring systems mainly rely on single-machine SCADA (Supervisory Control And Data Acquisition) data for analysis, and construct a health assessment model through local operating state parameters of the wind turbine such as temperature and power. This single-machine perspective monitoring paradigm has significant limitations: firstly, as an organic whole, global factors such as grid dispatching instructions, atmospheric turbulence propagation, and wake effects in a wind farm will affect the operating state of a single machine through energy transfer; secondly, the hidden performance degradation of multiple wind turbines may not reach the alarm threshold of single-machine parameters, resulting in limited sensitivity of early fault warning.

[0003] Therefore, an optimized control system for the operating state of a wind turbine is desired. Summary of the Invention

[0004] To solve the above technical problems, this application is proposed. Embodiments of this application provide an intelligent control system for the operating state of a wind turbine based on multi-source data. First, it collects a time series set of operating parameter data of a target wind turbine object and a time series set of electric field outlet power data, and then constructs a two-level information fusion mechanism for the field and equipment, performing depth interaction guided by mutual information between the global time series characteristics of the electric field outlet power and the local time series characteristics of single-machine operating parameters, so as to use the time series characteristics of the electric field outlet power to optimize the expression of the self-information of the wind turbine operating state, determine the detection result of the wind turbine operating state, and determine the warning level and control strategy, and then generate a wind turbine operation control signal, which is used to break through the technical bottleneck of the decoupling of the global context information and local equipment state in the operation of a wind farm in traditional methods, and is conducive to more accurately detecting and controlling the operating state of a wind turbine.

[0005] According to one aspect of this application, there is provided an intelligent control system for the operating state of a wind turbine based on multi-source data, which includes:

[0006] A wind turbine operating parameter acquisition module, configured to collect a time series set of operating parameter data of a target wind turbine object from a wind turbine SCADA system, where the operating parameter data includes a wind turbine temperature value, a wind turbine rotation speed value, a wind turbine power value, and a wind turbine voltage value;

[0007] The electric field outlet power data acquisition module is used to acquire the time series set of the electric field outlet power data from the energy management platform;

[0008] The fan operation state characterization module is used to perform an optimized characterization of the fan operation state based on time series mutual information perception on the time series set of the operation parameter data and the time series set of the electric field outlet power data to obtain the optimized characterization information of the fan operation state;

[0009] The fan early warning control module is used to determine the fan operation state detection result based on the optimized characterization information of the fan operation state, and determine the early warning level and control strategy;

[0010] The operation control signal generation module is used to generate the fan operation control signal based on the early warning level and the control strategy.

[0011] Compared with the prior art, an intelligent control system for the fan operation state based on multi-source data provided by the present application first acquires the time series set of the operation parameter data of the target fan object and the time series set of the electric field outlet power data, and then constructs a two-level information fusion mechanism of the field - device, performs deep interaction guided by mutual information between the global time series characteristics of the electric field outlet power and the local time series characteristics of the single - machine operation parameters, so as to use the time series characteristics of the electric field outlet power to optimize the expression of the self - information of the fan operation state, determine the fan operation state detection result, and determine the early warning level and control strategy, and further generate the fan operation control signal, which is used to break through the technical bottleneck of the decoupling of the global context information and the local device state in the operation of the wind farm in the traditional method, and is beneficial to more accurately detect and control the fan operation state. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] By describing the embodiments of the present application in more detail in conjunction with the drawings, the above - mentioned and other objects, features, and advantages of the present application will become more obvious. The drawings are used to provide a further understanding of the embodiments of the present application, and constitute a part of the specification, and are used to explain the present application together with the embodiments of the present application, and do not constitute a limitation to the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0013] Figure 1 FIG. is a system block diagram of an intelligent control system for the fan operation state based on multi - source data according to an embodiment of the present application.

[0014] Figure 2 FIG. is a schematic diagram of data flow of an intelligent control system for the fan operation state based on multi - source data according to an embodiment of the present application.

[0015] Figure 3 FIG. is a block diagram of the fan operation state characterization module in an intelligent control system for the fan operation state based on multi - source data according to an embodiment of the present application.

[0016] Figure 4 It is a block diagram of the fan operation state optimization characterization unit in the intelligent control system for fan operation state based on multi-source data according to an embodiment of the present application. Detailed implementation manners

[0017] Various exemplary embodiments, features and aspects of the present application will be described in detail below with reference to the accompanying drawings. The same reference numerals in the drawings denote elements having the same or similar functions. Although various aspects of the embodiments are shown in the drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0018] The special term "exemplary" here means "serving as an example, an embodiment or illustrative". Any embodiment described as "exemplary" here is not necessarily to be construed as superior or better than other embodiments.

[0019] In addition, for a better illustration of the present application, numerous specific details are given in the following detailed implementation manners. Those skilled in the art should understand that the present application can also be implemented without some specific details. In some instances, methods, means, elements and circuits well known to those skilled in the art are not described in detail so as to highlight the gist of the present application.

[0020] Furthermore, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present application, "a plurality of" means two or more unless otherwise specifically defined.

[0021] With the transformation of the energy structure towards renewable energy, wind power generation, as an important part of clean energy, has an increasing demand for large-scale and intelligent operation and maintenance. Modern wind farms consist of numerous wind turbines, forming a complex dynamic energy system. However, the traditional state monitoring method based on single-machine SCADA data only constructs a health assessment model by analyzing local parameters such as temperature and power, which has obvious limitations. This method ignores the influence of global factors such as grid dispatch instructions, atmospheric turbulence and wake effect on the operation state of single machines, and the latent performance degradation of multiple wind turbines may occur before the single-machine parameters reach the alarm threshold, which limits the sensitivity of early fault warning. Therefore, the traditional single-machine perspective monitoring paradigm is difficult to meet the needs of overall optimization and maintenance of modern wind farms.

[0022] In recent years, with the integrated development of industrial big data and artificial intelligence technologies, the state monitoring and control of wind turbines with multi-source data fusion has become the mainstream. In the existing technologies, there are mainly two technical routes: one is the single-machine multi-parameter time-series modeling based on deep learning, such as using the LSTM network to capture the equipment degradation trend; the other is the wind farm-level energy flow analysis based on physical simulation. However, the former lacks explicit modeling of the coupling effect of the field environment, and the latter is limited by the computational complexity and difficult to achieve real-time state perception. It is worth noting that the time-series data of the power at the exit of the power station, as the core index of the interaction between the wind farm and the power grid, naturally contains the state information of the collaborative operation of the equipment group. That is to say, the power at the exit of the power station is the result of the combined action of all the wind turbines in the entire wind farm, which provides a global perspective and context information for the operation of the wind farm. That is, the operating state of a single wind turbine is affected not only by its own factors but also by the overall wind farm environment (such as the overall wind resource and the power grid state). The time-series characteristics of the power at the exit of the power station can reflect these global influencing factors. Among them, the change trend of the power at the exit of the power station can reveal the mutual influence between wind turbines caused by uneven wind resource distribution, wake effect, etc.; for example, in some cases, although the operating parameters (such as temperature and speed) of a single wind turbine may seem normal, the total output power of the entire wind farm shows abnormal fluctuations, which may be due to the fact that some wind turbines are affected by the wake of the upstream wind turbines and result in a decrease in efficiency. In addition, the power at the exit of the power station can also reflect the impact of the power grid dispatching instructions on the operation mode of the wind farm. For example, when the power grid demand changes, the wind farm needs to adjust the power generation strategies of each wind turbine to match the requirements of the power grid, and this adjustment will leave traces in the time-series data of the power at the exit of the power station. By analyzing these time-series characteristics, the state information of the collaborative operation of the equipment group in the wind farm can be captured, so as to provide a more comprehensive operating state representation for each wind turbine, considering not only its own physical parameters but also the overall environmental background in which it is located, making the state monitoring more accurate and helping to predict potential failures and optimize control strategies. In addition, some wind turbine failures may not immediately show obvious abnormalities in the parameters of a single wind turbine, but may have a perceptible impact on the power at the exit of the power station, especially when similar problems occur in multiple wind turbines at the same time, the decrease in the power at the exit of the power station will be more obvious. For example, when multiple wind turbines simultaneously experience gearbox wear or pitch system response lag, although the single-machine power fluctuation is still within the normal range, abnormal patterns can be detected 3-6 months in advance through the frequency-domain feature analysis of the field-level power time series.

[0023] In view of the above technical problems, in the technical solution of the present application, an intelligent control system for the operating state of a wind turbine based on multi-source data is proposed. By constructing a two-level information fusion mechanism of field - equipment, it conducts deep interaction guided by mutual information between the global temporal characteristics of the power at the electric field outlet and the local temporal characteristics of the single - machine operating parameters, thereby using the temporal characteristics of the power at the electric field outlet to optimize the expression of the self - information of the operating state of the wind turbine, so as to break through the technical bottleneck of the decoupling of the global context information of the wind farm operation and the local equipment state in the traditional method, which is beneficial to more accurately detecting and controlling the operating state of the wind turbine.

[0024] The present application proposes an intelligent control system for the operating state of a wind turbine based on multi - source data. Figure 1 FIG. is a system block diagram of an intelligent control system for the operating state of a wind turbine based on multi - source data according to an embodiment of the present application. Figure 2 FIG. is a schematic diagram of data flow of an intelligent control system for the operating state of a wind turbine based on multi - source data according to an embodiment of the present application. As Figure 1 and Figure 2 shown, the intelligent control system 100 for the operating state of a wind turbine based on multi - source data according to an embodiment of the present application includes: a wind turbine operating parameter acquisition module 110, configured to acquire a time - series set of operating parameter data of a target wind turbine object from a wind turbine SCADA system, where the operating parameter data includes a wind turbine temperature value, a wind turbine rotation speed value, a wind turbine power value, and a wind turbine voltage value; an electric field outlet power data acquisition module 120, configured to acquire a time - series set of electric field outlet power data from an energy management platform; a wind turbine operating state characterization module 130, configured to perform an optimized characterization of the wind turbine operating state based on temporal mutual information perception on the time - series set of the operating parameter data and the time - series set of the electric field outlet power data to obtain optimized wind turbine operating state characterization information; a wind turbine early warning control module 140, configured to determine a wind turbine operating state detection result based on the optimized wind turbine operating state characterization information, and determine an early warning level and a control strategy; and an operating control signal generation module 150, configured to generate a wind turbine operating control signal based on the early warning level and the control strategy.

[0025] In the above intelligent control system 100 for the operating state of a wind turbine based on multi-source data, the wind turbine operating parameter acquisition module 110 is used to collect a time series set of operating parameter data of the target wind turbine object from the wind turbine SCADA system. The operating parameter data includes the wind turbine temperature value, the wind turbine speed value, the wind turbine power value, and the wind turbine voltage value. It should be understood that as the core equipment of a wind farm, the operating state of a wind turbine is directly affected not only by its own mechanical structure and electrical system, but also by the combined effects of various external environmental factors such as wind speed changes and grid dispatching instructions. Therefore, by collecting and analyzing the time series data of the above key operating parameters in real time, the actual operating conditions of the wind turbine under different working conditions can be comprehensively and carefully captured. Among them, the wind turbine temperature value can directly reflect the working temperature of mechanical components (such as bearings, gearboxes). Excessive or too low temperature may lead to a decline in equipment performance or even failure; the wind turbine speed value reflects the rotation speed of the wind turbine blades and is crucial for evaluating the power generation efficiency of the wind turbine and detecting potential mechanical problems; the wind turbine power value shows the current output capacity of the wind turbine and is one of the key indicators for measuring the working efficiency of the wind turbine. At the same time, it can also indirectly reflect whether the wind turbine is in the best working state; while the wind turbine voltage value helps to understand the stability of the wind turbine power system and ensure that the power quality meets the standard requirements. The process of collecting the time series set of operating parameter data of the target wind turbine object from the wind turbine SCADA system, especially the wind turbine temperature value, the wind turbine speed value, the wind turbine power value, and the wind turbine voltage value, is a precise and systematic operation process aimed at ensuring that the collected data is both accurate and highly time-consistent. First of all, an efficient and stable data transmission channel needs to be established, which is usually achieved through industrial Ethernet or dedicated communication protocols, such as standard protocols like Modbus TCP / IP or IEC 60870-5-104. These protocols can support high-speed data transmission and have good real-time performance and reliability, thus ensuring seamless docking of data between the wind turbine SCADA system and the central control system. Next, at the wind turbine end, sensors installed at various key positions (such as generators, gearboxes, main bearings, etc.) are responsible for continuously monitoring and recording the corresponding physical quantities. For example, temperature sensors are used to measure the working temperature of each component inside the wind turbine, and speed sensors are used to detect the rotation speed of the wind turbine blades. Power transmitters and voltage transformers are respectively used to obtain the active power output by the wind turbine and the voltage level on the grid side. These sensors convert physical signals into electrical signals, and after appropriate preprocessing steps such as amplification and filtering, they form a standard electrical signal format that can be recognized by digital systems. In order to ensure that the collected data can truly reflect the actual operating conditions of the wind turbine, great attention must be paid to the time synchronization problem during the collection process.Under normal circumstances, high-precision time synchronization technologies such as GPS-based NTP (Network Time Protocol) or PTP (Precision Time Protocol) are adopted, enabling wind turbine equipment distributed in different geographical locations to perform data sampling under a unified time reference. This not only helps improve the accuracy of subsequent data analysis but also facilitates cross-device comparative analysis. In addition, considering the complex and variable operating environment of wind turbines, the data acquisition system also needs to have a certain anti-interference ability to prevent electromagnetic noise or other external factors from affecting the signal quality. To this end, in addition to taking shielding measures at the hardware level, filtering algorithms such as Kalman filtering or adaptive filtering should also be added in software design to remove abnormal fluctuations and noise interference, ensuring that the data finally entering the database is pure and reliable. When it comes to specific data acquisition strategies, an appropriate sampling frequency needs to be set according to actual requirements. For rapidly changing parameters such as wind turbine speed and power, a relatively high sampling frequency is often required to capture their transient characteristics; while for relatively stable temperature parameters, a lower sampling interval can be selected, which can not only meet the monitoring requirements but also reduce unnecessary data storage pressure. At the same time, to cope with possible data loss or damage situations, a caching mechanism is generally set locally. Once it is found that the network connection is interrupted or data cannot be uploaded to the central server in a timely manner due to other reasons, the data can be temporarily stored in the local storage medium and uploaded later when conditions permit. During this process, the application of data compression technology also needs to be considered to reduce bandwidth occupancy and storage costs. For example, lossless compression algorithms are used to preliminarily process the original data to minimize the data volume as much as possible without affecting the integrity of information.

[0026] In the above intelligent control system 100 for the operating state of a wind turbine based on multi-source data, the electric field outlet power data acquisition module 120 is used to collect a time series set of electric field outlet power data from the energy management platform. It should be understood that the electric field outlet power, as one of the key indicators for measuring the power generation efficiency of the entire wind farm, not only reflects the working efficiency of a single wind turbine, but more importantly, it can reflect the comprehensive performance of the entire wind farm as an organic whole in the face of changes in the external environment and fluctuations in the operating conditions of internal equipment. By continuously and frequently collecting this data, detailed information on the dynamic changes in the energy output of the wind farm can be obtained. Specifically, the time series data of the electric field outlet power contains rich information content, which can not only reflect the instantaneous intensity and stability of the wind resources, but also reveal the energy convergence and distribution patterns caused by the collaborative operation of multiple wind turbines. First of all, the time series characteristics of the electric field outlet power data provide a direct basis for analyzing the overall performance of the wind farm. Since a wind farm usually consists of dozens or even hundreds of wind turbines, the output power of each wind turbine will fluctuate due to its own mechanical structure, electrical system, and external environmental factors (such as wind speed, temperature, etc.). However, when these individual differences are integrated to form the total electric field outlet power, the complex interaction relationships between them become apparent. For example, in some cases, although some wind turbines may be in a shutdown state due to maintenance or faults, the entire wind farm can still maintain a relatively stable power output level by adjusting the output strategies of other normally operating wind turbines. Such phenomena are difficult to comprehensively capture based on the data of a single wind turbine, but through the real-time monitoring of the electric field outlet power, such group behavior characteristics can be effectively identified, and based on this, the overall health status and operation efficiency of the wind farm can be evaluated. Secondly, the time series data of the electric field outlet power also provides important clues for predicting and coping with potential risks. In actual operation, the wind farm faces challenges from multiple aspects such as the natural environment and grid dispatching instructions. On the one hand, changes in weather conditions may cause a sudden increase or decrease in wind speed, which in turn leads to significant fluctuations in the electric field outlet power; on the other hand, the dispatching instructions issued by the grid operator according to the load demand and network conditions also require the wind farm to adjust its output power in a timely manner. In this context, it is particularly crucial to accurately grasp the historical trend of the electric field outlet power and its internal relationship with relevant variables. By deeply mining the patterns and rules in the time series data of the electric field outlet power, not only can abnormal situations that may occur, such as the risk of large-scale power outages caused by extreme weather events, be pre-warned in advance, but it also helps to formulate more scientific and reasonable emergency plans to ensure that in case of emergencies, a rapid response can be made and appropriate measures can be taken to protect the equipment safety and ensure the continuity of power supply.Specifically, first of all, a stable and efficient data transmission channel needs to be established, which usually involves the application of industrial Ethernet or dedicated communication protocols, such as standard protocols like IEC 61850, Modbus TCP / IP or DNP3. These protocols not only support high-speed data transmission but also possess good real-time performance and reliability, thus ensuring seamless docking of the electric field export power data between the energy management platform and the central control system. In addition, multiple sensors and measurement devices deployed within the wind farm are responsible for continuously monitoring and recording the actual power output value at the electric field export. These devices convert physical quantities into electrical signals, and after appropriate preprocessing steps such as amplification and filtering, they form a standard electrical signal format recognizable by the digital system. To ensure that the collected data can truly reflect the changes in the electric field export power, the issue of time synchronization must be addressed. Generally, high-precision time synchronization technologies, such as the Network Time Protocol (NTP) or Precision Time Protocol (PTP) based on GPS, are adopted, enabling energy management platforms located in different geographical positions to perform data sampling under a unified time reference. This time synchronization mechanism not only improves the consistency and comparability of data but also makes it possible to conduct comparative analysis across devices. At the same time, considering that the electric field export power is affected by various factors, such as grid dispatching instructions and wind speed changes, the data acquisition system also needs to have a certain anti-interference ability to prevent electromagnetic noise or other external factors from affecting the signal quality. For this purpose, in addition to taking shielding measures at the hardware level, filtering algorithms, such as Kalman filtering or adaptive filtering, should also be added in software design to remove abnormal fluctuations and noise interference, ensuring that the data finally entering the database is pure and reliable. In terms of specific acquisition strategies, it is crucial to set an appropriate sampling frequency according to actual needs. For rapidly changing electric field export power data, a relatively high sampling frequency is often required to capture its transient characteristics; while for relatively stable long-term trends, a lower sampling interval can be selected, which can not only meet the monitoring requirements but also reduce unnecessary data storage pressure. At the same time, to cope with possible data loss or damage, a caching mechanism is generally set locally. Once it is found that the network connection is interrupted or data cannot be uploaded to the central server in a timely manner due to other reasons, the data can be temporarily saved in the local storage medium and then re-transmitted when conditions permit. During this process, the application of data compression technology also needs to be considered to reduce bandwidth occupancy and storage costs. For example, lossless compression algorithms are used to preliminarily process the original data to ensure that the data volume is reduced as much as possible without affecting the integrity of information.

[0027] In the above intelligent control system 100 for the operating state of a wind turbine based on multi-source data, the wind turbine operating state characterization module 130 is configured to perform an optimized characterization of the operating state of the wind turbine based on time-series mutual information perception on the time-series set of the operating parameter data and the time-series set of the electric field outlet power data to obtain optimized characterization information of the operating state of the wind turbine. Figure 3 It is a block diagram of the wind turbine operating state characterization module in the intelligent control system for the operating state of a wind turbine based on multi-source data according to an embodiment of the present application. As Figure 3 shown, in the embodiment of the present application, the wind turbine operating state characterization module 130 includes: a wind turbine operating state time-series joint encoding unit 131, configured to divide the time-series set of the operating parameter data of the target wind turbine object by the operating parameter sample dimension and then perform joint encoding based on the time series to obtain a self-information characterization vector of the wind turbine operating state; an electric field outlet power data time-series encoding unit 132, configured to obtain an electric field outlet power data time-series encoding vector by using an LSTM-based electric field outlet power time-series encoder for the time-series set of the electric field outlet power data; a wind turbine operating state optimized characterization unit 133, configured to obtain a mutual information-guided wind turbine operating state optimized characterization matrix as the optimized characterization information of the wind turbine operating state by using a fine-grained global interaction network based on time-series feature information for the self-information characterization vector of the wind turbine operating state and the electric field outlet power data time-series encoding vector.

[0028] Specifically, the wind turbine operating state time-series joint encoding unit 131 is configured to divide the time-series set of the operating parameter data of the target wind turbine object by the operating parameter sample dimension and then perform joint encoding based on the time series to obtain a self-information characterization vector of the wind turbine operating state. In the embodiment of the present application, the wind turbine operating state time-series joint encoding unit 131 is configured to: divide the time-series set of the operating parameter data of the target wind turbine object by the operating parameter sample dimension, and then input it into a time-series encoder based on a multi-scale causal convolutional neural network to obtain a wind turbine temperature time-series encoding vector, a wind turbine speed time-series encoding vector, a wind turbine power time-series encoding vector, and a wind turbine voltage time-series encoding vector; perform joint encoding on the wind turbine temperature time-series encoding vector, the wind turbine speed time-series encoding vector, the wind turbine power time-series encoding vector, and the wind turbine voltage time-series encoding vector to obtain the self-information characterization vector of the wind turbine operating state.

[0029] Specifically, after dividing the time series set of the operation parameter data of the target wind turbine object according to the operation parameter sample dimension, it is input into a time series encoder based on a multi-scale causal convolutional neural network to obtain a wind turbine temperature time series encoding vector, a wind turbine speed time series encoding vector, a wind turbine power time series encoding vector, and a wind turbine voltage time series encoding vector. It should be understood that considering that parameters such as wind turbine temperature, speed, power, and voltage have different dynamic response characteristics in physical mechanisms (such as temperature changes lagging behind mechanical load fluctuations), directly performing global time series modeling is likely to cause feature confusion. Therefore, in order to break through the modeling limitation of the coupling relationship of multi-dimensional time series features in traditional single-machine monitoring systems, in the technical solution of this application, after dividing the time series set of the operation parameter data of the target wind turbine object according to the operation parameter sample dimension, it is input into a time series encoder based on a multi-scale causal convolutional neural network to obtain a wind turbine temperature time series encoding vector, a wind turbine speed time series encoding vector, a wind turbine power time series encoding vector, and a wind turbine voltage time series encoding vector. By dividing the time series set of the operation parameter data of the target wind turbine object according to the operation parameter sample dimension and using a multi-scale causal convolutional neural network to perform time series encoding on the time series data of each wind turbine operation parameter item, an adaptive feature extraction channel can be constructed for the unique time evolution law of different operation parameter physical quantities. In particular, by using a time series encoder based on a multi-scale causal convolutional neural network, time series causality can be ensured, and leakage of future time series information can be avoided. In addition, the multi-scale convolutional kernel group can synchronously capture second-level vibration spikes, minute-level heat conduction processes, and hour-level performance degradation trends. This processing method enables each parameter encoding vector to retain its independent dynamic features and, through subsequent joint encoding, form a self-consistent representation of the equipment operation state, laying a foundation for fine-grained interaction with the global features of the power at the exit of the electric field.

[0030] Specifically, the fan temperature time-series encoded vector, the fan speed time-series encoded vector, the fan power time-series encoded vector, and the fan voltage time-series encoded vector are jointly encoded to obtain the self-information representation vector of the fan operating state. It should be understood that due to the complex non-linear coupling relationship among parameters such as fan temperature, speed, power, and voltage in the time dimension (such as the energy transfer hysteresis effect between the temperature rise of the gearbox and the speed fluctuation), constructing parameter features separately will result in the loss of the intrinsic correlation information of the mechanical-electrical-thermodynamic system. Therefore, in order to overcome the problem of feature fragmentation caused by the analysis of isolated parameters in traditional single-machine monitoring systems, in the technical solution of this application, the fan temperature time-series encoded vector, the fan speed time-series encoded vector, the fan power time-series encoded vector, and the fan voltage time-series encoded vector are further jointly encoded to obtain the self-information representation vector of the fan operating state. Through the joint encoding operation, it is possible to perform dynamic weight assignment and non-linear fusion of the parameter-specific time-series features extracted by the multi-scale causal convolutional neural network (such as the thermal inertia feature of the temperature curve and the mutation response mode of the speed signal) in the feature space, forming a self-consistent vector space that characterizes the overall operating state of the device. This fusion mechanism not only retains the independent evolution laws of each parameter, but also captures the implicit correlations between parameters through feature interaction (for example, the abnormal pitch system implied by the phase difference between the voltage dip and the power oscillation), thereby constructing a device self-information representation with physical interpretability. Specifically, the joint encoding operation is achieved by using a method based on the Attention Mechanism or the Gated Recurrent Unit (GRU). The Attention Mechanism allows the model to pay different degrees of attention to different input features at different time points, thereby enhancing the information that is more critical for the current task while suppressing the irrelevant parts. Specifically, in this context, the similarity scores between each time-series encoded vector and all other vectors can be calculated, and the weight distribution of each vector can be adjusted according to these scores, thereby realizing the dynamic weighted fusion of information. This method can not only effectively handle the noise interference in the data, but also improve the sensitivity of the model to abnormal patterns. On the other hand, as a special recursive neural network structure, GRU can well capture the long-term dependence relationships in the sequence data, which is very useful for understanding the process of the fan operating state evolving over time. During the joint encoding process, each time-series encoded vector can be used as the input sequence of the GRU, allowing the network to automatically learn how to combine historical information with the current observation to predict the future state. In particular, the update gate and reset gate mechanisms inside the GRU can flexibly control the information flow path, ensuring that only the most useful historical records are passed to the next time step, which can not only avoid the problem of gradient disappearance, but also improve the memory ability of the model.In addition, by introducing the Layer Normalization technique, the optimization problems encountered during training can be alleviated to a certain extent, making it easier for the entire network to converge to the global optimal solution. It is worth mentioning that during the entire joint coding process, when selecting the activation function, ReLU (Rectified Linear Unit) is preferentially used in most cases due to its simplicity and efficiency. However, in some specific scenarios, such as when the continuity and differentiability of the output signal need to be maintained, other forms such as tanh or sigmoid can be selected. In addition, regularization techniques such as L2 regularization or Dropout are also indispensable components, which can help prevent overfitting and ensure that the model has good generalization ability on unseen data. Especially when facing large-scale and high-dimensional datasets, appropriate regularization strategies are particularly important.

[0031] Specifically, the electric field outlet power data time series coding unit 132 is used to obtain the electric field outlet power data time series coding vector by passing the time series set of the electric field outlet power data through the LSTM-based electric field outlet power time series encoder. It should be understood that since the electric field outlet power is a comprehensive representation of the field-level energy output, its time series data not only contains minute-level active power fluctuations but also implicitly contains long-term state degradation information of the coordinated operation of multiple wind turbines (such as the output power attenuation mode caused by quarterly gearbox wear). Therefore, in the technical solution of this application, the time series set of the electric field outlet power data is further passed through the LSTM-based electric field outlet power time series encoder to obtain the electric field outlet power data time series coding vector. The unique gating mechanism of the LSTM network can effectively capture multi-time scale dependencies from seconds to months, dynamically screen key events in the historical state (such as power mutations caused by grid frequency regulation) through the forget gate, and fuse environmental context information such as current wind speed prediction using the input gate. Compared with traditional physical simulation methods, while maintaining the time series causality, this encoder adaptively learns the mapping relationship between abnormal fluctuations in the power curve and the health status of the device group. For example, when the pitch control systems of multiple wind turbines show synchronous response delays, the LSTM coding vector will form a specific harmonic resonance pattern in the frequency domain feature space.

[0032] Specifically, the wind turbine operation state optimization characterization unit 133 is used to obtain the mutual information-guided wind turbine operation state optimization characterization matrix as the wind turbine operation state optimization characterization information by passing the wind turbine operation state self-information characterization vector and the electric field outlet power data time series coding vector through the fine-grained global interaction network based on time series feature information. Figure 4 Block diagram of the wind turbine operation state optimization characterization unit in the intelligent control system for wind turbine operation state based on multi-source data according to an embodiment of the present application. As Figure 4As shown, in the embodiment of the present application, the fan operating state optimization characterization unit 133 includes: a core correlation coding sub-unit 1331 between the main components of the multi-dimensional characterization features of the fan operating state, which is used to perform core correlation information coding based on principal component analysis on the self-information characterization vector of the fan operating state and the time-series coding vector of the electric field outlet power data to obtain a set of core correlation coding vectors between the main components of the multi-dimensional characterization features of the fan operating state; a core correlation performance analysis sub-unit 1332 of the multi-dimensional characterization of the fan operating state, which is used to calculate the core correlation performance analysis factors between every two core correlation coding vectors between the main components of the multi-dimensional characterization features of the fan operating state in the set of core correlation coding vectors between the main components of the multi-dimensional characterization features of the fan operating state to obtain a core correlation performance analysis factor association topology matrix; a mutual information-guided fan operating state optimization characterization sub-unit 1333, which is used to perform graph-structure-based association coding on the set of core correlation coding vectors between the main components of the multi-dimensional characterization features of the fan operating state based on the core correlation performance analysis factor association topology matrix of the multi-dimensional characterization of the fan operating state to obtain the mutual information-guided fan operating state optimization characterization matrix. It should be understood that since the self-information characterization vector of the fan operating state (including device-level features such as temperature and speed) and the time-series coding vector of the electric field outlet power data (reflecting the field-level energy dynamics) carry state information of different spatio-temporal scales respectively, direct linear fusion will ignore the complex non-linear coupling mechanism between the mechanical system and the energy system. At the same time, due to the time-series characteristics of the electric field outlet power providing a global perspective and context information of the wind farm operation. The operating state of a single fan is affected not only by its own factors but also by the entire wind farm environment (such as the overall wind resource and grid state). The time-series characteristics of the electric field outlet power can reflect these global influencing factors. In order to use the time-series characteristics of the electric field outlet power to optimize the expression of the self-information of the fan operating state, so as to more accurately detect and control the fan operating state, in the technical solution of the present application, the self-information characterization vector of the fan operating state and the time-series coding vector of the electric field outlet power data are further passed through a fine-grained global interaction network based on time-series feature information to obtain a mutual information-guided fan operating state optimization characterization matrix as the fan operating state optimization characterization information. The processing process of the fine-grained global interaction network based on time-series feature information is to perform decoupling and reconstruction on the self-information characterization vector of the fan operating state and the time-series coding vector of the electric field outlet power data through feature principal component analysis, eliminate the pseudo-correlation between the daily periodic fluctuations of the fan operating state such as the fan bearing temperature and the power fluctuations caused by grid scheduling (such as the accidental synchronization between the daily periodic changes in environmental temperature and the power fluctuations caused by grid scheduling), and at the same time use the principal component core correlation coding network to capture the cross-scale resonance mode between the gearbox vibration harmonics and the field power harmonics.The construction of the performance operator correlation topology matrix quantifies the implicit device-field interaction relationship (e.g., the icing of the blades of a certain fan affects the output of adjacent units through the wake effect) into a computable graph structure topology. Then, through the neighborhood information aggregation mechanism of the graph convolutional neural network, the weak representation of the single-unit state anomaly in the electric field outlet power (such as the distortion of the power curve) is spatio-temporally correlated with the collaborative decline mode of multiple fans (such as the cumulative effect of the yaw angle deviation of the whole field). The dynamic optimization mechanism of the cross-section function matrix effectively suppresses the dimensional fracture problem in the feature space projection process, reducing the false alarm rate in the wind speed mutation scenario. This deep feature interaction and fusion mechanism enables the single-unit state representation to not only retain its own operating characteristics but also incorporate the temporal context information of the field-level energy dynamics, so as to optimize the self-information of the fan operating state using the temporal characteristics of the electric field outlet power, thereby more accurately representing the operating state of the fan and providing a basis for optimizing the collaborative control strategy of the wind farm.

[0033] Specifically, the core correlation encoding subunit 1331 between the main components of the multi-dimensional representation features of the fan operating state is used to perform core correlation information encoding based on principal component analysis on the self-information representation vector of the fan operating state and the temporal encoding vector of the electric field outlet power data to obtain a set of core correlation encoding vectors between the main components of the multi-dimensional representation features of the fan operating state. In an embodiment of the present application, the core correlation encoding subunit 1331 between the main components of the multi-dimensional representation features of the fan operating state is used to: respectively perform principal component analysis on the self-information representation vector of the fan operating state and the temporal encoding vector of the electric field outlet power data to obtain a set of self-information feature principal component encoding vectors of the fan operating state and a set of temporal feature principal component encoding vectors of the electric field outlet power data; perform principal component core correlation encoding on each corresponding self-information feature principal component encoding vector of the fan operating state and temporal feature principal component encoding vector of the electric field outlet power data in the set of self-information feature principal component encoding vectors of the fan operating state and the set of temporal feature principal component encoding vectors of the electric field outlet power data to obtain a set of core correlation encoding vectors between the main components of the multi-dimensional representation features of the fan operating state.

[0034] Specifically, principal component analysis is respectively performed on the self-information representation vector of the fan operating state and the temporal encoding vector of the electric field outlet power data to obtain a set of self-information feature principal component encoding vectors of the fan operating state and a set of temporal feature principal component encoding vectors of the electric field outlet power data, which is expressed by the principal component analysis formula as:

[0035]

[0036] Among them, V1 is the self - information characterization vector of the fan operation state, V2 is the time - series coding vector of the electric field outlet power data, PCA(·) is the principal component analysis of features, X is the set of principal component coding vectors of the self - information features of the fan operation state, y is the set of principal component coding vectors of the time - series features of the electric field outlet power data, x1, x2, x i , x m are respectively the 1st, 2nd, ith, and mth principal component coding vectors of the self - information features of the fan operation state in the set of principal component coding vectors of the self - information features of the fan operation state, y1, y2, y i , y m are respectively the 1st, 2nd, ith, and mth principal component coding vectors of the time - series features of the electric field outlet power data in the set of principal component coding vectors of the time - series features of the electric field outlet power data, Λ1 is the diagonal matrix of the self - information features of the fan operation state, λ 11 and λ 1m are respectively the eigenvalues corresponding to x1 and x m , Λ2 is the diagonal matrix of the time - series features of the electric field outlet power data, λ 21 and λ 2m are respectively the eigenvalues corresponding to y1 and y m It should be understood that the core motivation for performing principal component analysis (PCA) on the self - information characterization vector of the fan operation state and the time - series coding vector of the electric field outlet power data is to solve the problems of raw data dimension redundancy and feature coupling. Since there are multi - scale coupling relationships between the fan operation parameters (temperature, speed, power, voltage) and the electric - field - level power data, for example, environmental disturbances (such as sudden changes in wind speed) may simultaneously trigger temperature fluctuations and power oscillations, resulting in significant linear correlations between the two types of data at the time - series feature level. This multicollinearity will not only increase the computational complexity of subsequent model training, but may also lead to a decline in the generalization ability of the model due to interference from redundant information between features, and even produce misleading associations. By performing orthogonal transformation on the two types of data through PCA, the original high - dimensional features can be mapped to a low - dimensional principal component space, and mutually orthogonal principal components can be extracted. This process is essentially a decoupling and reconstruction of the internal structure of the data, and the feature dimensions with the highest contribution to the system state representation are selected through the principle of maximizing variance retention. For example, the frequency - domain features of the fan vibration signal and the harmonic components of the power curve may be separated in different principal component directions, thereby eliminating the pseudo - correlations caused by the periodic instructions of power grid scheduling and the daily variation of environmental temperature. By regulating the number of retained principal components, while reducing the data dimension, the sensitivity to key fault modes such as fan gearbox wear and ice can be maintained, and finally noise reduction and redundancy removal of feature expression can be achieved, providing high - quality feature inputs for the subsequent deep interaction network based on mutual information.

[0037] Specifically, the set of principal component encoding vectors of the self-information features of the fan operating state and the set of principal component encoding vectors of the time-series features of the electric field outlet power data are respectively subjected to principal component core correlation encoding for each corresponding pair of principal component encoding vectors of the self-information features of the fan operating state and the principal component encoding vectors of the time-series features of the electric field outlet power data to obtain a set of core correlation encoding vectors between the principal components of the multi-dimensional representation features of the fan operating state, which is expressed by the principal component core correlation encoding formula as:

[0038]

[0039] where, ||·|| represents the one-norm of the vector, α and β respectively represent trainable weighted hyperparameters, and v i is the i-th core correlation encoding vector between the principal components of the multi-dimensional representation features of the fan operating state in the set of core correlation encoding vectors between the principal components of the multi-dimensional representation features of the fan operating state. It should be understood that although PCA realizes the linear decoupling and dimensionality reduction of features through orthogonal transformation, there may be complex non-linear coupling relationships between the fan operating state and the electric field power data (such as the phase coupling between the gearbox vibration frequency and the power harmonic). This relationship often shows a linearly inseparable pattern in the original low-dimensional principal component space. By introducing a kernel function (such as a radial basis kernel, polynomial kernel) to map the principal components to a high-dimensional feature space, it is possible to convert the originally linearly independent low-dimensional features into a linearly separable representation form in the high-dimensional space while maintaining the computational efficiency, thereby activating the deep neural network's ability to capture the implicit non-linear association patterns between features. This step constructs the dynamic association weights of the principal component pairs and uses the non-linear activation function in the forward propagation process of the network to encode the interaction relationship of each pair of principal components as a similarity measure in the high-dimensional space, so that the feature pairs with strong correlation are close in distance in the mapped space, and vice versa. The process of generating the set of core correlation encoding vectors between the principal components of the multi-dimensional representation features of the fan operating state essentially constructs a bridging mechanism that semantically aligns the principal component space after PCA dimensionality reduction with the graph structure space processed by the subsequent graph neural network, providing a computable mathematical basis for fine-grained interaction between heterogeneous components, and finally significantly improving the model's representation accuracy for composite fault modes such as fan gearbox wear and blade icing.

[0040] Specifically, the multi-dimensional representation core correlation performance analysis sub-unit 1332 of the fan operating state is used to calculate the core correlation performance analysis factor between every two core correlation encoding vectors between the principal components of the multi-dimensional representation features of the fan operating state in the set of core correlation encoding vectors between the principal components of the multi-dimensional representation features of the fan operating state to obtain a core correlation performance analysis factor association topology matrix, which is expressed by the multi-dimensional representation core correlation performance analysis formula of the fan operating state as:

[0041]

[0042] wherein, v i,k and v j,k are respectively the eigenvalues at the k-th position in the core correlation coding vectors of the multi-dimensional characterization feature principal components of the operating states of the i-th and j-th wind turbines, n is the number of eigenvalues in the core correlation coding vectors of the multi-dimensional characterization feature principal components of the operating states of the wind turbines, s(v i , v j ) is the core correlation performance analysis factor of the multi-dimensional characterization of the operating state of the wind turbine between v i and v j , and M A is the correlation topology matrix of the core correlation performance analysis factors of the multi-dimensional characterization of the operating state of the wind turbine. It should be understood that although the core correlation coding of the principal components solves the problem of linear inseparability in the low-dimensional space through kernel function mapping, the correlation relationship between the principal component pairs may still exhibit high non-linearity and dynamic time-variation (such as the instantaneous phase coupling between gearbox vibration and power harmonics). By defining customizable performance metric functions (such as mutual information, time series synchronization coefficient, or prediction error correlation), the semantic association strength between feature pairs can be quantified in specific task scenarios. For example, in the fault warning task, the co-variation pattern between equipment parameters and power fluctuations is emphasized, and in the energy efficiency optimization task, the statistical consistency of the wind speed-power curve is concerned. This numerical expression of the association strength transforms the abstract feature interaction into a structured topological relationship, enabling the graph neural network to perform efficient reasoning based on the topological features of the adjacency matrix (such as node centrality, community division). The correlation topology matrix of the core correlation performance analysis factors of the multi-dimensional characterization of the operating state of the wind turbine, as the input of the graph neural network, essentially constructs a semantic bridge between the device state features and the field energy features, enabling the model to capture the cross-scale fault propagation paths (such as the icing of the blades of a single wind turbine causing power resonance of adjacent units through the wake effect) through the message passing mechanism, and ultimately improving the system's early recognition ability for complex faults and the global state assessment accuracy.

[0043] Specifically, the mutual information-guided fan operation state optimization characterization subunit 1333 is configured to perform graph-structure-based association coding on the set of core association coding vectors among the multi-dimensional characterization feature principal components of the fan operation state based on the fan operation state multi-dimensional characterization core association performance analysis factor association topology matrix to obtain the mutual information-guided fan operation state optimization characterization matrix. In an embodiment of the present application, the mutual information-guided fan operation state optimization characterization subunit 1333 is configured to: perform compensation for the disorder of the core spatial distribution of the multi-dimensional characterization of the fan operation state on each core association coding vector among the multi-dimensional characterization feature principal components of the fan operation state to obtain an optimized set of core association coding vectors among the multi-dimensional characterization feature principal components of the fan operation state; input the optimized set of core association coding vectors among the multi-dimensional characterization feature principal components of the fan operation state and the fan operation state multi-dimensional characterization core association performance analysis factor association topology matrix into a graph convolutional neural network model to obtain the mutual information-guided fan operation state optimization characterization matrix.

[0044] Specifically, the mutual information-guided fan operation state optimization characterization subunit 1333 is configured to perform graph-structure-based association coding on the set of core association coding vectors among the multi-dimensional characterization feature principal components of the fan operation state based on the fan operation state multi-dimensional characterization core association performance analysis factor association topology matrix to obtain the mutual information-guided fan operation state optimization characterization matrix, which is expressed by the mutual information-guided fan operation state optimization characterization formula as:

[0045]

[0046] where GCN(·,·) represents the graph convolutional neural network model, M cIt is a mutual information-guided optimization characterization matrix for the operating state of the fan. It should be understood that by constructing a graph neural network model, the core correlation performance analysis factor correlation topology matrix of the multi-dimensional characterization of the fan operating state can be mapped to the connection relationship between graph nodes, and the graph convolution operation is used to dynamically capture cross-scale interaction features during the local neighborhood information aggregation process. Compensating for the spatial distribution disorder of the core correlation coding vectors between the principal components of the multi-dimensional characterization features of the fan operating state is to solve the problem of distribution deviation that may occur in high-dimensional features under a randomly initialized adjacency matrix. The mutual information-guided optimization characterization matrix for the fan operating state enables the features to have a more consistent geometric distribution characteristic in the graph space, thus adapting to the propagation mechanism of graph convolution. This step uses multi-hop graph convolution operations to enable each feature node to aggregate information from multi-level neighbors, effectively modeling the temporal causal relationship and spatial propagation path between the equipment state and the field energy dynamics (such as blade icing of a single fan causing power resonance of adjacent units through wake effects), and finally generating an optimized characterization matrix that fuses global mutual information, significantly improving the early recognition accuracy of complex faults and the reliability of control decisions in the system.

[0047] In a preferred example, during the calculation process of the above graph convolutional neural network model, the core correlation performance analysis factor correlation topology matrix M of the multi-dimensional characterization of the fan operating state A As the topological structure space, each core correlation coding vector v between the principal components of the multi-dimensional characterization features of the fan operating state i will serve as the master and slave nodes and follow the spatial distribution rule, that is, v i →M A And, considering the dimensional discontinuity between the core correlation coding vectors between the principal components of the multi-dimensional characterization features of the fan operating state and the core correlation performance analysis factor correlation topology matrix of the multi-dimensional characterization of the fan operating state It is necessary to construct an initial fan operating state multi-dimensional characterization feature cross-section function matrix to achieve dimensional calibration.

[0048] On the other hand, it is also necessary to correct the problem of spatial distribution disorder caused by the random potential field in the initial fan operating state multi-dimensional characterization feature cross-section function matrix M Bi First, multiply each core correlation coding vector v between the principal components of the multi-dimensional characterization features of the fan operating state i with the corresponding initial fan operating state multi-dimensional characterization feature cross-section function matrix M Bi to obtain the fan operating state multi-dimensional characterization feature cross-section compactified vector v Bi , and then the fan operating state multi-dimensional characterization feature cross-section compactified vector v corresponding to each core correlation coding vector v between the principal components of the multi-dimensional characterization features of the fan operating state i ​Bi Two-dimensional splicing to obtain the compact matrix M of the multi-dimensional characterization feature section of the fan operation state C :

[0049]

[0050] M C =(v B1 T ,v B2 T ,…)

[0051] where v i is the i-th core correlation coding vector between the main components of the multi-dimensional characterization features of the fan operation state in the set of core correlation coding vectors between the main components of the multi-dimensional characterization features of the fan operation state, and M Bi is the initial function matrix of the multi-dimensional characterization feature section of the fan operation state, is matrix multiplication, v Bi is the i-th compactification vector of the multi-dimensional characterization feature section of the fan operation state in the sequence of compactification vectors of the multi-dimensional characterization feature section of the fan operation state, (·,·,…) is two-dimensional splicing processing, and M C is the compact matrix of the multi-dimensional characterization feature section of the fan operation state.

[0052] In this way, the Gaussian similarity coefficient between the compact matrix M of the multi-dimensional characterization feature section of the fan operation state C and the associated topological matrix M of the core correlation performance analysis factor of the multi-dimensional characterization of the fan operation state A can be calculated, and an iterative optimization process that drives the high Gaussian similarity coefficient to approach zero is used for the initial function matrix M of the multi-dimensional characterization feature section of the fan operation state Bi to obtain the optimized function matrix M of the multi-dimensional characterization feature section of the fan operation state Bi ':

[0053]

[0054] where is subtraction by position, ||·|| F is the F-norm of the matrix, σ 2 is the variance of the set composed of all matrix values of M C and M A , and exp is the value of the natural exponential function with the natural constant e as the base.

[0055] Thus, by further optimizing the function matrix M of the multi-dimensional characterization feature section of the fan operation state Bi ', the core correlation coding vector between the main components of the multi-dimensional characterization features of the fan operation state is optimized:

[0056]

[0057] Among them, M Bi ' is the optimized multi-dimensional characterization feature cross-section function matrix of the fan operation state, and v' i is the optimized core correlation coding vector between the main components of the multi-dimensional characterization of the fan operation state corresponding to the core correlation coding vector between the main components of the multi-dimensional characterization of the i-th fan operation state.

[0058] In this way, in the operation process of the graph convolution inference architecture, by dynamically adjusting the topological adaptability of feature embedding, the phenomenon of topological structure spatial distribution deviation caused by the random initialization perturbation of high-dimensional feature embedding vectors can be effectively alleviated, thereby improving the discrimination efficiency of the model for composite fault modes. This process constructs a collaborative optimization mechanism for feature embedding and associated topology, converts the distribution deviation caused by random noise into quantifiable topological constraint conditions, and finally realizes the collaborative optimization of the model convergence accuracy and generalization ability.

[0059] In the above-mentioned intelligent control system 100 for the fan operation state based on multi-source data, the fan early warning control module 140 is used to determine the fan operation state detection result based on the optimized characterization information of the fan operation state, and determine the early warning level and control strategy. In the embodiment of the present application, the fan early warning control module 140 is used to: pass the mutual information-guided fan operation state optimization characterization matrix through a fan operation state detector based on a classifier to obtain the fan operation state detection result, and the fan operation state detection result is a fan operation state label; based on the fan operation state detection result, determine the early warning level and the control strategy.

[0060] Specifically, the mutual information-guided fan operation state optimization characterization matrix is passed through a classifier-based fan operation state detector to obtain the fan operation state detection result, which is a fan operation state label. In a specific embodiment of the present application, the fan operation state detection results include healthy, sub-healthy, slightly abnormal, and severely abnormal. It should be understood that the previously constructed mutual information-guided fan operation state optimization characterization matrix integrates various key parameters from the fan SCADA system (such as temperature, speed, power, voltage, etc.) and the temporal characteristics of the electric field outlet power data. It not only contains the working state information of individual fan internal components but also reflects the impact of the entire wind farm environment on it. To convert these rich characterization information into specific fan operation state labels, advanced machine learning techniques, especially classifier models, need to be introduced. Selecting an appropriate classifier is one of the key steps to achieve this goal. In practical applications, commonly used classifiers include Support Vector Machine (SVM), Random Forest, Neural Networks, etc. These classifiers have their own advantages. For example, SVM is good at dealing with non-linear problems in high-dimensional spaces, while Random Forest is widely used in data analysis in complex environments due to its good robustness and noise resistance. Regardless of which classifier is used, its core task is to establish a mapping relationship model from input features to output categories through learning a large amount of labeled historical data. In this process, the mutual information-guided fan operation state optimization characterization matrix serves as the input feature, providing a detailed description of the current state of the fan, and the expected output is the clear fan operation state label, such as healthy, sub-healthy, slightly abnormal, severely abnormal, etc. Specifically, in the training phase, the classifier will receive a large number of sample data, each sample consisting of an optimized characterization matrix and the corresponding fan operation state label. Through learning these samples, the classifier gradually adjusts its own parameter settings so that it can accurately predict the corresponding fan operation state when given new unseen data. For example, when encountering a new mutual information-guided fan operation state optimization characterization matrix, the trained classifier can, based on the knowledge patterns learned before, determine whether the fan represented by this matrix is in a normal working state or what potential problems exist. If it is found that certain specific eigenvalue deviates from the normal range, such as too high temperature or abnormal power fluctuation, then the classifier may classify it as "slightly abnormal" or "severely abnormal".

[0061] Specifically, the warning level and control strategy are determined based on the wind turbine operating status detection results. In an embodiment of the present application, the control strategies include speed limiting, capacity reduction, and shutdown. It should be understood that the wind turbine operating status detection results provide key information about the wind turbine's current health status. These results are typically derived by a classifier model based on mutual information-guided optimization of the wind turbine operating status representation matrix and are classified as "healthy," "sub-healthy," "slightly abnormal," or "severely abnormal." Each status label corresponds to a different risk level and potential impact range, requiring the development of corresponding warning levels and control strategies. Specifically, when a wind turbine is determined to be in a "healthy" state, it means that all its key parameters are within normal ranges, capable of stable operation, and with no obvious signs of failure. At this point, there is no need to trigger any warning signals or perform additional control operations; only regular monitoring is required. However, if the wind turbine's status deviates from the ideal range and enters a state such as "sub-healthy," "slightly abnormal," or "severely abnormal," the corresponding warning mechanism and control strategy must be immediately activated. For example, in a "sub-healthy" state, while the wind turbine hasn't yet shown obvious performance degradation or malfunction, certain key parameters have shown potential risk trends, such as a slight increase in temperature or slight fluctuation in power output. In this case, a lower-level warning signal (such as a yellow alert) should be set to alert operations and maintenance personnel to the turbine's operation and recommend preventive maintenance checks. At the same time, more moderate control measures can be considered, such as adjusting the turbine's operating mode to keep it operating within a safe range and prevent further deterioration. As the turbine's operating status deteriorates further to a "minor abnormality," the situation becomes more complex and urgent. At this point, some key parameters of the wind turbine have exceeded normal ranges, potentially indicating issues such as local component aging or minor faults. To prevent these issues from spreading and impacting the stability of the entire system, the warning level must be raised (such as an orange alert) and a more stringent control strategy must be immediately implemented. Speed limiting is a common control measure in this situation. By limiting the rotational speed of the wind turbine blades, mechanical loads can be effectively reduced, mitigating the risk of equipment damage due to overload. In addition, other auxiliary measures can be combined, such as optimizing the pitch angle or adjusting the power distribution strategy of the grid access point, to try to keep the wind turbine in a relatively safe state and continue to generate electricity, while buying time for subsequent maintenance work. When a wind turbine is diagnosed as a "serious abnormality" state, it indicates that there are serious hidden faults inside it. If not handled in time, it may cause major accidents or even shutdowns. At this time, not only is it necessary to issue the highest level of early warning signal (such as a red alert), but also to quickly implement more radical control strategies. Derating is an effective emergency measure, which is to proactively reduce the rated power output of the wind turbine to reduce the pressure on the damaged components and prevent the fault from further expanding. At the same time, professional technicians should be arranged to arrive at the site as soon as possible for a comprehensive inspection and repair.If a fault is found to be unrepairable or poses a serious safety hazard, decisive shutdown measures must be taken to completely cut off the power supply and ensure the safety of personnel and equipment. Although shutdown may result in certain economic losses, in the long run, it is undoubtedly the best option for ensuring the overall operational safety of the wind farm. Throughout the entire process, the setting of warning levels and control strategies is not static and needs to be flexibly adjusted according to specific circumstances. On the one hand, it is necessary to fully consider different types of faults and their impact on the system; on the other hand, it is also necessary to balance economic benefits and safety. For example, in certain special scenarios, such as extreme weather conditions, even if the wind turbine is in a "minor abnormality" state, immediate shutdown measures may be required to prevent the harsh environment from exacerbating the fault. In a specific embodiment of the present application, when the wind turbine is connected to the grid, based on a model such as the temperature approaching the alarm threshold or the temperature rise rate exceeding a certain value, the wind turbine is determined to require speed limiting. An audible and visual alarm is used to notify the user of abnormal critical temperature information and to initiate a capacity reduction operation. Furthermore, after the wind turbine is reduced in capacity or shut down and the critical temperature information returns to a certain value, it is restarted or restored to full power.

[0062] In the above intelligent control system 100 for the operating state of a wind turbine based on multi-source data, the operating control signal generation module 150 is used to generate a wind turbine operating control signal based on the warning level and the control strategy. It should be understood that the key to generating the control signal lies in constructing an efficient conversion mechanism that can map the abstract warning level and control strategy into specific control commands. This usually involves a series of complex calculation and logical judgment steps. First, it is necessary to quantify the warning level and convert it into a numerical index for subsequent arithmetic operations. For example, the "healthy" state can correspond to 0, the "sub-healthy" state to 1, the "slightly abnormal" state to 2, and the "severely abnormal" state to 3. Then, according to a preset rule base or algorithm model, combined with the current operating parameters of the wind turbine (such as temperature, speed, power, etc.), the most suitable control strategy is determined. In this process, various advanced mathematical tools and technical means may be applied, such as fuzzy logic, neural networks, genetic algorithms, etc., to search for the optimal solution. For example, using fuzzy logic can flexibly adjust the output result according to different input variables and their weight coefficients to adapt to complex actual situations. Once the specific control strategy is determined, the next task is to convert it into a practical and feasible control signal. This step usually relies on a specially designed control system that can receive instructions from the upper-level decision-making module and convert them into specific operations that the wind turbine can understand and execute. For example, in the speed limit scenario, the control system will send a pulse signal with a specific frequency to the frequency converter of the wind turbine to instruct it to reduce the rotation speed of the blades; in the derating mode, the output voltage and current of the inverter may be adjusted to reduce the overall power output of the wind turbine to a safe range. In addition, for the shutdown operation, the control system not only needs to cut off the power supply but also ensure that all related devices (such as the braking system, cooling system, etc.) are turned off in sequence according to the predetermined order to prevent equipment damage or other accidents caused by sudden power failure. It should be noted that to ensure the accuracy and timeliness of the control signal, the entire generation process must have a high degree of real-time performance and reliability. This means that the time delay from receiving the warning information to finally issuing the control instruction should be as short as possible, and at the same time, it should have a certain fault tolerance ability to cope with possible communication interruptions or data loss and other problems. For this reason, the method of redundant design is usually adopted, that is, backup channels or standby systems are set at key nodes to ensure that even if the main system fails, it can quickly switch to the standby solution to continue to perform tasks. In addition, a distributed computing architecture can also be introduced to disperse some computing tasks to multiple nodes for parallel processing, thereby further improving the response speed and processing ability of the system.

[0063] In summary, the intelligent control system 100 for the operating state of a wind turbine based on multi-source data according to the embodiments of the present application is elucidated. It first collects the time series set of the operating parameter data of the target wind turbine object and the time series set of the electric field outlet power data, and then constructs a two-level information fusion mechanism for the field-device, performing a depth interaction guided by mutual information between the global time series characteristics of the electric field outlet power and the local time series characteristics of the single-unit operating parameters, so as to optimize the expression of the self-information of the wind turbine operating state by using the time series characteristics of the electric field outlet power, determine the detection result of the wind turbine operating state, and determine the warning level and control strategy, and then generate a wind turbine operation control signal, which is used to break through the technical bottleneck of the decoupling of the global context information of the wind farm operation and the local device state in the traditional method, and is beneficial to more accurately detect and control the wind turbine operating state.

[0064] As described above, the intelligent control system 100 for the operating state of a wind turbine based on multi-source data according to the embodiments of the present application can be implemented in various terminal devices. In one example, the intelligent control system 100 for the operating state of a wind turbine based on multi-source data can be integrated into the terminal device as a software module and / or a hardware module. For example, the intelligent control system 100 for the operating state of a wind turbine based on multi-source data can be a software module in the operating system of the terminal device, or can be an application program developed for the terminal device; of course, the intelligent control system 100 for the operating state of a wind turbine based on multi-source data can also be one of the many hardware modules of the terminal device.

[0065] Alternatively, in another example, the intelligent control system 100 for the operating state of a wind turbine based on multi-source data and the terminal device can also be separate devices, and the intelligent control system 100 for the operating state of a wind turbine based on multi-source data can be connected to the terminal device through a wired and / or wireless network, and transmit and interact information according to a predefined data format.

Claims

1. An intelligent control system for the operating state of a fan based on multi-source data, characterized in that, Including: A fan operation parameter acquisition module, configured to acquire a time series set of operation parameter data of a target fan object from a fan SCADA system, where the operation parameter data includes a fan temperature value, a fan rotation speed value, a fan power value, and a fan voltage value; An electric field outlet power data acquisition module, configured to acquire a time series set of electric field outlet power data from an energy management platform; A fan operation state characterization module, configured to perform an optimized characterization of the fan operation state based on time series mutual information perception on the time series set of the operation parameter data and the time series set of the electric field outlet power data to obtain fan operation state optimized characterization information; A fan early warning control module, configured to determine a fan operation state detection result based on the fan operation state optimized characterization information, and determine an early warning level and a control strategy; An operation control signal generation module, configured to generate a fan operation control signal based on the early warning level and the control strategy.

2. The intelligent control system for the operating state of a fan based on multi-source data according to claim 1, wherein The fan operation state characterization module includes: A fan operation state time series joint encoding unit, configured to divide the time series set of the operation parameter data of the target fan object according to the operation parameter sample dimension, and perform joint encoding based on a time series on it to obtain a fan operation state self-information characterization vector; An electric field outlet power data time series encoding unit, configured to obtain an electric field outlet power data time series encoding vector by using an electric field outlet power time series encoder based on LSTM for the time series set of the electric field outlet power data; A fan operation state optimized characterization unit, configured to obtain a mutual information-guided fan operation state optimized characterization matrix as the fan operation state optimized characterization information by using a fine-grained global interaction network based on time series feature information for the fan operation state self-information characterization vector and the electric field outlet power data time series encoding vector.

3. The intelligent control system for the operating state of a fan based on multi-source data according to claim 2, wherein The fan operation state time series joint encoding unit is configured to: After dividing the time series set of the operation parameter data of the target fan object according to the operation parameter sample dimension, input it into a time series encoder based on a multi-scale causal convolutional neural network to obtain a fan temperature time series encoding vector, a fan rotation speed time series encoding vector, a fan power time series encoding vector, and a fan voltage time series encoding vector; Perform joint encoding on the fan temperature time series encoding vector, the fan rotation speed time series encoding vector, the fan power time series encoding vector, and the fan voltage time series encoding vector to obtain the fan operation state self-information characterization vector.

4. The intelligent control system for the operating state of a fan based on multi-source data according to claim 3, characterized in that, The fan operation state optimized characterization unit includes: A core correlation encoding subunit between main components of multi-dimensional characterization features of the fan operation state, configured to perform core correlation information encoding based on principal component analysis on the fan operation state self-information characterization vector and the electric field outlet power data time series encoding vector to obtain a set of core correlation encoding vectors between main components of multi-dimensional characterization features of the fan operation state; The core correlation performance analysis subunit of the multi-dimensional characterization of the fan operation state is used to calculate the core correlation performance analysis factors between each two core correlation coding vectors of the core correlation coding vector set of the multi-dimensional characterization feature principal components of the fan operation state, so as to obtain the core correlation performance analysis factor association topology matrix of the multi-dimensional characterization of the fan operation state; The mutual information-guided optimization characterization subunit of the fan operation state is used to perform graph-structure-based association coding on the set of core correlation coding vectors between the multi-dimensional characterization feature principal components of the fan operation state based on the core correlation performance analysis factor association topology matrix of the multi-dimensional characterization of the fan operation state, so as to obtain the mutual information-guided optimization characterization matrix of the fan operation state.

5. The intelligent control system for the operating state of a fan based on multi-source data according to claim 4, characterized in that, The core correlation coding subunit between the multi-dimensional characterization feature principal components of the fan operation state is used for: Performing principal component analysis on the self-information characterization vector of the fan operation state and the time-series coding vector of the electric field outlet power data respectively to obtain a set of self-information feature principal component coding vectors of the fan operation state and a set of time-series feature principal component coding vectors of the electric field outlet power data; Performing principal component core correlation coding on each group of corresponding self-information feature principal component coding vectors of the fan operation state and time-series feature principal component coding vectors of the electric field outlet power data in the set of self-information feature principal component coding vectors of the fan operation state and the set of time-series feature principal component coding vectors of the electric field outlet power data, so as to obtain a set of core correlation coding vectors between the multi-dimensional characterization feature principal components of the fan operation state.

6. The intelligent control system for the operating state of a wind turbine based on multi-source data according to claim 5, wherein The mutual information-guided optimization characterization subunit of the fan operation state is used for: Performing compensation for the disorder of the core space distribution of the multi-dimensional characterization of the fan operation state on each core correlation coding vector between the multi-dimensional characterization feature principal components of the fan operation state in the set of core correlation coding vectors between the multi-dimensional characterization feature principal components of the fan operation state, so as to obtain a set of optimized core correlation coding vectors between the multi-dimensional characterization feature principal components of the fan operation state; Inputting the set of optimized core correlation coding vectors between the multi-dimensional characterization feature principal components of the fan operation state and the core correlation performance analysis factor association topology matrix of the multi-dimensional characterization of the fan operation state into a graph convolutional neural network model to obtain the mutual information-guided optimization characterization matrix of the fan operation state.

7. The intelligent control system for the operating state of a fan based on multi-source data according to claim 6, characterized in that, The fan early warning control module is used for: Passing the mutual information-guided optimization characterization matrix of the fan operation state through a fan operation state detector based on a classifier to obtain the fan operation state detection result, and the fan operation state detection result is a fan operation state label; Based on the fan operation state detection result, determining the early warning level and the control strategy.

8. The intelligent control system for the operating state of a wind turbine based on multi-source data according to claim 7, characterized in that, The control strategies are speed limit, capacity reduction, and shutdown.

Images