Fan dormancy control method and system based on load prediction

Abstract

The invention relates to a fan dormancy control method and system based on load prediction. The method comprises the following steps: inputting historical fan operation data into a gradient neural network for feature extraction and state space modeling to obtain temperature prediction error data and heat dissipation load prediction error data; inputting the temperature prediction error data and the heat dissipation load prediction error data into a load prototype library for mode analysis to obtain a fan working condition vector and a working condition state transition matrix; inputting a fan working condition vector and the working condition state transition matrix into a transferability evaluation model for multi-dimensional similarity calculation to obtain heat dissipation load prediction data, and generating a sleep control instruction according to the heat dissipation load prediction data; and performing collaborative optimization control on the fan group according to the dormancy control instruction, and generating a fan group dormancy execution scheme. According to the method, the airflow influence and spatial layout between the fans are comprehensively considered, the overall heat dissipation efficiency of the group is optimized, and coordinated control from a single machine to the group is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of load prediction, and in particular to a fan sleep control method and system based on load prediction. Background Art

[0002] The fan system faces a complex and changing thermal load environment and different heat dissipation requirements, which makes fan load prediction and sleep control extremely challenging. The traditional control method based on fixed thresholds is difficult to adapt to the dynamic changes of the thermal environment and cannot effectively handle the collaborative work between fan groups.

[0003] At present, fan sleep control mainly relies on simple temperature thresholds and fixed switching strategies, lacking the ability to accurately predict and dynamically adjust real-time load status. Due to the volatility of ambient temperature and the nonlinear characteristics of heat dissipation load, a single control strategy is difficult to achieve optimal operation of fan groups. In addition, existing control methods often ignore the airflow interference and spatial layout relationship between fan groups, resulting in reduced overall heat dissipation efficiency and energy waste. In terms of fan load prediction, traditional methods mainly rely on real-time data from temperature sensors, making it difficult to predict load change trends and the influence of environmental factors in advance. At the same time, the differences in heat dissipation requirements and the diversity of operating conditions between different equipment areas pose serious challenges to the applicability of prediction models. These problems seriously restrict the improvement of the intelligent control level and energy-saving efficiency of fan groups. Summary of the invention

[0004] The main purpose of the present invention is to provide a fan sleep control method and system based on load prediction. The present invention comprehensively considers the airflow influence and spatial layout between fans, optimizes the overall heat dissipation efficiency of the group, and realizes coordinated control from a single machine to a group.

[0005] To achieve the above object, the present invention provides a fan sleep control method based on load prediction, comprising the following steps: The fan operation history data is input into the gradient neural network for feature extraction and state space modeling to obtain temperature prediction error data and heat dissipation load prediction error data; Inputting the temperature prediction error data and the heat dissipation load prediction error data into a load prototype library for pattern analysis to obtain a fan operating state vector and an operating state transfer matrix; Inputting the fan operating condition vector and the operating condition state transfer matrix into a transferability evaluation model to perform multi-dimensional similarity calculation to obtain heat dissipation load prediction data, and generating a sleep control instruction according to the heat dissipation load prediction data; The fan group is collaboratively optimized and controlled according to the sleep control instruction to generate a sleep execution plan for the fan group.

[0006] The present invention also provides a fan dormancy control system based on load prediction, comprising: A modeling module, used to input the fan operation history data into the gradient neural network for feature extraction and state space modeling, and obtain temperature prediction error data and heat dissipation load prediction error data; A mode analysis module, used for inputting the temperature prediction error data and the heat dissipation load prediction error data into a load prototype library for mode analysis to obtain a fan operating state vector and an operating state transfer matrix; A calculation module, used for inputting the fan operating condition vector and the operating condition state transfer matrix into a transferability evaluation model to perform multi-dimensional similarity calculation, obtain heat dissipation load prediction data, and generate a sleep control instruction according to the heat dissipation load prediction data; A generation module is used to perform collaborative optimization control on the fan group according to the sleep control instruction and generate a sleep execution plan for the fan group.

[0007] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0008] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0009] In summary, the technical solution provided by the present invention realizes the accurate extraction of heat dissipation load characteristics and the accurate calculation of multi-dimensional coupling errors through the combination of gradient neural network and state space model, thereby improving the accuracy of load prediction. Based on the design of load prototype library and working state transfer matrix, the adaptive classification and dynamic prediction of fan operating conditions are realized, and the adaptability of control strategy to temperature environment changes is enhanced. The transferability evaluation model is used for multi-dimensional similarity calculation, which solves the problem of dynamic adjustment of heat dissipation requirements in different regions and improves the practicality of the prediction model. The event trigger mechanism and sliding mode control strategy are introduced to realize the real-time optimization of fan sleep control and reduce the energy consumption of the system. Through the design of group collaborative control model, the airflow influence and spatial layout between fans are comprehensively considered, and the overall heat dissipation efficiency of the group is optimized. A complete exception handling and emergency response mechanism is established to improve the stability of the fan group sleep control system. The multi-objective optimization method is adopted to take into account both heat dissipation effect and energy consumption, thereby improving the energy saving level of the system. Through the hierarchical control architecture, coordinated control from single machine to group is realized, ensuring the balanced performance of the heat dissipation system. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1It is a schematic diagram of the steps of a fan sleep control method based on load prediction in one embodiment of the present invention; Figure 2 It is a structural block diagram of a fan dormancy control system based on load prediction in one embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0011] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0012] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0013] Reference Figure 1 , this embodiment provides a fan sleep control method based on load prediction, comprising the following steps: S1, input the fan operation history data into the gradient neural network for feature extraction and state space modeling, and obtain temperature prediction error data and heat dissipation load prediction error data; The fan operation history data is divided into an environmental parameter sequence and a heat dissipation parameter sequence. The environmental parameter sequence includes operating conditions related to the external environment such as ambient temperature parameters, humidity parameters and airflow parameters, while the heat dissipation parameter sequence includes equipment temperature parameters and heat dissipation load parameters, which directly reflect the heat exchange and load characteristics of the fan operation. The environmental parameter sequence is input into the input mapping unit of the gradient neural network for feature extraction. In the input mapping unit, a fully connected parameter layer and a feature extraction layer are designed. The fully connected parameter layer contains 4 neuron nodes, corresponding to the ambient temperature, humidity, airflow and time parameters respectively. The structure extracts the basic characteristics of each environmental parameter through preliminary weighting and bias adjustment. The feature extraction layer processes the output of the fully connected layer. This layer consists of 8 neuron nodes and uses the ReLU activation function, so that the nonlinear characteristics between environmental parameters can be effectively captured and the environmental mapping vector is generated. The environmental mapping vector is merged with the heat dissipation parameter sequence and input into the double-layer hidden unit of the gradient neural network to complete feature fusion. In the double-layer hidden unit, a deep residual layer and an attention calculation layer are designed. The deep residual layer contains 16 neuron nodes, which are used to retain the key information of environmental parameters during the feature extraction process and alleviate the gradient vanishing problem introduced by the deep network. At the same time, the attention calculation layer can dynamically focus on the important relationship between the heat dissipation parameters and environmental characteristics through the 12 neuron nodes it contains, thereby improving the effectiveness of feature fusion. The heat dissipation mapping vector is input into the feature fusion unit of the gradient neural network for multi-layer feature fusion. Through the synergy of multiple layers of neurons, the network can comprehensively model the input data from a higher dimension and a wider perspective, extract the deep feature relationship in the heat dissipation mapping vector, and generate the fan heat dissipation feature vector. The fan heat dissipation feature vector is input into the state space model to complete the multi-dimensional coupling calculation. In the state space model, using the time series and multi-dimensional features contained in the feature vector, the model can establish a dynamic expression of the fan operation status, and obtain the temperature prediction error data and the heat dissipation load prediction error data through repeated iterative calculations.

[0014] In this embodiment, the fan heat dissipation feature vector is decomposed in state space to obtain a temperature feature sequence and a heat dissipation load feature sequence. The temperature feature sequence and the heat dissipation load feature sequence are used to construct a state space model and generate a state equation. In the process of constructing the state equation, the temperature prediction error variable and the heat dissipation load prediction error variable are introduced to describe the deviation between the system prediction and the actual operating state. These variables are embedded in the state equation in a specific mathematical form to ensure that the model can capture the coupled dynamics of temperature and load during the operation of the fan. The state equation is input into the state transfer calculation unit to construct the coupling feature and generate a state transfer matrix. The state transfer matrix is ​​an important parameter in the state space model to describe the dynamic changes of the system, and its determinant value is constrained between 0 and 1 to ensure the stability and reversibility of the matrix. The eigenvalues ​​of the matrix are all set to positive real numbers, reflecting that the system described by the model has stable dynamic evolution characteristics. Through this construction method, the state transfer matrix can not only effectively capture the interactive relationship between temperature and heat dissipation load, but also avoid instability in numerical calculations to a certain extent. The observation equation is constructed by quantitatively calculating the temperature data and heat dissipation load data collected in real time. The observation equation is another key component in the state space model, which is used to describe the relationship between the real-time observation data of the system and the predicted state of the model. By combining the observation equation with the state equation, a covariance matrix is ​​established. The covariance matrix is ​​used to quantify the error between the predicted state and the observed data, and provides a reference for subsequent error correction. The construction of the covariance matrix needs to fully consider the dynamic characteristics of temperature and heat dissipation load and the correlation between the two to ensure the accuracy of error analysis. The error quantization result is subjected to Kalman filtering operation to obtain the state update vector. Kalman filtering is a recursive optimal estimation method that reduces the random error in system prediction by updating the state estimate in real time. The Kalman filtering operation combines the state equation, observation equation and covariance matrix to optimize the error quantization result and obtain the state update vector. The state update vector reflects the state adjustment information of the system in the current time step. The state update vector is subjected to coordinate mapping operation to obtain the temperature prediction error data and the heat dissipation load prediction error data.

[0015] S2, inputting the temperature prediction error data and the heat dissipation load prediction error data into the load prototype library for pattern analysis to obtain the fan operating vector and operating state transfer matrix; Specifically, the temperature prediction error data and the heat dissipation load prediction error data are preprocessed, and the original data is decomposed into multiple time series segments by time series segmentation to capture the operating characteristics of the system in different time periods, while retaining the time dependence and change trend of the data. A correlation coefficient matrix is ​​constructed for the time series segments. The elements of the matrix represent the correlation between each time series segment, and the diagonal elements of the matrix are fixed to 1 to reflect the complete correlation between the sequence and itself. By analyzing the correlation coefficient matrix, a time series association matrix is ​​generated to extract the potential correlation characteristics between the data. Based on the time series association matrix, the heat dissipation feature analysis is performed to convert the time series segments into a more physically meaningful heat dissipation feature sequence. The heat dissipation feature sequence includes a temperature change sequence, a heat dissipation efficiency sequence, and a heat dissipation trend sequence, which respectively describe the temperature change pattern of the fan during operation, the efficiency fluctuation of the heat dissipation system, and the long-term trend of the heat dissipation capacity. The heat dissipation feature sequence is input into the load prototype library for feature matching to generate a prototype feature set. The load prototype library is a database containing a variety of typical heat dissipation features, and each feature prototype represents a common heat dissipation mode or working condition characteristic. In the matching process, the similarity between the heat dissipation feature sequence and the prototype in the library is compared one by one, and the closest feature prototype set is finally determined. The prototype feature set contains M heat dissipation feature prototypes, where M is an integer greater than 1, indicating that the current heat dissipation characteristics of the system are described by M different prototype feature combinations. K-means clustering operation is performed on the prototype feature set to find the commonalities between the feature prototypes and classify them into several cluster centers. The cluster center set represents the core distribution points of the heat dissipation characteristics in the data space. The cluster center set is input into the Markov state transfer unit for state calculation to generate an M×M-dimensional operating state transfer matrix. Each element of the matrix represents the transition probability from one heat dissipation operating state to another. Through the calculation of the Markov model, the dynamic change law of the fan between different operating conditions is captured, and a complete state transfer picture is generated. According to the operating state transfer matrix, the current heat dissipation operating condition data is classified and calculated to generate a fan operating condition vector. Using the dynamic characteristics of the transfer matrix, the current operating condition data is matched with the possible operating condition state to determine the operating state of the fan at the current time point. The fan operating condition vector contains a comprehensive description of information such as temperature, heat dissipation efficiency and trend.

[0016] S3, inputting the fan operating condition vector and the operating condition state transfer matrix into the transferability evaluation model to perform multi-dimensional similarity calculation, obtaining heat dissipation load prediction data, and generating a sleep control instruction according to the heat dissipation load prediction data; It should be noted that the fan operating vector and operating state transfer matrix are input into the feature decomposition unit of the transferability evaluation model for data splitting. The feature decomposition unit extracts the environmental feature sequence and the heat dissipation feature sequence respectively by analyzing the structure of the operating vector and the state transfer matrix. The environmental feature sequence describes the external environmental conditions during the operation of the fan, including parameters such as temperature, humidity and airflow, while the heat dissipation feature sequence focuses on key data such as heat dissipation performance and load characteristics, such as heat distribution and system efficiency. The environmental feature sequence is input into the multidimensional similarity calculation unit of the transferability evaluation model for feature mapping. The multidimensional similarity calculation unit maps high-dimensional environmental features into a similarity space through an embedded learning method to generate environmental similarity parameters. The environmental similarity parameters quantify the similarity between the current environmental conditions and the typical environment in the historical data. At the same time, the heat dissipation feature sequence is input into the load distribution calculation unit of the transferability evaluation model to analyze the distribution characteristics of the heat dissipation load through probability calculation. The load distribution calculation unit uses a statistical model to evaluate the probability distribution of heat dissipation features under different states, and generates heat dissipation distribution features to capture the heat dissipation behavior law of the current operating state. The heat dissipation distribution characteristics are input into the distribution consistency calculation unit for similarity calculation to obtain the distribution similarity parameters. The distribution consistency calculation unit compares the difference between the current heat dissipation distribution characteristics and the historical distribution characteristics, calculates the similarity between them, and reveals the consistency between the current heat dissipation state and the historical typical mode. At the same time, the environmental similarity parameters and the distribution similarity parameters are input into the weight fusion unit of the transferability evaluation model. These parameters are combined by weight adjustment according to their importance to obtain a transferability weight coefficient that comprehensively reflects the similarity between the environment and the heat dissipation characteristics, indicating the applicability of the current fan working condition in the prediction model. The transferability weight coefficient is input into the parameter optimization unit of the transferability evaluation model for coefficient update. The parameter optimization unit uses a dynamic optimization algorithm, such as gradient descent or adaptive learning rate algorithm, to optimize the transferability weight coefficient through iterative update to obtain a more accurate prediction parameter sequence. These prediction parameter sequences contain the key variables required for fan heat dissipation load prediction. The optimized prediction parameter sequence is input into the load prediction unit of the transferability evaluation model for calculation to generate heat dissipation load prediction data, which describes the change of the fan heat dissipation load in the future time period. The heat dissipation load forecast data is input into the event-triggered controller, and the sleep control instructions are generated through sliding mode control optimization. The event-triggered controller determines whether the fan needs to remain running or enter a sleep state within a specific time period based on the heat dissipation load forecast data. The sliding mode control optimization method generates accurate control instructions by dynamically adjusting the forecast data and the current system status to ensure the efficiency and reliability of the sleep strategy.

[0017] In this embodiment, the heat dissipation load prediction data is subjected to sliding interval operation to dynamically capture the changing trend of the heat dissipation load in different time periods, and to construct a trigger condition function. The trigger condition function consists of a temperature error threshold variable and a heat dissipation state change rate variable. The former quantifies the deviation between the current system temperature and the set target, and the latter reflects the dynamic change rate of the heat dissipation system. The sliding interval operation smoothes the data fluctuations through a gradually moving time window to ensure that the trigger condition function can accurately describe the operating state of the system. The trigger condition function is input into the event trigger controller for error calculation to generate an error product matrix. During the error calculation process, the temperature error variable and the heat dissipation deviation variable are extracted respectively and a two-dimensional matrix is ​​constructed through a product operation to describe the coupling relationship between the system temperature and the heat dissipation deviation. The Lyapunov-Krasovskii function is constructed according to the error product matrix to analyze the stability of the system. The Lyapunov-Krasovskii function is a stability analysis tool for delay systems, and its purpose is to evaluate whether the current system can maintain stable operation under input disturbances. By constructing a stability function under the Lyapunov-Krasovskii framework, the dynamic behavior of the system is converted into a mathematically solvable problem. The stability function is input into the linear matrix inequality (LMI) solver for parameter calculation to generate the controller parameter matrix. The linear matrix inequality solver uses optimization algorithms, such as semidefinite programming (SDP), to extract the optimal control parameters from the stability conditions to ensure that the system is robust while achieving the control target. The controller parameter matrix is ​​adaptively constructed to generate an adaptive control function. The adaptive control function can dynamically respond to changes in the system operating environment by adjusting the control parameters in real time to improve the accuracy and adaptability of the control. The adaptive control function is input into the state feedback control unit for parameter adjustment to generate a sleep control parameter sequence. The sleep control parameter sequence includes a sleep start time variable, a sleep duration variable, and a wake-up trigger variable, which correspond to when the fan enters sleep, the length of sleep duration, and when it wakes up from sleep. Through the parameter adjustment of the state feedback control unit, these variables are optimized to the optimal values ​​that meet the system performance requirements, thereby ensuring that the fan group minimizes energy consumption while maintaining the heat dissipation requirements. The sleep control parameter sequence is input into the stability constraint analysis unit to calculate the stable control interval to ensure that the execution of the control instruction will not cause instability or transition shocks in the system. Through constraint analysis, it is determined whether the current parameter sequence falls within the safe and feasible control range, and the parameters that do not meet the conditions are adjusted. After the stable control interval analysis is completed, the optimized sleep control parameter sequence is converted into specific sleep control instructions. The instruction conversion process maps the parameter sequence into the operation instructions of the fan group, including which fans enter sleep, the specific sleep duration, and the wake-up time point.

[0018] S4, performing collaborative optimization control on the fan group according to the sleep control instruction, and generating a sleep execution plan for the fan group.

[0019] Specifically, the sleep control instruction is input into the decision processing unit of the group collaborative control model to extract and calculate the fan impact data. In the decision processing unit, the spatial layout calculation module and the airflow interference calculation module respectively undertake different calculation tasks. Based on the layout relationship of the fans in the physical space, the spatial layout calculation module uses the radial basis function (RBF) to calculate the influence coefficient between adjacent fans, thereby quantifying the interaction between the fans. At the same time, the airflow interference calculation module simulates the airflow distribution during fan operation and its influence on the heat dissipation efficiency through the fluid mechanics model, and calculates the heat dissipation efficiency loss caused by airflow interference. These calculation results are integrated into the fan impact data to describe the spatial coupling and airflow interaction characteristics of each fan in the group. The fan impact data is input into the objective function construction unit of the group collaborative control model to construct the multi-objective optimization objective function. In the objective function construction unit, the heat dissipation efficiency calculation module and the energy consumption calculation module model and quantify the system performance from different perspectives. The heat dissipation efficiency calculation module evaluates the overall heat dissipation of each fan in the group based on the heat exchange model to ensure that the heat dissipation requirements of the system can be met under different sleep strategies. The energy consumption calculation module uses a load model to quantify the energy consumption level of the system under different strategies, thereby providing support for energy efficiency optimization. By integrating the results of these two modules, the objective function construction unit generates a multi-objective optimization function to weigh the relationship between heat dissipation efficiency and energy consumption. The multi-objective genetic algorithm optimization is performed according to the optimization objective function to obtain the Pareto optimal solution set. The genetic algorithm simulates the natural selection process to generate a set of solutions that achieve a balance between different optimization objectives. These solutions form a solution set on the Pareto frontier, and each solution represents a possible sleep execution strategy. The Pareto optimal solution set is input into the timing planning unit of the group collaborative control model, and an execution timing plan is generated according to the time series requirements of the fan operation. The execution timing plan clarifies when each fan in the fan group sleeps, when it starts, and the operation order in different time periods, providing a specific plan in the time dimension for collaborative control. The execution timing plan is input into the heat dissipation optimization unit for air volume allocation to obtain a heat dissipation control plan. The heat dissipation optimization unit optimizes the overall heat dissipation capacity of the group by allocating the air volume of each fan according to the operating characteristics and timing requirements of the fan, ensuring that the system can maintain stable heat dissipation performance during the execution of the sleep strategy. The heat dissipation optimization unit constructs the state monitoring rules for the heat dissipation control scheme and generates a monitoring parameter sequence. The monitoring parameter sequence includes temperature deviation variables, heat dissipation efficiency variables, and energy intensity variables, which are used to monitor the system status in real time to ensure the safety and stability of the control strategy during execution. The monitoring parameter sequence is input into the exception handling unit, and an emergency control plan is generated by analyzing potential abnormal situations. The emergency control plan includes specific response measures for fan failures or sudden changes in heat dissipation demand to ensure that the system can still maintain normal operation under unexpected circumstances.The execution timing plan, heat dissipation control plan and emergency control plan are integrated and optimized to generate the final fan group sleep execution plan.

[0020] In one example, the steps of inputting wind turbine operation history data into a gradient neural network for feature extraction and state space modeling include: The fan operation history data is divided into an environmental parameter sequence and a heat dissipation parameter sequence, wherein the environmental parameter sequence includes an environmental temperature parameter, a humidity parameter and an airflow parameter, and the heat dissipation parameter sequence includes a device temperature parameter and a heat dissipation load parameter; The environmental parameter sequence is input into the input mapping unit of the gradient neural network for feature extraction to obtain an environmental mapping vector, wherein the input mapping unit includes a fully connected parameter layer and a feature extraction layer. The fully connected parameter layer contains 4 neuron nodes corresponding to the environmental temperature, humidity, airflow and time parameters respectively, and the feature extraction layer contains 8 neuron nodes and uses the ReLU activation function; The environment mapping vector and the heat dissipation parameter sequence are input into the double-layer hidden unit of the gradient neural network for feature fusion to obtain the heat dissipation mapping vector, wherein the double-layer hidden unit includes a deep residual layer and an attention calculation layer, the deep residual layer contains 16 neuron nodes for retaining environmental feature information, and the attention calculation layer contains 12 neuron nodes; The heat dissipation mapping vector is input into the feature fusion unit of the gradient neural network for multi-layer feature fusion to obtain the fan heat dissipation feature vector; The fan heat dissipation characteristic vector is input into the state space model for multi-dimensional coupling calculation to obtain temperature prediction error data and heat dissipation load prediction error data.

[0021] In this example, the fan operation history data is divided into an environmental parameter sequence and a heat dissipation parameter sequence. The environmental parameter sequence includes the ambient temperature ,humidity , air flow velocity , and time characteristics , these variables reflect the external conditions when the fan is running. The heat dissipation parameter sequence includes the equipment temperature and cooling load , which is the core performance indicator of fan operation. By dividing, the input data is modularized, which makes it easier for the neural network to process environmental characteristics and heat dissipation characteristics respectively. Input the input mapping unit of the gradient neural network for feature extraction. The input mapping unit consists of two main parts: the fully connected parameter layer and the feature extraction layer. In the fully connected parameter layer, each input parameter is connected through the weight of a neuron node. and bias Perform linear transformation, the calculation formula is: ; in is the output after linear transformation, is an input parameter (such as ), is the weight, is the bias. The output of the fully connected layer is input to the feature extraction layer, which contains 8 neuron nodes. Each node uses the ReLU activation function, and the calculation formula is: ; Through ReLU activation, the feature extraction layer captures the nonlinear characteristics of the input data and generates an environment mapping vector , where each is the extracted high-dimensional feature. The environment mapping vector and cooling parameter sequence Input into the double-layer hidden unit of the gradient neural network for feature fusion. The double-layer hidden unit consists of a deep residual layer and an attention calculation layer. The deep residual layer retains the environmental feature information through 16 neuron nodes, and the residual calculation formula is: ; in is a nonlinear transformation of the features, is the residual output. In this way, the information flow of environmental features is retained in the deep network to avoid information loss. The attention calculation layer performs weighted processing on the heat dissipation characteristics through 12 neuron nodes. The attention weight The calculation formula is: ; in and are the query and key features in the attention mechanism, Indicates The final output heat map vector is , which is used to comprehensively describe the environment and heat dissipation characteristics. Input into the feature fusion unit of the gradient neural network, perform multi-layer feature fusion operation, and generate the fan heat dissipation feature vector The fusion process uses multiple fully connected layers and activation functions to further compress and refine high-dimensional features so that they can characterize the overall heat dissipation behavior of the fan. The state space model is input for multi-dimensional coupling calculation to generate temperature prediction error data and heat dissipation load prediction error data. In the state space model, the system state is represented by the state variable The dynamic equation is: ; in is the state transfer matrix, is the control matrix, is the control matrix, is the process noise. The output equation of the system is: ; in is the output matrix, The temperature error is predicted by performing Kalman filter calculation on the dynamic equation and output equation of the system. and heat dissipation load error , the formula is: ; in and are the temperature and load values ​​predicted by the model.

[0022] In one example, the steps of inputting the fan heat dissipation characteristic vector into the state space model for multi-dimensional coupling calculation to obtain temperature prediction error data and heat dissipation load prediction error data include: Decompose the fan heat dissipation feature vector in state space to obtain temperature feature sequence and heat dissipation load feature sequence; A state space model is constructed according to the temperature characteristic sequence and the heat dissipation load characteristic sequence to obtain a state equation, wherein the state equation includes a temperature prediction error variable and a heat dissipation load prediction error variable; The state equation is input into the state transfer calculation unit for coupling feature construction to obtain a state transfer matrix, wherein the determinant value of the state transfer matrix is ​​between 0 and 1 and the eigenvalues ​​are all positive real numbers; The real-time collected temperature data and heat dissipation load data are quantified and calculated to obtain the observation equation, and the covariance matrix is ​​constructed based on the state equation and the observation equation to obtain the error quantification result; A Kalman filter operation is performed on the error quantization result to obtain a state update vector, and coordinate mapping is performed on the state update vector to obtain temperature prediction error data and heat dissipation load prediction error data.

[0023] In this example, the fan cooling feature vector Perform state space decomposition to divide the high-dimensional feature vector into two main parts: temperature feature sequence and heat dissipation load characteristic sequence The temperature characteristic sequence describes the thermal characteristics of the system, including variables such as heat distribution and temperature gradient, while the heat dissipation load characteristic sequence reflects the load characteristics of the system, such as energy consumption and heat dissipation efficiency. The decomposition process is achieved through the linear mapping matrix and accomplish: ; in and They are mapping matrices, which are used to project the original eigenvectors into the corresponding eigenspace. and heat dissipation load characteristic sequence Construct a state space model and generate state equations. The core of the state space model is to describe the dynamic evolution of the system, and its state equation is defined as: ; in, is a state variable, including the temperature characteristic state and load characteristic status ; is the state transition matrix, describing the system from time step arrive dynamic relationships; is the control input matrix; is the control input; is the process noise, which is assumed to be a zero-mean Gaussian distribution. The state transfer matrix The determinant value of is constrained to be in the range of [0,1], and its eigenvalue is required to be a positive real number to ensure the stability and reversibility of the system. In order to complete the state estimation, the real-time collected temperature data and heat dissipation load data are input into the observation equation, and the observation equation is in the form of: ; in, is the observed variable, which contains the temperature and heat dissipation load data at the current moment; is the observation matrix, connecting state variables and observation variables; is the observation noise, which is assumed to be a zero-mean Gaussian distribution. The dynamic characteristics of the system are combined with the observation data through the state equation and the observation equation. After the state and observation models are established, the covariance matrix is ​​constructed based on the two. , the covariance matrix is ​​used to quantify the prediction error and observation error of the system, which is defined as: ; in, is the estimated value of the state. The covariance matrix is ​​iteratively updated through the Kalman filter algorithm. The prediction steps of the Kalman filter are: ; ; in, is the time step The predicted state of is the forecast covariance matrix, is the process noise covariance matrix. After observing new data, the update steps of the Kalman filter are: ; ; ; in, is the Kalman gain matrix, is the observation noise covariance matrix. Through the prediction and update steps, the Kalman filter generates the state update vector . Update the state vector Coordinate mapping is performed to convert it into temperature prediction error data and heat dissipation load prediction error data. The formula is: ; in, and denote the prediction errors of temperature and cooling load, respectively.

[0024] In one example, the steps of inputting temperature prediction error data and heat dissipation load prediction error data into a load prototype library for pattern analysis to obtain a fan operating vector and an operating state transfer matrix include: The temperature prediction error data and the heat dissipation load prediction error data are segmented into time series to obtain time series segments, and a correlation coefficient matrix is ​​constructed for the time series segments to obtain a time series correlation matrix, wherein the diagonal elements of the correlation coefficient matrix are 1; Perform heat dissipation feature analysis on the time series correlation matrix to obtain a heat dissipation feature sequence, wherein the heat dissipation feature sequence includes a temperature change sequence, a heat dissipation efficiency sequence, and a heat dissipation trend sequence; Input the heat dissipation feature sequence into the load prototype library for feature matching to obtain a prototype feature set, wherein the prototype feature set includes M heat dissipation feature prototypes, where M is an integer greater than 1; Perform K-means clustering operation on the prototype feature set to obtain a cluster center set, and input the cluster center set into the Markov state transfer unit for state calculation to obtain an M×M-dimensional operating state transfer matrix, where each element of the operating state transfer matrix represents the transition probability between heat dissipation operating conditions; The current heat dissipation condition data is classified and calculated according to the condition state transfer matrix to obtain the fan condition vector.

[0025] In this example, the temperature prediction error data and cooling load prediction error data Arranged into time series and based on fixed time windows Segment to form time series segments , where each segment Contains the error data within the window. The segmentation method can effectively capture the dynamic changes of heat dissipation characteristics within the time window. Construct a correlation coefficient matrix for the time series segment and define the correlation coefficient matrix , whose elements Represents a time series segment and The correlation between them is calculated as: ; in, It is a sequence segment and The covariance of and are the standard deviations of the sequence segments. The diagonal elements of the matrix , indicating that the sequence segment is completely correlated with itself. Based on the correlation coefficient matrix , analyze the heat dissipation characteristics of the time series. Through the eigendecomposition method, the main eigenvalues ​​and eigenvectors of the matrix are extracted to generate the heat dissipation characteristic sequence. The heat dissipation characteristic sequence consists of three main parts: temperature change sequence , heat dissipation efficiency series and heat dissipation trend series The temperature change sequence describes the fluctuation characteristics of the temperature error in each time period, and its calculation formula is: ; in, is the mean value of the temperature error. Heat dissipation efficiency series It is calculated by the ratio of heat dissipation load to temperature change, and the formula is: ; Cooling trend sequence Use regression analysis to calculate the changing trend of temperature error. Input the heat dissipation feature sequence into the load prototype library for feature matching. The load prototype library stores a variety of typical heat dissipation feature prototypes, each of which represents a common heat dissipation working mode. Feature matching is achieved by calculating the similarity between the heat dissipation feature sequence and the prototype feature. The similarity calculation formula is: ; in, is the input heat dissipation feature sequence, It is Prototype features, It is a scale parameter that controls the similarity distribution. By matching, a prototype feature set is generated. ,in Indicates the number of matched prototypes. Perform K-means clustering on the prototype feature set. The K-means algorithm generates a cluster center set by minimizing the sum of the squared distances between each cluster center and its members. Each cluster center Represents the average characteristics of a group of similar heat dissipation characteristics. The optimization goal of the clustering process is: ; in, It is The cluster center set is input into the Markov state transfer unit for state calculation to generate Dimensional working condition state transfer matrix . Matrix elements Indicates the working state Transfer to state The probability of satisfying: ; The state transition probability is calculated through the frequency statistics of the time series. According to the working state transition matrix, the current heat dissipation working condition data is classified and calculated to generate the fan working condition vector Each component of the operating condition vector represents the characteristic weight of the current wind turbine state.

[0026] In one example, the steps of inputting the fan operating condition vector and the operating condition state transfer matrix into the transferability evaluation model to perform multi-dimensional similarity calculation to obtain heat dissipation load prediction data, and generating a sleep control instruction according to the heat dissipation load prediction data include: Input the fan operating condition vector and operating state transfer matrix into the characteristic decomposition unit of the transferability evaluation model to split the data, and obtain the environmental characteristic sequence and the heat dissipation characteristic sequence; Input the environmental feature sequence into the multidimensional similarity calculation unit of the transferability evaluation model for feature mapping to obtain environmental similarity parameters, and input the heat dissipation feature sequence into the load distribution calculation unit of the transferability evaluation model for probability calculation to obtain heat dissipation distribution features; Input the heat dissipation distribution feature into the distribution consistency calculation unit of the transferability evaluation model to perform similarity calculation to obtain the distribution similarity parameter, and input the environment similarity parameter and the distribution similarity parameter into the weight fusion unit of the transferability evaluation model to perform feature combination to obtain the transferability weight coefficient; Inputting the transferability weight coefficient into the parameter optimization unit of the transferability evaluation model to update the coefficient to obtain a prediction parameter sequence, and inputting the prediction parameter sequence into the load prediction unit of the transferability evaluation model to calculate the heat dissipation load to obtain heat dissipation load prediction data; The heat dissipation load prediction data is input into the event-triggered controller to perform sliding mode control optimization and obtain the sleep control instruction.

[0027] In this example, the fan operating vector And the operating state transfer matrix The input is sent to the feature decomposition unit of the transferability assessment model for data splitting. Through the matrix decomposition method, the two inputs are decomposed into environmental feature sequences and heat dissipation characteristic sequence The core formula of eigendecomposition is: ; in, and They are mapping matrices used to extract environmental features and heat dissipation features. The environmental feature sequence reflects the impact of the external environment on the operation of the fan, while the heat dissipation feature sequence describes the internal dynamic characteristics of the heat dissipation system. The multi-dimensional similarity calculation unit of the input transferability evaluation model is used for feature mapping. Through the kernel function method, the environmental features are mapped to a high-dimensional similarity space to generate environmental similarity parameters. The mapping formula is: ; in, is the reference environment feature sequence, is the scale parameter that controls the similarity distribution. The environmental similarity parameter indicates the similarity between the current environment and the historical environment data. Input load distribution calculation unit, calculate heat dissipation distribution characteristics through probability modeling method The calculation formula of the distribution characteristics is: ; in, is a parameter that controls weight decay. The heat dissipation distribution characteristic describes the position of the current heat dissipation state in the overall load distribution. Input the distribution consistency calculation unit and compare it with the historical distribution characteristics to obtain the distribution similarity parameter The calculation formula for distribution similarity is: ; in, is the historical heat dissipation distribution characteristic, The closer the value is to 1, the more consistent the current distribution is with the historical distribution. and distribution similarity parameter Input to the weight fusion unit for feature combination to generate transferability weight coefficient The fusion process adopts the weighted average method, and the formula is: ; in, and are weight parameters corresponding to the environment and distribution similarity respectively. Input to the parameter optimization unit for dynamic update to generate a prediction parameter sequence . Parameter optimization is iteratively updated by gradient descent: ; in, is the learning rate, is the objective function, usually the sum of squares of load forecast errors. Input the load prediction unit to calculate the heat dissipation load prediction data The prediction formula is: ; in, is a feature mapping function that combines environmental and heat dissipation characteristics to generate load predictions. The input event triggers the controller to optimize the sliding mode control and generate the sleep control command. The core of sliding mode control is to design a switching surface , and adjust the control strategy according to its dynamic behavior, which is defined as: ; in, and is the sliding mode control parameter, is the reference load value. When , adjust the control strategy to reduce the load; when The strategy is adjusted to increase the load. The final generated sleep control instruction includes the specific fan sleep time, duration and triggering conditions.

[0028] In one example, the step of inputting heat dissipation load prediction data into an event-triggered controller to perform sliding mode control optimization and obtain a sleep control instruction includes: The heat dissipation load prediction data is subjected to sliding interval calculation to obtain a trigger condition function, wherein the trigger condition function includes a temperature error threshold variable and a heat dissipation state change rate variable; Inputting the trigger condition function into the event trigger controller to perform error calculation to obtain an error product matrix, wherein the error product matrix includes a temperature error variable and a heat dissipation deviation variable; According to the error product matrix, a Lyapunov-Krasovskii function is constructed to obtain a stability function, and the stability function is input into a linear matrix inequality solving unit for parameter calculation to obtain a controller parameter matrix; An adaptive rule is constructed for the controller parameter matrix to obtain an adaptive control function, and the adaptive control function is input into the state feedback control unit for parameter adjustment to obtain a sleep control parameter sequence, wherein the sleep control parameter sequence includes a sleep start time variable, a sleep duration variable and a wake-up trigger variable; A stability constraint analysis is performed on the sleep control parameter sequence to obtain a stable control interval, and an instruction conversion is performed on the sleep control parameter sequence according to the stable control interval to obtain a sleep control instruction.

[0029] In this example, the cooling load forecast data According to the fixed length time window Sliding interval operation is performed to capture the short-term dynamic characteristics of load changes. Sliding interval operation is achieved by calculating the sliding average and the rate of change, the formula is: ; in, is the sliding mean load, is the load change rate. Based on these calculation results, the trigger condition function is constructed , which includes the temperature error threshold variable and the rate of change of thermal state variable , defined as: ; in, Indicates the measured temperature With reference temperature The conditional function will be triggered. Input to the event-triggered controller for error calculation and generate the error product matrix The elements of the error product matrix are calculated using the following formula: ; in, and They are the trigger condition function and components. Error product matrix Is a symmetric matrix used to describe the interaction between temperature error and heat dissipation deviation. According to the error product matrix Construct the Lyapunov-Krasovskii function to analyze the stability of the system. Lyapunov-Krasovskii function The general form is: ; in, is the system state vector, and is a positive definite matrix, is the system delay. By multiplying the error product matrix Integrate the Lyapunov-Krasovskii framework to obtain the stability function of the system. Input the stability function into the linear matrix inequality (LMI) solving unit for parameter calculation to generate the controller parameter matrix The basic form of LMI is: ; Through numerical optimization algorithms, such as semidefinite programming (SDP), the controller parameter matrix that satisfies the stability condition is solved. Used to construct adaptive rules and generate adaptive control functions The adaptive control function is defined as: ; in, is the control bias, which is used to dynamically adjust the control input. The adaptive control function is input into the state feedback control unit for parameter adjustment to obtain the sleep control parameter sequence The sleep control parameter sequence includes the sleep start time variable , Sleep duration variable and wake-up trigger variables . Sequence of sleep control parameters Stability constraint analysis is performed to ensure that the control instructions will not cause system instability during execution. By analyzing the control constraint area of ​​the system, the stable control interval is determined, and the constraint conditions are: ; in, is the stable control interval. According to the analysis results, the sleep control parameter sequence is optimized and adjusted. The optimized sleep control parameter sequence is converted into an actually executable sleep control instruction. The instruction format includes the specific fan group number, sleep start time, duration and wake-up condition.

[0030] In one example, the steps of performing collaborative optimization control on the wind turbine group according to the sleep control instruction and generating a sleep execution plan for the wind turbine group include: Input the sleep control command into the decision processing unit of the group collaborative control model for data preprocessing to obtain fan impact data, wherein the decision processing unit includes a space layout calculation module and an airflow interference calculation module. The space layout calculation module uses a radial basis function to calculate the influence coefficient of adjacent fans, and the airflow interference calculation module calculates the heat dissipation efficiency loss based on a fluid mechanics model. Input the fan impact data into the objective function construction unit of the group collaborative control model for multi-objective construction to obtain an optimized objective function, wherein the objective function construction unit includes a heat dissipation efficiency calculation module and an energy consumption calculation module. The heat dissipation efficiency calculation module calculates the heat dissipation of the group based on the heat exchange model, and the energy consumption calculation module uses a load model to calculate the system energy consumption. Perform multi-objective genetic algorithm optimization according to the optimization objective function to obtain the Pareto optimal solution set, and input the Pareto optimal solution set into the timing planning unit of the group collaborative control model for sequence generation to obtain the execution timing plan; The execution timing plan is input into the heat dissipation optimization unit of the group collaborative control model to distribute the air volume, and a heat dissipation control plan is obtained. The state monitoring rule of the heat dissipation control plan is constructed to obtain a monitoring parameter sequence, wherein the monitoring parameter sequence includes a temperature deviation variable, a heat dissipation efficiency variable, and an energy consumption intensity variable; The monitoring parameter sequence is input into the exception handling unit to generate an emergency strategy to obtain an emergency control plan. The execution timing plan, heat dissipation control plan and emergency control plan are integrated and optimized to obtain a fan group sleep execution plan.

[0031] In this example, the sleep control command is input into the decision processing unit of the group collaborative control model for data preprocessing. The spatial layout calculation module and the airflow interference calculation module are used in the decision processing unit to evaluate the mutual influence between fans and the overall heat dissipation performance respectively. The spatial layout calculation module uses the radial basis function (RBF) to evaluate the influence coefficient of adjacent fans, which is defined as: ; in, Indicates fan For fans The influence coefficient of and Fan and The position vector of The airflow interference calculation module evaluates the heat dissipation efficiency loss caused by airflow interference based on the fluid mechanics model. The formula is: ; in, It is a fan The heat dissipation efficiency loss, is the proportionality coefficient, It is a fan Through the above calculation, the fan impact data is generated. and airflow disturbance data The fan impact data is input into the objective function construction unit of the group collaborative control model to construct the multi-objective optimization objective function. The objective function construction unit includes a heat dissipation efficiency calculation module and an energy consumption calculation module. The heat dissipation efficiency calculation module evaluates the heat dissipation of the group based on the heat exchange model and is defined as: ; in, is the total heat dissipation, It is a fan The heat dissipation efficiency, is the specific heat capacity of air, It is a fan The air volume, It is a fan The energy consumption calculation module uses a load model to evaluate the total energy consumption of the system, which is defined as: ; in, is the total energy consumption, Indicates fan The power consumption, and is the power consumption model parameter. Combining these two modules, we get the multi-objective optimization function: ; The multi-objective genetic algorithm is used to solve the optimization objective function and generate the Pareto optimal solution set. The multi-objective genetic algorithm optimizes the objective function through the evolutionary process (selection, crossover, mutation), and finally obtains multiple solution sets that balance the heat dissipation efficiency and energy consumption. Pareto optimal solution set It contains multiple optimal solutions, each of which corresponds to a sleep strategy. The Pareto optimal solution set is input into the timing planning unit of the group collaborative control model, and the execution timing plan is generated through time series analysis. The timing planning unit allocates the operation and sleep time periods for each fan according to the operating characteristics of the fan group. The timing planning formula is: ; in, Indicates fan The operating time period, and are the start and end time of the operation respectively. The execution timing plan is input into the heat dissipation optimization unit to distribute the air volume and generate a heat dissipation control plan. The heat dissipation optimization optimizes the air volume distribution of each fan through the linear programming method to maximize the total heat dissipation of the group. The optimization goal is: ; in, is the upper limit of the total air volume of the fan group. At the same time, the state monitoring rules are constructed for the heat dissipation control scheme to generate a monitoring parameter sequence, including the temperature deviation variable , heat dissipation efficiency variables and energy intensity variables The monitoring rules evaluate the operating status of the fan group by detecting these variables in real time. The monitoring parameter sequence is input into the abnormal processing unit to generate an emergency control plan. The emergency control plan responds to emergencies (such as fan failure or temperature abnormality) through the rule engine and threshold analysis. For example, when the temperature deviation of a fan When the set threshold is exceeded, the emergency control plan adjusts the air volume of other fans or starts the backup fans. The execution timing plan, heat dissipation control plan and emergency control plan are integrated and optimized to generate the final fan group sleep execution plan. The plan includes the sleep time, operating air volume and abnormal handling strategy of each fan.

[0032] Reference Figure 2 This embodiment provides a fan dormancy control system based on load prediction, including: Modeling module 1, used for inputting the fan operation history data into the gradient neural network for feature extraction and state space modeling, and obtaining temperature prediction error data and heat dissipation load prediction error data; Mode analysis module 2, used for inputting temperature prediction error data and heat dissipation load prediction error data into the load prototype library for mode analysis, and obtaining the fan operating vector and operating state transfer matrix; The calculation module 3 is used to input the fan operating condition vector and the operating condition state transfer matrix into the transferability evaluation model to perform multi-dimensional similarity calculation, obtain the heat dissipation load prediction data, and generate the sleep control instruction according to the heat dissipation load prediction data; The generating module 4 is used to perform collaborative optimization control on the fan group according to the sleep control instruction and generate a sleep execution plan for the fan group.

[0033] In this embodiment, for the specific implementation of each unit in the above system embodiment, please refer to the above method embodiment, which will not be repeated here.

[0034] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0035] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0036] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0037] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0038] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0039] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A fan sleep control method based on load prediction, characterized in that: The following steps are involved: The fan operation history data is input into the gradient neural network for feature extraction and state space modeling to obtain temperature prediction error data and heat dissipation load prediction error data; Inputting the temperature prediction error data and the heat dissipation load prediction error data into a load prototype library for pattern analysis to obtain a fan operating state vector and an operating state transfer matrix; Inputting the fan operating condition vector and the operating condition state transfer matrix into a transferability evaluation model to perform multi-dimensional similarity calculation to obtain heat dissipation load prediction data, and generating a sleep control instruction according to the heat dissipation load prediction data; The fan group is collaboratively optimized and controlled according to the sleep control instruction to generate a sleep execution plan for the fan group.

2. The method according to claim 1, characterized in that The step of inputting the wind turbine operation history data into the gradient neural network for feature extraction and state space modeling includes: Dividing the fan operation history data into an environmental parameter sequence and a heat dissipation parameter sequence, wherein the environmental parameter sequence includes an environmental temperature parameter, a humidity parameter and an airflow parameter, and the heat dissipation parameter sequence includes an equipment temperature parameter and a heat dissipation load parameter; Inputting the environmental parameter sequence into the input mapping unit of the gradient neural network for feature extraction to obtain an environmental mapping vector, wherein the input mapping unit includes a fully connected parameter layer and a feature extraction layer, the fully connected parameter layer includes 4 neuron nodes corresponding to environmental temperature, humidity, airflow and time parameters respectively, and the feature extraction layer includes 8 neuron nodes and adopts a ReLU activation function; Inputting the environment mapping vector and the heat dissipation parameter sequence into a double-layer hidden unit of a gradient neural network for feature fusion to obtain a heat dissipation mapping vector, wherein the double-layer hidden unit includes a deep residual layer and an attention calculation layer, the deep residual layer includes 16 neuron nodes for retaining environmental feature information, and the attention calculation layer includes 12 neuron nodes; Inputting the heat dissipation mapping vector into the feature fusion unit of the gradient neural network for multi-layer feature fusion to obtain a fan heat dissipation feature vector; The fan heat dissipation characteristic vector is input into a state space model for multi-dimensional coupling calculation to obtain the temperature prediction error data and the heat dissipation load prediction error data.

3. The method according to claim 2, characterized in that The step of inputting the fan heat dissipation characteristic vector into a state space model for multi-dimensional coupling calculation to obtain the temperature prediction error data and the heat dissipation load prediction error data comprises: Decomposing the fan heat dissipation feature vector in state space to obtain a temperature feature sequence and a heat dissipation load feature sequence; Constructing a state space model according to the temperature characteristic sequence and the heat dissipation load characteristic sequence to obtain a state equation, wherein the state equation includes a temperature prediction error variable and a heat dissipation load prediction error variable; Inputting the state equation into a state transfer calculation unit for coupling feature construction to obtain a state transfer matrix, wherein the determinant value of the state transfer matrix is ​​between 0 and 1 and the eigenvalues ​​are all positive real numbers; Quantitatively calculate the temperature data and heat dissipation load data collected in real time to obtain an observation equation, and construct a covariance matrix based on the state equation and the observation equation to obtain an error quantification result; A Kalman filter operation is performed on the error quantization result to obtain a state update vector, and coordinate mapping is performed on the state update vector to obtain the temperature prediction error data and the heat dissipation load prediction error data.

4. The method according to claim 3, characterized in that: The step of inputting the temperature prediction error data and the heat dissipation load prediction error data into a load prototype library for pattern analysis to obtain a fan operating state vector and an operating state transfer matrix comprises: The temperature prediction error data and the heat dissipation load prediction error data are segmented into time series to obtain time series segments, and a correlation coefficient matrix is ​​constructed for the time series segments to obtain a time series correlation matrix, wherein the diagonal elements of the correlation coefficient matrix are 1; Performing heat dissipation feature analysis on the time series correlation matrix to obtain a heat dissipation feature sequence, wherein the heat dissipation feature sequence includes a temperature change sequence, a heat dissipation efficiency sequence, and a heat dissipation trend sequence; Inputting the heat dissipation feature sequence into a load prototype library for feature matching to obtain a prototype feature set, wherein the prototype feature set includes M heat dissipation feature prototypes, where M is an integer greater than 1; Performing a K-means clustering operation on the prototype feature set to obtain a cluster center set, and inputting the cluster center set into a Markov state transfer unit for state calculation to obtain an M×M-dimensional operating state transfer matrix, wherein each element of the operating state transfer matrix represents a transition probability between heat dissipation operating conditions; The current heat dissipation working condition data is classified and calculated according to the working condition state transfer matrix to obtain a fan working condition vector.

5. The method according to claim 4, characterized in that The step of inputting the fan operating condition vector and the operating condition state transfer matrix into a transferability evaluation model to perform multi-dimensional similarity calculation to obtain heat dissipation load prediction data, and generating a sleep control instruction according to the heat dissipation load prediction data includes: Inputting the fan operating condition vector and the operating condition state transfer matrix into the feature decomposition unit of the transferability evaluation model for data splitting to obtain an environmental feature sequence and a heat dissipation feature sequence; Input the environmental feature sequence into the multidimensional similarity calculation unit of the transferability evaluation model for feature mapping to obtain environmental similarity parameters, and input the heat dissipation feature sequence into the load distribution calculation unit of the transferability evaluation model for probability calculation to obtain heat dissipation distribution features; Input the heat dissipation distribution feature into the distribution consistency calculation unit of the transferability evaluation model to perform similarity calculation to obtain a distribution similarity parameter, and input the environment similarity parameter and the distribution similarity parameter into the weight fusion unit of the transferability evaluation model to perform feature combination to obtain a transferability weight coefficient; Inputting the transferability weight coefficient into the parameter optimization unit of the transferability evaluation model to update the coefficient to obtain a prediction parameter sequence, and inputting the prediction parameter sequence into the load prediction unit of the transferability evaluation model to calculate the heat dissipation load to obtain heat dissipation load prediction data; The heat dissipation load prediction data is input into an event trigger controller to perform sliding mode control optimization to obtain a sleep control instruction.

6. The method according to claim 5, characterized in that The step of inputting the heat dissipation load prediction data into an event triggering controller to perform sliding mode control optimization to obtain a sleep control instruction comprises: Performing a sliding interval operation on the heat dissipation load prediction data to obtain a trigger condition function, wherein the trigger condition function includes a temperature error threshold variable and a heat dissipation state change rate variable; Inputting the trigger condition function into an event trigger controller to perform error calculation to obtain an error product matrix, wherein the error product matrix includes a temperature error variable and a heat dissipation deviation variable; Constructing a Lyapunov-Krasovskii function according to the error product matrix to obtain a stability function, and inputting the stability function into a linear matrix inequality solving unit for parameter operation to obtain a controller parameter matrix; An adaptive rule is constructed for the controller parameter matrix to obtain an adaptive control function, and the adaptive control function is input into a state feedback control unit for parameter adjustment to obtain a sleep control parameter sequence, wherein the sleep control parameter sequence includes a sleep start time variable, a sleep duration variable, and a wake-up trigger variable; A stability constraint analysis is performed on the sleep control parameter sequence to obtain a stable control interval, and an instruction conversion is performed on the sleep control parameter sequence according to the stable control interval to obtain a sleep control instruction.

7. The method according to claim 6, characterized in that The step of performing collaborative optimization control on the wind turbine group according to the sleep control instruction and generating a sleep execution plan for the wind turbine group comprises: Input the sleep control instruction into the decision processing unit of the group collaborative control model for data preprocessing to obtain fan influence data, wherein the decision processing unit includes a space layout calculation module and an airflow interference calculation module, the space layout calculation module uses a radial basis function to calculate the influence coefficient of adjacent fans, and the airflow interference calculation module calculates the heat dissipation efficiency loss based on a fluid mechanics model; Inputting the fan impact data into the objective function construction unit of the group collaborative control model for multi-objective construction to obtain an optimized objective function, wherein the objective function construction unit includes a heat dissipation efficiency calculation module and an energy consumption calculation module, the heat dissipation efficiency calculation module calculates the group heat dissipation based on the heat exchange model, and the energy consumption calculation module uses the load model to calculate the system energy consumption; Perform multi-objective genetic algorithm optimization according to the optimization objective function to obtain a Pareto optimal solution set, and input the Pareto optimal solution set into the timing planning unit of the group collaborative control model for sequence generation to obtain an execution timing plan; The execution timing scheme is input into the heat dissipation optimization unit of the group collaborative control model to distribute the air volume, so as to obtain a heat dissipation control scheme, and a state monitoring rule is constructed for the heat dissipation control scheme to obtain a monitoring parameter sequence, wherein the monitoring parameter sequence includes a temperature deviation variable, a heat dissipation efficiency variable and an energy consumption intensity variable; The monitoring parameter sequence is input into the abnormality processing unit to generate an emergency strategy to obtain an emergency control plan, and the execution timing plan, the heat dissipation control plan and the emergency control plan are integrated and optimized to obtain a fan group sleep execution plan.

8. A fan dormancy control system based on load prediction, characterized in that: For implementing the steps of the method according to any one of claims 1 to 7, the system comprises: A modeling module, used to input the fan operation history data into the gradient neural network for feature extraction and state space modeling, and obtain temperature prediction error data and heat dissipation load prediction error data; A mode analysis module, used for inputting the temperature prediction error data and the heat dissipation load prediction error data into a load prototype library for mode analysis to obtain a fan operating state vector and an operating state transfer matrix; A calculation module, used for inputting the fan operating condition vector and the operating condition state transfer matrix into a transferability evaluation model to perform multi-dimensional similarity calculation, obtain heat dissipation load prediction data, and generate a sleep control instruction according to the heat dissipation load prediction data; A generation module is used to perform collaborative optimization control on the fan group according to the sleep control instruction and generate a sleep execution plan for the fan group.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images